Download Clinical Activity in General Practice and Cancer

 Clinical Activity in General Practice and Cancer PhD dissertation Peter Hjertholm
Faculty of Health Aarhus University 2015 Clinical Activity in General Practice and Cancer
PhD‐student: Peter Hjertholm, MD Research Unit for General Practice and Research Centre for Cancer Diagnosis in Primary Care (CaP), Department of Public Health, Aarhus University, Denmark Supervisors: Peter Vedsted, MD, PhD, Professor (main supervisor) Research Unit for General Practice and Research Centre for Cancer Diagnosis in Primary Care (CaP), Department of Public Health, Aarhus University, Denmark Morten Fenger‐Grøn, MSc, Biostatistician Research Unit for General Practice, Department of Public Health, Aarhus University, Denmark Morten Bondo Christensen, GP, PhD, Senior researcher Research Unit for General Practice, Department of Public Health, Aarhus University, Denmark Mogens Vestergaard, MD, PhD, Professor Research Unit for General Practice and Section for General Medicine, Department of Public Health, Aarhus University, Denmark Assessment committee: Henrik A. Kolstad, Professor, Department of Occupational Medicine, Danish Ramazzini Centre, Aarhus University Hospital, Aarhus, Denmark (Chairman) Therese Stukel, PhD, Senior Scientist, Professor, Institute for Health Policy, Management and Evaluation, University of Toronto, Canada Debbie Sharp, Professor, School of Social and Community Medicine, University of Bristol, England Financial support: This project was funded by Aarhus University, the Novo Nordic Foundation, the Danish Cancer Society, the Danish National Research Foundation for Primary Care, Helga and Peter Korning’s Foundation and The Hede Nielsen Family Foundation II Acknowledgements
ACKNOWLEDGEMENTS Sitting in my living room with a cup of coffee, looking at the Christmas tree. From upstairs I can hear my beloved daughter shouting, “I do not want to sleeeeep!!!” while her mother patiently sings lullabies. Everything is just perfect. An ideal moment to look back on the previous 4 years. About 4 years ago, Morten Bondo introduced me to Peter Vedsted and sparked an interesting collaboration. For that I am very grateful, Morten, and thank you for keeping my research on track and putting it into a clinical context. Peter, you open‐mindedly allowed me to develop my own protocol for this PhD project which was an interesting and sometimes frustrating process. You have answered ridiculous questions day and night and your energy has been very inspiring. Thank you for all our discussions; I am impressed that you have endured all my incidents of “talk‐before‐you‐think”. I am working on it… I also appreciate the trust you have shown in me by giving me the responsibility for overseeing the work of two research year students, Kasper and Pernille. I owe special gratitude to Morten Fenger‐Grøn. Thank you for lending me your brain. Without your statistical support and astonishing wisdom, my life had not been the same. Without your insight, I would most likely have finished my PhD two years ago, written five more papers and probably been rich, but I have enjoyed every frustrating challenge you have given me. It has made me a much better researcher. Special thanks go to Mogens Vestergaard for always being there. Your epidemiologic knowledge has improved this thesis and I admire the way you work and your ability to draw a line in the sand when it is needed. Furthermore, I would like to thank all my colleagues at the Research Unit, especially “Jomfruburet” (Marie, Margrethe, Jacob, Marlene, Karen, Anette), who endured a lot of frustrations and “beating the lamp”. And “Riddersalen” for providing good coffee and keeping my anthropological skills just above the III Clinical Activity in General Practice and Cancer
level of sheer ignorance. Thank you, Kaare, for always being helpful the numerous times I have asked for more data, and thank you, Thomas Mukai for collecting data on PSA tests. Additionally, I want to thank Bjørn for the “Cambridge style walks” around Campus and a fun course in Copenhagen. I would also like to express my gratitude to Willie Hamilton, Christine Campbell and David Weller for allowing me to visit them in Exeter and Edinburgh. Furthermore, I am grateful for the support from my co‐authors Mads Lind Ingeman, Grete Moth, Lene Hjerrild Iversen, Henrik Møller and Michael Borre. I also wish to thank Lone Niedziella and Morten Pilegaard for linguistic support and for changing my “Danglish” into proper English in the thesis and papers. Finally, a special thanks to my fantastic wife, Jannie, and my daughter, Johanne. Thank you for reminding me what life is all about and that sometimes Winnie the Pooh and Curious George have higher impact factors than JAMA and New England Journal of Medicine. From upstairs there is no more shouting, only the cute sleeping sound from the future big sister. Peter, Aarhus, Christmas 2014 IV Preface
PREFACE To test or not to test… Working as a general practitioner (GP) is to navigate in the area of tension between testing (or referring) the right patients at the right time and keeping healthcare expenses and patient harms to a minimum. As potential cancer symptoms are frequently seen in general practice and the predictive values of the symptoms are low, GPs face a very challenging task which I have experienced in my training as a GP. I am going to be a GP; and I believe that the important role of general practice in Denmark makes our healthcare system among the best in the world. However, I am discouraged by the poor cancer survival in Denmark and other countries with strong gatekeeper functions in general practice. We need to do better. To optimise the support to general practice in this difficult mission; we need to know how a suspicion of cancer or other serious diseases is handled in general practice. Another pertinent question is whether some GPs are handling the task more appropriately than others. Can we learn from the differences that we believe exist between general practices? Variation in the delivery of care has been investigated for decades, and thoughtful studies have proven unexplained and unwarranted variation in numerous aspects of health care. However, studies on the consequences of this variation are lacking in general and from a GP perspective, in particular. Therefore, this dissertation aims to investigate how often GPs suspect cancer and other serious diseases, and how they deal with this suspicion (Study I). Furthermore, it explores whether variations between GPs in the use of diagnostic investigations influence the prognosis of cancer patients (Study II and Study III). Before going into detail with each study, a general introduction to cancer, general practice and variation in healthcare is warranted. V Clinical Activity in General Practice and Cancer
THE THREE PAPERS OF THE DISSERTATION This thesis is based on the following papers: Paper I Hjertholm P, Moth G, Ingeman ML, Vedsted P. Predictive values of GPsʹ suspicion of serious disease: A population‐based follow‐up study. Br J Gen Pract. 2014;64(623):e346‐53. Paper II Hjertholm P, Fenger‐Gron M, Vestergaard M, Christensen MB, Borre M, Møller H, Vedsted P. Variation in general practice prostate‐specific antigen testing and prostate cancer outcomes: An ecological study. Int J Cancer. 2015;136(2):435‐442. Paper III Hjertholm P, Fenger‐Gron M, Vestergaard M, Christensen MB, Iversen LH, Vedsted P. Variation in general practitioners’ propensity to refer patients to lower endoscopy and the outcomes of colorectal cancer patients. Submitted VI Abbreviations
ABBREVIATIONS CI: Confidence interval CPR number: Personal identification number DRG: Diagnosis‐related groups GP: General practitioner ICD‐10: International Classification of Diseases version 10 ICPC: International Classification of Primary Care iFOBT: Immunochemical faecal occult blood test IRR: Incidence rate ratio LABKA: Laboratory information system PPV: Positive predictive value PSA: Prostate‐specific antigen TNM: Tumour classification system, T denotes tumour size, N denotes the presence of regional lymph nodes and M denotes the existence of distant metastasis TRUSP: Trans‐rectal ultrasound of the prostate UICC: Union for International Cancer Control VII Clinical Activity in General Practice and Cancer
VIII Contents
CONTENTS Contents ............................................................................................................................................ IX Chapter 1 ........................................................................................................................................... 1 Background ...................................................................................................................................... 1 Cancer in Denmark .................................................................................................... 2 General practice in Denmark .................................................................................... 4 Diagnostic intervals................................................................................................ 4 General practice and cancer .................................................................................. 5 The challenge in general practice ......................................................................... 7 Variation in health care ............................................................................................. 9 Variation in general practice ................................................................................. 9 Prostate cancer .......................................................................................................... 12 Colorectal cancer ...................................................................................................... 13 Background at a glance ............................................................................................ 15 Aim ............................................................................................................................. 16 Chapter 2 ......................................................................................................................................... 17 Materials and Methods ............................................................................................................... 17 Data sources and key variables .............................................................................. 19 The CPR number and the Danish Civil Registration System ......................... 19 The Danish National Patient Register ............................................................... 19 The Charlson Comorbidity Index ...................................................................... 20 The Danish Cancer Register ................................................................................ 21 The Danish National Health Service Register .................................................. 22 The Provider Number and the Patient Lists ..................................................... 23 IX Clinical Activity in General Practice and Cancer
Statistics Denmark ................................................................................................ 23 The KOS 2008 survey ........................................................................................... 24 LABKA – A clinical laboratory information system ........................................ 25 The Danish Register of Causes of Death ........................................................... 25 The Danish Colorectal Cancer Group database ............................................... 26 Methods Study I ........................................................................................................ 27 Study design .......................................................................................................... 27 Study participants................................................................................................. 27 Exposure ................................................................................................................ 27 Outcomes ............................................................................................................... 27 Statistical analyses ................................................................................................ 28 Methods Study II ...................................................................................................... 30 Study design .......................................................................................................... 30 Study participants................................................................................................. 30 Exposure ................................................................................................................ 30 Outcomes ............................................................................................................... 31 Statistical analyses ................................................................................................ 33 Methods Study III ..................................................................................................... 35 Study design .......................................................................................................... 35 Study participants................................................................................................. 35 Exposure (step 1) .................................................................................................. 35 Outcomes (step 2) ................................................................................................. 36 Statistical analyses ................................................................................................ 37 Lower endoscopies and mortality ...................................................................... 38 Ethics and approvals ................................................................................................ 39 Paper I .................................................................................................................... 39 Paper II and III ...................................................................................................... 39 Chapter 3 ........................................................................................................................................ 41 Results ............................................................................................................................................... 41 Study I ........................................................................................................................ 42 Prevalence of suspicion ........................................................................................ 42 Actions following the consultation .................................................................... 42 X Contents
Subsequent diagnoses .......................................................................................... 43 Use of health care services .................................................................................. 45 Study II ....................................................................................................................... 47 PSA test rates ........................................................................................................ 47 Associations between PSA test rates and outcomes ........................................ 48 Study III ..................................................................................................................... 50 Lower endoscopy rates ........................................................................................ 50 Associations between propensity to refer for lower endoscopy and outcomes ................................................................................................................ 51 Chapter 4 ......................................................................................................................................... 53 Discussion of methods ................................................................................................................ 53 Design ........................................................................................................................ 54 Bias, statistical methods and generalisability ....................................................... 57 Selection bias ......................................................................................................... 57 Information bias ................................................................................................... 58 Confounding ......................................................................................................... 66 Statistical methods and statistical precision ..................................................... 68 External validity ................................................................................................... 69 Variation in healthcare and outcomes ................................................................... 70 Chapter 5 ......................................................................................................................................... 73 Discussion of results ..................................................................................................................... 73 Results in general ..................................................................................................... 74 Study I ........................................................................................................................ 74 Study II ....................................................................................................................... 76 Study III ..................................................................................................................... 79 Chapter 6 ......................................................................................................................................... 83 Main conclusions ........................................................................................................................... 83 Study I ........................................................................................................................ 84 Study II ....................................................................................................................... 84 Study III ..................................................................................................................... 85 Chapter 7 ......................................................................................................................................... 87 Perspectives for the healthcare system and future research .................................... 87 English summary ........................................................................................................................... 91 XI Clinical Activity in General Practice and Cancer
Background and aims .............................................................................................. 92 Methods ..................................................................................................................... 92 Results ........................................................................................................................ 92 Conclusions and perspectives ................................................................................. 93 Dansk resume ................................................................................................................................ 95 Baggrund og formål ................................................................................................. 96 Metoder ...................................................................................................................... 96 Resultater ................................................................................................................... 96 Konklusion og perspektiver .................................................................................... 97 References ...................................................................................................................................... 99 Appendix I .................................................................................................................................... 119 Registration form for GPs in the KOS 2008 survey ...................................................... 119 Appendix II ................................................................................................................................... 121 Diagnoses included in Study I............................................................................................... 121 Paper I ............................................................................................................................................ 129 Paper II ........................................................................................................................................... 139 Paper III.......................................................................................................................................... 151 XII Background
CHAPTER 1 BACKGROUND 1 Clinical Activity in General Practice and Cancer
CANCER IN DENMARK Cancer has been the leading cause of death in Denmark since 20001. The incidence of cancer increased by 11% from 2003 to 20122 and it is predicted that the cancer incidence will increase by 20% from 2002 to 2020 because of the ageing of the population3. A total of almost 37,000 new cancer diagnoses were registered in the Danish Cancer Registry in 2012 (excl. basal‐cell carcinoma), and the prevalence was estimated to be 257,619 in the Danish population (5.6 million) at the end of 2012. Cancer diagnosis, treatment and the related expenses are estimated to amount to 2.2 billion Euros, about 1% of the Danish gross domestic product4. Additionally, the personal and psychological costs are immense. About one third of the Danish population will be diagnosed with cancer during their lifetime3; hence, cancer affects the lives of the majority of the population, directly or indirectly. Both nationally and internationally, the prognosis has improved for most cancers5. However, international comparisons have revealed significant differences in cancer survival between countries5, 6. Survival after a cancer diagnosis in Denmark and the United Kingdom, for example, is inferior to the survival in comparable western countries and has consistently been so for most cancer types throughout the past decades (Figure 1.1). These differences seem to be rooted partly in more advanced tumour stages at diagnosis and later treatment initiation in Denmark and the United Kingdom7‐14. The disparities in cancer survival have caused much concern about cancer diagnostics and have sparked various political initiatives in Denmark. In 2001, a law was passed comprising a 2‐week waiting time guarantee15. Furthermore, in 2007, the government and the five Danish regions (who own and run the hospitals) launched a new diagnostic strategy that classified cancer as an acute condition and curtailed waiting time in the cancer care pathway to that which was necessary only for medical purposes16. By the spring of 2009, 2 Background
multidisciplinary groups had outlined urgent referral pathways for patients suspicious of cancer for diagnosis and treatment of the most common cancers (cancer fast‐tracks)17. In 2012, an additional fast‐track referral for non‐specific cancer or serious disease was introduced. Figure 1.1: Age‐standardised 1‐year and 5‐year relative survival trends 1995‐2007, by cancer and country. Data are for adults (15‐99 years) diagnosed with colorectal, lung, breast or ovarian cancer in 1995–99 and 2000–02 (cohort approach), and short‐term prediction of survival for those diagnosed in 2005–07 (period approach). Ovarian cancer data were not supplied by Sweden.5 3 Clinical Activity in General Practice and Cancer
GENERAL PRACTICE IN DENMARK The Danish healthcare system is tax‐funded and free of charge at the point of care18. All medical expenses are paid by the government except medication and dental care which are only partly covered. General practitioners (GPs) play a pivotal role in the Danish healthcare system where they act as gatekeepers to the secondary healthcare and manage most chronic and acute diseases. More than 98% of the Danish citizens are listed at a general practice which they must attend for medical advice, and 86% of the population does so within a year19. The average GP list comprises around 1600 patients, and GPs work either in single‐handed or partnership practices. In total, there are about 3600 GPs in Denmark in 2100 GP practices of which 40% are partnership practices20. Approximately one third of a GP’s income comes from capitation payment for patients on the GP’s list and two‐thirds come from fee‐for‐service payments. About 90% of the contacts to general practice are managed solely by general practice without any referral to other parts of the healthcare system. This system seems cost‐effective for most diseases, but a study from 2011 has questioned the benefits of the system regarding cancer21. The study showed that countries with healthcare systems where GPs serve as gatekeepers had worse outcomes for cancer patients than countries operating no such system. The data for the study covered patients diagnosed in 1995‐1999 (Eurocare 4)22, but the picture is the same in more recent comparisons5, 6. Diagnostic intervals The pathway to diagnosis for cancer patients, the initial part of “the cancer journey”, comprises several steps, conceptualised in the Aarhus statement (Figure 1.2)23. Ideally, a shortening of each of these intervals in combination with improved prevention and better treatment would optimise the prognosis for cancer patients. Each of these factors also offers a possible explanation for the 4 Background
less favourable prognosis in Denmark and the United Kingdom, especially the primary care interval as indicated by Vedsted et al21. Figure 1.2: Categorisation of intervals in the cancer care pathway23 General practice and cancer GPs are involved in the diagnostic process of about 85% of all cancer patients24, 25
. This entails that GPs have a significant responsibility for cancer diagnostics. A study investigating the primary care interval found that the median time from first presentation of the symptom in general practice to referral was 0 days26; still, some patients experienced very long primary care intervals (Figure 1.3). Research also shows that more than 6 months before diagnosis, cancer patients have increasing consultation rates compared with the background population27 (Figure 1.4). These findings indicate a potential for earlier diagnosis in primary care. 5 Clinical Activity in General Practice and Cancer
Figure 1.3: Primary care interval (GP delay) for 1877 newly diagnosed cancer patients in 2004‐200526 Figure 1.4: Incidence rates of health services received per month by cancer patients and reference population. Cancer patients (n = 63,362 women and 63,848 men). Reference population (n = 633,620 women and 638,480 men). Incidence rates were adjusted for time at risk. Vertical line indicates date of diagnosis. GP: General practitioner; Diag.: Diagnostic investigations; Hosp.: Hospital contacts.27 The introduction of the fast‐tracks gave GPs new opportunities for referring patients to diagnostic workup in cases where patients present with pre‐specified symptoms suspicious of cancer (alarm symptoms) and, hence, an opportunity to shorten the primary care interval arose. The potential effects of these efforts have yet to be scrutinised in more updated comparisons. But research shows 6 Background
that the use of fast‐tracks varies among GPs28 and the GP’s symptom interpretation is a stronger predictor of the diagnostic interval than the use of a fast‐track29, which evidently questions the potential effect of the fast‐track strategy. The challenge in general practice The strategy of cancer fast‐tracks may also be weakened by three major challenges that GPs are facing concerning cancer. ‐ Firstly, potential cancer symptoms are frequently seen in the general population and in general practice. A Danish survey among more than 13,000 persons found that 15.3% of Danish females and 12.7% of the males had experienced at least one cancer alarm symptom within the past 12 months30. Among their patients, Norwegian GPs registered warning signs of cancer in 12.4% of the consultations,31 but GPs only suspected cancer among 24% of these, indicating that GPs do not solely rely on alarm symptoms when diagnosing cancer. ‐ Secondly, positive predictive values of potential cancer symptoms are low; this is in part due to the high prevalence of the symptoms and the low prevalence of cancer. The positive predictive value (PPV) of a certain symptom is the proportion of patients with that symptom who are diagnosed with cancer32. If 100 patients present with rectal bleeding and five of these turn out to have colorectal cancer, the PPV of rectal bleeding for colorectal cancer is 5%. A review by Shapley et al.33 found that only eight potential cancer symptoms had PPVs in general practice above 5%. Many other symptoms have PPVs between 1% and 5%34‐40, which means that GPs need to refer between 20 and 100 patients to find one cancer patient. The PPV increases as symptoms appear simultaneously, e.g. abnormal rectal examination and rectal bleeding combined has a PPV of 8.5%34. ‐ Thirdly, not all cancer patients present with cancer alarm symptoms. A Danish study41 showed that among cancer patients, the first symptom was interpreted 7 Clinical Activity in General Practice and Cancer
as an alarm symptom by the GP in 49% of the patients, as a general symptom indicating serious disease in 24% and as a non‐characteristic symptom in 27% of the patients. GPs are accordingly faced with a very complex task regarding cancer diagnostics. For example, coughing is the most common symptom of lung cancer present among 30‐65% of lung cancer patients34, 42, 43, but it is also a frequent symptom in the population and among patients consulting their GP43. This leads to a low PPV of cough (<1%)34. On the other hand, haemoptysis has a higher PPV (2.4 to above 10%)34, 38 because its prevalence in the population is lower, but focusing on haemoptysis will identify only about 20% of the lung cancer patients. Hence, to rely only on cancer fast‐tracks in the diagnosis of cancer may be questionable because the cancer fast‐tracks focus exclusively on patients with cancer alarm symptoms. GPs must daily balance the information on PPVs, disease prevalence, etc., in order to diagnose cancer patients as early as possible and to avoid unnecessary investigations in other patients44. Nylenna showed that GPs suspected cancer in 4.2% of the consultations in the 1980s45 and that 7.8% of these patients were subsequently diagnosed with cancer46. To achieve an optimal organisation of our healthcare system, we need updated knowledge on how often GPs suspect cancer and other serious diseases. We need to know how they react upon this suspicion, especially because the GP’s symptom interpretation strongly affects the diagnostic interval29. 8 Background
VARIATION IN HEALTH CARE How symptoms are interpreted and managed varies among GPs47‐49; this fact may open a window of opportunity to learn from “the best practice”. The variation in health care services and activity provided by general practices raises questions about equity, quality and efficiency of resource allocation50. If patients registered at GPs with a higher diagnostic activity (e.g. more diagnostic testing) have better outcomes, then other patients may benefit from similar care. Adversely, if more diagnostic activity does not entail better outcomes, then there may be opportunities for reorganising the services provided. Variation in health care has been a focus of much attention since Glover showed large variations in tonsillectomy rates in adjacent school districts in 1938 without obvious explanations or differences in outcomes51. John E. Wennberg,52 who for decades has investigated variation in health care delivery especially in the secondary sector in the United States, has contributed to the field with numerous studies. Acknowledging the importance of these studies, this thesis, however, will focus on variation in general practice. Variation in general practice The studies exploring variation in general practice are diverse. Some describe the variation, some try to explain it, and only few studies aim to investigate the potential prognostic impact of such variation for patients53. The subjects of the studies also vary. Most focus on referrals 50, 54‐68, whereas a few focus on referrals for the suspicion of cancer28, 69, the use of x‐ray70‐72, ordering of laboratory tests70, 72‐74
, appropriateness of referrals75, 76, and medicine prescriptions70, 77. Studies exploring the reasons for variation have only been able to explain up to 50% of the variation examined53, 58, leaving most variation unexplained. Variation has been attributed to both the actors (patient, GP, and specialist) and the circumstances (GP practice, access to specialist care) involved. The findings have 9 Clinical Activity in General Practice and Cancer
been equivocal, but most studies conclude that patient‐related factors (age, socio‐economic situation and differences in morbidity (case‐mix)) are important to take into account when studying variation. Regarding the consequences of variation in general practice, O’Kane et al73 found that differences in the use of HbA1c and thyroid function tests were not associated with disease prevalence or Quality and Outcomes Framework (QOF) scores in diabetes and hypothyroidism. Hippisley‐Cox et al78 made a small study on 300 colorectal cancer patients and 131 breast cancer patients and found that late‐stage disease was not associated with being registered at practices with low referral rates. In contrast, the large study by Shawihdi et al.79 showed that a low gastroscopy referral rate in general practice was associated with a lower rate of major surgery for oesophagogastric cancer, a higher rate of emergency admission and a higher mortality. A study by Mäntynen et al80 investigating gastroscopy referrals showed a positive relation between referral volume and the proportion of gastrooesophageal reflux disease (GERD) patients with abnormal findings on gastroscopy. The latter study did not include the number of patients in the practices and hence could not calculate referral rates; furthermore, only one patient was diagnosed with adenocarcinoma during the study period which compromises the strength of the study. As the above paragraph illustrates, the possible consequences of variation in general practice have not been systematically and thoroughly investigated and the methods have been of varying quality53, 81. As GPs hold a crucial role in cancer patients’ diagnostic process, it is important to know whether differences between GPs’ actions are related to the prognosis of their cancer patients. The healthcare system as a whole is remarkably complex, and variation between GPs is shaped by a myriad of factors (waiting lists, GP characteristic, patient factors, organisation of the health care system, etc.)53. However, studies of variation can 10 Background
take all this into account if they investigate how different levels of diagnostic activity affect outcomes in comparable populations (Figure 1.5). Figure 1.5: Simplified illustration of factors influencing referral variation and the effects of such variation on outcomes Factors related to GP, patients, or health care system Variation in activity
Outcome
Regarding variation, this thesis focuses on two cancers, prostate cancer and colorectal cancer. Prostate and colorectal cancers are frequent and important, and collectively they represented 35% of cancer cases among male patients in 2012, whereas colorectal cancer represented 12% of female cancer patients. Both patient groups present with symptoms often seen in general practice. Moreover, both diseases may be diagnosed with tests that can be initiated by the GP, but with different levels of invasiveness (blood test vs. referral for endoscopy). Both cancers have been investigated intensively regarding screening, and the studies performed indicate that an earlier diagnosis is possible82‐88, but whether this applies to patients with symptomatic presentation is unknown. 11 Clinical Activity in General Practice and Cancer
PROSTATE CANCER Prostate cancer is the second most frequent cancer in Denmark among males, (excl. non‐melanoma skin cancer) with 4316 incident cases in 2012,2 and it is the second most common cause of cancer‐related death in Danish men1. The incidence of prostate cancer in Denmark and the United States showed dramatic increases that were temporally linked to the introduction of prostate‐specific antigen (PSA) testing89, 90; and worldwide the incidence correlates closely with the use of PSA testing91. The 5‐year prostate cancer survival is estimated at 53% in Denmark compared with 78% or above in the other Nordic populations for patients diagnosed in 1999–200392, a difference partly caused by lead time bias as the uptake of screening with PSA is lower in Denmark93, 94. PSA is a protein secreted by the prostate to nourish sperm, and elevated levels of PSA indicate that “something is wrong with the prostate”. This includes prostate cancer, but also benign diseases like benign prostatic hyperplasia, prostatitis and urinary retention95. This compromises the diagnostic value of the PSA test by lowering its specificity; and, moreover, the sensitivity of the PSA test is low because some prostate cancers do not produce PSA96, 97. Various trials98‐102 have investigated PSA as a tool for screening; and findings have been ambiguous, but a Cochrane review from 2013 concluded that screening with PSA does not reduce prostate cancer mortality82. However, discussions are ongoing as the European Randomised Study of Screening for Prostate Cancer recently reported substantial reductions in prostate cancer mortality after 13 years of follow‐up83. Early diagnosis of prostate cancer is desirable, but there is an unresolved balance between the benefits of screening and the harm of diagnosing and treating men with low‐risk prostate cancer, who are unlikely to profit from treatment. Danish guidelines for PSA testing therefore recommend against the screening of asymptomatic men103 (Figure 1.6), but studies indicate that opportunistic screening is being performed104‐106. If 12 Background
there is a variation between GPs in their use of PSA tests, it may have downstream consequences mimicking the effects of a screening program with more diagnostic investigations, higher prostate cancer incidence and consequences for the stage distribution and outcomes. Figure 1.6: Danish guidelines for PSA testing, 2012103 The Danish Urological Society recommends PSA testing restricted to men with: ‐ Abnormal digital rectal examination ‐
Lower urinary tract symptoms or other symptoms that may be due to prostate cancer ‐
Two or more first‐ or second‐degree relatives with prostate cancer
COLORECTAL CANCER Colorectal cancer is the third most common cancer among Danish men after lung and prostate cancer, and the third most common cancer among women after breast and lung cancer (excluding non‐melanoma skin cancer)2. The age‐
standardised incidence of colorectal cancer has risen 20% over the past 20 years in Denmark. The lifetime risk of colorectal cancer in the Western world is approximately 3‐4%107. The colorectal cancer incidence increases with age with more than 90% of all diagnoses occurring in individuals older than 50 years108. Survival after a colorectal cancer diagnosis has increased over the past decade in Denmark, but it is the lowest in international and Nordic comparisons with a 5‐
year relative survival that is 5‐8 percentage points lower than in the other Nordic countries5, 109 (Figure 1.1). One possible explanation for this disparity seems to be more advanced disease stages at diagnosis in Denmark8, 12. To improve colorectal cancer outcomes, a national screening program in Denmark for colorectal cancer with biennial tests using the immunochemical 13 Clinical Activity in General Practice and Cancer
test for occult blood in stools (iFOBT) was launched in March 2014110. A positive test result will be followed by a colonoscopy. Randomised, controlled trials have shown reductions in colorectal cancer mortality rates using this screening scenario111‐113, and a Danish report from 2008 showed that a participation rate above 40% would be cost‐effective114. However, with participation rates down to 40‐50%, the proportion of colorectal cancer diagnosed through the screening program is estimated to be below 25%115. This highlights that the role of general practice remains crucial, also in the future. Survival from colorectal cancer is highly dependent on stage at diagnosis with a 5‐year survival above 80% for stage I tumours (T1‐T2, N0, M0) and less than 20% in stage IV (any T, any N, M1)116. Furthermore, the choice of treatment also depends on the stage at diagnosis with patients in stage I‐III usually having surgery with a curative intent, whereas this only befalls to a few patients with stage IV cancers. Most colorectal carcinomas are derived from precursor lesions, commonly referred to as polyps, and resection of polyps can prevent the development of colorectal cancer117. Screening trials using sigmoidoscopy have also proven to reduce the cancer incidence and the number of patients with late‐
stage disease85, 86, 88, 118, 119 and thereby to reduce mortality by up to 31%. In observational studies involving screening endoscopies, the reduction in mortality has reached 60%87. Another important prognostic factor associated with the timeliness of diagnosis is acute vs. elective surgery. Patients with emergency presentation and hence acute surgery are known to have worse outcomes120, 121. The above paragraph indicates that variation in the referral for lower endoscopy in response to symptoms could influence the timeliness of diagnosis and hence the incidence of colorectal cancer and the prognosis. 14 Background
BACKGROUND AT A GLANCE Globally, cancer is an immense and growing challenge to healthcare systems, and cancer survival in Denmark compares unfavourably with that of other western countries across most cancers. Earlier diagnosis may be one way to close the survival gap, and general practice plays a pivotal role in this effort. General practice is challenged by a high prevalence of cancer alarm symptoms with only low PPVs. Knowledge about GPs’ suspicion of cancer and related activities can therefore guide the healthcare system and policy makers in optimising the support to GPs. Furthermore, we may learn from the variation in GPs’ cancer‐related diagnostic workup. Variation in the use of PSA tests and lower endoscopy in general practice may affect the outcomes of prostate and colorectal cancer patients, respectively, and this knowledge may help inform and refine GPs’ diagnostic processes for cancer patients in the future. 15 Clinical Activity in General Practice and Cancer
AIM The overall aim of this thesis is to investigate 1) how often GPs suspect cancer and other serious diseases after a consultation, how they respond to this suspicion, and the consequences of this suspicion and 2) to investigate whether variation in GPs’ diagnostic activity influences cancer patients’ prognosis. In Paper I, we aimed to describe how often Danish GPs suspect cancer or other serious diseases after a consultation and to characterise the patients in whom suspicion was raised. We wanted to describe how the GPs acted on their suspicion and to analyse how a suspicion may influence the demand for healthcare services and predict a future diagnosis of serious disease. In Paper II, we explored the association between variations in the use of PSA tests among general practices in the Central Denmark Region in 2004–2009 and prostate cancer‐related outcomes, i.e. cancer incidence, stages at diagnosis, survival and mortality. In Paper III, we investigated the variation in the propensity to refer to lower endoscopy among general practices and whether this variation was associated with the stage at diagnosis for colorectal cancer patients, and the proportions of patients subjected to elective surgery, being treated with curative intent, or having a poor‐prognosis colorectal cancer. 16 Materials and Methods
CHAPTER 2 MATERIALS AND METHODS 17 Clinical Activity in General Practice and Cancer
Table 2.1: Overview of study populations, designs, data sources, exposures and outcomes in Papers I‐III Paper I Paper II Study 18+ patients in the population KOS 2008 survey, Central Denmark Region Study Prospective design population‐based cohort study Period 2008‐2009: survey data and 6 months of follow‐up Data ‐ The Danish National Sources Patient Register ‐ The Danish National Health Service Register ‐ KOS 2008 ‐ Statistics Denmark Exposure GP suspicion of cancer or another serious disease Outcomes ‐ Actions taken by GP ‐ New diagnoses of serious disease ‐ Use of healthcare Paper III 40+ men registered with 40+ patients registered GPs in Central Denmark with a Danish GP Region Population‐based register study Population‐based register study 2004‐2009: follow‐up until January 2012 2002‐2010: for exposure 2004‐2011: main analysis ‐ The Danish National Patient Register ‐ The Danish Cancer Register ‐ Patient lists ‐ LABKA ‐ The Danish Register of Causes of Death ‐ Statistics Denmark ‐ The Danish National Patient Register ‐ The Danish Cancer Register ‐ Patient lists ‐ The Danish National Health Service Register ‐ The Danish Colorectal Cancer Group database ‐ Statistics Denmark GPs propensity to refer to lower endoscopy GP PSA test rate ‐ Mean PSA level ‐ Age at diagnosis ‐ Incidence of TRUSP ‐ Incidence of biopsies ‐ Incidence of PC ‐ Stages of PC at diagnosis ‐ Incidence of radiotherapy ‐ Incidence of radical prostatectomy ‐ Relative survival ‐ Mortality, PC‐specific and overall ‐ Incidence of CRC ‐ Stages of CRC at diagnosis ‐ Proportion with of elective surgery ‐ Proportion of surgery with curative intent ‐ Proportion with poor‐
prognosis CRC (no surgery, acute surgery, stage IV disease) KOS 2008: Survey on reasons for encounter and disease patterns in Danish general practice, GP: General practitioner; LABKA: Clinical laboratory information system, PSA: Prostate specific antigen; TRUSP: Trans‐rectal ultrasound of the prostate; PC: Prostate Cancer, CRC: Colorectal Cancer 18 Materials and Methods
The studies in this thesis differ in design, study populations, data sources and outcome measures (Table 2.1). Even though some methodological issues in Studies II and III are similar, there are also important differences. First some mutual data sources and key variables will be described and then the three studies will be explained individually. DATA SOURCES AND KEY VARIABLES The CPR number and the Danish Civil Registration System The Danish Civil Registration System assigns a unique 10‐digit identifier (the CPR number) to each Danish citizen at birth or immigration. This identifier is used in all contacts with public authorities, including health care contacts122, 123. This facilitates linkage of information from various registries at an individual level. The civil registration system was established in 1968, and it keeps daily updated information on vital status, emigrations and residential addresses for all Danish inhabitants. The information on vital status, emigrations and deaths was used to define the study populations and the corresponding risk time (Studies I, II and III). The Danish National Patient Register The Danish National Patient Register holds information on all non‐psychiatric hospitalisations in Denmark since 1977. From 1995 all outpatient visits were also registered124, 125
. Registrations are mandatory and used for administrative purposes, and since 2000 the Danish National Patient Register has formed the basis of payment to public as well as private hospitals using the Diagnosis‐
Related Group (DRG) system. For each hospital contact, the Danish National Patient Register holds a record with information on CPR number, date of admission and discharge, procedure codes, surgical codes and diagnoses. The records hold up to 20 diagnosis coded 19 Clinical Activity in General Practice and Cancer
with the International Classification of Diseases coding system (ICD) version 8 until 1993 and version 10 onwards126. Information was obtained from the Danish National Patient Register on new diagnoses, hospital contacts, diagnostic imaging and endoscopic investigations (Study I), diagnostic, surgical and radio‐therapeutic procedures on the prostate (Study II) and sigmoidoscopies and colonoscopies (Study III). Furthermore, diagnoses used for calculation of the Charlson comorbidity index were also obtained from the Danish National Patient Register (Studies II and III). The Charlson Comorbidity Index The Charlson comorbidity index is presented here as it builds entirely on data from the Danish National Patient Register. In Studies II and III, the Charlson comorbidity index127 was included as a measure for the burden of pre‐existing comorbidity. It is based on all available primary and secondary diagnoses in the Danish National Patient Register (both inpatient and outpatient hospital diagnoses). If a patient was registered with one or more of the diseases in Table 2.2, the corresponding scores were added together. Regarding liver disease, diabetes and cancer, only the highest score was applied. Persons were grouped based on their individual score (0, 1‐2 and ≥3). Included diagnosis codes (ICD‐
10) are based on the slightly modified version described by Sundararajan et al 128 and on the codes proposed by the Department of Clinical Epidemiology, Aarhus University Hospital129. In Study II, the Charlson comorbidity index was based on all diagnoses since 1994 until censoring and used for comparison of comorbidity level in different practice populations. Prostate cancer diagnoses were not included. In Study III, the index was calculated each year for every person using a 10‐year history of diagnoses. Colorectal cancers were not included. 20 Materials and Methods
Table 2.2: ICD‐10 diagnosis codes included in the calculations of the Charlson comorbidity index Score Condition ICD‐10 codes 1 Myocardial infarction I21; I22; I23 Congestive heart failure I50; I11.0; I13.0; I13.2 Peripheral vascular disease I70; I71; I72; I73; I74; I77 Cerebrovascular disease I60‐I69; G45; G46 Dementia F00‐F03; F05.1; G30 Chronic pulmonary disease J40‐J47; J60‐J67; J68.4; J70.1; J70.3; J84.1; J92.0; J96.1; J98.2; J98.3 Connective tissue disease M05; M06; M08; M09; M30; M31; M32; M33; M34; M35; M36; D86 Ulcer disease K22.1; K25‐K28 Mild liver disease B18; K70.0‐K70.3; K70.9; K71; K73; K74; K76.0 Diabetes without end‐organ E10.0, E10.1; E10.9; E11.0; E11.1; damage E11.9 2 Diabetes with end‐organ damage E10.2‐E10.8, E11.2‐E11.8 Hemiplegia Moderate to severe renal disease G81; G82 I12; I13; N00‐N05; N07; N11; N14; N17‐N19; Q61 Non‐metastatic solid tumour C00‐C75 Leukaemia C91‐C95 Lymphoma C81‐C85; C88; C90; C96 3 Moderate to severe liver disease B15.0; B16.0; B16.2; B19.0; K70.4; K72; K76.6; I85 6 Metastatic cancer C76‐C80 AIDS B21‐B24 The Danish Cancer Register The Danish Cancer Register was founded in 1942 and covers the entire population of Denmark124, 130. It contains information on all incident malignant neoplasms diagnosed at Danish hospitals including information on diagnosis (reconstructed to ICD‐10 back to 1978), date of diagnosis and tumour stage at diagnosis. To optimise the completeness of the Danish Cancer Register, various data sources are used. The Danish Cancer Register is mainly based on 21 Clinical Activity in General Practice and Cancer
mandatory electronic reports from the hospitals (via the Danish National Patient Register), complemented with linkage to the Danish Pathology Register, the Danish Register of Causes of Death and registrations from the primary healthcare sector131. If a patient develops more than one cancer, each cancer will be registered in an individual record and numbered according to order of diagnosis. The date of diagnosis in the Cancer Register is recorded as the first date from either hospital registrations (the Danish National Patient Register) or registrations from the Danish Pathology Register132. Dates from the latter register are preferred, because it is the date at which the diagnosis is verified. Dates from the Danish National Patient Register are coded as the first date of the actual hospital record where the cancer is coded for the first time. Since 2004, stage at diagnosis has been coded using the TNM system (T denotes tumour size, N denotes the presence of regional lymph nodes, and M denotes the existence of distant metastasis133). Stage at diagnosis in the Danish Cancer Register is the first registered TNM stage (can be clinical or pathological); however, if information modifies the stage within the first 4 months (e.g. by additional investigations, multidisciplinary team conference, etc.), the TNM stage is changed in the Register. From the Danish Cancer Register, we identified all prostate cancers including information on stages at diagnosis (Study II) and all colorectal cancers (Study III). The Danish National Health Service Register The Danish National Health Service Register covers activities of health professionals contracted with the tax‐funded public healthcare system134 including GPs and private practising medical specialists (e.g. radiologists and gastroenterologists). Professionals report patient contacts and related 22 Materials and Methods
procedures to the Regional Health Administrations, and registrations form the basis for remuneration. From this Register, information was collected on contacts to GPs and practising specialists, diagnostic imaging at private practising radiologists and endoscopies at gastroenterologists and surgeons (Study I). In Study III, this Register was used to obtain information on non‐hospital sigmoidoscopies and colonoscopies. The Provider Number and the Patient Lists More than 98% of the Danish population are listed at a general practice (see Chapter 1), and the list system enables linkage of patients to a specific general practice at any time. Danish healthcare professionals contracted with the public healthcare system are conferred a provider number. GPs in partnership practices work under a shared provider number. In Studies II and III, the provider number and the list system were used to link patients and general practices, and to obtain information about whether practices were single‐handed or partnership practices. The register information enabled identification of patients shifting to another practice, and consequently calculation of accurate risk time at the level of each general practice. Statistics Denmark Statistics Denmark was founded in 1850 and serves as an independent supplier of statistics concerning the Danish society. Furthermore, Statistics Denmark holds information on all Danish citizens available for researchers. In the studies of this thesis, we used information on ethnicity, income, labour market status, education, urbanisation and marital status135. Data on ethnicity, income, labour market status and education are all available from the Danish Integrated Database for Labour Market Research (IDA)136, 137. Ethnicity is based on information on each individual’s country of origin and coded as Denmark (including Greenland and Faroe Island), western, or non‐
23 Clinical Activity in General Practice and Cancer
western countries. Descendants of persons without Danish citizenship are also coded as non‐Danish. Income is based on taxable income and defined as the Organisation for Economic Cooperation and Development adjusted household income, adjusted for number of persons in the household138, 139. Labour market status is based on each person’s most important source of income each year computed at the end of November every year136. Educational level was defined as each individual’s highest completed education which is available in the Population’s Education Register as an eight‐digit code140. Urbanisation indicates the city size in which each person is living. Marital status is based on information on marriage and addresses, the latter enabling registration of cohabitation. Statistics Denmark also hosted the data for this thesis. The data were accessed through a secure remote access. This enabled completely anonymised linkage of the register data141. The KOS 2008 survey The KOS 2008 survey (Kontakt‐ og sygdomsmønsteret i almen praksis) was a survey on reasons for encounter and disease patterns in Danish general practice used in Study I. All 845 GPs serving approximately 1.2 million inhabitants in the Central Denmark Region were invited to participate and 404 GPs (48%) concurred 142, 143. During the 12‐month period from December 2008 to December 2009, participating GPs were randomly assigned one work‐day on which they had to record all patient contacts. The GPs received payment for their participation (€32) and for each registered contact (€3). The recorded information included background information such as date of the consultation, the patient’s age, gender and CPR number; the latter enabled linkage to various registers. Furthermore, clinical information was registered including reason for encounter, chronic diseases and actions taken by the GP (referrals, tests in the clinic, follow‐up appointments). Reason for encounter was 24 Materials and Methods
written in text or stated by codes using the International Classification of Primary Care coding system (ICPC144). Diagnoses in text were subsequently coded by an experienced medical student who was trained in ICPC coding. All codes were subsequently validated by one of the authors (Grete Moth). Additionally, the registration form included the question “Are you left with the slightest suspicion of cancer or another serious disease (new)?”. The registration form was based on ad hoc questions and scrutinised by several researchers before being piloted among nine GPs. The questionnaire (in Danish) is shown in Appendix I. LABKA – A clinical laboratory information system The clinical laboratory information system, LABKA, is an electronic database for all laboratories in the Central Denmark Region145. Most GPs analyse C‐reactive protein, haemoglobin and blood glucose in the clinics; however, almost all other analyses are made at hospital laboratories. The results of biochemical analyses performed at the hospitals are electronically transferred directly to the database, which holds information on all PSA tests including test date, test requestor and test result. For Study II, we included all PSA tests ordered in general practice. The Danish Register of Causes of Death All deceased persons in Denmark are registered with a death certificate in the Danish Register of Causes of Death124, 146
. Persons dying at hospitals are registered by hospital physicians, and persons dying outside hospitals are registered by their GP. The death certificate contains the CPR number of the deceased person and information on one to four causes leading to death, one coded as the underlying cause of death. Before 2007, the registrations were sent to the National Board of Health where the recordings were completed and the cause of death was coded (ICD‐10 from 1994 and onwards). Since 2007, the death certificate has been electronic, and the 25 Clinical Activity in General Practice and Cancer
data registered by the physician are automatically transferred to the Register, including ICD‐10 codes. In Study II, we looked at prostate‐cancer‐specific deaths (ICD‐10 code: C61)126. The Danish Colorectal Cancer Group database The Danish Colorectal Cancer Group established a national database in May 2001. It covers all patients with a first‐time diagnosis of colorectal adenocarcinoma treated or diagnosed in Danish surgical departments. Surgeons report clinical data on patients to the database encompassing year of surgery, tumour stage (Union for International Cancer Control133 (UICC)), tumour site (colon/rectum), intent of surgery (curative/palliative), whether surgery was acute or elective, and other clinical information. The database is validated on a regular basis by linkage to the Danish National Patient Register and the Danish Cancer Register147‐149. The details on colorectal cancers from the database were used in Study III. 26 Materials and Methods
METHODS STUDY I Study design We conducted a prospective population‐based cohort study in the Central Denmark Region during 2008–2009 combining data from the KOS 2008 survey on patient contacts to general practice with registry data. Study participants We included 4518 patients aged 18 years and above and registered with a face‐
to‐face consultation and a complete CPR number in the KOS 2008 survey. Exposure The question “Are you left with the slightest suspicion of cancer or another serious disease (new)?” formed two exposure groups: patients where the GP had a suspicion and patients where the GP had no suspicion. Outcomes The following outcomes were compared in the two exposure groups: Actions taken by the GP after the consultation. Information on referrals, laboratory tests in the clinic and follow‐up appointments was retrieved from the registration form. Serious diagnoses. Patients were followed for 6 months to identify all new serious hospital‐based diagnoses in The Danish National Patient Register. Only new diagnoses were included. Thus, we excluded diagnoses already registered between January 2000 and the index consultation. All cancers except non‐
melanoma skin cancers (C44) were included. Serious diseases other than cancer were defined by the authors (PH, MI, PV) by independently reviewing all 1023 new diagnoses (4‐digit ICD‐10 code, e.g. A415) for the patients while blinded to the GP registrations. Disagreements were discussed and consensus reached. The full list of the 279 included serious diseases is shown in Appendix II. 27 Clinical Activity in General Practice and Cancer
Use of health care services. We included hospital services from the Danish National Patient Register for the 6 months following the index consultation. The information included inpatient stays, outpatient visits, diagnostic imaging (ultrasound, conventional x‐ray, CT and MR scan) and endoscopies (gastroscopy, colonoscopy and sigmoidoscopy). Data on diagnostic imaging and endoscopic investigations performed by primary care specialists were obtained from the Danish National Health Service Register134 as was also the use of general practice and other practicing specialists. Contacts to general practice were defined as face‐to‐face consultations, including home visits. Contacts to practicing specialists included all contacts to practising specialists in dermatology, neurology, surgery, gynaecology, psychiatry, otorhinolaryngology and internal medicine. Patient characteristics. The two exposure groups were compared regarding the following variables: Age at consultation, chronic diseases categorised according to the number of diseases. Other descriptive variables were obtained from Statistic Denmark including income, marital status (married and cohabiting or living alone) and labour market status (working, retirement pension or out of the workforce (unemployment/incapacity/sickness)). Statistical analyses We used generalised linear models from the binomial family with either identity link (for prevalence differences) or logarithmic link (for prevalence ratios) to investigate associations between patient characteristics and suspicion of serious disease (the STATA command binreg). We performed robust variance estimation accounting for clustering at GP level. Associations between suspicion and actions taken during the consultation were tested in a corresponding model. The risk of being diagnosed with a new serious disease in the period from the index consultation until 2 months after (61 days) and 2‐6 months after (62‐183 days) was analysed using Cox proportional hazard models with the first 28 Materials and Methods
diagnosis as the outcome variable. Patients were censored at the date of diagnosis, at death or 6 months after the consultation, whichever came first. The assumption of proportional hazards was fulfilled. The assumption was tested with visual inspection of a plot of the log(‐log) survival curves and inspection whether there was interaction between time and the explanatory variable (suspicion or not). Persons experiencing an event within the first 2 months were not included in the analysis of months 2‐6. The use of health care services was compared by using a generalised linear model with a dichotomised outcome (consultation or no consultation). The analyses were repeated with the number of GP consultations as a count variable using a negative binomial regression model to test the consistency of the results. PPVs were calculated as the proportion of patients in whom a suspicion was raised and confirmed, divided by the total number of suspected patients. Confidence intervals were calculated using the STATA command diagt. Sensitivity analyses were performed including only 58 definitely serious diseases (marked * in Appendix II) to test the possible effect on PPVs and hazard ratios of adding disputable serious diseases as outcomes. Multivariate analyses were adjusted for the following confounders: age group (18‐39, 40‐54, 55‐69, ≥70), gender, marital status, income, chronic diseases on the registration form (0, 1‐2, 3 or more) and risk time. In the analyses of healthcare use, we also adjusted for the use of each service in the year preceding the index consultation (dichotomous). This was done to account for possible pre‐existing differences in the use of healthcare. In the Cox regression, the number of outcomes was smaller than in the other analyses which has implications for the number of parameters that can be included in the model. Thus, the Cox regressions were adjusted for age as a continuous variable, gender and chronic diseases coded as dichotomous variables (0, 1 or more). 29 Clinical Activity in General Practice and Cancer
METHODS STUDY II Study design We conducted an observational population‐based ecological study with individual‐level analyses where general practices were ranked according to their PSA test level and divided into quartiles. Prostate cancer‐related outcomes in the four practice populations were compared. Study participants We included men residing in the Central Denmark Region and registered at a general practice from 1 January 2004 to 31 December 2009. Men without a previous diagnosis of prostate cancer in the Danish Cancer Register were included in the study from 1 January 2004, when turning 40 years old or when moving to the Central Denmark Region, whichever occurred last. Individuals were censored at the end of the study period, upon death, when moving out of the Central Denmark Region or when leaving the list system, whichever occurred first. We included only general practices and their patients if the practice had at least one PSA test registered in LABKA at one of the included hospitals during the study period (Aarhus, Horsens, Holstebro, Herning, Viborg and Skive). Furthermore, we excluded practices and patients if the practice commenced or ended its activities during the study period. Exposure We calculated the overall PSA test rate for each general practice in the study period as the number of PSA tests ordered in the practice on men free of prostate cancer divided by the sum of person years contributed by all eligible prostate cancer‐free men listed in the practice adjusted for relevant confounders. Follow‐up time was calculated from study entry to the point of prostate cancer diagnosis, leaving the region, leaving the list system, death or end of study, whichever came first. 30 Materials and Methods
Based on the PSA test rates, practices were ranked and divided into quartiles. Consequently, an incident prostate cancer patient was exposed to the quartile in which his practice population was categorised. Outcomes Each of the following outcomes was compared in the four quartiles of practice populations: Mean PSA value. The value of each PSA test was available in μg/L from LABKA. Age at diagnosis. Information on age at diagnosis was collected from the Danish Cancer Register. The incidence of trans‐rectal ultrasound of the prostate (TRUSP) and biopsy of the prostate. These diagnostic procedures were identified in the Danish National Patient Register based on register‐specific procedure codes. TRUSP procedures were identified as UXUD92, and biopsies were identified using the codes KTKE00, KTKE05, KTKE10 and KKEB00. As mentioned in the description of the Danish Cancer Register, the date of diagnosis is often the first date of the hospital contact in which the diagnosis is made; hence, we also included TRUSP and biopsies performed within 14 days after the date of diagnosis as this was expected to be a pre‐diagnostic procedure. The incidence of prostate cancer and the incidence of the different disease stages at diagnosis. Stage at diagnosis was categorised based on the TNM classification from The Danish Cancer Register using the clinical categories local, regional, distant and unknown as proposed by Nguyen‐Nielsen et al. (Table 2.3).150 31 Clinical Activity in General Practice and Cancer
Table 2.3: Algorithm for prostate cancer staging according to the TNM classification in the Danish Cancer Register150 Stage TNM codes Localised T1–4,x N0 M0 T1–2 N0 Mx T1–2 Nx M0,x Regional T1–4,x N1, M0 Distant T1–4,x N1, M1 T1–4,x N0 M1 T1–4,x Nx M1 Unknown T3–4,x Nx M0,x T3–4,x N0 Mx T1‐4,x N1 Mx Note: Tis, Ta, N2 and N3 are not used for prostate cancer, and such registrations were categorised as “unknown.” Cases registered with “T0” were also categorised as “unknown”. The incidence of radical prostatectomy and radiotherapy. Radical prostatectomy was identified in the Danish National Patient Register using the codes KKEC00‐
KKEC01C. We only included primary radiotherapy, i.e. patients who received radiotherapy within 6 months after diagnosis and who had not been treated with prostatectomy (codes BWGC1, BWGC4, BWGC4A, BWGC5 and BWGC5A). Relative survival is a net survival measure which measures the survival of prostate cancer patients relative to the survival expected had they been subject to the background mortality given the same demographic factors (year of birth, gender and calendar time). Information on the general population mortality was collected from national life tables available from Statistics Denmark. Mortality was calculated as prostate cancer‐specific and all‐cause deaths in the entire study population per person year. Prostate cancer‐specific mortality was included to complement the survival analysis because relative survival is highly influenced by lead‐time bias when investigating prostate cancer and PSA 32 Materials and Methods
testing. Calculation of the prostate cancer‐specific mortality circumvents this problem as it does not include “survival time”, but is calculated as the incidence of death from the relevant disease. Information on date of death was retrieved from the Civil Registration System, and information on the cause of death was retrieved from the Danish Register of Causes of Death.146 Statistical analyses The analyses consisted of two main steps; 1) categorisation of general practices and their patients into four groups (quartiles) based on their adjusted PSA test rates, 2) computation of associations between PSA test rate groups and outcomes. First step: PSA test rates at practice level were calculated using a Poisson regression adjusting for relevant confounders chosen a priori based on existing literature and discussions in the author group. We included adjustment for calendar year in 2‐year periods (2004‐2005, 2006‐2007, 2008‐2009), age in 10‐year age groups (40‐49, 50‐59… 90‐99, >100), educational level (<10 years, 10 ‐ 12 years, > 12 years), ethnic origin (Danish, western immigrant, and non‐western immigrant) and marital status (married/cohabitating and living alone). The general practices were ranked according to this adjusted PSA test rate and categorised into four groups. Second step: We compared age at prostate cancer diagnosis and mean PSA values using one‐way analysis of variance. Incidence rate ratios in the period 2004‐2009 of TRUSP and biopsy of the prostate, prostate cancer, each disease stage at diagnosis, prostatectomy, radiotherapy and mortality among all included men were calculated using Poisson regressions with person years in the four groups as the denominator. The practice group with least testing (group 1) was the reference group. Adjusted models included calendar year, age group, educational level, ethnic origin and marital status as covariates in the same categories as in the first step. Robust variance estimation was performed to 33 Clinical Activity in General Practice and Cancer
account for possible cluster effects at practice level151. Relative survival was calculated by using the STATA procedure provided by Dickman (Ederer II method)152. Additionally, the relative survival was age‐standardised using the International Cancer Survival Standard (ICSS) weight distribution slightly modified to our age groups (40‐49 years: weight 0.09, 50‐59 years: 0.17, 60‐69 years: 0.27, 70‐79 years: 0.29, 80 years+: 0.15)153. To be able to adjust survival models for additional confounders, the relative survival can be modelled using excess mortality rates in a Poisson model. This model was adjusted for calendar year, age group, educational level, ethnic origin and marital status. For survival and mortality, the included men were followed until 1 January 2012. 34 Materials and Methods
METHODS STUDY III Study design We conducted an observational population‐based ecological study with individual‐level analyses to investigate the association between the propensity to refer for lower endoscopies in general practice and colorectal cancer‐related outcomes. Study participants The study built on a two‐step approach based on two study periods. The first study period (step 1) was from 1 January 2002 to 31 December 2010 and was used to estimate the propensity to refer for lower endoscopies (exposure). The second period (step 2) in which outcomes were investigated ran from 1 January 2004 to 31 December 2011. In both cohorts, we included all Danish citizens aged 40 or older who were listed with a general practice. Persons with a colorectal cancer diagnosis before their study entry were excluded. Persons had to be living in Denmark at study start. Persons entered at study start or when turning 40 years, whichever came last. Risk time was calculated as the time from entering the study until a colorectal cancer diagnosis, a diagnosis of chronic inflammatory bowel disease (only step 1), death, leaving the list system, emigration or end of study, whichever came first. Exposure (step 1) Compared to Study II, the calculation of the exposure was slightly modified as the temporal consistency of the practices was lower for endoscopies than for PSA tests. This means that the Spearman correlation between e.g. 2006/2007 quartiles and 2008/2009 quartiles was 0.58 (corresponding figure for PSA quartiles was 0.75). Consequently, we aimed to make an exposure that was more time‐dependent, allowing general practices to change behaviour. General practices were therefore characterised based on their propensity to refer for 35 Clinical Activity in General Practice and Cancer
lower endoscopies defined as their endoscopy rates in each of the preceding overlapping 2‐year periods (2002‐2003, 2003‐2004…2009‐2010). Practices were ranked and divided into quartiles based on this propensity. These quartiles were used as the exposure in the following year, e.g. the quartile based on the 2002/2003‐rate in a practice comprised the exposure for its patients in 2004, which ensured exposure before outcome (Figure 2 in Paper III). In the calculation of lower endoscopies rates, we included sigmoidoscopies and colonoscopies performed at public hospitals, private hospitals and at private, practicing gastroenterologists. Only outpatient lower endoscopies performed after a referral from general practice were included. Because the dates of diagnosis in the Danish Cancer Register are generated as explained above, we included GP‐ordered endoscopies up to 30 days after the diagnosis, but only if registered at the same contact as the first appearance of the cancer diagnosis (n= 13,292 scopies (6197 after GP referral)). Outcomes (step 2) In the main analyses, the following outcomes were compared between the quartiles of practice populations: Colorectal cancer incidence. Colorectal cancers were identified in the Danish Cancer Register using the ICD‐10 codes C18‐20. Disease stage at diagnosis. Information on stages at diagnosis was collected from the Danish Colorectal Cancer Group database and divided into four groups based on UICC classification (Table 2.4). The database also contained information on type of cancer (colon or rectal) used in the stratified analyses. We compared the proportions of patients diagnosed in stage I and II together (early) and stage I‐III (curable) combined. In a subanalysis, we compared the proportions of stage III among the curable (stage I, II and III). This was included 36 Materials and Methods
only as a subanalysis as it ignores the proportions of stage IV cancers and demands careful interpretation. Table 2.4: The association between UICC and TNM staging for colorectal cancer149 TNM UICC T1‐2, N0, M0 I T3‐4, N0, M0 II Tany, N1‐2, M0 III Tany, Nany, M1 IV Elective surgery and treatment with curative intent. Information on surgery and related details was available from the Danish Colorectal Cancer Group database. To increase statistical power, we also constructed a composite measure of poor‐
prognosis colorectal cancer for persons experiencing acute surgery, no surgery at all, or stage IV disease at diagnosis. Statistical analyses In step 1 (exposure), we calculated the number of expected lower endoscopies for each general practice population in overlapping 2‐year periods (2002‐03, 2003‐04, 2004‐05 …2009‐10) using a Poisson regression model which included the following potential predictors of endoscopy use: sex, educational level (<10 years, 10 ‐ 12 years, > 12 years), ethnic origin (Danish, western immigrant, and non‐western immigrant), marital status (married/cohabitating or living alone) and Charlson comorbidity index (0, 1‐2, ≥3). Age was included in the analysis and modelled using restricted cubic splines with 6 knots according to Harrell’s recommended percentiles154. For each practice, an observed/expected ratio was calculated by dividing the observed number of endoscopies by the expected number. Practices were ranked and divided into quartiles based on this ratio in each 2‐year period. Consequently, practices were excluded if they existed for less than two consecutive years (n=181) or had discontinuous observation time (n=3). Furthermore, we excluded practices with less than 100 person years of 37 Clinical Activity in General Practice and Cancer
observation time (n=8). A total of 2564 general practices were included in the analysis. In step 2 (outcome), age at diagnosis in the four quartiles was compared using one‐way analysis of variance. To further compare the four quartiles, incidence rate ratios of colorectal cancer diagnoses were calculated using Poisson regressions with person years as the denominator. Differences in the proportion of patients with each disease stage at diagnosis, the proportion receiving elective surgery, the proportion treated with curative intent and the proportion with poor‐prognosis colorectal cancer were compared in the four quartiles using logistic regression. The practices with the lowest propensity to refer represent the reference group (quartile 1). Adjusted models included calendar year (2004, 2005…2011), gender, educational level (<10 years, 10 ‐ 12 years, > 12 years), ethnic origin (Danish, western immigrant, and non‐western immigrant), marital status (married/cohabitating or living alone) and Charlson comorbidity index (0, 1‐2, ≥3). Age was included in the analysis and modelled using restricted cubic splines with 6 knots according to Harrell’s recommended percentiles154. Robust variance estimation was performed to account for possible cluster effects at practice level. All analyses in step 2 were stratified according to type of cancer (rectal or colon). Lower endoscopies and mortality Ultimately, the goal is to reduce mortality from colorectal cancer, and we wished to include analyses on mortality as well, but the results were probably unreliable. We found a dose‐response association between more endoscopies and lower overall mortality. As no screening program offering lower endoscopies has decreased the overall mortality (only the cause‐specific), we believe that these results are biased by residual confounding. We found this dose‐response association even though the analyses were adjusted for all possible confounders from Table 1 in Paper III. Inclusion of additional 38 Materials and Methods
information on income did not change the findings. Therefore, we believe that the differences in Table 1 in Paper III cover other unknown factors that are strongly related to overall mortality. Furthermore, inclusion of Charlson comorbidity index amplified the association which could indicate that general practices referring more patients to endoscopies are referring more patients in general. This could increase the patients’ risk of being diagnosed with a disease included in Charlson comorbidity index, maybe even at a lower morbidity level. As we excluded these analyses from Paper III, this issue will not be further discussed, but the challenges associated with the use of the Charlson comorbidity index should be considered in future studies. ETHICS AND APPROVALS Paper I The project was approved by the Danish Data Protection Agency (J no. 2008‐41‐
2195 and J.no. 2009‐41‐3471) and by the Danish Health and Medicines Agency (J.no. 7‐604‐ 04‐2/49/EHE). According to Danish law, approval from the National Committee on Health Research Ethics was not required as no biomedical intervention was performed. Paper II and III These projects were approved by the Danish Data Protection Agency (J. no. 2009‐41‐3471). The approval obtained applies to the entire database on the Research Centre for Cancer Diagnosis in Primary Care (CaP). Approval from the National Committee on Health Research Ethics was not required as de‐
identified register data were used for analyses. 39 Clinical Activity in General Practice and Cancer
40 Results
CHAPTER 3 RESULTS 41 Clinical Activity in General Practice and Cancer
The following chapter will present the results of the three studies. The chapter is devoted to the main results and any supplemental results not included in the papers. For further details, please see the individual papers. STUDY I Prevalence of suspicion In Study I, we included 4518 face‐to‐face contacts with patients aged 18 years or more from the KOS 2008 survey. The GPs had a suspicion of cancer or another serious disease in 256 (5.7%, 95% confidence interval [CI]: 5.0‐6.4) of the consultations. In the adjusted analyses, we saw that a suspicion was statistically significantly associated with increasing age with a prevalence ratio of suspicion of 3.07 (CI: 1.98‐4.76) among patients above 70 years compared with patients aged 18‐39 years. Actions following the consultation Overall, 52% of the patients whom the GP suspected had cancer or serious disease were referred after the registered consultation, especially to outpatient clinics (16%) and diagnostic imaging (17%). The prevalence ratio of referral for diagnostic imaging among patients whom the GP suspected had cancer or serious disease was 3.95 (CI: 2.80‐5.57) compared with patients in whom the GP had no suspicion, and the overall prevalence ratio for referral was 2.56 (CI: 2.22‐
2.96). Patients in whom the GP suspected cancer or another serious disease were more often tested in the clinic than other patients (prevalence ratio= 1.29, CI: 1.16‐
1.44), and they were more often scheduled for follow‐up (prevalence ratio= 1.15, CI: 1.05‐1.26). Forty‐three (17%) of the patients in whom the GP suspected cancer or serious disease had no scheduled follow‐up, but 33 of these patients were referred for diagnostic work‐up. Overall, 8 (3%) of the 256 patients suspected to harbour serious disease or cancer had no scheduled follow‐up, and they were 42 Results
not referred or tested in the clinic. This indicates that actions were taken in almost all patients suspected of harbouring cancer or serious disease. Subsequent diagnoses Within the first 6 months following the consultation, patients suspected of cancer or serious disease were more likely to be diagnosed with cancer or serious disease than patients not suspected to have cancer or serious disease. The difference was most pronounced within the first 2 months where the hazard ratio of a new serious diagnosis or cancer was 2.98 (CI: 1.93‐4.62). In months 2‐6, the difference was not statistically significant (hazard ratio=1.52, CI: 0.92‐2.53). A corresponding pattern was seen for cancer and serious diseases separately, with hazard ratios within the first 2 months of 7.55 (CI: 2.66‐21.39) for cancer and 2.51 (CI: 1.53‐4.11) for other serious diseases. The proportion of patients suspected of cancer or serious disease who eventually had a diagnosis within the next 2 months (the PPV) was 9.8% (CI: 6.4‐
14.1) for all diagnoses combined, 2.3% (CI: 0.9‐5.0) for cancer and 7.4% (CI: 4.5‐
11.3) for serious diagnoses other than cancer. The corresponding negative predictive values were 97.2% (CI: 96.6‐97.6) for all diagnoses, 99.8% (CI: 99.6‐
99.9) for cancer and 97.4% (CI: 96.9‐97.9) for other serious diseases. Thus, 2.8% of the patients with no suspicion had a serious diagnosis anyway. Inclusion of only definitely serious diseases (see Appendix II) changed the estimates regarding PPV and hazard ratios (Table 3.1). The hazard ratio changed from 2.98 to 4.69, and the PPV decreased from 9.8% to 5.5% for all diagnoses combined. 43 Clinical Activity in General Practice and Cancer
Table 3.1: A comparison of hazard ratios and positive predictive values for new diagnoses in the main analyses and the sensitivity analyses including only definitely serious diagnoses. Combined All serious diagnoses Multivariate PPV hazard ratio (95% CI) (95%CI) Only definitely serious diagnoses Multivariate PPV hazard ratio (95% CI) (95%CI) 0‐2 months 2.98 (1.93‐4.62) 9.8 (6.4‐14.1) 4.69 (2.51‐8.75) 5.5 (3.0‐9.0) 2‐6 months 1.52 (0.92‐2.53) 16.4 (12.1‐21.5) 1.13 (0.45‐2.85) 7.4 (4.5‐11.3) 0‐2 months 7.55 (2.66‐21.39) 2.3 (0.9‐5.0) 8.68 (3.21‐23.52) 2.7 (1.1‐5.6) 2‐6 months 1.82 (0.40‐8.29) 3.1 (1.4‐6.1) 1.77 (0.39‐8.05) 3.5 (1.6‐6.6) Other serious diseases 0‐2 months 2.51 (1.53‐4.11) 7.4 (4.5‐11.3) 3.23 (1.40‐7.48) 2.7 (1.1‐5.6) 2‐6 months 1.49 (0.87‐2.54) 13.3 (9.4‐18.1) 0.91 (0.28‐2‐96) 3.9 (1.9‐7.1) Cancer PPV: Positive predictive value, CI: Confidence interval To investigate whether the subsequent diagnosis was related to the reason for encounter, we compared the ICD‐10 codes from the Danish National Patient Register with the ICPC codes from the survey for each patient who had a new serious diagnosis or cancer. Of the 42 patients who got a diagnosis and in whom the GP had a suspicion of cancer or serious disease, 22 (52%) had a reason for encounter that seemed related to the later diagnosis. The corresponding figure for the non‐suspected patients was 62/279 (22%). Some examples are mentioned in Table 3.2. 44 Results
Table 3.2: Examples of patients in whom the reason for encounter (ICPC) and the subsequent diagnosis (ICD‐10) were related ICPC code K02 Pressure/tightness of heart K89 Transient cerebral ischaemia R02 D21 X08 U08 W03 S09 L86 ICD‐10 code I509 Heart failure, unspecified I649 Stroke, not specified as haemorrhage or infarction Shortness of breath/dyspnoea J969 Respiratory failure, unspecified Swallowing problem K222 Oesophageal obstruction Intermenstrual bleeding C549 Malignant neoplasm of corpus uteri, unspecified Urinary retention C679 Malignant neoplasm of bladder, unspecified Antepartum bleeding O200 Threatened abortion Infected finger/toe M009 Pyogenic arthritis Back syndrome with radiating M511 Lumbar and other pain intervertebral disc disorders with radiculopathy Use of health care services The use of GP, primary‐care specialist and diagnostic imaging increased in the 2‐month period after the index consultation among patients in whom the GP had a suspicion compared with patients in whom the GP had no suspicion. But from 2 to 6 months, there was no difference between the two patient groups. Regarding the use of hospital services (inpatient and outpatient), the difference remained after 2 months. In Paper I, we report the proportion of patients who used the different types of healthcare services. To test the effect of dichotomising the use of general practice (visit/no‐visit), we repeated the analysis using a negative binomial regression model including the number of consultations as a continuous variable. The results from the different multivariable analyses were quite similar, see Table 3.3. 45 Clinical Activity in General Practice and Cancer
Table 3.3: A comparison of the use of general practice calculated as a continuous outcome or a dichotomous outcome Suspicion No suspicion Univariate Multivariate present n= 256
n= 4262 RR (95%CI) RR (95%CI) Count mean(sd) mean(sd) 0‐2 months 1.61 (1.84) 1.33 (1.94) 1.21 (1.04‐1.42) 1.13 (0.96‐1.32) 2‐6 months 1.76 (3.00) 1.61 (2.89) 1.11 (1.01‐1.37) 0.94 (0.77‐1.15) Dichotomised n (%) n (%) 0‐2 months 172 (67.2) 2522 (59.2) 1.13 (1.04‐1.24) 1.14 (1.05‐1.23) 2‐6 months 132 (52.0) 2159 (50.8) 1.03 (0.91‐1.16) 0.98 (0.88‐1.11) RR: Relative risk 46 Results
STUDY II PSA test rates Among the 498 general practices that ordered PSA tests from 2004 to 2009, 368 (74%) were active throughout the entire study and ordered 106,882 PSA tests among 303,098 males. The practices were divided into four groups based on their adjusted PSA test rate. In the four groups of practices, the populations were comparable concerning socio‐demographic composition and Charlson comorbidity index (Table 1 in Paper I); however, the PSA testing rates varied 3.6‐fold between the practices that tested the most (Group 4) compared with practices that tested the least (Group 1) (Figure 3.1). Mean PSA values varied among groups of practices with the highest values in group 1 (5.37 mg/L) and the lowest in group 4 (3.97 mg/L). Considering the 169 single‐handed practices only, a 4.3‐fold variation was observed between the lowest and the highest testing groups. Figure 3.1: 368 general practices in Central Denmark Region ranked according to their PSA test rate in the period 2004‐2009. (Light grey lines indicate 95% confidence intervals)
47 Clinical Activity in General Practice and Cancer
Associations between PSA test rates and outcomes After the publication of Paper II, we received updated files from the Danish National Patient Register which changed the estimates on TRUSPs. In Table 3.4, the updated figures are reported. Patients had more trans‐rectal ultrasounds in group 4 compared with group 1 (incidence rate ratio (IRR): 1.30, CI: 1.11‐1.51) and more prostate biopsies (IRR: 1.76, CI: 1.54‐2.01). In total, 4199 men were diagnosed with prostate cancer during the study period. Patients in group 4 had more cancer diagnoses (IRR: 1.37, CI: 1.23‐1.52). The patients in group 1 were older at diagnosis. Regarding the stage distribution, the incidence of local prostate cancer was higher in group 4 (IRR: 1.61, CI: 1.37‐1.89) than in the other groups, but there was no difference in the incidence of regional and distant disease. Men listed at practices in group 4 were subjected to more treatments such as prostatectomies (IRR: 2.25, CI: 1.72‐2.94) and radiotherapy (IRR: 1.28, CI: 1.02‐
1.62) than men in the practices in group 1. The 1‐year relative survival and excess mortality rate ratios showed no statistically significant difference between the four groups. The 5‐year relative survival was higher in group 4 than in group 1 (83.4% vs 74.2%), which corresponds to a relative excess mortality rate ratio in group 4 of 0.70 (CI: 0.50‐
0.97) compared with group 1. The all‐cause mortality among all the men was identical in the four groups, and there was no statistically significant difference in prostate cancer‐specific mortality (IRR in group 4: 1.11, CI: 0.92‐1.33). All analyses were tested for consistency by looking at the 169 single‐handed practices only, by exclusion of outliers, and by defining the PSA test rates based only on the first PSA test for each man. Neither of these analyses notably altered the results (see Table 3.4) 48 Results
Table 3.4: Selected outcomes with data from sensitivity analyses (solo practices only, PSA test rates based on first PSA tests only, and without outliers) Outcome Quartile 1 Quartile 2 TRUSPa Number (9234) IRR, unadjusted IRR, adjusted IRR, no outliersb IRR, solo only First onlyc 1689
Ref
Ref
Ref
Ref
Ref
2320
1.09 (0.92‐1.29)
1.11 (0.95‐1.31)
1.11 (0.93‐1.32)
1.03 (0.83‐1.28)
1.13 (0.97‐1.32)
1016
Ref
Ref
Ref
Ref
Ref
1446
1.13 (0.99‐1.28)
1.14 (1.01‐1.29)
1.12 (0.98‐1.27)
1.02 (0.81‐1.28)
1.18 (1.04‐1.33)
1943 1.48 (1.30‐1.68) 1.44 (1.28‐1.63) 1.41 (1.24‐1.60) 1.52 (1.23‐1.87) 1.44 (1.27‐1.64) 1968
1.79 (1.56‐2.06)
1.76 (1.54‐2.01)
1.73 (1.49‐2.00)
1.56 (1.25‐1.96)
1.74 (1.53‐1.99)
780
Ref
Ref
Ref
Ref
Ref
1053
1.07 (0.96‐1.19)
1.12 (1.02‐1.25)
1.13 (1.01‐1.25)
0.95 (0.77‐1.17)
1.14 (1.03‐1.26)
1233 1.22 (1.10‐1.37) 1.24 (1.13‐1.37) 1.24 (1.12‐1.38) 1.22 (1.02‐1.47) 1.22 (1.11‐1.35) 1133
1.34 (1.20‐1.50)
1.37 (1.23‐1.52)
1.37 (1.22‐1.54)
1.23 (1.02‐1.48)
1.40 (1.26‐1.54)
206
Ref
Ref
Ref
Ref
Ref
234
0.90 (0.74‐1.09)
0.97 (0.80‐1.17)
1.00 (0.82‐1.22)
0.76 (0.52‐1.11)
0.96 (0.79‐1.17)
237 0.89 (0.73‐1.09) 0.94 (0.78‐1.14) 0.97 (0.79‐1.18) 1.03 (0.70‐1.52) 1.02 (0.85‐1.23) 237
1.07 (0.88‐1.29)
1.11 (0.92‐1.33)
1.23 (1.02‐1.49)
0.83 (0.58‐1.18)
1.03 (0.85‐1.24)
9734
0.92 (0.86‐0.98)
1.01 (0.98‐1.05)
1.02 (0.98‐1.06)
1.02 (0.95‐1.08)
0.98 (0.95‐1.02)
9999 0.92 (0.87‐0.98) 1.00 (0.96‐1.04) 1.00 (0.96‐1.04) 0.98 (0.92‐1.05) 1.01 (0.97‐1.05) 8630
0.95 (0.90‐1.01)
1.01 (0.97‐1.05)
1.01 (0.97‐1.06)
1.00 (0.95‐1.07)
1.00 (0.96‐1.03)
Cancer incidence Number (4199) IRR, unadjusted IRR, adjusted IRR, no outliersb IRR, solo only IRR, first onlyc Mortality from prostate cancer Number (914) IRR, unadjusted IRR, adjusted IRR, no outliersb IRR, solo only IRR, first onlyc Mortality, all cause Number (36,742) IRR, unadjusted IRR, adjusted IRR, no outliersb IRR, solo only IRR, first onlyc Quartile 4 2848 1.30 (1.11‐1.53) 1.29 (1.11‐1.51) 1.28 (1.08‐1.52) 1.32 (1.05‐1.66) 1.30 (1.12‐1.51) Biopsya Number (6373) IRR, unadjusted IRR, adjusted IRR, no outliersb IRR, solo only First onlyc Quartile 3 8379
Ref
Ref
Ref
Ref
Ref
2377
1.30 (1.11‐1.53)
1.30 (1.11‐1.51)
1.30 (1.09‐1.55)
1.34 (1.09‐1.65)
1.43 (1.23‐1.65)
IRR: Incidence rate ratio, TRUSP: Trans‐rectal ultrasound of prostate a Results do not correspond with Paper II as we have received new, updated data b Without the 5% most testing and the 5% least testing practices c The exposure calculation is based only on the first PSA test for each person in the study period 49 Clinical Activity in General Practice and Cancer
STUDY III Lower endoscopy rates The endoscopy rates are depicted in three different ways in Table 3.5. It is seen that the variation was highest between the quartiles when they were based on yearly rates. This variation diminished when the estimates were based on 2‐year rates. As expected, the variation was further reduced when 2‐year‐based quartiles (propensity) were applied as the exposure in the following year. The crude endoscopy rates in 2004‐2011 were 9.9 scopies/1000 person years in the lowest quartile and 20.3 scopies/1000 person years in the highest. This corresponded to an adjusted IRR of lower endoscopies in quartile 4 of 2.02 (CI: 2.00‐2.04). The four populations showed small differences, especially regarding urbanisation (Table 1 in Paper III) indicating geographical differences in supply and use of health care services as also shown by the different proportions of endoscopies performed at private clinics. 50 Results
Table 3.5: Variation in endoscopy referrals among quartiles of practices based on three different calculations Quartile 1 Quartile 2
Quartile 3 Quartile 4
28,522
71,431
99,958 124,246
Crude rate of endoscopies, mean (tests/1000 years) 6.0
11.2
15.6 23.0
IRR, unadjusted Ref
1.87 (1.84‐1.89)
2.60 (2.56‐2.63) 3.83 (3.78‐3.88)
IRR, adjusted Ref
1.85 (1.93‐1.88)
2.56 (2.52‐2.59) 3.76 (3.72‐3.81)
Quartiles based on 1‐year periods 2002‐2010 Number of lower endoscopies (324,157) Quartiles based on 2‐year periods 2002‐2010 Number of lower endoscopies (322,616*) 34,249
72,745
97,145 11,8477
Crude rate of endoscopies, mean (tests/1000 years) 6.9
11.8
15.5 22.0
IRR, unadjusted Ref
1.70 (1.68‐1.72)
2.24 (2.22‐2.27) 3.19 (3.15‐3.23)
IRR, adjusted Ref
1.68 (1.66‐1.70)
2.21 (2.18‐2.23) 3.13 (3.09‐3.16)
Quartiles based on the calculations in step 1 and applied to the following year 2004‐2011 Number of lower endoscopies (301,792) 42,511
72,885
89,885 96,511
Crude rate of endoscopies, mean (tests/1000 years) 9.9
13.4
16.1 20.3
IRR, unadjusted Ref
1.35 (1.34‐1.37)
1.62 (1.61‐1.64) 2.05 (2.02‐2.07)
IRR, adjusted Ref
1.34 (1.33‐1.36)
1.61 (1.59‐1.63) 2.02 (2.00‐2.04)
IRR: Incidence rate ratio *Two‐year periods are based on 2002‐2003, 2004‐2005, 2006‐2007 2008‐2009 and 2010. Some endoscopies are therefore not counted if a practice is not active in both years. This is not the same rates as those used in the exposure calculation as the latter involve overlapping periods (2002‐
2003, 2003‐2004, etc.) Associations between propensity to refer for lower endoscopy and outcomes With quartile 1 as the reference, the adjusted IRR of colorectal cancer was 1.04 (CI: 1.00‐1.07) in quartile 2, 1.03 (CI: 1.00‐1.07) in quartile 3 and 1.04 (CI: 1.00‐
1.08) in quartile 4. Stratification showed that the differences in incidence among the four quartiles were evident for colon cancers only (Table 3 in Paper III). The proportion of cancers diagnosed in early stage (stage I‐II) was 44.0% in quartile 1 and 45.9% in quartile 4, corresponding to an OR in quartile 4 of 1.08 51 Clinical Activity in General Practice and Cancer
(CI: 1.01‐1.16); this was mainly caused by a bigger difference for rectal cancers with an OR in quartile 4 of 1.19 (CI: 1.05‐1.34). Overall, the proportions with stage I‐III disease were almost similar, with an OR of 1.04 (CI: 0.96‐1.13) in quartile 4 compared with quartile 1; this difference was caused mainly by rectal cancers (Tables 2 and 3). We found that the proportion of patients having elective surgery was in general higher in quartile 3 and 4 than in other two quartiles, although not statistically significantly so except for colon cancer in quartile 3 (OR: 1.12, CI: 1.01‐1.24). The proportions treated with curative intent were quite similar across the groups. The proportion of patients having a poor‐prognosis colorectal cancer (acute surgery, no surgery at all, or stage IV disease at diagnosis) was lower in quartile 4 than in the other quartiles, especially for rectal cancer patients, although the difference was not statistically significant; the OR (rectal) in quartile 4 was 0.89 (CI: 0.78‐1.03). 52 Discussion of methods
CHAPTER 4 DISCUSSION OF METHODS 53 Clinical Activity in General Practice and Cancer
This chapter discusses important methodological challenges. All results (Chapter 3) and conclusions (Chapter 6) should be evaluated in the light of the strengths and limitations of each individual study. Finally, the chapter recapitulates some of the main lessons learned concerning studies of variation and outcomes in Danish general practice. DESIGN The studies in this thesis are observational population‐based cohort studies. Observational studies may have different designs, but they all investigate the effects of an exposure without interference from the researcher. In Study I, data from a cross‐sectional survey were combined with register data to enable follow‐up. The study was completed during the course of a full year and comprised more than 400 GPs, which minimises the effects of seasonal and individual GP variation. The survey’s cross‐sectional design was suitable for description of the prevalence of GPs’ suspicion of cancer and other serious diseases following a consultation and for characterisation of the persons whom the GP suspected had serious disease or cancer. However, the cross‐sectional design was less appropriate for studying the consequences of a suspicion for future diagnoses and health care use. To study these issues optimally, information on subsequent consultations and any new suspicion raised had to be obtained. In most of the patients in whom a serious disease was diagnosed during follow‐up, the GP suspected no serious disease after the index consultation. However, no inference can be drawn about the GP’s possible lack of attention at the index consultation because the patients could have consulted later with new symptoms that raised a relevant suspicion. This issue could have been studied in other ways than the one chosen here; for instance by obtaining information by reviewing the GPs’ records or by asking the GPs to make other and more registrations during any subsequent consultations. Both of these 54 Discussion of methods
methods would, however, be time‐consuming, and the latter would impose a substantial workload on the GPs. Moreover, these alternative approaches would compromise an advantage of the current design because they might introduce bias by making GPs pay special attention to serious disease and cancer. This bias is related to the phenomenon called the Hawthorne effect which reflects that people tend to change behaviour just because they are being studied, regardless of the intervention they may receive155. Studies II and III represent another type of observational study that exploits natural variation in GP test rates. In a corresponding randomised, controlled trial, GPs should be randomised to different rates of PSA testing or different endoscopy referral thresholds; yet, such a study would be difficult to implement. The present studies are ecological32, 156 because different levels of exposure are applied to a large population; hence, the studies do not investigate whether a specific individual was exposed or not. However, the studies deviate from the classic ecological design in important ways as we had information about outcomes and possible confounders at the level of the individual patient. The studies could also be characterised as cohort studies with an ecological exposure, but we prefer the term “ecological study with individual‐level analyses”. The design minimised the risk of reversed causality (i.e. that PSA testing or the use of endoscopy causes cancer), which could occur if exposed persons were compared with non‐exposed persons at the individual level (PSA‐
test vs. no PSA‐test or endoscopy vs. no endoscopy) like in a conventional cohort study. An element of reversed causation may, however, be present in Study II; importantly, though, PSA testing does not cause the disease, but causes the disease to be diagnosed. The risk of reversed causality was further reduced by adopting the approach used in Study III where the exposure was defined as the propensity to refer. 55 Clinical Activity in General Practice and Cancer
We used the GPs’ test rates as a proxy for their threshold for testing. Alternatively, to study the effect of referral timeliness, we could have investigated the time from symptom presentation to PSA test or endoscopy for each individual person. However, this approach would introduce confounding by severity157, and a quicker referral may be associated with, e.g., poorer stage at diagnosis. This corresponds to the “waiting time paradox”158 where urgently referred cancer patients have a higher mortality, probably because advanced disease comes with more evident symptoms. This problem could have been minimised by calculating the “mean GP interval” for each general practice instead of intervals at the individual level, but this was not possible as we have no access to information on the date of symptom presentation. The methods in Studies II and III invite a discussion whether instrumental variable analysis could have been used32. However, the use of an instrumental variable analysis would express, e.g., the effect of a colonoscopy or not, which was not our aim159. We aimed at evaluating the effects of different levels of exposure as markers of different ways to deliver healthcare. Our study design assumes that men’s choice of practice is largely independent of the practice’s PSA test rate or endoscopy rate as no such information is publicly available in Denmark. The present studies are therefore to some extent similar to natural experiments. In Study III, we defined the propensity to refer to lower endoscopy as the referral rate in the preceding 2 years. The 2‐year referral rate varied more than three‐fold between the most and the least testing quartiles of practices as illustrated in Chapter 3. As these quartiles were applied to the following year in Study III, the variation was reduced to a two‐fold variation between quartiles. We do, however, consider this to be a strength as it seems that our measure represents an underlying propensity. If the variation had not reflected an 56 Discussion of methods
inherent propensity but was caused by chance alone, the variation in the referral rates would have been non‐existing in the subsequent year. BIAS, STATISTICAL METHODS AND GENERALISABILITY Selection bias A general strength of using national population‐based registries in research is that selection bias is limited when study subjects are included. However, important aspects of selection bias deserve to be mentioned. In Study I, the results may be affected by selection bias because 53.6% of the invited GPs chose not to participate. Details on the non‐participating GPs have been published previously142. There were no statistically significant differences between the participating GPs and all GPs in the Central Denmark Region with regard to the type of practice and the number of listed patients. The age and gender distributions of the listed patients of participating GPs and the patients of the whole group of GPs were similar. Female GPs and GPs with less than 20 years of practice tenure were overrepresented in the group of participating GPs. This may indicate that non‐participation was not entirely random, and this could potentially have influenced the results. However, we find it unlikely that non‐participation would be strongly related to the prevalence of suspicion or the predictive value of the GPs’ suspicion. Selection bias could emerge from the fact that some GPs did not register the patients’ CPR number. This was the case in 525 of the consultations (10.4%), and these patients were therefore not included in the study. The GPs explained that the omissions were a consequence of their own principles of confidentiality, and this might have introduced selection bias. However, a suspicion was present in 5.6% of the cases with missing CPR number and, hence, it seems unlikely that it could bias the results. 57 Clinical Activity in General Practice and Cancer
Studies II and III used register‐based data only and therefore enjoyed the strength of the completeness of the Danish population registers with registration of all citizens with full follow‐up. In Study II, we only included general practices that were active during the entire period from 2004 to 2009. This could introduce selection bias as a practice closing e.g. due to GP retirement could possibly be serving a practice population with an overrepresentation of elderly citizens. Therefore, this exclusion was not entirely random; nonetheless, inclusion of all practices in the analyses did not alter the conclusions. Information bias The Achilles’ heel of register‐based research is the data quality. Data are often collected for administrative, reimbursement or demographic purposes; and collection procedures, coverage, quality and classifications may change over time and may not fulfil the needs of specific research projects. Conversely, it is a strength of the present studies that data collection is unrelated to the research projects which reduces the risk of differential misclassification. An additional strength was the prospective nature of the registrations which eradicates recall bias. Nevertheless, information bias may occur if information used in the study is erroneous due to systematic measurement error or misclassification of subjects32. Data linkage between the Danish registers is possible as all registers are based on the Danish citizens’ unique CPR number. The quality of registration varies across the different registers and among different variables within each register. As the studies of this thesis are largely register‐based, I feel compelled to discuss some of the key variables in more detail in order to clarify the choices, considerations and potential shortcomings in relation to each variable. But first I will focus on possible information bias in the survey data. 58 Discussion of methods
Study I - The KOS 2008 survey
The two main issues of information bias in the KOS 2008 survey concern the question of suspicion and the definition of serious disease. The question “Are you left with the slightest suspicion of cancer or another serious disease (new)?” was developed ad hoc for the purpose of the KOS 2008 survey and has not been validated and especially content validity is important160. GPs may differ in their understanding of the word “suspicion”, which could mean that different thresholds were applied when GPs registered a “suspicion”. This was partly confirmed by the broad prevalence range of GPs’ suspicion (0% to 33% of the consultations) (data not shown). Misclassification of patients may have taken place if some GPs were unaware of their suspicion, but instinctively reacted as if a suspicion was present. In contrast, it seems unlikely that a GP who was aware of a suspicion deliberately indicated “no suspicion”. Consequently, assuming that a suspicion led to referral and not vice versa, such misclassification would be non‐differential and this would result in an underestimation of the investigated associations. Furthermore, it would cause an underestimation of the true prevalence of suspicion. Nevertheless, it is possible that the misclassification might be differential if ticking “referral” increases the risk of ticking “suspicion”, but this risk is expected to be negligible compared with the opposite situation. The outcomes ‘risk of new diagnoses’ and ‘use of healthcare services’ were assessed after the exposure, and misclassification concerning these outcomes is therefore most likely non‐differential. Similarly, the definition of serious disease could vary among GPs which would influence the prevalence of the suspicion. To avoid this situation, we could have provided the GPs with a pre‐specified list of diagnoses to be considered serious. This would, however, not portray the prevalence of the GPs’ subjective feeling that “this could be serious”. In general practice, you are more often left with the 59 Clinical Activity in General Practice and Cancer
vague feeling that “something is wrong” than with a suspicion of the presence of a specific disease161, 162, and our approach may therefore be the more suitable for the present purpose. The lack of definition of serious disease could impact both the hazard ratios and the PPVs. To estimate the effect of this lack of definition, we performed sensitivity analyses including only definitely serious diagnoses, which revealed a PPV that remained above 5% and an increased hazard ratio. The absence of a definition of serious disease and possible disagreements among GPs therefore seemed not to substantially affect our conclusions. Another source of information bias was missing data. In 191 (4.2%) of the contacts, the GPs did not indicate whether a suspicion was present or not. We expected that the missing values would be associated with “no suspicion”, which was supported by the similarity between the socio‐demographic data of the cases for whom data were missing and the data of those who belonged to the non‐suspected group (data not shown). In our main analyses, these ‘missings’ were therefore included in the non‐suspected group. Exclusion of these 191 patients from the analyses did not alter the results. Overall, we have no reason to believe that the results in Study I were compromised by important information bias and, if anything, the bias would most likely be towards null. Study II - PSA tests and diagnostic procedures
The clinical laboratory information system, LABKA, holds high‐quality clinical data on blood tests in the Central Denmark Region145, but the data on the PSA tests have never been validated. However, the validity and completeness of the PSA test data are likely to be high. The proportion of missing data in LABKA is expected to be very low because the LABKA system functions as a crucial daily routine diagnostic tool for the medical personnel and because it is based on instantaneous entry of approved results into the system. Furthermore, we 60 Discussion of methods
believe that the PPVs of registrations in LABKA are very high. Any misclassification of GPs based on missing or erroneous data in LABKA seems doubtful and, if present, such misclassification would most likely not cause the variation and effects seen in Study II. In contrast, a registration problem may exist regarding diagnostic procedures. In Table 2 in Paper II, there is a lack of correspondence between the number of TRUSPs and the number of biopsies. At the included urology departments, biopsies should not be performed without a TRUSP (personal correspondence with Michael Borre, professor at one of the two hospitals performing the procedures). The higher number of biopsies than of TRUSPs in Group 4 therefore indicates a registration inaccuracy. However, after publication of the study, we have received updated files concerning biopsies and TRUSPs. Using these updated data led to the results shown in Table 3.4 in Chapter 3 where the number of TRUSPs is higher than the number of biopsies in all four groups. The editor of International Journal of Cancer has been informed about this finding. Study II – The Register of Causes of Death
In Study II, we used information on the cause of death for prostate cancer patients. The validity of the cause of death relies on the quality of the registrations of the doctors who complete the death certificates. As the autopsy rate in Denmark is low (<10%), Danish mortality statistics are not regularly validated; hence, the presence of important data discontinuity could invalidate specific research projects. A Danish study validated acute myocardial infarction in Danish mortality statistics and found a relatively high sensitivity compared with clinical records163, but prostate cancer recorded in the Register of Causes of Death has not been validated. Our results could be biased as more PSA tests led to more prostate cancer diagnoses, which may have increased the risk of prostate cancer being registered as the cause of death. However, it is a strength 61 Clinical Activity in General Practice and Cancer
that registrations of cause of death are done independently of the exposure and the study. Study III - Sigmoidoscopies and colonoscopies
Payment to the public and private hospitals has been based on diagnosis‐related groups (DRG) since 2000, and the registrations from the hospitals are ultimately registered in the Danish National Patient Register. Previous studies validating this register have focused mainly on specific diagnoses, and they have shown a high completeness and validity, often exceeding 90%, with some variation between studies164‐172. One study, however, found a PPV of only 32% for vitamin B12 deficiency anaemia,173 which underpins the importance of validation of the codes used in each study. The administrative validity is believed to be high regarding dates of hospitalisation and discharge etc. Diagnostic and surgical procedures (like lower endoscopies) need to be coded to obtain reimbursement, and the completeness of these registrations is hence also believed to be high; however, a high completeness does not equate a high validity. The hospitals are only reimbursed for the registration of the actual colonoscopy performed and not for correct registration of other information (e.g. data on referrals) We included lower endoscopies registered as outpatient procedures and performed after referral from a GP. To test whether regional differences existed in the coding of this additional information (origin of referral), we compared the total endoscopy rates and the GP‐ordered outpatient‐endoscopy rates between the five administrative Regions (Table 4.1). This comparison showed a consistent pattern. We also investigated the variation within the regions and over time. These variations were of a similar magnitude, which indicates that apart from a small decrease in variation over time (Table 4.2), the overall differences do not conceal large differences within the regions or across time trends. Throughout the study period, a consistent proportion of all endoscopies comprised 62 Discussion of methods
outpatient endoscopies that were performed after a GP‐referral (data not shown). A systematic validation would be the best way to investigate the quality of the registrations. To calculate the PPVs of registrations in the Danish National Patient Register would require a review of a number of patient records at all the hospitals performing these procedures. To estimate the completeness would be an even greater task that would demand a review of thousands of medical records to find lower endoscopies and to identify whether each procedure was recorded in the Register. This effort seems disproportionate to the information gained as the internal validation demonstrated considerable consistency (Tables 4.1 and 4.2). Table 4.1: Rates of lower endoscopies in the five administrative Regions in Denmark in 2004‐2011 Northern Denmark Region Central Denmark Region Region of Southern Denmark Capital Region of Denmark Region Zealand GP‐ordered outpatient All Ref 0.92 (0.90‐0.93) 1.07 (1.06‐1.09) 1.06 (1.04‐1.07) 1.02 (1.01‐1.04) Ref 0.95 (0.94‐0.96) 1.06 (1.05‐1.07) 1.05 (1.04‐1.06) 1.01 (1.00‐1.02) Table 4.2: Examples of variation in lower endoscopy rates over time and within regions (crude) IRR of lower endoscopies 2004 2007 2011 Northern Denmark Region Central Denmark Region Region of Southern Denmark Capital Region of Denmark Region Zealand Q1 Ref Quartile 2 1.49 (1.44‐1.54) Quartile 3 1.85 (1.79‐1.92) Quartile 4 2.47 (2.38‐2.55) Ref Ref Ref Ref Ref Ref Ref 1.29 (1.25‐1.34) 1.34 (1.29‐1.38) 1.32 (1.27‐1.36) 1.28 (1.25‐1.31) 1.28 (1.25‐1.32) 1.49 (1.45‐1.53) 1.38 (1.34‐1.43) 1.54 (1.49‐1.59) 1.59 (1.54‐1.65) 1.57 (1.52‐1.62) 1.52 (1.49‐1.56) 1.54 (1.50‐1.58) 1.76 (1.72‐1.81) 1.62 (1.58‐1.66) 1.96 (1.90‐2.02) 1.96 (1.90‐2.03) 1.97 (1.90‐2.03) 1.87 (1.82‐1.92) 1.95 (1.90‐2.00) 2.25 (2.20‐2.30) 1.95 (1.90‐2.01) During 2005 and 2006, the former counties of Vejle and Copenhagen conducted a randomised trial inviting 50% of persons aged 50‐74 years to be screened with 63 Clinical Activity in General Practice and Cancer
a faecal occult blood test. According to the protocol, a positive faecal occult blood test led to a colonoscopy; and to exclude any misclassification of GPs caused by these trials, we performed sub‐analyses without persons eligible for the screening trial according to their age and municipality. Furthermore, entire general practices were excluded if more than 50 patients in the practice were eligible for the trial. These analyses supported the main results. Overall, we believe that the completeness and validity of the included endoscopies in Study III are good, which is also substantiated by the consistency of our findings when including not only GP‐ordered but all registered lower endoscopies. Furthermore, the quality of the registrations is most likely unrelated to the outcomes investigated in our study. Study II and III - The Danish Cancer Register
The validity of the Danish Cancer Register is ascertained through quality control routines applied in the daily data production and in the completion of the yearly publications130. The completeness is ensured by the automated cancer logic using various sources of information131 (see Chapter 2). In Study II, prostate cancers were characterised by their stage at diagnosis based on the TNM classification, and they were categorised according to the clinical groups described by Nguyen‐Nielsen et al150. The overall completeness of T, N and M was only 34% from 2004‐2009, but use of the clinical categories decreased the percentage of patients with unknown TNM stage to 24% in our study. It seems implausible that missing registrations of stage at diagnosis should be associated with the GP’s PSA test rate and, correspondingly, the proportion with missing information was distributed equally among the four groups (Table 2 in Paper II). In Study III, we used the Danish Cancer Register to identify previous and incident colorectal cancer cases and therefore the results rely on the completeness of this Register which is known to be high. Information on stage, 64 Discussion of methods
location and treatment was retrieved from the Danish Colorectal Cancer Group database. Study III - The Danish Colorectal Cancer Group database
The comparison of stage distribution, the proportion receiving elective surgery and treatment with curative intent in Study III, relied on the validity of the Danish Colorectal Cancer Group database. The database is validated on a regular basis by linkage to the Danish National Patient Register and the Danish Cancer Register. The completeness of patient registration is estimated annually and has been close to 95% since 2002149, which corroborates our findings. In a random sample of 86 patients selected from May 2001 to December 2002, the validity of data on treatment was 94%, and on postoperative data it was 90%120 in a comparison of two independent registrations of the 86 patients. We found that information on stage was missing for 6% and information on surgery was missing for 3% of the patients in the database. Further details on surgery (elective, curative intent) were missing for 11% of those who underwent surgery. The missing information was almost equally distributed across the four groups, except from surgery which was missing more frequently in quartile 1 (3.5%) than in quartile 4 (2.2%). In the unlikely event that these extra ‘missings’ in quartile 1 all received elective surgery, our result might have been biased towards an inflation of the associations. However, the differences would be small, and our conclusions would still be that the effect of more endoscopies on surgery was modest, if evident at all. Study II and III - The Provider Number and Patient Lists
The calculations of risk time and hence the classification of general practices in Studies II and III rely on information from the patient lists. The register on patient lists is not a validated database, but it forms the basis for remuneration of GPs, and its validity is therefore assumed to be high. The patient lists are used for payment of the per‐capita fee for GPs18 (Chapter 1). 65 Clinical Activity in General Practice and Cancer
A limitation of using the provider number is that it covers a variable number of GPs. This could potentially conceal greater variations than those reported here as differences between GPs within a practice can be masked by aggregation at the practice level. In both Study II and Study III, we made sub‐analyses restricted to single‐handed practices, and the results of these analyses did not differ from the main results. In conclusion, the studies of this thesis are based on high‐quality registers and are largely unbiased; and any existing bias is believed to cause an underestimation of the reported associations. Confounding A possible drawback of using registers is the limited availability of data on important confounding factors; however, we had information on age, gender, education, cohabitation status, income and comorbidity. This information was included in multivariable regression models to ensure that any findings could not be explained by these factors. To further minimise the risk of confounding, we performed stratified sub‐analyses and restricted our study populations to certain age groups. Nevertheless, the observational design carries an inherent risk of residual confounding. Residual confounding exists when possible confounders are not accounted for (e.g. lifestyle factors) or when included confounders are measured inaccurately or categorised improperly. To account for comorbidity in Study III, we used the Charlson comorbidity index. This index is based on an algorithm first published in 1987127. Originally, it was developed on 604 patients admitted to medical service and applied to 685 female breast cancer patients to validate the ability of the index to predict 1‐year survival. During the following decades, it was applied to several other study designs, settings, patients and outcomes; and though it has basically remained 66 Discussion of methods
unchanged since 1987, its ability to predict various outcomes has proven to be among the best indexes 174‐177
. Still, the appropriateness of the Charlson comorbidity index related to our exposure calculation and outcomes is debatable, and we cannot rule out residual confounding, e.g. because the index includes only hospital‐based diagnoses. A risk in ecological studies like Studies II and III is that of ecological fallacy32, 155. This fallacy implies that any associations found may be confounded and not be causal in nature. It is possible that the populations differed in other respects than PSA test rates or lower endoscopy rates, which could have confounded the results. Furthermore, the GPs’ different PSA test rates and endoscopy rates may conceal different clinical behaviours (use of digital rectal examination, other blood samples, referrals in general, etc.). In Study II, the baseline characteristics were similar (Table 1 in Paper II), and causality is supported by the echoing of the results of previous studies. In contrast, the practice populations were more diverse in Study III (Table 1 in Paper III). This indicates that the adjustments in step 1 did not generate comparable populations; however, we did not want to adjust for urbanisation, for instance, as distance to hospital does not justify differences in the GPs’ symptom appraisal. Therefore, adjustment of urbanisation in step 1 could have concealed some of the associations of interest. Observational studies may suffer from confounding by indication32, 157
, especially in a traditional cohort design, as patients later diagnosed with colorectal cancer are likely to be facing a higher risk of having a lower endoscopy than other persons. But by using the variation in propensity to refer to lower endoscopy as the exposure, we exploit that the differences in patient characteristics seen in Table 1 in Paper III do not warrant a two‐fold variation in the use of lower endoscopies. Therefore, patients in group 1 and group 4 seem to be exposed to different behaviours in respect to endoscopy use, which dilutes 67 Clinical Activity in General Practice and Cancer
the differences at patient level and minimises the risk of confounding by indication. Furthermore, encountering patients with late‐stage cancer may affect the GP’s test rate and lead to reversed causality. However, regarding PSA tests, it seems unlikely that the rate is strongly influenced by previous late‐stage prostate cancers as the overall PSA test rate is approximately 25 times higher than the cancer incidence. In contrast, the number of endoscopies per colorectal cancer is lower, and the GPs’ endoscopy rates may hence more easily be influenced by recently diagnosed cancers. The two‐year period approach in Study III reduces this risk by ensuring that the exposure is calculated before each cancer patient is diagnosed. Statistical methods and statistical precision The use of national, population‐based registers strengthens the statistical precision as indicated by narrow confidence intervals in our main analyses. The narrow confidence intervals minimise the risk of falsely accepting the null hypothesis (type II error). However, this invites careful interpretation of the results regarding clinical relevance. The 95% confidence interval was chosen to indicate statistical significance. In Studies II and III, we used a Poisson model to generate adjusted rates at the general practice level. Data on PSA tests and lower endoscopies are non‐
normally distributed. A Poisson model handles discrete variables and is able to process different time scales (calendar time, age and time from diagnosis). Other models are also able to manage this (stratified Cox, negative binomial), but these models would be even more time‐consuming. In step 2 in Studies II and III, we divided the general practices into quartiles. This categorisation reduces the statistical power compared with analysis of a continuous exposure178. Nevertheless, we chose this approach to obtain a more 68 Discussion of methods
communicable result and a result that appears more relevant from a health services research perspective. We repeated the analyses in Study III with general practices divided into deciles to investigate whether the quartiles covered greater underlying differences. Even though the IRR of lower endoscopies in decile 10 was 3.00 (2.94‐3.06), the findings of these analyses did not affect our main conclusions (data not shown). In all studies, we accounted for clustering at the general practice level as patients within a practice cannot be assumed to be entirely independent179. We adjusted for clustering by using robust variance estimation151. This does not change the estimates, but it broadens the confidence intervals. External validity The GPs participating in Study I were almost comparable to the non‐
participating GPs, and the study period covered an entire year and comprised more than 400 GPs. We therefore expect the findings to be generalisable to the rest of the Central Denmark Region and to the entire country. Extrapolation of the study results to other countries requires careful consideration of differences in healthcare systems and education of GPs. Furthermore, differences in populations, traditions, symptom interpretation and accessibility of diagnostic investigations may also influence the generalisability. The large variation found in Study II is probably present in many healthcare systems, and the downstream effects of PSA testing are also universal. We expect the findings to be generalisable to all of Denmark; however, there may be differences in the overall level of PSA testing in different countries that compromise the generalisability across borders. Extrapolation of the findings in the nation‐wide Study III to other countries demands the presence of similar contextual conditions and healthcare structures. Differences in access to endoscopy services and overall endoscopy 69 Clinical Activity in General Practice and Cancer
rates may influence the possible associations between endoscopy rates and outcomes. Considering the differences in the extent of the variation and the effects of the variation between Study II and Study III, we do not expect the findings to be generalisable across different diagnostic investigations and diseases. VARIATION IN HEALTHCARE AND OUTCOMES This thesis presents studies of variation in clinical activity in Danish general practice which, to the best of my knowledge, has not previously been done in a similar manner. Study II on the PSA tests revealed that the method is capable of identifying effects of variation on patient outcomes. In Study III, the findings were less prominent. As further discussed in Chapter 5, this may be due to absence of any association or to methodological shortcomings, e.g. insufficient variation or an inadequate number of endoscopies per GP and overall. To further improve future studies on variation and outcomes in the healthcare system, the following questions should be addressed early in the process of each study: 1. Is there sufficient variation between the GPs regarding the investigated exposure? 2. Is the characterisation of the GPs’ level of activity precise and valid? 3. Does the chosen exposure occur frequently among the diseased individuals? The first two questions are closely connected as they involve concepts like standard error of the measurement, latent variables, reliability and the intra‐
class correlation coefficient (ICC). The ICC is a measure of the variance attributable to e.g. the GP level (variance between GPs/total variance)180. Total variance is the sum of the variance between and within GPs. Reliability or 70 Discussion of methods
precision of the ranking can be calculated as (n x ICC)/(1 + [(n‐1) x ICC]). This formula implies that if the variation between GPs is high (high ICC), the number of observations (n, e.g. expected number of colonoscopies) per GP can be lower. It follows that a lower variation necessitates more observations to safely classify GP activity as high or low181. Regarding the third question, the exposure needs to influence a sufficient proportion of the patients to cause an effect that is not diluted by the bulk of un‐
exposed cancer patients. For example, it could be relevant to investigate the variation in the prescriptions rates of medicine for haemorrhoids, but only 5‐
10% of the colorectal cancer patients have a prescription for haemorrhoids in the year preceding diagnosis. Therefore, any change in stage distribution among the affected individuals is most likely mitigated when studying the stage distribution among all colorectal cancer patients. Evaluation of these factors prior to future studies may help researchers avoid initiating studies that are prone to suffer from methodological and informational shortcomings. But more studies are needed to determine which levels of reliability and ICC will be sufficient to produce trustworthy results. Furthermore, the two different approaches in the exposure calculations in Studies II and III need to be investigated further to study the strengths and limitations of each approach. 71 Clinical Activity in General Practice and Cancer
72 Discussion of results
CHAPTER 5 DISCUSSION OF RESULTS 73 Clinical Activity in General Practice and Cancer
RESULTS IN GENERAL We showed that GPs on average have a suspicion of cancer or serious disease once per day and that a suspicion influences both the use of diagnostic tests and the number of referrals, and that it is associated with future diagnoses. Furthermore, we found that variation in diagnostic activity among GPs depends on the activity of interest both in terms of the magnitude of the variation and its consequences. STUDY I We found that GPs suspected cancer or another serious disease in almost 6% of the daytime consultations. This corresponds well with the 1986 study by Nylenna45 in which 4.2% of the consultations led to follow‐up for suspected cancer. Nevertheless, our inclusion of other serious diseases may invalidate this comparison as we do not know whether the object of the GP’s suspicion was cancer or another serious disease. Both studies, however, show that suspicion of serious disease including cancer is present on a daily basis. In the study by Scheel et al182, GPs registered ‘cancer possible, follow‐up needed’ in 3.0% of all consultations, but they were only asked to indicate this for the 12.4% of the patients presenting with cancer alarm symptoms; hence, the true prevalence is unknown. A study found a prevalence of chest pain of 1.5% in GP consultations; and about half of those patients who presented with chest pain were suspected as having ischaemic heart disease183. In our study, we asked for a suspicion in consecutive patients without pre‐defining the symptoms or any signs of relevance. This implies that 5.7% is probably a sound measure of the overall burden of suspicion of cancer and other serious diseases in daily practice. Age was the patient factor most strongly associated with the presence of a suspicion; this finding is in line with other studies45, 182, 183. The digestive system, 74 Discussion of results
blood and blood‐forming organs and female genitals most frequently gave rise to a suspicion, which also corresponds very well with the findings by Nylenna45. A suspicion led to a follow‐up, referral or test in 97% of the patients in our study compared with 89.5% in the study by Scheel et al182, which indicates that actions are taken in almost all patients whom the GP suspects have cancer or serious disease. In total, 41% of suspected patients were referred in Scheel et al compared with 52.3% in our study. The PPV of a suspicion was 7.8% for cancer within 2 years in Nylenna’s study46 compared with an overall PPV for serious disease and cancer of 16.4% after 6 months in our study. The PPV for cancer in our study was 2.3% within 2 months and 3.1% within 6 months. Scheel et al found a PPV of 3.8% for cancer (after approximately 6 months). These numbers are not entirely comparable because of important differences in the follow‐up time and in the included diagnoses. But in all studies, the suspicion substantially increased the risk of being diagnosed with cancer, six times in Scheel et al182, eight times in our study (hazard ratio for cancer within 2 months: 7.55) and nine times in Nylenna’s46 study (own calculation). The increased use of health care services following a consultation in which a suspicion of serious disease was present has, to the best of our knowledge, not been investigated before. But studies on diabetes and cancer have shown that a new diagnosis leads to increasing use of health care services, including general practice27, 184. Overall, the increased use of referrals, laboratory tests, diagnostic imaging and health care services indicates that the GP’s suspicion has a strong impact on the entire healthcare system. However, the increased use seems justified as the PPVs for new diagnoses in our study are of a magnitude corresponding to symptoms that normally warrants a fast‐track referral17, 33. Therefore, it is important to have a healthcare system that supports general practice in the diagnostic work‐up of 75 Clinical Activity in General Practice and Cancer
cancer and serious disease185. An older study found that GPs with direct access to diagnostic facilities had lower outpatient referral rates, which indicates a tolerable cost‐benefit balance186. A recent Scottish study showed that the colorectal cancer detection rate in direct‐access colonoscopies was comparable to that of conventional colonoscopies in secondary care, which indicates a reasonable use of colonoscopies by the GPs187. These two studies also underpin that easier access from general practice could be feasible and relevant. STUDY II We found a substantial variation in PSA test levels among the four quartiles of practices; to our knowledge, this has not been investigated before; but regional and international differences in PSA testing levels are known to exist91, 188‐191. The four populations of men were comparable regarding the measured covariates, and the variation detected may be a result of GPs’ different attitudes towards PSA testing. A Danish survey104 among GPs in Northern Denmark Region showed that only 28% of the GPs tested men with lower urinary tract symptoms as recommended by existing guidelines103. Furthermore, 14% indicated that they performed opportunistic screening, which could explain some of the contrasts observed in our study. The mean PSA level in group 4 was lower than that in group 1, which also suggests that diverse clinical approaches are being used. Correspondingly, a comparison of prostate cancer patients in Denmark, Sweden and Iceland showed that PSA levels in Denmark were higher, depicting the lower uptake of PSA testing in Denmark in 199793. We found strong associations between PSA test levels and various clinical outcomes. This points to a downstream effect of higher test levels. Because of the nature of the study, we cannot infer causality, but the similarity of the four populations indicates that the PSA test levels may be an important explanation 76 Discussion of results
for the differences seen. Furthermore, most of our findings are supported by other studies. We found an IRR of prostate cancer diagnoses in group 4 of 1.37 compared with group 1 which corroborates with both screening trials and observational studies83, 100, 190, 192. In the European Randomised Study of Screening for Prostate Cancer, the authors found an IRR of 1.57 in the screened group compared with the non‐screened group after 13 years of follow‐up83. In the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial, the corresponding figure was 1.22 in the screened group after 2 years100. In Denmark, Outzen et al. showed an increase in prostate cancer incidence from 29.2 per 100 000 person years in 1978‐
1982 to 76.2 per 100 000 person years in 2008‐2009, with the increase primarily starting in the 1990s when the PSA test was introduced192. In an ecologic study comparing two areas in the US with a five‐fold difference in PSA testing levels in the early era of PSA testing (1988‐1990), Lu‐Yao et al. found a 93% higher cumulative incidence of prostate cancer in the Seattle‐Puget Sound (high testing) area than in Connecticut in 1987‐90190. In the same period, the rate ratio of biopsies was 2.20 in Seattle compared with Connecticut, echoing the findings in our study. Our findings confirm the association between higher PSA test levels and younger age at diagnosis shown by Sandblom et al101. In Denmark, Outzen et al found that the median age at diagnosis decreased from 75.1 years in 1988–1992 to 69.7 in 2008–2009; a period in which PSA test rates rose more than 40‐fold192. International comparisons also report an association between more PSA tests and younger age at diagnosis91. The findings of more localised cancers and no difference in incidence of distant tumours are in agreement with several previous studies93, 192, 193. However, our findings contrast the results from the Göteborg randomised population‐based prostate‐cancer screening trial by Hugosson et al194. They showed a lower 77 Clinical Activity in General Practice and Cancer
number of more advanced prostate cancers in the screened group than in the control group which indicates a more favourable stage distribution. The increased rate of prostatectomies in the highest testing quartile corroborates previous results. Hugosson et al194 found that 41% in the screened group had a prostatectomy compared with 34% in the control group. Even more pronounced were the differences between the high‐testing Seattle and the low‐testing Connecticut areas (5.9 fold)190. In the study by Sandblom et al., 19% of the screen‐detected cases had a prostatectomy compared with 8% of the cancers in the control arm. Sandblom et al. also found a tendency towards better relative survival among the screened patients than among members of the control group101. Bray et al. found an increasing 5‐year relative survival in the Nordic countries which intensified during the period when PSA testing picked up92. Furthermore, the association between PSA testing and survival is known from international comparisons91. These survival differences and our findings are probably caused by lead‐time and length‐time biases93, 94, 195, which is why randomised controlled trials focus on mortality. We found no statistically significant differences in prostate cancer‐specific mortality rates between the four groups of practices. This is supported by the study from Seattle and Connecticut, even after an extension of the follow‐up period to 15 years196. A Cochrane review from 2013, which included five randomised trials with PSA testing, showed a risk ratio of prostate cancer‐
specific mortality of 1.00 (95% CI 0.86 to 1.17) based on a meta‐analysis. However, this contrasts with the findings from the recently updated European Randomised Study of Screening for Prostate Cancer in which a risk ratio for prostate cancer death of 0.79 (0.69‐0.91) was found in the screened group after 13 years of follow‐up. The European trial is probably the best of the studies included in the Cochrane review with more than 13,000 prostate cancer cases 78 Discussion of results
and less contamination of the control group than in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. In our study, the follow‐up time may therefore be too short to show a difference as the European Randomised Study of Screening for Prostate Cancer shows that more than 9 years are needed for a difference to emerge. Still, the conclusion from the screening trial was that “further quantification of harms and their reduction are still considered a prerequisite for the introduction of population‐based screening” as the number‐
needed‐to‐invite is 781 and the number‐needed‐to‐diagnose is 27 to avert one prostate cancer death. STUDY III To our knowledge, this is the first study investigating the variation in GPs’ propensity to refer to lower endoscopy facilities in a healthcare system not encouraging screening with endoscopy. Jacob et al197 found variation in referrals for colonoscopies among >11,000 primary care physicians in Ontario with a median rate of 47 colonoscopies per 1000 patients and an interquartile range of 27‐75 in 2005. In another study by Jacob et al, the authors estimated that more than 95% of the colonoscopies were performed for screening purposes159, and they found a median screening rate of 3.7% in the lowest testing quintile of primary care physicians and 12.1% among the highest testing quintile during 1996‐2000. Both studies were, however, unable to identify screening colonoscopies with certainty. The variation in these studies corresponds fairly well with the variations found in our studies when looking at the yearly rates (Table 3.5 in Chapter 3). We found no studies investigating the effects of variation in the use of lower endoscopies on colorectal cancer outcomes in a non‐screening setting. Our results are therefore neither comparable to the results from observational studies from the United States and Canada nor to results from screening trials. 79 Clinical Activity in General Practice and Cancer
However, these studies (including studies on the FOBT) support that colorectal cancer diagnosis can be made earlier and that earlier detection improves prognosis85‐88, 111, 113, 118, 198‐201, but it is unknown whether this also a applies to a setting focusing on symptomatic patients. The studies also found a decreased incidence of colorectal cancer among people who were screened, but we found no such correlation with a higher endoscopy rate. This discrepancy could reflect the non‐screening setting in our study as a decreased cancer incidence is achieved primarily through polypectomies which may be asymptomatic. Another explanation could be insufficient follow‐up as illustrated by Holme et al86, who showed that the decrease in incidence did not appear until after 8 years of follow‐up. We saw a small positive effect regarding earlier stage at diagnosis, but the absolute differences were small and only evident for rectal cancer. Gross et al. found an increased proportion of cancers diagnosed in early stage with increased colonoscopy rates over time, but the number of patients with missing stage was not reported200. Screening trials have also found an improvement in stage distribution in the intervention group, mainly for the patients choosing to participate. In the Minnesota trial, screening with FOBT led to 47% fewer patients being diagnosed in Duke’s D in the annual screening group than in the non‐screened group113, and a Danish screening trial also found more patients diagnosed in early stage111. Similarly, randomised trials with sigmoidoscopy also report improved stage distributions85, 86, 118, mainly in per‐protocol analyses. The difference we found between colon and rectum is not directly supported by other studies. Gross et al200 found the effect to be most pronounced among proximal colon cancers; Schoen et al85 and Baxter et al201 found it to be most pronounced for distal colon cancer, and Segnan et al118 found no difference according to site. 80 Discussion of results
We found only small, statistically non‐significant differences according to treatment received; an issue which has not been examined before in relation to colorectal cancer. In the study of variation in gastroscopy rates, Shawihdi et al79 found that oesophagogastric cancer patients from a practice in the lowest referring tertile of practices experienced more emergency admissions and less major surgery. There may be several explanations as to why the associations are so modest in our study. Firstly, the magnitude of the variation investigated needs to be sufficient to cause differences in outcomes70. We do now know whether a two‐
fold variation in lower endoscopy rates is inadequate, and efforts to identify studies addressing this aspect have been fruitless. Secondly, a robust classification of GPs based on lower endoscopy rates, for example, calls for an adequate amount of procedures to avoid that differences are due only to chance (reliability, see Chapter 4)180, 181. This could be a challenge in Study III, as indicated by the modest consistency over time (Chapter 2), but inclusion of the additional scopies not ordered in general practice doubled the number of endoscopies without changing the findings. Thirdly, the colonoscopy rates in Denmark could be too low for a substantial benefit to be present. Compared with the United States and Canada, we found colonoscopy rates of less than 30/1000 in Denmark, while they were above 50/1000 in Canada in 2006199, 200. A threshold may exist above which GPs need to refer for an effect to be apparent. 81 Clinical Activity in General Practice and Cancer
82 Main conclusions
CHAPTER 6 MAIN CONCLUSIONS 83 Clinical Activity in General Practice and Cancer
STUDY I GPs suspected cancer or another serious disease after 5.7% of consultations with patients aged 18 years and above. The GP’s suspicion was strongly associated with the patient’s age and with symptoms from the digestive system, blood or blood‐forming organs, or female genitals. A suspicion led to increased use of tests in the clinic and an increase in referrals for further investigation. In particular, the use of diagnostic imaging and endoscopies rose after the index consultation. Patients suspected of harbouring serious disease consumed more health care services in the subsequent 6 months than patients in whom no serious disease was expected. The risk of being diagnosed with a new diagnosis of cancer or another serious disease was increased if the GP had a suspicion, particularly within the first 2 months after the index consultation. The PPV of a GP suspicion for getting a new serious diagnosis was 9.8% within the first 2 months and 16.4% within 6 months after the index consultation. STUDY II A higher use of PSA testing in general practice was associated with a higher risk of being investigated with a trans‐rectal ultrasound of the prostate and biopsies. Furthermore, men listed with the quartile of practices performing the highest number of tests had an increased risk of getting a prostate cancer diagnosis and were more often diagnosed with local‐stage disease, but had the same risk of distant disease as men listed with the quartile of practices performing the lowest number of tests. In the highest testing practices, more men experienced a radical prostatectomy or radiotherapy. However, we found no relation between PSA test rate and prostate cancer‐specific or overall mortality. 84 Main conclusions
STUDY III A higher propensity to refer for lower endoscopies was associated with a higher incidence of colon cancer and a higher proportion of rectal cancers being diagnosed at an early stage (UICC I+II). However, the differences were small and the clinical relevance is debatable. No clear association was seen between the propensity to refer to a lower endoscopy and the proportion of patients treated with elective surgery or with a curative intent. This thesis has provided updated information on how often GPs suspect cancer and other serious diseases and how they deal with this suspicion. We now know that a suspicion is associated with actions being taken, and the GPs’ access to diagnostic tests has to be supported as such suspicion substantially predicts future diagnoses. We have also shown that general practices differ in their PSA testing and lower endoscopy testing rates. Our findings show that whether variation in general practice has consequences for patients depends on the investigated activity. 85 Clinical Activity in General Practice and Cancer
86 Perspectives for the healthcare system and future research
CHAPTER 7 PERSPECTIVES FOR THE HEALTHCARE SYSTEM AND FUTURE RESEARCH 87 Clinical Activity in General Practice and Cancer
Cancer is a common disease, the incidence of cancer is increasing and cancer is the leading cause of death in Denmark. Earlier diagnosis is critical to any improvement in cancer patients’ outcomes, and the GP plays a pivotal role in achieving this goal. In this thesis, we explored aspects of the diagnostic work performed in general practice in relation to cancer and serious disease. Suspicion of cancer and other serious diseases is a prerequisite for initiating diagnostic work‐up, and we showed that such work‐up in general practice is instituted almost on a daily basis. The growing use of tests and referrals following a suspicion underscores the importance of supporting GPs by not restraining their access to relevant diagnostic tools. The high PPV for getting a serious diagnosis or cancer further indicates that a strategy that allows easier access to secondary sector diagnostic work‐up may be cost‐effective, but our study was not designed to adequately address this particular issue. Our findings lend strong support for the new initiatives encompassed by the Danish Government, which states that “GPs should have greater access to directly referring patients for diagnostic investigations”. This appears to be a reasonable initiative as many patients do not fulfil the current criteria for an urgent referral in a cancer fast‐track. Future studies should aim to develop and evaluate the effects of such new initiatives, and a study similar to Study I with more patients and more details on diagnostic reasoning is also highly relevant. Such a study would, furthermore, generate knowledge about the GPs’ more specific needs in relation to investigations for cancer and serious diseases. We found variation in the GPs’ diagnostic activity regarding PSA testing and use of lower endoscopies. Our study on PSA testing indicates variation in the adherence to current guidelines. We do not know which group of GPs is best at complying with the prevailing guidelines, but previous research indicates that opportunistic screening is ongoing in Denmark105, 106. Therefore, we think that it 88 Perspectives for the healthcare system and future research
is important for GPs to know that differences in PSA test levels are not without consequences, and we want to emphasise the importance of adhering to the current guidelines. PSA screening experts do not agree on the effects of PSA screening on mortality, but they are unanimous that knowledge about the adverse effects of screening is too poor to recommend screening at the moment202. Thus, to optimise the diagnosis for prostate cancers, PSA alone is not the answer, but a combination of tests could prove beneficial. However, today’s most important issue in prostate cancer research is to develop diagnostic methods to differentiate indolent cancers from aggressive cancers202. Furthermore, it would be relevant to reanalyse the study in 5‐7 years to be able to obtain knowledge of mortality in the longer run. In Denmark, GPs make much less use of lower endoscopies than primary care physicians in the United States, for example, because screening has never been encouraged in Denmark. Lower endoscopies are only performed on the basis of symptoms, and this may explain the small variation in the use of lower endoscopies observed in the present study. Our findings could encourage health system administrators and politicians to demand GPs with the highest propensity to refer to decrease their use of referral because the extra costs could seem disproportionate to the benefits. However, our study cannot reject that an even higher rate may be beneficial. This issue could be investigated in an intervention study aiming at lowering the referral threshold in general practice. This type of intervention implies careful considerations about adverse effects of more colonoscopies203. A study on the compliance with the existing referral guidelines may be relevant prior to such intervention studies. Furthermore, future research should investigate the effectiveness of alternative strategies in the early detection of CRC. National screening with iFOBT (immune‐chemical faecal occult blood test) was initiated in Denmark in 2014, but the majority of CRC patients still depend on timely investigation and action 89 Clinical Activity in General Practice and Cancer
in general practice. Currently, a study is being prepared on the use of iFOBT on symptomatic patients in general practice. This may be a relevant tool as the threshold for applying this test may be lower than for referring to lower endoscopy, but only little knowledge exists about its effectiveness in symptomatic patients. The results of studies on variation demand careful interpretation as what may seem like obvious over‐use or poor clinical practice may be rooted in methodological shortcomings. Consequently, the use of variation as a publicly available performance indicator180, as a tool for resource allocation204 or as a basis for reprimanding GPs, is not our aim. From our perspective, research on variation in diagnostic activity can be an effective tool in the pursuit of ways of improving clinical practice, but such research needs to be done meticulously. Therefore, future focus should be on improving our models and enhancing our knowledge regarding quantification of the magnitude of variation. Intra‐class correlation coefficient and reliability are two concepts we need to test in this context as well. Used judiciously, we intent to expand the focus of these studies to multiple activities in general practice which may help us to learn from the best. 90 English summary
ENGLISH SUMMARY 91 Clinical Activity in General Practice and Cancer
BACKGROUND AND AIMS Cancer is a common, serious disease and early diagnosis is a cornerstone in the effort to improve the outcome from cancer disease. The general practitioner (GP) plays a crucial role in achieving this goal. Little is known about GPs’ suspicion of cancer and the activities the GPs institute in relation to such suspicion. Knowledge is also sparse on any effects of different diagnostic activities in general practice. The overall aims of this thesis were therefore: ‐
to describe how often Danish GPs suspected cancer or other serious diseases and how they acted on the suspicion, and to analyse how a suspicion influenced the demand for health care services and predicted a future diagnosis of serious disease ‐
to investigate whether variation in GPs’ diagnostic activity influences cancer patients’ prognosis in relation to prostate‐specific antigen (PSA) testing and prostate cancer and lower endoscopies and colorectal cancer METHODS In Study I, survey data from more than 400 GPs and 4000 consultations were combined with registry data on serious disease. Study II and Study III were based only on registry data. RESULTS In Study I, we saw that a suspicion of cancer or another serious disease was present in 6% of GP consultations, and the suspicion was associated with an increased use referrals and diagnostic imaging. The suspicion was associated with an increased risk of being diagnosed with serious disease including cancer and had a positive predictive value for a new diagnosis within 2 months of 9.8%. In Study II, we found a 3.6‐fold variation in PSA test rates among the highest testing and the lowest testing quartiles of general practices in the Central 92 English summary
Denmark Region. Compared with men in the lowest testing quartile of practices, men in the highest quartile had an increased risk of being diagnosed with prostate cancer and an increased risk of having a radical prostatectomy. We found no differences in prostate cancer or overall mortality between the groups. In Study III, the quartile of general practices with the highest propensity to refer for lower endoscopies ordered twice as many endoscopies as the least referring quartile of practices. A positive correlation was seen between a higher propensity to refer and more rectal cancers being diagnosed in early stages. The differences were, however, small, and the clinical significance may be debatable. CONCLUSIONS AND PERSPECTIVES GPs need to be supported by the secondary healthcare sector in investigating patients suspected of serious disease because the GPs’ suspicion has a significant predictive potential. General practice should play a key role in the diagnostic evaluation of patients suspected of serious disease to reduce patients’ mortality and morbidity from cancer, but also because it probably pays to do so from a cost‐effectiveness perspective. The findings in Study II underscore that the current guidelines regarding PSA testing for general practice should be followed to avoid inflicting iatrogenic harm. In the future, we need to learn to differentiate between indolent and aggressive prostate cancers. This may, again, leave room for PSA testing as a screening tool. Our findings do indicate a small positive effect of more lower endoscopies, which could be investigated in an intervention study where some GPs are randomised to a more liberal access to lower endoscopies. Alongside this, we need to keep on exploring alternative approaches including the use of iFOBT in symptomatic patients. Overall, this thesis indicates that the role of GPs in the diagnosis of cancer should be strengthened through easy access to diagnostic investigations, but at the same time we need research on the possible consequences of variation in the use of these activities in order to optimise the GPs’ work. 93 Clinical Activity in General Practice and Cancer
94 Dansk resume
DANSK RESUME 95 Clinical Activity in General Practice and Cancer
BAGGRUND OG FORMÅL Kræft er en hyppig, alvorlig sygdom, og tidlig diagnostik er vigtigt i indsatsen for en bedre prognose efter en kræftdiagnose. Almen praksis (AP) i Danmark spiller en helt central rolle i dette. Vi mangler viden om hyppigheden af mistanke om kræft eller anden alvorlig sygdom i AP samt information om, hvorledes den praktiserende læge handler på denne mistanke. Vi mangler desuden viden om de mulige konsekvenser af variation mellem lægers diagnostiske aktivitet. Derfor var formålene med dette ph.d.‐studie: ‐
At undersøge hyppigheden af praktiserende lægers mistanke om kræft eller anden alvorlig sygdom efter en konsultation samt at analysere, hvordan en mistanke influerede på forbruget af sundhedsydelser og prædikterede fremtidige alvorlige diagnoser ‐
At undersøge om variation mellem praktiserende lægers diagnostiske aktivitet influerede på kræftpatienters prognose i forhold til brug af prostata‐specifik antigen (PSA)‐tests og prostatakræft og brug af endoskopier og kolorektal METODER I studie I blev audit‐registreringer fra almen praksis (>4000 konsultationer) kombineret med registerdata om alvorlig sygdom. Studie II og III var baseret udelukkende på registerdata. RESULTATER I studie I fandt vi, at mistanke om kræft eller anden alvorlig sygdom var til stede efter ca. 6% af konsultationerne, og at en mistanke var relateret til øget forbrug af henvisninger og radiologiske ydelser. Desuden medførte en mistanke en øget risiko for at blive diagnosticeret med en alvorlig lidelse, og den positive prædiktive værdi af en læges mistanke var 9,8% for en ny alvorlig diagnose 96 Dansk resume
inden for 2 måneder. I studie II fandt vi en 3,6‐fold variation i brug af PSA tests mellem den fjerdedel af praktiserende læger i Region Midtjylland, der brugte færrest, og den fjerdedel der brugte flest PSA tests i 2004‐2009. Mænd, hvis praktiserende læge var i den højest PSA‐test forbrugende fjerdedel, fik flere prostatakræftdiagnoser og fik oftere fjernet prostata. Vi fandt ingen forskel i prostatakræftspecifik eller overordnet dødelighed imellem de 4 grupper. I studie III fandt vi en 2‐fold variation i tilbøjeligheden til at henvise til skopier imellem den mest henvisende fjerdedel af almen praksis og den mindst henvisende fjerdedel. Blandt den fjerdedel, der henviste mest, var der en tendens til, at flere blev diagnosticeret i tidligt stadie blandt patienter med endetarmskræft. Forskellene var små, og den kliniske relevans diskutabel. KONKLUSION OG PERSPEKTIVER Denne afhandling viser, at praktiserende læger mistænker kræft eller anden alvorlig sygdom dagligt, og den efterfølgende udredning bør understøttes af det sekundære sundhedsvæsen, eftersom mistanken havde en signifikant prædiktiv værdi. Resultaterne i studie II understreger vigtigheden af at efterleve retningslinjer vedrørende PSA‐brug for at undgå, at patienter udsættes for unødvendig overlast i form af diagnostiske og kirurgiske procedurer. Aktuelt forskes der intenst i metoder til at differentiere imellem indolente og aggressive prostatakræfttilfælde og opnås dette kan PSA potentielt få en rolle som screeningsværktøj. Om flere skopier kunne have en positiv effekt på prognosen efter kolorektalkræft, kan efterprøves i et studie, hvor nogle praktiserende lægers adgang til skopier udvides. Desuden bør effekterne af den nyindførte screening for kolorektalkræft monitoreres, og effekten af at bruge iFOBT (påvisning af blod i afføringen) blandt symptomatiske patienter undersøges. Overordnet fandt vi, at praktiserende lægers rolle i kræftdiagnostik bør styrkes, eksempelvis gennem adgang til undersøgelser. Brugen af disse undersøgelser skal desuden monitoreres for at lære af variationen mellem praktiserende læger. 97 Clinical Activity in General Practice and Cancer
98 References
REFERENCES 99 Clinical Activity in General Practice and Cancer
1. The Danish National Board of Health. Cause of Death Register 2012 [report in Danish]. ; 2013. 2. The Danish National Board of Health. Cancerregisteret 2012 [report in Danish]. Copenhagen: ; 2013. 3. Albreht T, McKee M, Alexe D, Coleman MP, Martin‐Moreno JM. Making progress against cancer in Europe in 2008. Eur J Cancer 2008; 7;44(10):1451‐6. 4. Luengo‐Fernandez R, Leal J, Gray A, Sullivan R. Economic burden of cancer across the European Union: a population‐based cost analysis. The Lancet Oncology 2013; 11;14(12):1165‐74. 5. Coleman MP, Forman D, Bryant H, Butler J, Rachet B, Maringe C, et al. Cancer survival in Australia, Canada, Denmark, Norway, Sweden, and the UK, 1995‐
2007 (the International Cancer Benchmarking Partnership): an analysis of population‐based cancer registry data. Lancet 2011; Jan 8;377(9760):127‐38. 6. De Angelis R, Sant M, Coleman MP, Francisci S, Baili P, Pierannunzio D, et al. Cancer survival in Europe 1999–2007 by country and age: results of EUROCARE‐5—a population‐based study. The Lancet Oncology 2014; 1;15(1):23‐
34. 7. Walters S, Maringe C, Coleman MP, Peake MD, Butler J, Young N, et al. Lung cancer survival and stage at diagnosis in Australia, Canada, Denmark, Norway, Sweden and the UK: a population‐based study, 2004‐2007. Thorax 2013; Jun;68(6):551‐64. 8. Maringe C, Walters S, Rachet B, Butler J, Fields T, Finan P, et al. Stage at diagnosis and colorectal cancer survival in six high‐income countries: A population‐based study of patients diagnosed during 2000–2007. Acta Oncol 2013; 06/01; 2014/07;52(5):919‐32. 9. Walters S, Maringe C, Butler J, Rachet B, Barrett‐Lee P, Bergh J, et al. Breast cancer survival and stage at diagnosis in Australia, Canada, Denmark, Norway, Sweden and the UK, 2000‐2007: a population‐based study. Br J Cancer 2013; Mar 19;108(5):1195‐208. 10. Christensen LH, Engholm G, Ceberg J, Hein S, Perfekt R, Tange UB, et al. Can the survival difference between breast cancer patients in Denmark and Sweden 1989 and 1994 be explained by patho‐anatomical variables?‐‐a population‐based study. Eur J Cancer 2004; May;40(8):1233‐43. 100 References
11. Storm H, Dickman P, Engeland A, Haldorsen T, Hakulinen T. Do morphology and stage explain the inferior lung cancer survival in Denmark?. European Respiratory Journal 1999; February 01;13(2):430‐5. 12. Folkesson J, Engholm G, Ehrnrooth E, Kejs AM, Pahlman L, Harling H, et al. Rectal cancer survival in the Nordic countries and Scotland. Int J Cancer 2009; Nov 15;125(10):2406‐12. 13. Sant M, Allemani C, Capocaccia R, Hakulinen T, Aareleid T, Coebergh JW, et al. Stage at diagnosis is a key explanation of differences in breast cancer survival across Europe. Int J Cancer 2003; Sep 1;106(3):416‐22. 14. Engholm G, Kejs AM, Brewster DH, Gaard M, Holmberg L, Hartley R, et al. Colorectal cancer survival in the Nordic countries and the United Kingdom: excess mortality risk analysis of 5 year relative period survival in the period 1999 to 2000. Int J Cancer 2007; Sep 1;121(5):1115‐22. 15. Bekendtgørelse om behandling af patienter med livstruende kræftsygdomme m.v. [in Danish]. Available at: https://www.retsinformation.dk/forms/R0710.aspx?id=21709. Accessed 08/29, 2014. 16. Aftale om gennemførelse af målsætningen om akut hand‐ling og klar besked til kræftpatienter [in Danish]. Available at: https://sundhedsstyrelsen.dk/~/media/A92B68A03752493FA62EA08BA169FC05.
ashx. Accessed 08/29, 2014. 17. Beskrivelse af pakkeforløb 2012 [in Danish]. Available at: http://sundhedsstyrelsen.dk/da/sundhed/folkesygdomme/kraeft/pakkeforloeb/p
akkeforloeb‐2012. Accessed 08/29, 2014. 18. Pedersen KM, Andersen JS, Søndergaard J. General Practice and Primary Health Care in Denmark. The Journal of the American Board of Family Medicine March‐April 2012; March‐April 2012;25(Suppl 1):S34‐8. 19. Vedsted P, Olesen F. The use of Danish day‐time general practice. Status and perspectives. Ugeskr Laeger 2005; Aug 29;167(35):3280‐2. 20. Faktaark om Almen Praksis 2013 [in Danish]. Available at: http://www.laeger.dk/portal/pls/portal/!PORTAL.wwpob_page.show?_docnam
e=10474986.PDF. 101 Clinical Activity in General Practice and Cancer
21. Vedsted P, Olesen F. Are the serious problems in cancer survival partly rooted in gatekeeper principles? An ecologic study. Br J Gen Pract 2011; Aug;61(589):e508‐12. 22. Moller H, Linklater KM, Robinson D. A visual summary of the EUROCARE‐
4 results: a UK perspective. Br J Cancer 2009; Dec 3;101 Suppl 2:S110‐4. 23. Weller D, Vedsted P, Rubin G, Walter FM, Emery J, Scott S, et al. The Aarhus statement: improving design and reporting of studies on early cancer diagnosis. Br J Cancer 2012; Mar 27;106(7):1262‐7. 24. Allgar VL, Neal RD. General practictionersʹ management of cancer in England: secondary analysis of data from the National Survey of NHS Patients‐
Cancer. Eur J Cancer Care (Engl) 2005; Dec;14(5):409‐16. 25. Hansen RP, Olesen F, Sorensen HT, Sokolowski I, Sondergaard J. Socioeconomic patient characteristics predict delay in cancer diagnosis: a Danish cohort study. BMC Health Serv Res 2008; Feb 28;8:49. 26. Hansen R, Vedsted P, Sokolowski I, Sondergaard J, Olesen F. Time intervals from first symptom to treatment of cancer: a cohort study of 2,212 newly diagnosed cancer patients. BMC Health Services Research 2011;11(1):284. 27. Christensen KG, Fenger‐Gron M, Flarup KR, Vedsted P. Use of general practice, diagnostic investigations and hospital services before and after cancer diagnosis ‐ a population‐based nationwide registry study of 127,000 incident adult cancer patients. BMC Health Serv Res 2012; Jul 28;12:224,6963‐12‐224. 28. Meechan D, Gildea C, Hollingworth L, Richards MA, Riley D, Rubin G. Variation in use of the 2‐week referral pathway for suspected cancer: a cross‐
sectional analysis. Br J Gen Pract 2012; Sep;62(602):e590‐7. 29. Jensen H, Torring ML, Olesen F, Overgaard J, Vedsted P. Cancer suspicion in general practice, urgent referral and time to diagnosis: a population‐based GP survey and registry study. BMC Cancer 2014; Aug 30;14(1):636,2407‐14‐636. 30. Svendsen RP, Stovring H, Hansen BL, Kragstrup J, Sondergaard J, Jarbol DE. Prevalence of cancer alarm symptoms: a population‐based cross‐sectional study. Scand J Prim Health Care 2010; Sep;28(3):132‐7. 31. Ingebrigtsen SG, Scheel BI, Hart B, Thorsen T, Holtedahl K. Frequency of ‘warning signs of cancer’ in Norwegian general practice, with prospective recording of subsequent cancer. Family Practice 2012; October 23;. 102 References
32. Rothman KJ, Greenland S,1951‐, Lash TL. Modern epidemiology. 3. ed. ed. Philadelphia: Lippincott Williams & Wilkins; 2008. 33. Shapley M, Mansell G, Jordan JL, Jordan KP. Positive predictive values of >/=5% in primary care for cancer: systematic review. Br J Gen Pract 2010; Sep;60(578):e366‐77. 34. Hamilton W. The CAPER studies: five case‐control studies aimed at identifying and quantifying the risk of cancer in symptomatic primary care patients. Br J Cancer 2009; Dec 3;101 Suppl 2:S80‐6. 35. Hamilton W, Round A, Sharp D, Peters T. Clinical features of colorectal cancer before diagnosis: a population‐based case‐control study. Br J Cancer 2005;93:399‐405. 36. Hamilton W, Sharp DJ, Peters TJ, Round AP. Clinical features of prostate cancer before diagnosis: a population‐based, case‐control study. Br J Gen Pract 2006; Oct;56(531):756‐62. 37. Astin M, Griffin T, Neal RD, Rose P, Hamilton W. The diagnostic value of symptoms for colorectal cancer in primary care: a systematic review. Br J Gen Pract 2011; May;61(586):231‐43. 38. Jones R, Latinovic R, Charlton J, Gulliford MC. Alarm symptoms in early diagnosis of cancer in primary care: cohort study using General Practice Research Database. BMJ 2007; May 19;334(7602):1040. 39. Hamilton W, Lancashire R, Sharp D, Peters TJ, Cheng KK, Marshall T. The importance of anaemia in diagnosing colorectal cancer: a case‐control study using electronic primary care records. Br J Cancer 2008; Jan 29;98(2):323‐7. 40. Hamilton W, Lancashire R, Sharp D, Peters T, Cheng KK, Marshall T. The risk of colorectal cancer with symptoms at different ages and between the sexes: a case‐control study. BMC Medicine 2009;7(1):17. 41. Nielsen TN, Hansen RP, Vedsted P. Symptom presentation in cancer patients in general practice]. Ugeskr Laeger 2010; Oct 11;172(41):2827‐31. 42. Bjerager M, Palshof T, Dahl R, Vedsted P, Olesen F. Delay in diagnosis of lung cancer in general practice. Br J Gen Pract 2006; Nov;56(532):863‐8. 103 Clinical Activity in General Practice and Cancer
43. Ades AE, Biswas M, Welton NJ, Hamilton W. Symptom lead time distribution in lung cancer: natural history and prospects for early diagnosis. International Journal of Epidemiology 2014; August 29;. 44. Vedsted P, Hansen RP, Bro F. General practice and early cancer diagnosis. Ugeskr Laeger 2011; Jun 13;173(24):1712‐5. 45. Nylenna M. Diagnosing cancer in general practice: when is cancer suspected?. Br Med J (Clin Res Ed) 1986; Jul 26;293(6541):245‐8. 46. Nylenna M. Diagnosing cancer in general practice: from suspicion to certainty. Br Med J (Clin Res Ed) 1986; Aug 2;293(6542):314‐7. 47. Earwicker SC, Whynes DK. General practitionersʹ referral thresholds and choices of referral destination: an experimental study. Health Econ 1998;7(8):711‐
22. 48. Cummins RO, Jarman B, White PM. Do general practitioners have different ʺreferral thresholdsʺ?. Br Med J (Clin Res Ed) 1981;282:1037‐9. 49. Sirovich BE, Gottlieb DJ, Welch HG, Fisher ES. Variation in the Tendency of Primary Care Physicians to Intervene. Arch Intern Med 2005; October 24;165(19):2252‐6. 50. Reid FDA, Cook DG, Majeed A. Explaining variation in hospital admission rates between general practices: cross sectional study. BMJ 1999; July 10;319(7202):98‐103. 51. Glover JA. The Incidence of Tonsillectomy in School Children: (Section of Epidemiology and State Medicine). Proc R Soc Med 1938; Aug;31(10):1219‐36. 52. Wennberg JE. Tracking Medicine ‐ A Researcherʹs Guide to Understand Health Care. 1st ed. New York: Oxford University Press; 2010. 53. Appleby J, Raleigh V, Frosini F, Bevan G, Gao H, Lyscom T. Variations in health care The good, the bad and the inexplicable . 1st ed. The King’s Fund 11–13 Cavendish Square London W1G 0AN: Kingʹs Fund; 2011. 54. Ringberg U, Fleten N, Frøde OH. Examining the variation in GPsʹ referral practice: A cross‐sectional study of GPsʹ reasons for referral. British Journal of General Practice 2014;64(624):e426‐33. 104 References
55. Carlsen F, Grytten J, Kjelvik J, Skau I. Better primary physician services lead to fewer hospital admissions. European Journal of Health Economics 2007;8(1):17‐
24. 56. McBride D, Hardoon S, Walters K, Gilmour S, Raine R. Explaining variation in referral from primary to secondary care: cohort study. BMJ 2010; November 30;341. 57. Ringberg U, Fleten N, Deraas TS, Hasvold T, Førde O. High referral rates to secondary care by general practitioners in Norway are associated with GPsʹ gender and specialist qualifications in family medicine, a study of 4350 consultations. BMC Health Services Research 2013;13(1). 58. OʹDonnell CA. Variation in GP referral rates: what can we learn from the literature?. Fam Pract 2000; Dec;17(6):462‐71. 59. Franks P, Mooney C, Sorbero M. Physician referral rates: Style without much substance?. Med Care 2000;38(8):836‐46. 60. Franks P, Zwanziger J, Mooney C, Sorbero M. Variations in primary care physician referral rates. Health Serv Res 1999; Apr;34(1 Pt 2):323‐9. 61. Ambery P, Donald I. Variation in general practice medical admission rates for elderly people. Journal of Public Health 2000; September 01;22(3):422‐6. 62. Rossdale M, Kemple T, Pyne S, Calnan M, Greenwood R. An observational study of variation in GPsʹ out‐of‐hours emergency referrals. British Journal of General Practice 2007;57(535):152‐4. 63. Ferguson B, Gravelle H, Dusheiko M, Sutton M, Johns R. Variations in practice admission rates: the policy relevance of regression standardisation. J Health Serv Res Policy 2002; July 1;7(3):170‐6. 64. Hungin AP, Bramble MG, OʹCallaghan H. Reasons for variations in the use of open access gastroscopy by general practitioners. Gut 1995; Feb;36(2):180‐2. 65. Christensen B, Sørensen HT, Mabeck CE. Differences in Referral Rates from General Practice. Family Practice 1989; March 01;6(1):19‐22. 66. Christensen B, Sorensen HT, Mabeck CE. Referral from general practice to dermatologists. Scand J Prim Health Care 1988; Feb;6(1):29‐32. 105 Clinical Activity in General Practice and Cancer
67. Sørensen TH, Olsen KR, Vedsted P. Association between general practice referral rates and patients’ socioeconomic status and access to specialised health care: A population‐based nationwide study. Health Policy 2009; 10;92(2‐3):180‐6. 68. Rose Olsen K, Dansk Sundhedsinstitut. Analyse af henvisningsmønstret i dansk almen praksis : delanalyse 2 ‐ en registerundersøgelse. Dansk Sundhedsinstitut; 2008. 69. Baughan P, Keatings J, OʹNeill B. Urgent suspected cancer referrals from general practice: audit of compliance with guidelines and referral outcomes. Br J Gen Pract 2011; Nov;61(592):e700‐6. 70. Davis P, Gribben B, Lay‐Yee R, Scott A. How much variation in clinical activity is there between general practitioners? A multi‐level analysis of decision‐making in primary care. J Health Serv Res Policy 2002; October 1;7(4):202‐8. 71. Verstappen W, ter Riet G, van der Weijden T, Hermsen J, Grol R. Variation in requests for imaging investigations by general practitioners: a multilevel analysis. J Health Serv Res Policy 2005; Jan;10(1):25‐30. 72. Kristiansen IS, Hjortdahl P. The general practitioner and laboratory utilization: why does it vary?. Fam Pract 1992; Mar;9(1):22‐7. 73. OʹKane MJ, Casey L, Lynch PLM, McGowan N, Corey J. Clinical outcome indicators, disease prevalence and test request variability in primary care. Ann Clin Biochem 2011; March 1;48(2):155‐8. 74. Verstappen WH, ter Riet G, Dubois WI, Winkens R, Grol RP, van der Weijden T. Variation in test ordering behaviour of GPs: professional or context‐
related factors?. Family Practice 2004; August 01;21(4):387‐95. 75. Knottnerus JA, Joosten J, Daams J. Comparing the quality of referrals of general practitioners with high and average referral rates: an independent panel review. Br J Gen Pract 1990; May;40(334):178‐81. 76. Fertig A, Roland M, King H, Moore T. Understanding variation in rates of referral among general practitioners: are inappropriate referrals important and would guidelines help to reduce rates?. BMJ 1993;307:1467‐70. 77. Ward PR, Noyce PR, St Leger AS. Differential associations between actual and expected GP practice prescribing rates for statins, ACE inhibitors, and beta‐
blockers: a cross‐sectional study in England. Ther Clin Risk Manag 2005; Mar;1(1):61‐8. 106 References
78. Hippisley‐Cox J, Hardy C, Pringle M, Fielding K, Carlisle R, Chilvers C, et al. Are patients who present late with cancer registered with low referring practices?. Br J Gen Pract 1997; Nov;47(424):731‐2. 79. Shawihdi M, Thompson E, Kapoor N, Powell G, Sturgess RP, Stern N, et al. Variation in gastroscopy rate in English general practice and outcome for oesophagogastric cancer: retrospective analysis of Hospital Episode Statistics. Gut 2014; Feb;63(2):250‐61. 80. Mantynen T, Farkkila M, Kunnamo I, Mecklin JP, Juhola M, Voutilainen M. The impact of upper GI endoscopy referral volume on the diagnosis of gastroesophageal reflux disease and its complications: a 1‐year cross‐sectional study in a referral area with 260,000 inhabitants. Am J Gastroenterol 2002; Oct;97(10):2524‐9. 81. Corallo AN, Croxford R, Goodman DC, Bryan EL, Srivastava D, Stukel TA. A systematic review of medical practice variation in OECD countries. Health Policy 2014; 1;114(1):5‐14. 82. Ilic D, Neuberger MM, Djulbegovic M, Dahm P. Screening for prostate cancer. Cochrane Database Syst Rev 2013; Jan 31;1:CD004720. 83. Schroder FH, Hugosson J, Roobol MJ, Tammela TL, Zappa M, Nelen V, et al. Screening and prostate cancer mortality: results of the European Randomised Study of Screening for Prostate Cancer (ERSPC) at 13 years of follow‐up. Lancet 2014; Aug 6;. 84. Hewitson P, Glasziou P, Watson E, Towler B, Irwig L. Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (Hemoccult): An update. Am J Gastroenterol 2008;103(6):1541‐9. 85. Schoen RE, Pinsky PF, Weissfeld JL, Yokochi LA, Church T, Laiyemo AO, et al. Colorectal‐cancer incidence and mortality with screening flexible sigmoidoscopy. N Engl J Med 2012;366(25):2345‐57. 86. Holme O, Loberg M, Kalager M, Bretthauer M, Hernan MA, Aas E, et al. Effect of flexible sigmoidoscopy screening on colorectal cancer incidence and mortality: a randomized clinical trial. JAMA 2014; Aug 13;312(6):606‐15. 87. Hermann Brenner, Christian Stock, Michael Hoffmeister. Effect of screening sigmoidoscopy and screening colonoscopy on colorectal cancer incidence and mortality: systematic review and meta‐analysis of randomised controlled trials and observational studies. BMJ 2014; BMJ Publishing Group Ltd;348. 107 Clinical Activity in General Practice and Cancer
88. Atkin WS, Edwards R, Kralj‐Hans I, Wooldrage K, Hart AR, Northover JM, et al. Once‐only flexible sigmoidoscopy screening in prevention of colorectal cancer: a multicentre randomised controlled trial. The Lancet 2010;375(9726):1624‐33. 89. Welch HG, Albertsen PC. Prostate Cancer Diagnosis and Treatment After the Introduction of Prostate‐Specific Antigen Screening: 1986–2005. Journal of the National Cancer Institute 2009; October 07;101(19):1325‐9. 90. Mukai TO, Bro F, Pedersen KV, Vedsted P. Use of prostate‐specific antigen testing]. Ugeskr Laeger 2010; Mar 1;172(9):696‐700. 91. Baade PD, Youlden DR, Krnjacki LJ. International epidemiology of prostate cancer: geographical distribution and secular trends. Mol Nutr Food Res 2009; Feb;53(2):171‐84. 92. Bray F, Klint Å, Gislum M, Hakulinen T, Engholm G, Tryggvadóttir L, et al. Trends in survival of patients diagnosed with male genital cancers in the Nordic countries 1964–2003 followed up until the end of 2006. Acta Oncol 2010; 06/01; 2012/11;49(5):644‐54. 93. Brasso K, Ingimarsdottir IJ, Rusch E, Engholm G, Adolfsson J, Tryggvadottir L, et al. Differences in survival from prostate cancer in Denmark, Iceland and Sweden. Eur J Cancer 2013; May;49(8):1984‐92. 94. Finne P, Fallah M, Hakama M, Ciatto S, Hugosson J, de Koning H, et al. Lead‐time in the European Randomised Study of Screening for Prostate Cancer. Eur J Cancer 2010; 11;46(17):3102‐8. 95. Carter HB, Pearson JD, Metter EJ, Brant LJ, Chan DW, Andres R, et al. Longitudinal evaluation of prostate‐specific antigen levels in men with and without prostate disease. JAMA 1992; Apr 22‐29;267(16):2215‐20. 96. Thompson IM, Ankerst DP, Chi C, Lucia MS, Goodman PJ, Crowley JJ, et al. Operating characteristics of prostate‐specific antigen in men with an initial PSA level of 3.0 ng/ml or lower. JAMA 2005; Jul 6;294(1):66‐70. 97. Sharifi N, Kramer BS. Screening for prostate cancer: current status and future prospects. Am J Med 2007; Sep;120(9):743‐5. 98. Schroder FH, Hugosson J, Roobol MJ, Tammela TL, Ciatto S, Nelen V, et al. Prostate‐cancer mortality at 11 years of follow‐up. N Engl J Med 2012; Mar 15;366(11):981‐90. 108 References
99. Labrie F, Candas B, Cusan L, Gomez JL, Belanger A, Brousseau G, et al. Screening decreases prostate cancer mortality: 11‐year follow‐up of the 1988 Quebec prospective randomized controlled trial. Prostate 2004; May 15;59(3):311‐
8. 100. Andriole GL, Crawford ED, Grubb RL,3rd, Buys SS, Chia D, Church TR, et al. Mortality results from a randomized prostate‐cancer screening trial. N Engl J Med 2009; Mar 26;360(13):1310‐9. 101. Sandblom G, Varenhorst E, Löfman O, Rosell J, Carlsson P. Clinical Consequences of Screening for Prostate Cancer: 15 Years Follow‐up of a Randomised Controlled Trial in Sweden. Eur Urol 2004; 12;46(6):717‐24. 102. Kjellman A, Akre O, Norming U, Tornblom M, Gustafsson O. 15‐Year Followup of a Population Based Prostate Cancer Screening Study. J Urol 2009; Apr;181(4):1615,21; discussion 1621. 103. DACROPA. Available at: http://ducg.dk/prostata/retningslinjer. Accessed 09/05, 2013. 104. Jønler M, Eddy B, Poulsen J. Prostate‐specific antigen testing in general practice: A survey among 325 general practitioners in Denmark. Scand J Urol Nephrol 2005; 01/01; 2012/08;39(3):214‐8. 105. Karlsen RV, Larsen SB, Christensen J, Brasso K, Friis S, Tjonneland A, et al. PSA testing without clinical indication for prostate cancer in relation to socio‐
demographic and clinical characteristics in the Danish Diet, Cancer and Health Study. Acta Oncol 2013; Nov;52(8):1609‐14. 106. Borre M. Screening by lower urinary tract symptoms vs asymptomatic prostate‐specific antigen levels leading to radical prostatectomy in Danish men: tumour characteristics and treatment outcome. BJU Int 2009; Jul;104(2):205‐8. 107. Ferlay J, Shin HR, Bray F, Forman D, Mathers C, Parkin DM. Estimates of worldwide burden of cancer in 2008: GLOBOCAN 2008. Int J Cancer 2010; Dec 15;127(12):2893‐917. 108. <br />NORDCAN: Cancer Incidence, Mortality, Prevalence and Survival in the Nordic Countries, Version 4.0. Available at: http://www‐
dep.iarc.fr/NORDCAN/DK/frame.asp. 109. Klint A, Engholm G, Storm HH, Tryggvadottir L, Gislum M, Hakulinen T, et al. Trends in survival of patients diagnosed with cancer of the digestive 109 Clinical Activity in General Practice and Cancer
organs in the Nordic countries 1964‐2003 followed up to the end of 2006. Acta Oncol 2010; Jun;49(5):578‐607. 110. The Danish National Board of Health. Recommendations for colorectal cancer screening [report in Danish], available from http://sundhedsstyrelsen.dk/publ/Publ2012/09sep/ScreeningTarmkraeftAn
bef2udg.pdf. ; 2012. 111. Kronborg O, Fenger C, Olsen J, Jorgensen OD, Sondergaard O. Randomised study of screening for colorectal cancer with faecal‐occult‐blood test. Lancet 1996; Nov 30;348(9040):1467‐71. 112. Kewenter J, Brevinge H, Engaras B, Haglind E, Ahren C. Results of screening, rescreening, and follow‐up in a prospective randomized study for detection of colorectal cancer by fecal occult blood testing. Results for 68,308 subjects. Scand J Gastroenterol 1994; May;29(5):468‐73. 113. Mandel JS, Church TR, Ederer F, Bond JH. Colorectal cancer mortality: effectiveness of biennial screening for fecal occult blood. J Natl Cancer Inst 1999; Mar 3;91(5):434‐7. 114. The Danish Institute of Health Technology Assessment. Colorectal Cancer Screening: The impact of participation rate [report in Danish], available from http://sundhedsstyrelsen.dk/publ/Publ2008/MTV/screening_tarmkraeft/M
TV_tarmkraeft_net_final_version2.pdf. ; 2008. 115. Hamilton W. Five misconceptions in cancer diagnosis. Br J Gen Pract 2009; Jun;59(563):441,5, 447; discussion 446. 116. The Danish Colorectal Cancer Group. Annual report 2013 [Danish]. Available at: http://www.dccg.dk/03_Publikation/2013.pdf. Accessed 11/03, 2014. 117. Zauber AG, Winawer SJ, OʹBrien MJ, Lansdorp‐Vogelaar I, Van Ballegooijen M, Hankey BF, et al. Colonoscopic polypectomy and long‐term prevention of colorectal‐cancer deaths. N Engl J Med 2012;366(8):687‐96. 118. Segnan N, Armaroli P, Bonelli L, Risio M, Sciallero S, Zappa M, et al. Once‐
only sigmoidoscopy in colorectal cancer screening: Follow‐up findings of the italian randomized controlled trial ‐ SCORE. J Natl Cancer Inst 2011;103(17):1310‐
22. 110 References
119. Hoff G, Grotmol T, Skovlund E, Bretthauer M, Norwegian Colorectal Cancer Prevention Study Group. Risk of colorectal cancer seven years after flexible sigmoidoscopy screening: randomised controlled trial. BMJ 2009; May 29;338:b1846. 120. Iversen LH, Bulow S, Christensen IJ, Laurberg S, Harling H, Danish Colorectal Cancer Group. Postoperative medical complications are the main cause of early death after emergency surgery for colonic cancer. Br J Surg 2008; Aug;95(8):1012‐9. 121. McArdle CS, Hole DJ. Emergency presentation of colorectal cancer is associated with poor 5‐year survival. Br J Surg 2004; May;91(5):605‐9. 122. Pedersen CB. The Danish Civil Registration System. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):22‐5. 123. Pedersen CB, Gotzsche H, Moller JO, Mortensen PB. The Danish Civil Registration System. A cohort of eight million persons. Dan Med Bull 2006; Nov;53(4):441‐9. 124. Toft Sørensen H, Aarhus Universitet. Institut for Epidemiologi og Socialmedicin. Klinisk‐Epidemiologisk Afdeling. Use of medical databases in clinical epidemiology. [Aarhus]: Department of Clinical Epidemiology, Aarhus University Hospital; 2008. 125. Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient Register. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):30‐3. 126. International Classification of Diseases. Available at: http://www.who.int/classifications/icd/en/. Accessed 01/28, 2013. 127. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40(5):373‐83. 128. Sundararajan V, Quan H, Halfon P, Fushimi K, Luthi JC, Burnand B, et al. Cross‐national comparative performance of three versions of the ICD‐10 Charlson index. Med Care 2007; Dec;45(12):1210‐5. 129. Schmidt M, Jacobsen JB, Lash TL, Bøtker HE, Sørensen HT. 25 year trends in first time hospitalisation for acute myocardial infarction, subsequent short and long term mortality, and the prognostic impact of sex and comorbidity: a Danish nationwide cohort study. BMJ 2012; BMJ Publishing Group Ltd;344. 111 Clinical Activity in General Practice and Cancer
130. Gjerstorff ML. The Danish Cancer Registry. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):42‐5. 131. The Danish National Board of Health. Det moderniserede Cancerregister – metode og kvalitet [report in Danish] 2009;. 132. Bjerregaard B, Larsen OB. The Danish Pathology Register. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):72‐4. 133. Sobin LH, Gospodarowicz MK, Wittekind C, International Union against Cancer. TNM classification of malignant tumours. 7. ed. ed. Chichester, West Sussex, UK: Wiley‐Blackwell; 2010. 134. Sahl Andersen J, De Fine Olivarius N, Krasnik A. The Danish National Health Service Register. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):34‐7. 135. Quality Declarations. Available at: http://www.dst.dk/en/Statistik/dokumentation/Declarations.aspx. Accessed 10/20, 2014. 136. Petersson F, Baadsgaard M, Thygesen LC. Danish registers on personal labour market affiliation. Scand J Public Health 2011; Jul;39(7 Suppl):95‐8. 137. Timmermans B. The Danish Integrated Database for Labor Market Research: Towards Demystification for the English Speaking Audience. Aalborg University: Druid.dk; 2010. 138. What are equivalence scales? Available at: http://www.oecd.org/dataoecd/61/52/35411111.pdf. Accessed January, 2013. 139. Baadsgaard M, Quitzau J. Danish registers on personal income and transfer payments. Scand J Public Health 2011; Jul;39(7 Suppl):103‐5. 140. Jensen VM, Rasmussen AW. Danish Education Registers. Scand J Public Health 2011; Jul;39(7 Suppl):91‐4. 141. Thygesen LC, Daasnes C, Thaulow I, Brønnum‐Hansen H. Introduction to Danish (nationwide) registers on health and social issues: Structure, access, legislation, and archiving. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):12‐6. 112 References
142. Moth G, Olesen F, Vedsted P. Reasons for encounter and disease patterns in Danish primary care: Changes over 16 years. Scand J Prim Health Care 2012; 06/01; 2012/11;30(2):70‐5. 143. Moth G, Vedsted P, Olesen F. <br />Kontakt‐ og sygdomsmønsteret i almen praksis. KOS 2008. Aarhus: Forskningsenheden for Almen Praksis i Aarhus, Aarhus Universitet; 2010. 144. WONCA. International Classification of Primary Care (ICPC‐2‐R). Revised 2nd Edition ed.Oxford University Press; 2005. 145. Grann AF, Erichsen R, Nielsen AG, Froslev T, Thomsen RW. Existing data sources for clinical epidemiology: The clinical laboratory information system (LABKA) research database at Aarhus University, Denmark. Clin Epidemiol 2011; Apr 1;3:133‐8. 146. Helweg‐Larsen K. The Danish Register of Causes of Death. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):26‐9. 147. The Danish Colorectal Cancer Group. Annual report 2012 [Danish]. Available at: http://www.dccg.dk/pdf/Aarsrapport_2012_dccg.pdf. Accessed 09/04, 2014. 148. Gogenur I, Ingeholm P, Iversen LH. Danish Colorectal Cancer Database. Ugeskr Laeger 2012; Oct 15;174(42):2525. 149. Iversen LH. Aspects of survival from colorectal cancer in Denmark. Dan Med J 2012; Apr;59(4):B4428. 150. Nguyen‐Nielsen M, Froslev T, Friis S, Borre M, Harving N, Sogaard M. Completeness of prostate cancer staging in the Danish Cancer Registry, 2004‐
2009. Clin Epidemiol 2012;4 Suppl 2:17‐23. 151. Williams RL. A note on robust variance estimation for cluster‐correlated data. Biometrics 2000; Jun;56(2):645‐6. 152. Dickman PW, Sloggett A, Hills M, Hakulinen T. Regression models for relative survival. Stat Med 2004;23(1):51‐64. 153. Corazziari I, Quinn M, Capocaccia R. Standard cancer patient population for age standardising survival ratios. Eur J Cancer 2004; Oct;40(15):2307‐16. 113 Clinical Activity in General Practice and Cancer
154. Frank E. Harrell J. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. 1st ed. New York: Springer Science & Business Media; 2001. 155. Fletcher RH, Fletcher SW. Clinical epidemiology : the essentials. 4. ed. ed. Baltimore, Md.: Lippincott Williams & Wilkins; 2005. 156. Juul S,f.1940. Epidemiologi og evidens. 1. udgave ed. Kbh.: Munksgaard Danmark; 2004. 157. Salas M, Hofman A, Stricker BH. Confounding by indication: an example of variation in the use of epidemiologic terminology. Am J Epidemiol 1999; Jun 1;149(11):981‐3. 158. Torring ML. <br />Time from first presentation of symptoms in primary care until diagnosis of cancer: Association with mortality [PhD dissertation]. Aarhus: Research Unit and Department of General Practice, Faculty of Health Sciences, Aarhus University; 2011. 159. Jacob BJ, Moineddin R, Sutradhar R, Baxter NN, Urbach DR. Effect of colonoscopy on colorectal cancer incidence and mortality: An instrumental variable analysis. Gastrointest Endosc 2012;76(2):355,364.e1. 160. Terwee CB, Bot SD, de Boer MR, van der Windt DA, Knol DL, Dekker J, et al. Quality criteria were proposed for measurement properties of health status questionnaires. J Clin Epidemiol 2007; Jan;60(1):34‐42. 161. Johansen M, Holtedahl KA, Rudebeck CE. How does the thought of cancer arise in a general practice consultation? Interviews with GPs. Scand J Prim Health Care 2012; 09/01; 2012/11;30(3):135‐40. 162. Stolper E, van Bokhoven M, Houben P, Van Royen P, van dW, van dW, et al. The diagnostic role of gut feelings in general practice A focus group study of the concept and its determinants. BMC Family Practice 2009;10(1):17. 163. Madsen M, Davidsen M, Rasmussen S, Abildstrom SZ, Osler M. The validity of the diagnosis of acute myocardial infarction in routine statistics: A comparison of mortality and hospital discharge data with the Danish MONICA registry. J Clin Epidemiol 2003; 2;56(2):124‐30. 164. Vestergaard M, Obel C, Henriksen TB, Christensen J, Madsen KM, Ostergaard JR, et al. The Danish National Hospital Register is a valuable study 114 References
base for epidemiologic research in febrile seizures. J Clin Epidemiol 2006; Jan;59(1):61‐6. 165. Rix TA, Riahi S, Overvad K, Lundbye‐Christensen S, Schmidt EB, Joensen AM. Validity of the diagnoses atrial fibrillation and atrial flutter in a Danish patient registry. Scand Cardiovasc J 2012; Jun;46(3):149‐53. 166. Lund JL, Froslev T, Deleuran T, Erichsen R, Nilsson T, Pedersen AN, et al. Validity of the Danish National Registry of Patients for chemotherapy reporting among colorectal cancer patients is high. Clin Epidemiol 2013; Aug 30;5:327‐34. 167. Thomsen RW, Lange P, Hellquist B, Frausing E, Bartels PD, Krog BR, et al. Validity and underrecording of diagnosis of COPD in the Danish National Patient Registry. Respir Med 2011; Jul;105(7):1063‐8. 168. Christensen J, Vestergaard M, Olsen J, Sidenius P. Validation of epilepsy diagnoses in the Danish National Hospital Register. Epilepsy Res 2007; Jul;75(2‐
3):162‐70. 169. Djurhuus BD, Skytthe A, Faber CE. Validation of the cholesteatoma diagnosis in the Danish National Hospital Register. Dan Med Bull 2010; Oct;57(10):A4159. 170. Wermuth L, Lassen CF, Himmerslev L, Olsen J, Ritz B. Validation of hospital register‐based diagnosis of Parkinsonʹs disease. Dan Med J 2012; Mar;59(3):A4391. 171. Thygesen SK, Olsen M, Christian FC. Positive predictive value of the infant respiratory distress syndrome diagnosis in the Danish National Patient Registry. Clin Epidemiol 2013; Aug 12;5:295‐8. 172. Vest‐Hansen B, Riis AH, Christiansen CF. Registration of acute medical hospital admissions in the Danish National Patient Registry: a validation study. Clin Epidemiol 2013; Apr 26;5:129‐33. 173. Ben Ghezala I, Arendt JF, Erichsen R, Zalfani J, Gammelager H, Froslev T, et al. Positive predictive value of the diagnosis coding for vitamin B12 deficiency anemia in the Danish National Patient Register. Clin Epidemiol 2012;4:333‐8. 174. de Groot V, Beckerman H, Lankhorst GJ, Bouter LM. How to measure comorbidity: a critical review of available methods. J Clin Epidemiol 2003; 3;56(3):221‐9. 115 Clinical Activity in General Practice and Cancer
175. Needham DM, Scales DC, Laupacis A, Pronovost PJ. A systematic review of the Charlson comorbidity index using Canadian administrative databases: a perspective on risk adjustment in critical care research. J Crit Care 2005; Mar;20(1):12‐9. 176. Sarfati D. Review of methods used to measure comorbidity in cancer populations: No gold standard exists. J Clin Epidemiol 2012; Sep;65(9):924‐33. 177. Brilleman SL, Salisbury C. Comparing measures of multimorbidity to predict outcomes in primary care: a cross sectional study. Fam Pract 2013; Apr;30(2):172‐8. 178. Streiner DL. Breaking up is hard to do: the heartbreak of dichotomizing continuous data. Can J Psychiatry 2002; Apr;47(3):262‐6. 179. Campbell MK, Mollison J, Steen N, Grimshaw JM, Eccles M. Analysis of cluster randomized trials in primary care: a practical approach. Family Practice 2000; April 01;17(2):192‐6. 180. Fung V. Meaningful Variation in Performance A Systematic Literature Review. Med Care 2010; ‐02;48(2):140‐8. 181. Moore AT, Roland MO. How much variation in referral rates among general practitioners is due to chance?. BMJ 1989; Feb 25;298(6672):500‐2. 182. Scheel BI, Ingebrigtsen SG, Thorsen T, Holtedahl K. Cancer suspicion in general practice: the role of symptoms and patient characteristics, and their association with subsequent cancer. Br J Gen Pract 2013; Sep;63(614):627‐35. 183. Nilsson S, Scheike M, Engblom D, Karlsson LG, Molstad S, Akerlind I, et al. Chest pain and ischaemic heart disease in primary care. Br J Gen Pract 2003; May;53(490):378‐82. 184. Dalsgaard E, Vedsted P, Fenger‐Grøn M, Sandbaek A, Vestergaard M. Socioeconomic position and contact to general practice among persons with diabetes. Primary Care Diabetes 2012; 12;6(4):313‐8. 185. Rubin G, Vedsted P, Emery J. Improving cancer outcomes: better access to diagnostics in primary care could be critical. Br J Gen Pract 2011; May;61(586):317‐8. 186. Jones D.T. A survey of hospital outpatient referral rates, Wales, 1985 1987;295:734. 116 References
187. Ahmed J, Mehmood S, Khan S, Rao M. Direct access colonoscopy in primary care: is it a safe and practical approach?. Scottish Medical Journal 2013; August 01;58(3):168‐72. 188. Jemal A, Ward E, Wu X, Martin HJ, McLaughlin CC, Thun MJ. Geographic Patterns of Prostate Cancer Mortality and Variations in Access to Medical Care in the United States. Cancer Epidemiology Biomarkers & Prevention 2005; March 01;14(3):590‐5. 189. Kvåle R, Auvinen A, Adami H, Klint Å, Hernes E, Møller B, et al. Interpreting Trends in Prostate Cancer Incidence and Mortality in the Five Nordic Countries. Journal of the National Cancer Institute 2007; December 19;99(24):1881‐7. 190. Lu‐Yao G, Albertsen PC, Stanford JL, Stukel TA, Walker‐Corkery ES, Barry MJ. Natural experiment examining impact of aggressive screening and treatment on prostate cancer mortality in twofixed cohorts from Seattle area and Connecticut. BMJ 2002; BMJ Publishing Group Ltd;325(7367):740. 191. Bynum J, Song Y, Fisher E. Variation in prostate‐specific antigen screening in men aged 80 and older in fee‐for‐service Medicare. J Am Geriatr Soc 2010; Apr;58(4):674‐80. 192. Outzen M, Brasso K, Martinussen N, Christensen J, Tjonneland A, Friis S, et al. Prostate cancer in Denmark 1978‐2009‐‐trends in incidence and mortality. Acta Oncol 2013; May;52(4):831‐6. 193. Sandblom G, Varenhorst E, Rosell J, Löfman O, Carlsson P. Randomised prostate cancer screening trial: 20 year follow‐up. BMJ 2011; BMJ Publishing Group Ltd;342. 194. Hugosson J, Carlsson S, Aus G, Bergdahl S, Khatami A, Lodding P, et al. Mortality results from the Göteborg randomised population‐based prostate‐
cancer screening trial. The Lancet Oncology 2010; 8;11(8):725‐32. 195. Törnblom M, Eriksson H, Franzén S, Gustafsson O, Lilja H, Norming U, et al. Lead time associated with screening for prostate cancer. International Journal of Cancer 2004;108(1):122‐9. 196. Lu‐Yao G, Albertsen PC, Stanford JL, Stukel TA, Walker‐Corkery E, Barry MJ. Screening, treatment, and prostate cancer mortality in the Seattle area and Connecticut: fifteen‐year follow‐up. J Gen Intern Med 2008; Nov;23(11):1809‐14. 117 Clinical Activity in General Practice and Cancer
197. Jacob B, Baxter N, Moineddin R, Sutradhar R, Del Giudice L, Urbach D. Social disparities in the use of colonoscopy by primary care physicians in Ontario. BMC Gastroenterology 2011;11(1):102. 198. Shaukat A, Mongin SJ, Geisser MS, Lederle FA, Bond JH, Mandel JS, et al. Long‐term mortality after screening for colorectal cancer. N Engl J Med 2013; Sep 19;369(12):1106‐14. 199. Rabeneck L, Paszat LF, Saskin R, Stukel TA. Association between colonoscopy rates and colorectal cancer mortality. Am J Gastroenterol 2010; Jul;105(7):1627‐32. 200. Gross CP, Andersen MS, Krumholz HM, McAvay GJ, Proctor D, Tinetti ME. Relation between Medicare screening reimbursement and stage at diagnosis for older patients with colon cancer. JAMA 2006; Dec 20;296(23):2815‐22. 201. Baxter NN, Warren JL, Barrett MJ, Stukel TA, Doria‐Rose VP. Association between colonoscopy and colorectal cancer mortality in a US cohort according to site of cancer and colonoscopist specialty. Journal of Clinical Oncology 2012;30(21):2664‐9. 202. Cuzick J, Thorat MA, Andriole G, Brawley OW, Brown PH, Culig Z, et al. Prevention and early detection of prostate cancer. Lancet Oncol 2014; /10/1;15(11):e484‐92. 203. Rabeneck L, Paszat LF, Hilsden RJ, Saskin R, Leddin D, Grunfeld E, et al. Bleeding and perforation after outpatient colonoscopy and their risk factors in usual clinical practice. Gastroenterology 2008; Dec;135(6):1899,1906, 1906.e1. 204. Schang L, Morton A, DaSilva P, Bevan G. From data to decisions? Exploring how healthcare payers respond to the NHS Atlas of Variation in Healthcare in England. Health Policy 2014; Jan;114(1):79‐87. 118 Appendix I
APPENDIX I REGISTRATION FORM FOR GPS IN THE KOS 2008 SURVEY 119 Bilag 2
Lægeregistreringsskema
Baggrundsoplysninger
-
1. CPR-nr:
Kendte kroniske lidelser
1.
2.
3.
Evt. ICPC-kode Kendte kroniske lidelser
4.
5.
6.
Evt. ICPC-kode
Patientens indgang til konsultationen (Sæt kryds ved A, B og/eller C)
A Alm. konsultation
(kun ét kryds)
Konsultation
Telefonkonsultation
Besøg
E-mailkonsultation
B Aftalte forebyggelseskonsultationer C Kontakt vedr. recept, attest mm.
Receptfornyelse pr. telefon/mail
Børneundersøgelse/vaccination
Social-medicinsk attest (LÆ)
Svangreundersøgelse
Sygemeldingsattest/fraværsmelding
Vaginal-cytologi (inviteret)
Motorattest
Vaccination (influenza, rejse mv.)
Forsikringsattest
Aftalt forebyggelseskons.
Anden attest
Klamydiapartneropsporing
Registreringen af B og C afsluttes her, MEDMINDRE der er en afledt
konsultation samme dag, dvs. problemer, der indebærer yderligere us. og lign.
Fortsæt
Fortsæt hvis der er en afledt konsultation
Indhold af kontakten
Evt. ICPC-kode
2 Vigtigste kontaktårsag:
3 Kontaktens hovedsymptom/-diagnose:
opfølgning
4 Er kontakten en ny episode
5 Indhold af kontakten (gerne flere krydser): 6 Henvisning til: diagnostik
Lægemiddelordination/receptudstedelse
Svar på undersøgelse mv.
Prøver/undersøgelser foretaget i klinikken
Forebyggelse/råd om:
Kost
Røg Alkohol
Motion
Andet
Biomed. problemer udover hoveddiag. Antal
og/eller behandling
(gerne flere krydser)
Speciallæge, speciale:_______________________
Hosp.amb.,
speciale:_______________________
Indlæggelse, speciale:_______________________
Fysioterapi
Ekstern lab.
Psykol. problemer udover hoveddiag. Antal
Billeddiagnostik: Røntgen
Henvisning til andet:
Sociale problemer udover hoveddiag. Antal
Kontakt til:
hj.pleje
CT
Anden
UL
sundhedspl.
7 Deltagelse af klinikpers:
med selvstændig del af konsultationen
ingen deltagelse
med hjælp
ja
nej
Kunne konsultationen have været foretaget af en konsultationssygepl./klinikpersonale?
8 Afslutning Aftalt opfølgning
Der ses an, henvendelse ved behov
Afsl. uden yderligere kontakt
Om kontakten
9 I hvor høj grad vurderer du som læge, at følgende faktorer spillede en rolle i kontakten? (Angiv i %)
Ved ikke
sociale?
%. psykologiske?
%.
%.
en
specifik
medicinsk
diagnose?
10 Er det din opfattelse, at den
endelige diagnose er:
en symptomdiagnose - forbigående?
(kun ét kryds)
en symptomdiagnose - længerevarende (funktionelt somatisk sygdom)
biomedicinske?
11 Overvejede du på noget tidspunkt kræft eller anden alvorlig sygdom (ny)?
Sidder du tilbage med den mindste mistanke om kræft eller anden alvorlig sygdom (ny)?
12 Tid benyttet i alt til kontakten
ja
ja
nej
nej
min. Vurder kontaktens tyngde fra 1 til 10 (1=let,10=tung)
Angiv evt. årsagen til den vurderede tyngde:
31623
89
Appendix II
APPENDIX II DIAGNOSES INCLUDED IN STUDY I 121 Clinical Activity in General Practice and Cancer
The 265 new diagnoses included in the Study I (58 definitely serious diseases for subanalyses are marked with *) ICD‐10 A030 A047 A152* A153* A162* A327 A403 A410 A415 A419 A469 A692 B659 D33 D421 D432 D459 D469* D487 D489 D629 D630 D659* D685 D860 E05 E050 E052 E059 E104 E105 E107 E108 E109 E113 E115 E116 E117 E118 E144 E259 E282 E662 E876 F001 F009 122 Diagnosis Shigellosis due to Shigella dysenteriae
Enterocolitis due to Clostridium difficile
Tuberculosis of lung, confirmed histologically
Tuberculosis of lung, confirmed by unspecified means
Tuberculosis of lung, without mention of bacteriological or histological confirmation Listerial sepsis Sepsis due to Streptococcus pneumonia
Sepsis due to Staphylococcus aureus
Sepsis due to other Gram‐negative organisms
Sepsis, unspecified Erysipelas
Lyme disease Schistosomiasis, unspecified
Benign neoplasm of brain and other parts of central nervous system
Neoplasm of uncertain or unknown behaviour: Spinal meninges
Neoplasm of uncertain or unknown behaviour: Brain, unspecified
Polycythaemia vera Myelodysplastic syndrome, unspecified
Neoplasm of uncertain or unknown behaviour: Other specified sites
Neoplasm of uncertain or unknown behaviour: Neoplasm of uncertain or unknown behaviour, unspecified Anaemia due to hemorrhage, acute
Anaemia in neoplastic disease
Disseminated intravascular coagulation [defibrination syndrome]
Primary Thrombophilia Sarcoidosis of lung Thyrotoxicosis [hyperthyroidism]
Thyrotoxicosis with diffuse goitre
Thyrotoxicosis with toxic multinodular goitre
Thyrotoxicosis, unspecified Insulin‐dependent diabetes mellitus: With neurological complications
Insulin‐dependent diabetes mellitus: With peripheral circulatory complications Insulin‐dependent diabetes mellitus: With multiple complications
Insulin‐dependent diabetes mellitus: With unspecified complications
Insulin‐dependent diabetes mellitus: Without complications
Non‐insulin‐dependent diabetes mellitus: With ophthalmic complications Non‐insulin‐dependent diabetes mellitus: With peripheral circulatory complications Non‐insulin‐dependent diabetes mellitus: With other specified complications Non‐insulin‐dependent diabetes mellitus: With multiple complications
Non‐insulin‐dependent diabetes mellitus: With unspecified complications Unspecified diabetes mellitus: With neurological complications
Adrenogenital disorder, unspecified
Polycystic ovarian syndrome
Extreme obesity with alveolar hypoventilation
Hypokalaemia Dementia in Alzheimer disease with late onset
Dementia in Alzheimer disease, unspecified
Appendix II
F011 F039 F059 F103 F104 F190 F239* F299 F321 F33 F331 F339 G049* G209 G219 G249 G40* G401* G402* G403* G409* G459* G562 G576 G900 G91 G911 G930 G952 H200 H260 H301 H330 H331 H333 H350 H353 H356 H403 H431 H492 H810 I062 I080 I120 I200* I208* Multi‐infarct dementia
Dementia Delirium, unspecified
Mental and behavioural disorders due to use of alcohol: Withdrawal state Mental and behavioural disorders due to use of alcohol: Withdrawal state with delirium
Mental and behavioural disorders due to multiple drug use and use of other psychoactive substances: Acute intoxication Acute and transient psychotic disorder, unspecified
Unspecified nonorganic psychosis
Moderate depressive episode
Recurrent depressive disorder
Recurrent depressive disorder, current episode moderate
Recurrent depressive disorder, unspecified
Encephalitis, myelitis and encephalomyelitis, unspecified
Parkinson disease Secondary parkinsonism, unspecified
Dystonia, unspecified
Epilepsy Localization‐related (focal)(partial) symptomatic epilepsy and epileptic syndromes with simple partial seizures Localization‐related (focal)(partial) symptomatic epilepsy and epileptic syndromes with complex partial seizures Generalized idiopathic epilepsy and epileptic syndromes
Epilepsy, unspecified
Transient cerebral ischaemic attack, unspecified
Lesion of ulnar nerve
Lesion of plantar nerve
Idiopathic peripheral autonomic neuropathy
Hydrocephalus Obstructive hydrocephalus
Cerebral cysts Cord compression, unspecified
Acute and subacute iridocyclitis
Infantile, juvenile and presenile cataract
Disseminated chorioretinal inflammation
Retinal detachment with retinal break
Retinoschisis and retinal cysts
Retinal breaks without detachment
Background retinopathy and retinal vascular changes
Degeneration of macula and posterior pole
Retinal haemorrhage
Glaucoma secondary to eye trauma
Vitreous haemorrhage
Sixth [abducent] nerve palsy
MÚniÞre disease Rheumatic aortic stenosis with insufficiency
Disorders of both mitral and aortic valves
Hypertensive renal disease with renal failure
Unstable angina Other forms of angina pectoris
123 Clinical Activity in General Practice and Cancer
I209* I211* I213* I214* I219* I249* I259 I269* I301* I311 I313 I340 I350 I352 I420* I429* I441 I442* I460 I461 I469 I470* I471 I472* I479 I490* I495* I500* I501* I509 I609* I611* I612* I613* I619* I620* I63* I633* I639* I64* I649* I652 I669 I714 I719 I743 I744 I850* J040 J139 124 Angina pectoris, unspecified
Acute transmural myocardial infarction of inferior wall
Acute transmural myocardial infarction of unspecified site
Acute subendocardial myocardial infarction
Acute myocardial infarction, unspecified
Acute ischaemic heart disease, unspecified
Chronic ischaemic heart disease, unspecified
Pulmonary embolism without mention of acute cor pulmonale
Infective pericarditis Chronic constrictive pericarditis
Pericardial effusion (noninflammatory)
Mitral (valve) insufficiency Aortic (valve) stenosis Aortic (valve) stenosis with insufficiency
Dilated cardiomyopathy Cardiomyopathy, unspecified
Atrioventricular block, second degree
Atrioventricular block, complete
Cardiac arrest with successful resuscitation
Sudden cardiac death, so described
Cardiac arrest, unspecified Re‐entry ventricular arrhythmia
Supraventricular tachycardia
Ventricular tachycardia Paroxysmal tachycardia, unspecified
Ventricular fibrillation and flutter
Sick sinus syndrome Congestive heart failure Left ventricular failure Heart failure, unspecified Subarachnoid haemorrhage, unspecified
Intracerebral haemorrhage in hemisphere, cortical
Intracerebral haemorrhage in hemisphere, unspecified
Intracerebral haemorrhage in brain stem
Intracerebral haemorrhage, unspecified
Subdural haemorrhage (acute)(nontraumatic)
Cerebral infarction Cerebral infarction due to thrombosis of cerebral arteries
Cerebral infarction, unspecified
Stroke, not specified as haemorrhage or infarction
Stroke, not specified as haemorrhage or infarction
Occlusion and stenosis of carotid artery
Occlusion and stenosis of unspecified cerebral artery
Abdominal aortic aneurysm, without mention of rupture
Aortic aneurysm of unspecified site, without mention of rupture
Embolism and thrombosis of arteries of lower extremities
Embolism and thrombosis of arteries of extremities, unspecified
Oesophageal varices with bleeding
Acute laryngitis Pneumonia due to Streptococcus pneumoniae
Appendix II
J158 J159 J188 J189 J369 J819 J851 J869 J939 J960 J969 K046 K113 K222 K254 K290* K35* K351* K359* K430 K501 K509 K562 K566 K567 K625 K673 K703 K800 K801 K819 K830* K859* K861 K920 K922 M009* M059 M060 M069 M073 M353 M472 M480 M501 M511 M512 M710 M808 M809 Other bacterial pneumonia
Bacterial pneumonia, unspecified
Other pneumonia, organism unspecified
Pneumonia, unspecified
Peritonsillar abscess Pulmonary oedema Abscess of lung with pneumonia
Pyothorax without fistula
Pneumothorax, unspecified
Acute respiratory failure
Respiratory failure, unspecified
Periapical abscess with sinus
Abscess of salivary gland
Oesophageal obstruction
Gastric ulcer: Chronic or unspecified with haemorrhage
Acute haemorrhagic gastritis
Acute appendicitis Acute appendicitis with peritoneal abscess
Acute appendicitis, other and unspecified
Ventral hernia with obstruction, without gangrene
Crohn disease of large intestine
Crohn disease, unspecified
Volvulus Other and unspecified intestinal obstruction
Ileus, unspecified Haemorrhage of anus and rectum
Tuberculous peritonitis
Alcoholic cirrhosis of liver
Calculus of gallbladder with acute cholecystitis
Calculus of gallbladder with other cholecystitis
Cholecystitis, unspecified
Cholangitis Acute pancreatitis, unspecified
Other chronic pancreatitis
Haematemesis Gastrointestinal haemorrhage, unspecified
Pyogenic arthritis, unspecified
Seropositive rheumatoid arthritis, unspecified
Seronegative rheumatoid arthritis
Rheumatoid arthritis, unspecified
Other psoriatic arthropathies
Polymyalgia rheumatica
Other spondylosis with radiculopathy
Spinal stenosis Cervical disc disorder with radiculopathy
Lumbar and other intervertebral disc disorders with radiculopathy Other specified intervertebral disc displacement
Abscess of bursa Other osteoporosis with pathological fracture
Unspecified osteoporosis with pathological fracture
125 Clinical Activity in General Practice and Cancer
N109* N130 N133 N178* N179* N180 N188 N189 N19 N199 N700 N719 N804 N835 O001 O009* O031 O14 O140 O149 O200 O211 O350 O359 O36 O360 O361 O363 O365 O472 O680 O681 O683 O911 O993 Q211 Q612 R100 R339 R392 R402* R570* R630 R634 R649 R989 R999 S024 S065 126 Acute pyelonephritis Hydronephrosis with ureteropelvic junction obstruction
Other and unspecified hydronephrosis
Other acute renal failure Acute renal failure, unspecified
End stage kidney disease Chronic kidney disease, others
Chronic kidney disease, unspecified
Unspecified kidney failure Unspecified kidney failure Acute salpingitis and oophoritis
Inflammatory disease of uterus, unspecified
Endometriosis of rectovaginal septum and vagina
Torsion of ovary, ovarian pedicle and fallopian tube
Tubal pregnancy Ectopic pregnancy, unspecified
Spontaneous abortion: Incomplete, complicated by delayed or excessive haemorrhage Gestational [pregnancy‐induced] hypertension with significant proteinuria Moderate pre‐eclampsia Pre‐eclampsia, unspecified Threatened abortion Hyperemesis gravidarum with metabolic disturbance
Maternal care for (suspected) central nervous system malformation in fetus Maternal care for (suspected) fetal abnormality and damage, unspecified Maternal care for other known or suspected fetal problems
Maternal care for rhesus isoimmunisation
Maternal care for other isoimmunisation
Maternal care for signs of fetal hypoxia
Maternal care for poor fetal growth
False labour before 37 completed weeks of gestation
Labour and delivery complicated by fetal heart rate anomaly
Labour and delivery complicated by meconium in amniotic fluid
Labour and delivery complicated by biochemical evidence of fetal stress Abscess of breast associated with childbirth
Mental disorders and diseases of the nervous system complicating pregnancy, childbirth and the puerperium Atrial septal defect Polycystic kidney, autosomal dominant
Acute abdomen Urine retention Extrarenal uraemia Coma, unspecified Cardiogenic shock Anorexia Abnormal weight loss Cachexia Unattended death Other ill‐defined and unspecified causes of mortality
Fracture of malar and maxillary bones
Traumatic subdural haemorrhage
Appendix II
S068 S320 S321 S325 S641 S651 S720 S721 S722 S728 S821 S823 S824 S825 S826 S827 S828 S829 T261 T400 T403 T459 T783* Other intracranial injuries
Fracture of lumbar vertebra
Fracture of sacrum Fracture of pubis Injury of median nerve at wrist and hand level
Injury of radial artery at wrist and hand level
Fracture of neck of femur
Pertrochanteric fracture
Subtrochanteric fracture
Fractures of other parts of femur
Fracture of upper end of tibia
Fracture of lower end of tibia
Fracture of fibula alone
Fracture of medial malleolus
Fracture of lateral malleolus
Multiple fractures of lower leg
Fractures of other parts of lower leg
Fracture of lower leg, part unspecified
Burn of cornea and conjunctival sac
Poisoning: Opium Poisoning: Methadone
Poisoning: Primarily systemic and haematological agent, unspecified Angioneurotic oedema
127 Clinical Activity in General Practice and Cancer
128 Paper I
PAPER I 129 Research
Peter Hjertholm, Grete Moth, Mads Lind Ingeman and Peter Vedsted
Predictive values of GPs’ suspicion
of serious disease:
a population-based follow-up study
Abstract
Background
Knowledge is sparse on the prevalence of
suspicion of cancer and other serious diseases in
general practice. Likewise, little is known about
the possible implications of this suspicion on
future healthcare use and diagnoses.
Aim
To study the prevalence of GPs’ suspicions of
cancer or other serious diseases and analyse
how this suspicion predicted the patients’
healthcare use and diagnoses of serious disease.
Design and setting
Prospective population-based cohort study of
4518 patients consulting 404 GPs in a mix of
urban, semi-urban and rural practices in Central
Denmark Region during 2008–2009.
Method
The GPs registered consultations in 1 work
day, including information on their suspicion
of the presence of cancer or another serious
disease. The patients were followed up for use of
healthcare services and new diagnoses through
the use of national registers.
Results
Prevalence of suspicion was 5.7%. Suspicion
was associated with an increase in referrals
(prevalence ratio [PR] = 2.56, 95% confidence
interval [CI] = 2.22 to 2.96), especially for
diagnostic imaging (PR = 3.95, 95% CI = 2.80 to
5.57), increased risk of a new diagnosis of cancer
or another serious disease within 2 months
(hazard ratio [HR] = 2.98, 95% CI = 1.93 to 4.62) —
especially for cancer (HR = 7.55, 95% CI = 2.66 to
21.39) — and increased use of general practice
(relative risk [RR] = 1.14, 95% CI = 1.06 to 1.24) and
hospital visits (RR = 1.90, 95% CI = 1.62 to 2.23).
The positive predictive value of a GP suspicion
was 9.8% (95% CI = 6.4 to 14.1) for cancer or
another serious disease within 2 months.
Conclusion
A GP suspicion of serious disease warrants
further investigation, and the organisation of the
healthcare system should ensure direct access
from the primary sector to specialised tests.
Keywords
Denmark; diagnosis; general practice;
neoplasm; referral and consultation.
e346 British Journal of General Practice, June 2014
INTRODUCTION
General practice forms the first line of
the healthcare system.1 When patients
present with symptoms and signs in
general practice, the positive predictive
values (PPVs) of serious disease are low
(often <5%), whereas the frequency of
‘low-risk-but-not-no-risk’ symptoms and
signs is high.2–7 This fundamental conflict
constitutes a major clinical challenge for
GPs and for the organisation of the entire
healthcare system.
In a Norwegian study, warning signs
of cancer were identified in 12.4% of GP
consultations and, among these, the
GPs suspected 24% to have cancer.8,9
This indicates that GPs do not always
use specific ‘alarm’ symptoms to identify
serious disease and start a diagnostic
process.9,10 A Danish study found that
approximately half of patients with cancer
(depending on cancer type) did not present
with alarm symptoms.11 This important
group of patients was not investigated
in, for example, the Norwegian studies;
knowledge is still lacking, therefore, about
how often GPs suspect serious disease
among all patients. Nylenna found that
a suspicion of cancer prompted GPs to
initiate further investigation in 4.2% of
patients in general practice of whom 7.8%
were later diagnosed with cancer.12,13
However, Nylenna’s study, and similar
studies, may be influenced by the
Hawthorne effect, and GPs’ awareness of
cancer could be influenced by the study
P Hjertholm, MD, PhD student, ML Ingeman,
MD, GP, PhD student; P Vedsted, MD, PhD,
professor, Research Unit for General Practice,
Aarhus, Denmark and Research Centre for Cancer
Diagnosis in Primary Care, Aarhus, Denmark.
G Moth, MHSc, PhD, postdoctoral fellow, Research
Unit for General Practice, Aarhus, Denmark.
Address for correspondence
Peter Hjertholm, Research Unit for General
Practice, Aarhus University, Bartholins Alle 2,
itself. How often GPs suspect cancer and
serious disease in daily practice among
all patients must be acknowledged; and
this knowledge should not be confounded
by the awareness that researchers are
looking for specific diseases. Further, it
is essential to know how GPs act when a
suspicion of serious disease is raised and
how such suspicion may predict serious
diagnoses. This knowledge is crucial in
order to optimise support for GPs when
patients are suspected of having a serious
disease.
The aims of this study were to:
• describe how often GPs in Denmark
suspect cancer or other serious disease
after a consultation;
• characterise the patients in whom
suspicion was raised;
• describe how the GPs acted on their
suspicion; and
• analyse how a suspicion may influence
the demand for healthcare services and
predict a future diagnosis of serious
disease.
METHOD
Study design
All 845 GPs serving approximately 1.2 million
inhabitants in the Central Denmark Region
were invited to participate in the KOS 2008
study, a survey on reasons for encounter
and disease patterns in Danish general
practice.14 During the 12-month period
Aarhus C, 8000, Denmark.
E-mail: [email protected]
Submitted: 24 October 2013; Editor’s response:
26 November 2013; final acceptance: 27 January
2014.
©British Journal of General Practice
This is the full-length article (published online
27 May 2014) of an abridged version published in
print. Cite this article as: Br J Gen Pract 2014;
DOI: 10.3399/bjgp14X680125.
How this fits in
Patients eligible for a diagnostic fast
track (for example for cancer) must have
specific symptoms. However, only half
of the patients who are later diagnosed
with cancer initially present with ‘alarm’
symptoms in general practice. This study
shows that GPs suspect cancer or another
serious disease, on average, once per
day. Such suspicion is associated with the
need for referral to specialised testing, a
hazard ratio of 7.6, and a positive predictive
value of 2.3% of cancer within 2 months.
Therefore, the healthcare system must
support the diagnostic work-up on patients
referred from general practice when
serious disease is suspected as the PPV
resembles that of alarm symptoms.
from December 2008 to December 2009,
participating GPs were randomly assigned
a work day on which they had to record all
patient contacts. The GPs received payment
for their participation (€32) and for each
registered contact (€3).
Data
The registration form included a range of
questions addressing the following themes:
• basic clinical information on the patient;
• chronic diseases;
• type of contact;
• reason for encounter;
• content of the contact;
• actions taken (referrals, clinic tests, and
follow-up appointments); and
• the question: ‘Are you left with the
slightest suspicion of cancer or another
serious disease (new)?’.
Reasons for encounter and diagnoses
were written in text or stated by codes using
the International Classification of Primary
Care (ICPC-2).15 Diagnoses in text were
coded subsequently by an experienced
medical student, who was trained in ICPC
coding. All codes were subsequently
validated by one of the authors. Information
on the number of chronic diseases was
collected from the registration form and
categorised as: 0, 1–2, or ≥3.
Outcome
The unique personal identification number
assigned to all Danish citizens at birth
enabled linkage of information from various
national health registries. The Danish
National Registry of Patients 16 was used
to register all new, serious, hospital-based
diagnoses and use of hospital services
for each patient during the 6 months
following the index consultation. Diagnoses
were coded using the tenth revision of
the International Classification of Diseases
(ICD-10).17 For each person only incident
diagnoses were included, so that diagnoses
registered between January 2000 and the
index consultation were excluded.
Serious diseases other than cancer were
defined by independently reviewing all new
diagnoses (four-digit code, for example,
A415) for the patients while blinded to the
GP registrations. Disagreements were
discussed and consensus reached. Nonmelanoma skin cancers (DC44) were
not included as they differ substantially
regarding treatment and prognosis
compared to other cancers. The full list of
included serious diseases is available from
the author. To test the possible effect on
PPVs and hazard ratios (HRs) of adding lessserious diseases as outcomes, sensitivity
analyses were performed including only
definite serious diseases.
Hospital services included inpatient stays,
outpatient visits, diagnostic imaging, and
endoscopies (gastroscopy, colonoscopy,
and sigmoidoscopy). Data on diagnostic
imaging and endoscopic investigations
performed by primary-care specialists, as
well as use of general practice and practising
specialists, were obtained from the
Danish National Health Insurance Service
Registry.18 Contacts to general practice
were defined as face-to-face consultations,
including home visits. Contacts to
practising specialists included all contacts
to practising specialists in dermatology,
neurology, surgery, gynaecology, psychiatry,
otorhinolaryngology, and internal medicine.
Data on sociodemographic variables from
Statistics Denmark were included on marital
status (married/cohabiting or living alone)
and labour-market status (working, retired/
receiving pension, or out of the workforce
[unemployment, incapacity, or sickness]).
Income was defined as the Organisation for
Economic Cooperation and Developmentadjusted household income for the year of
the consultation, adjusted for number of
persons in the household and divided into
quartiles based on included patients.19,20
Statistical analyses
Associations between patient characteristics
and suspicion of serious disease were
analysed using generalised linear models
(GLMs) from the binomial family, with either
identity link (for prevalence differences)
British Journal of General Practice, June 2014 e347
Figure 1. Flow chart of patient inclusion.
or logarithmic link (for prevalence ratios).
Robust variance estimation accounting for
clustering at GP level was performed. A
similar GLM model was used to investigate
associations between suspicion and actions
taken during the consultation.
Cox proportional hazard models,
with time to diagnosis as the outcome
variable, were used to calculate the risk
of being diagnosed with a new serious
disease in the period from the index
consultation until 2 months later (61 days)
and 2–6 months later (62–183 days).
Patients were censored at the date of
diagnosis, at death, or 6 months after the
consultation, whichever came first. The use
of healthcare services was compared using
a GLM, including adjustment for the use
of each service in the year preceding the
consultation (dichotomous). The outcome
was dichotomised into consultation or no
consultation. Use of GP was analysed both
as a continuous and dichotomous variable
to test for consistency of results.
Multivariate analyses were adjusted
for age group (18–39 years, 40–54 years,
55–69 years, ≥70 years), sex, marital status,
income, chronic diseases (0, 1–2, ≥3), and
risk time. The Cox regression was adjusted
for age as a continuous variable, with sex
and chronic diseases coded as dichotomous
variables (0, ≥1). All analyses were carried
Eligible patients, n = 16 680
Staff contact only, n = 3548
Eligible patients, n = 13
16680
132
Telephone, email contacts and
home visits, n = 5537
Eligible patients, n = 7595
Contacts due to prophylaxis or
certificates, n = 1353
Eligible patients, n = 6242
<18 years, n = 1192
Eligible patients, n = 5050
No information about the personal
registration number, n = 525
Eligible patients, n = 4525
Duplicates, n = 5
Eligible patients, n = 4520
Patients died, n = 2
Eligible patients, n = 4518
e348 British Journal of General Practice, June 2014
= the study population
out using the statistical software Stata
(version 12.1).
RESULTS
A total of 404 of the 845 invited GPs
participated. The proportion of female GPs
among participating GPs was higher than in
the Central Denmark Region (44.6% versus
38.9%, P = 0.002) and the proportion of
GPs with >20 years experience was lower
among participating GPs than in Central
Denmark Region (20.1% versus 25.5%, P
= 0.007). 14
In total, 16680 contacts were registered
(Figure 1); of these, 4518 patients had a
face–to-face consultation with a GP, were
aged ≥18 years and were registered with
their complete information on a personal
identification number. GPs suspected
cancer or another serious disease after the
consultation in 256 (5.7%, 95% confidence
interval [CI] = 5.0 to 6.4) encounters. In 191
(4.2%) of the consultations, this information
was missing. To preserve statistical power,
the study included these as having no
suspicion.
The prevalence of suspicion was highest
among males, older patients, patients with
chronic diseases, and retired individuals
(Table 1). There was no association with
patient income. The effect of age remained
statistically significant when associated with
suspicion in the multivariate analysis (Table
1). The prevalence of suspicion was highest
among patients presenting symptoms from
blood, blood-forming organs, digestive
organs, and female genital organs (Table
1). The suspicion was lowest in patients
presenting with psychological problems
or symptoms from the musculoskeletal,
endocrine, and cardiovascular systems.
The overall probability of being referred
was 2.56 times higher in patients for whom
serious disease was suspected, ranging
from 1.45 for other referrals to 3.95 for
diagnostic imaging (Table 2). Suspicion
increased the probability of having a test
conducted in the GP clinic and increased
the likelihood of a follow-up appointment
(Table 2).
The risk of a diagnosis of cancer or
another serious disease was higher
within the first 2 months after the index
consultation for patients with a suspicion
compared with those with no suspicion
(adjusted HR = 2.98, 95% CI = 1.93 to 4.62)
and remained increased during 2–6 months
after the consultation (HR = 1.52, 95% CI =
0.92 to 2.53, P = 0.103) (Table 2). This
pattern was the same when stratified into
cancer and other serious diseases. The
PPV of GPs’ suspicion for later diagnosis of
bleeding and endometrial cancer (data not
shown). Among the 279 patients with a later
serious diagnosis, in whom the GP had not
suspected serious disease, 62 (22.2%, 95%
CI = 17.7 to 27.5) had a diagnosis related to
their reason for encounter (data not shown) .
Inclusion of only definite serious diseases
in the analysis resulted in a PPV of 5.5%
serious disease or cancer was 9.8% (95% CI
= 6.4 to 14.1) within the first 2 months after
the index consultation (Table 3).
Of the 42 persons in whom the GP’s
suspicion was confirmed, 22 (52.4%,
95% CI = 37.7 to 66.6) had a reason for
encounter clearly related to the subsequent
diagnosis, for example, intermenstrual
Table 1. Patient characteristics and GPs’ suspicions of serious disease after consultation
All contacts
Characteristic
n
Prevalence of suspicion,a
n (%)
Prevalence
difference,
% (95% CI)
Crude
prevalence ratio
(95% CI)
Adjustedb
prevalence ratio
(95% CI)
All
4518
256 (5.7)
Age, yearsc
18–39 1367
38 (2.8)
Ref
1
1
40–54
1157
48 (4.2)
1.4 (0.0 to 2.8)
1.49 (0.98 to 2.27)
1.50 (0.93 to 2.41)
55–69
1153
90 (7.8)
5.0 (3.3 to 6.8)
2.81 (1.94 to 4.07)
2.73 (1.73 to 4.30)
≥70
841
80 (9.5)
6.7 (4.6 to 8.9)
3.42 (2.35 to 4.99)
3.07 (1.98 to 4.76)
Sex Female 2802
141 (5.0)
Ref
1
1
Male
1716
115 (6.7)
1.7 (0.2 to 3.1)
1.33 (1.05 to 1.69)
1.24 (0.98 to 1.56)
Marital status
Married/cohabiting
3007
154 (5.1)
Ref
1
1
Living alone
1499
101 (6.7)
1.6 (0.1 to 3.1)
1.32 (1.03 to 1.68)
1.30 (0.98 to 1.72)
Labour–market status
Working 2643
105 (4.0)
Ref
1
1
Retired/receiving pension 1600
136 (8.5)
4.5 (3.0 to 6.1)
2.14 (1.67 to 2.74)
1.24 (0.86 to 1.78)
Out of the workforce
263
14 (5.3)
1.4 (–1.5 to 4.2)
1.34 (0.78 to 2.31)
1.59 (0.93 to 2.73)
Income, quartiles
Lowest
1126
76 (6.8)
Ref
1
1
Second 1127
60 (5.3)
–1.4 (–3.4 to 0.5)
0.79 (0.57 to 1.09)
0.81 (0.59 to 1.13)
Third 1126
60 (5.3)
–1.4 (–3.4 to 0.6)
0.79 (0.57 to 1.10)
0.91 (0.66 to 1.26)
Highest 1127
59 (5.2)
–1.5 (–3.5 to 0.4)
0.78 (0.56 to 1.08)
0.81 (0.57 to 1.15)
Number of chronic diseases
0
1994
84 (4.2)
Ref
1
1
1 or 2
2085
134 (6.4)
2.2 (0.8 to 3.6)
1.53 (1.17 to 1.99)
1.08 (0.83 to 1.40)
1.17 (0.75 to 1.82)
≥’3
439
38 (8.7)
4.4 (1.7 to 7.2)
2.05 (1.42 to 2.97)
Reason for encounter
A: General and unspecified
416
37 (8.9)
Ref
1
1
(ICPC–2)
B: Blood, blood forming organs
39
8 (20.5)
11.6 (–1.3 to 24.6)
2.31 (1.16 to 4.60)
2.45 (1.22 to 4.92)
D: Digestive 218
33 (15.1)
6.2 (0.8 to 11.7)
1.70 (1.10 to 2.64)
1.85 (1.19 to 2.89)
F: Eye
43
1 (2.3)
–6.6 (–11.8 to –1.3)
0.26 (0.04 to 1.86)
0.34 (0.05 to 2.37)
H: Ear
71
1 (1.4)
–7.5 (–11.4 to –3.6)
0.16 (0.02 to 1.14)
0.18 (0.03 to 1.30)
K: Cardiovascular
561
17 (3.0)
–5.9 (–8.9 to –2.8)
0.34 (0.19 to 0.60)
0.26 (0.15 to 0.44)
L: Musculoskeletal
819
27 (3.3)
–5.6 (–8.6 to –2.6)
0.37 (0.23 to 0.60)
0.43 (0.26 to 0.70)
N: Neurological
174
7 (4.0)
–4.9 (–8.9 to –0.9)
0.45 (0.21 to 0.99)
0.52 (0.23 to 1.15)
P: Psychological
501
7 (1.4)
–7.5 (–10.4 to –4.6)
0.16 (0.07 to 0.35)
0.21 (0.09 to 0.47)
R: Respiratory
492
37 (7.5)
–1.4 (–5.0 to 2.2)
0.85 (0.55 to 1.31)
0.91 (0.55 to 1.50)
S: Skin
464
28 (6.0)
–2.9 (–6.3 to 0.6)
0.68 (0.42 to 1.09)
0.86 (0.54 to 1.37)
T: Endocrine/metabolic
193
6 (3.1)
–5.8 (–9.5 to –2.1)
0.35 (0.15 to 0.81)
0.29 (0.13 to 0.63)
U: Urinary tract
136
9 (6.6)
–2.3 (–7.3 to 0.3)
0.74 (0.37 to 1.50)
0.73 (0.34 to 1.56)
W: Pregnancy, contraception
116
3 (2.6)
–6.3 (–10.3 to –2.3)
0.29 (0.09 to 0.93)
0.84 (0.25 to 2.87)
X: Female genital
187
25 (13.4)
4.5 (–1.1 to 10.1)
1.50 (0.93 to 2.42)
2.82 (1.69 to 4.72)
Y: Male genital
55
10 (18.2)
9.3 (–1.3 to 19.8)
2.04 (1.08 to 3.87)
1.84 (0.97 to 3.49)
Z: Social problems
33
0 (0)
—
—
—
Prevalence of suspicion: 60.2 (mean), 18–97 (min–max), 17.5 (SD). Adjusted for sex, age group, chronic disease group, income quartile, and clustering. Age: 51 years
a
b
c
(mean), 18–100 years (min–max), 18.6 (SD). ICPC–2 = International Classification of Primary Care, second edition. SD = standard deviation.
British Journal of General Practice, June 2014 e349
Table 2. GPs’ actions following consultation prompted by suspicion of serious disease
Referral
Suspicion
present
(n = 256), n (%)
No suspicion
(n = 4262),
n (%)
Univariate,
PR (95% CI) Multivariate,a
PR (95% CI)
All referrals
134 (52.3)
878 (20.6)
2.54 (2.23 to 2.90)
2.56 (2.22 to 2.96)
40 (15.6)
207 (4.9)
3.22 (2.35 to 4.40)
3.27 (2.34 to 4.56)
9 (3.5)
45 (1.1)
3.33 (1.65 to 6.74)
3.17 (1.54 to 6.50)
Outpatient clinic
Hospital admission
Primary care specialist
28 (10.9)
215 (5.0)
2.17 (1.49 to 3.15)
2.35 (1.65 to 3.33)
Diagnostic imaging
43 (16.8)
181 (4.2) 3.96 (2.91 to 5.38)
3.95 (2.80 to 5.57)
Othersb
25 (9.8)
299 (7.0)
1.39 (0.94 to 2.05)
1.45 (0.98 to 2.16)
Tests in GP clinic 140 (54.7)
1744 (40.9)
1.34 (1.19 to 1.50)
1.29 (1.16 to 1.44)
Follow-up in
Scheduled follow-up
162 (63.3)
2082 (48.9)
1.30 (1.18 to 1.42)
1.15 (1.05 to 1.26)
general practice
New contact if needed
33 (12.9)
876 (20.6)
0.63 (0.46 to 0.87)
0.69 (0.50 to 0.95)
No follow-up scheduled
43 (16.8) 1011 (23.7)
0.71 (0.54 to 0.93)
0.78 (0.59 to 1.03)
Missing
18 (7.0)
c
293 (6.9)
PR (prevalence ratio) adjusted for sex, age group, chronic disease group, income quartile, and clustering. bSuch as physiotherapist, laboratory, psychologist, or dentist.
a
Such as blood samples or urine analysis.
c
(95% CI = 3.0 to 9.0) and an HR of 4.69 (95%
CI = 2.51 to 8.75) after 2 months, and a PPV
of 7.4% (95% CI = 4.5 to 11.3) and an HR of
1.13 (95% CI = 0.45 to 2.85) after 6 months
(data not shown).
The proportions of patients using the
different healthcare services after the index
consultation are seen in Table 4. The use of
GP, primary-care specialist, and diagnostic
imaging increased, especially in the
2-month period after the index consultation.
Use of hospital services (inpatient and
outpatient) remained increased after
2 months. Results were not altered when
analysing the number of GP visits instead
of the proportion of patients who had a GP
consultation (data not shown).
Exclusion of individuals with missing
information on suspicion did not significantly
change any results.
DISCUSSION
Summary
In nearly 6% of all consultations in general
practice, the GP had a suspicion of cancer
or another serious disease. Higher age and
presentation of symptoms from the digestive
system, blood or blood-forming organs,
or female genitals were associated with
suspicion of serious disease. A suspicion
Table 3. Risk of serious disease after index consultation and predictive values of suspicion
Time after
index
Suspicion
No
consultation, Risk time, present, suspicion,
Univariate
Multivariatea
PPV
NPV
months
months
n
n
HR (95% CI)
HR (95% CI)
(95% CI)
(95% CI)
Prevalence,
%
All diagnoses of
0–2 8853.6
25
122
serious disease
3.54
(2.30 to 5.45)
2.98
(1.93 to 4.62)
9.8
(6.4 to 14.1)
97.2
(96.6 to 97.6)
3.2
2–6 17 060.5
17
157
2.00
(1.21 to 3.30)
1.52
(0.92 to 2.53)
16.4
(12.1 to 21.5)b
93.5
(92.7 to 94.2)b
7.1b
New diagnoses
0–2 8853.6
6
10
of cancer
10.42
(3.79 to 28.67)
7.55
(2.66 to 21.39)
2.3
(0.9 to 5.0)
99.8
(99.6 to 99.9)
0.4
2–6 17 060.5
2
12
3.08
(0.69 to 13.78)
1.82
(0.40 to 8.29)
3.1
(1.4 to 6.1)b
99.5
(99.2 to 99.7)b
0.7 b
New diagnoses of
0–2 8853.6
19
112
another serious disease
2.93
(1.80 to 4.77)
2.51
(1.53 to 4.11)
7.4
(4.5 to 11.3)
97.4
(96.9 to 97.9)
2.9
2–6 17 060.5
15
145
1.91
(1.12 to 3.25)
1.49
(0.87 to 2.54)
13.3
(9.4 to 18.1)b
94.0
(93.2 to 94.7)b
6.4b
HR adjusted for age (continuous), sex, and chronic disease (dichotomous). bPrevalence: 0–6 months. HR = hazard ratio. NPV = negative predictive value. PPV = positive
a
predictive value.
e350 British Journal of General Practice, June 2014
Table 4. Use of healthcare services after index consultation, by GP suspicion
Time after
consultation, months
Suspicion present (n = 256) n (%)
No suspicion (n = 4262) n (%)
Univariate
RR (95% CI)
Adjusteda
RR (95% CI)
GP consultation
0–2
172 (67.2)
2522 (59.2)
1.13 (1.04 to 1.24)
1.14 (1.06 to 1.24)
2–6
132 (51.6) 2159 (50.6) 1.03 (0.91 to 1.16)
0.99 (0.88 to 1.11)
Primary care specialists
0–2
43 (16.8)
406 (9.5)
1.77 (1.33 to 2.36)
1.73 (1.32 to 2.26)
2–6
31 (12.1)
425 (10.0)
1.23 (0.88 to 1.73)
1.20 (0.86 to 1.68)
All hospital visits
0–2
95 (37.1)
714 (16.8)
2.22 (1.87 to 2.64)
1.90 (1.62 to 2.23)
2–6
93 (36.3) 982 (23.0)
1.60 (1.35 to 1.90)
1.40 (1.19 to 1.63)
Hospital admission
0–2
43 (16.8)
258 (6.1)
2.78 (2.07 to 3.75)
1.98 (1.49 to 2.63)
2–6
43 (16.8)
354 (8.3)
2.05 (1.53 to 2.74)
1.62 (1.24 to 2.12)
Outpatient clinic visits
0–2
81 (31.6)
604 (14.2)
2.24 (1.84 to 2.72)
1.99 (1.66 to 2.40)
2–6
78 (30.5) 872 (20.5)
1.51 (1.25 to 1.84)
1.34 (1.11 to 1.62)
All diagnostic imagingb
0–2
66 (25.8)
472 (11.1)
2.33 (1.87 to 2.92)
1.71 (1.27 to 2.29)
2–6
55 (21.5)
679 (15.9)
1.37 (1.07 to 1.74)
1.11 (0.81 to 1.52)
X-rayb
0–2
40 (15.6)
243 (5.7)
2.75 (2.02 to 3.75)
2.08 (1.53 to 2.81)
2–6
34 (13.3)
335 (7.9)
1.71 (1.23 to 2.38)
1.29 (0.94 to 1.77)
Ultrasound 0–2
30 (11.7)
239 (5.6)
2.10 (1.47 to 3.00)
1.89 (1.32 to 2.70)
2–6
28 (10.9) 362 (8.5)
1.30 (0.91 to 1.88)
1.20 (0.84 to 1.72)
b
CT or MRI scanningb
0–2
23 (9.0)
72 (1.7)
5.33 (3.39 to 8.38)
3.76 (2.37 to 5.98)
2–6
20 (7.8) 143 (3.4)
2.36 (1.51 to 3.71)
1.51 (0.98 to 2.34)
Endoscopies
0–2
17 (6.6)
59 (1.4)
4.81 (2.85 to 8.13)
3.76 (2.22 to 6.38)
2–6
18 (7.0) 89 (2.1)
3.42 (2.09 to 5.58)
2.90 (1.77 to 4.75)
Adjusted for sex, age group, chronic disease group, income quartile and use of relevant variable in the preceding year (dichotomous). bProcedures performed at hospitals
a
and by practising specialists. CT = computerised tomography. MRI = magnetic resonance imaging. RR = relative risk.
increased the risk of having a test performed
or being referred for further investigation. In
particular, the use of diagnostic imaging
and endoscopies was increased after the
index consultation. The risk of receiving a
new diagnosis of cancer or other serious
disease increased, particularly within the
first 2 months after the index consultation,
and the PPV of GP suspicion was 9.8%
within the first 2 months and 16.4% within
6 months after the consultation.
Strengths and limitations
The major strengths of this study are the
prospective design and the high number
of consecutive consultations at randomly
assigned working days in a non-selected
group of patients. Further, the ability to
link the registered patients with national
registries is an additional strength. The
authors believe the risk of Hawthorne
effect is minimal because the question
on suspicion only formed a small part of
a larger registration form and GPs were
informed only that this study concerned
disease and symptoms patterns, along with
activities in general practice.
One limitation is that GPs did not
register the patient’s personal identification
numbers in 525 of the consultations (10.4%),
(Figure 1). This omission occurred among
specific GPs as a consequence of their
principles of confidentiality; it is likely that
these omissions did not bias the results.
Another limitation was missing
information about suspicion. These
contacts were included in the ‘no-suspicion’
group to preserve statistical power and
because the researchers considered it
plausible that GPs would rarely miss this
question if they did have a suspicion. The
sociodemographic characteristics of these
patients was similar to that of those in the
‘no-suspicion’ group and excluding patients
with missing data on suspicion from the
analyses did not alter the results as there
was no reason to believe that this principle
is associated with the association between
suspicion and future healthcare use and
diagnoses.
British Journal of General Practice, June 2014 e351
Funding
This project has been supported by Central
Denmark Region (Ref number 1-30-11006-v) and the former Aarhus County, (Ref
number 4-01-3-04), the Danish National
Research Foundation for Primary Care,
(Ref number 05/4309), special grants from
the Novo Nordisk Foundation, and the
Danish Cancer Society.
Ethical approval
The project was approved by the Danish
Data Protection Agency (J.no. 2008-41-2195
and J.no. 2009-41-3471) and by the Danish
Health and Medicines Agency (J.no. 7-60404-2/49/EHE). According to Danish law,
approval from the National Committee on
Health Research Ethics was not required as
no biomedical intervention was performed.
Provenance
Freely submitted; externally peer reviewed.
Competing interests
The authors have declared no competing
interests.
Acknowledgements
The authors would like to thank all the GPs
and the clinical staff who took part in this
study. We thank Statistics Denmark for
providing data and Morten Fenger-Grøn for
invaluable statistical support.
Discuss this article
Contribute and read comments about this
article: www.bjgp.org/letters
e352 British Journal of General Practice, June 2014
A further limitation was that 53.6% of
the invited GPs chose not to participate.
However, the researchers have no reason to
believe that this influences the association
between suspicion and subsequent
diagnosis.
Serious diseases diagnosed and managed
in general practice were not included as
serious disease was identified in the Danish
National Registry of Patients, which includes
only hospital diagnoses. However, it is only
rarely that serious disease of relevance to
this study does not include hospitalisation.
The authors did not know whether the
diagnosed serious disease was directly
associated with the recorded consultation,
but 52.4% (22 of 42) of the patients in whom
a suspicion was confirmed had a reason
for encounter that was related to the
subsequent diagnosis. It remains unknown
whether the 22% with a reason for encounter
that was related to the later diagnosis, but in
whom GPs had no suspicion, represent
neglected seriousness of disease as there
is no information on possible subsequent
consultations. This lack of information
on patient courses is also important to
consider when noting that for 87.0% (279 of
321, Table 3) those diagnosed with serious
disease, this was not suspected by the GPs.
Another possible source of information
bias was the lack of definition of serious
diseases. The number of serious diseases
defined and identified in the hospital registry
influenced the prevalence of serious disease
and, hence, caused the PPV to change
from 9.8% to 5.5% when including only
definite serious diseases. Nevertheless, this
is still of a magnitude that corresponds
with the most important alarm symptoms.
The effect on the HR of including more
diseases was less predictable because of
possible differential misclassifications. The
HRs increased from 2.98 to 4.69, indicating
that the inclusion of more diseases caused
an underestimation of the associations
between suspicion and serious disease.
Comparison with existing literature
The high frequency of symptoms of
potentially serious disease is challenging
for GPs if they are to identify patients for
further investigation.21 This is confirmed
by the findings of Ingebrigtsen et al and
Scheel et al, who found that warning signs
of cancer were present in 12.4%8 of all
GP consultations and that suspicion was
raised in only 24%9 of these consultations.
However, these figures cannot be compared
with the 5.7% found in this study as the
authors of the former studies only asked for
information on suspicion if a warning sign
was present, whereas this study inquired
whether there was a suspicion of cancer
or another serious disease for all patients
seen.8,9 The authors believes that this
proportion of 5.7% illustrates the actual
load of serious disease suspicion in GPs’
daily work.
In this study, the PPV of a GP suspicion
(9.8%) corresponds well with the PPVs
of cancer alarm symptoms and the 7.8%
found in Nylenna’s study;13 however, the
present study included various other
serious diseases as well as cancer. Shapley
et al reviewed the literature on PPVs of
cancer alarm symptoms and signs in
general practice and found that only nine
symptoms and signs had PPVs of more
than 5%.7 A specific level of PPV prompting
referral cannot be established, but studies
indicate that levels above 1% should prompt
investigation.4,7 As in the study by Scheel
et al, 9 6.2% (279 of 4518) of the patients in
the current study were diagnosed with a
serious disease without the GP having a
suspicion of one being present after the
index consultation.
The increased use of referrals and
diagnostic tests among patients where the
GP had a suspicion is in line with the findings
by Scheel et al. 9 This emphasises that
support for further investigation initiated by
GPs is crucial when a suspicion emerges.22
The importance of GPs’ suspicion has also
been highlighted by Hamilton and the fact
that most patients start their diagnostic
pathway for cancer in primary care.6,23
GPs depend on relevant secondary care
investigations in order to, most often, reject
that the cause of symptoms is cancer or
another serious disease.
Implications for practice
The present study confirms that action
should be taken when the GP suspects
serious disease; PPVs are relatively high,
and the healthcare system should support
this investigation by providing access to,
for example, imaging and endoscopies.
The UK and Denmark have organised
cancer investigation as a fast-track system
(for example, 2-week wait referrals) that
requires patients to present with specific
alarm symptoms to qualify for immediate
referral. However, as many patients in
general practice present with vague or
unspecific symptoms, GP access to relevant
and speedy diagnostic investigations
is crucial. Organisation of the primary
diagnostic pathways and how to support
GPs should be a main focus in future studies
in this area.
REFERENCES
1.
Pedersen KM, Andersen JS, Søndergaard J. General practice and primary
health care in Denmark. J Am Board Fam Med 2012; 25(Suppl 1): S34–S38.
2.
Nilsson S, Scheike M, Engblom D, et al. Chest pain and ischaemic heart disease
in primary care. Br J Gen Pract 2003; 53(490): 378–382.
3.
Jones R, Latinovic R, Charlton J, Gulliford MC. Alarm symptoms in early
diagnosis of cancer in primary care: cohort study using General Practice
Research Database. BMJ 2007; 334(7602): 1040.
4.
Astin M, Griffin T, Neal RD, et al. The diagnostic value of symptoms for colorectal
cancer in primary care: a systematic review. Br J Gen Pract 2011; 61(586):
231–243.
5.
Meineche-Schmidt V, Jorgensen T. ‘Alarm symptoms’ in patients with dyspepsia:
a three-year prospective study from general practice. Scand J Gastroenterol
2002 ; 37(9): 999–1007.
6.
Hamilton W. Five misconceptions in cancer diagnosis. Br J Gen Pract 2009;
59(563): 441–445, 447; discussion 446.
7.
Shapley M, Mansell G, Jordan JL, Jordan KP. Positive predictive values of ≥5%
in primary care for cancer: systematic review. Br J Gen Pract 2010; DOI: 10.3399/
bjgp10X515412.
8.
Ingebrigtsen SG, Scheel BI, Hart B, et al. Frequency of ‘warning signs of cancer’
in Norwegian general practice, with prospective recording of subsequent cancer.
Fam Pract 2013;30(2):153–160
9.
Scheel BI, Ingebrigtsen SG, Thorsen T, Holtedahl K. Cancer suspicion in general
practice: the role of symptoms and patient characteristics, and their association
with subsequent cancer. Br J Gen Pract 2013; 63(614): 627–635.
10. Johansen M, Holtedahl KA, Rudebeck CE. How does the thought of cancer arise
in a general practice consultation? Interviews with GPs. Scand J Prim Health
Care 2012; 30(3): 135–140.
11. Nielsen TN, Hansen RP, Vedsted P. [Symptom presentation in cancer patients in
general practice] [In Danish]. Ugeskr Laeger 2010; 172(41): 2827–2831.
12. Nylenna M. Diagnosing cancer in general practice: when is cancer suspected?
Br Med J (Clin Res Ed) 1986; 293(6541): 245–248.
13. Nylenna M. Diagnosing cancer in general practice: from suspicion to certainty.
Br Med J (Clin Res Ed) 1986; 293(6542): 314–317.
14. Moth G, Olesen F, Vedsted P. Reasons for encounter and disease patterns in
Danish primary care: Changes over 16 years. Scand J Prim Health Care 2012;
30(2): 70–75.
15. WONCA. International Classification of Primary Care (ICPC-2-R). Revised 2nd
edn. Oxford: OUP, 2005.
16. Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient Register. Scand
J Pub Health 2011; 39(7 suppl): 30–33.
17. World Health Organization. International Classification of Diseases (ICD). http://
www.who.int/classifications/icd/en/ (accessed 12 May 2014).
18. Andersen JS, Olivarius Nde F, Krasnik A. The Danish National Health Service
Register. Scand J Pub Health 2011; 39(7 suppl): 34–37.
19. OECD Project on Income Distribution and Poverty. What are equivalence scales?
2009. www.oecd.org/eco/growth/OECD-Note-EquivalenceScales.pdf (accessed
12 May 2014).
20. Statistics Denmark. Quality Declarations. http://www.dst.dk/en/Statistik/
dokumentation/Declarations.aspx. (accessed 12 May 2014).
21. Svendsen RP, Stovring H, Hansen BL, et al. Prevalence of cancer alarm
symptoms: a population-based cross-sectional study. Scand J Prim Health Care
2010; 28(3): 132–137.
22. Rubin G, Vedsted P, Emery J. Improving cancer outcomes: better access to
diagnostics in primary care could be critical. Br J Gen Pract 2011; 61(586):
317–318.
23. Allgar VL, Neal RD. General practictioners’ management of cancer in England:
secondary analysis of data from the National Survey of NHS Patients-Cancer.
Eur J Cancer Care (Engl) 2005; 14(5): 409–416.
British Journal of General Practice, June 2014 e353
Clinical Activity in General Practice and Cancer
138 Paper II
PAPER II 139 IJC
International Journal of Cancer
Variation in general practice prostate-specific antigen testing
and prostate cancer outcomes: An ecological study
Peter Hjertholm1,2,3, Morten Fenger-Grïn1,2, Mogens Vestergaard2,3, Morten B. Christensen1,2, Michael Borre4,
Henrik Mïller1,2,5 and Peter Vedsted1,2
1
Research Center for Cancer Diagnosis in Primary Care, Aarhus University, Aarhus C, Denmark
Research Unit for General Practice, Aarhus University, Aarhus C, Denmark
3
Section for General Medical Practice, Aarhus University, Aarhus C, Denmark
4
Department of Urology, Aarhus University Hospital Skejby, Aarhus N, Denmark
5
King’s College London, Cancer Epidemiology and Population Health, Bermondsey Wing, Guy’s Hospital, Great Maze Pond, London, United Kingdom
2
Most Western countries have seen an increase in prostate
cancer incidence during the last 25 years. This is temporally linked with the introduction of the prostate-specific
antigen (PSA) test in the different countries.1 Differences
among countries in prostate cancer incidence, stage distribution, and survival are predominantly induced by differences in the use of PSA testing and to policies and
recommendations regarding its use.2 However, the question
remains whether these associations also exist within countries due to variations among individual practitioners and
general practices.
Public healthcare in Denmark is tax-funded and free at
the point of care. General practices act as gatekeepers to sec-
ondary care by managing referrals to specialists, hospital
admission and diagnostic and therapeutic services. Almost all
Danish citizens (99%) are listed with a general practice,
which they must seek for medical advice.3 Danish guidelines
for PSA testing recommend against screening of asymptomatic men (Fig. 1),4 but PSA testing remains an issue for discussion. The resulting variable frequency of PSA testing
provides a suitable study population for investigating effects
of PSA testing in a primary health care setting.
The aim of this study was to investigate the association
between variations in the use of PSA tests among general
practices in the Central Denmark Region in 2004–2009 and
prostate cancer-related outcomes.
Material and Methods
Key words: neoplasm, primary health care, general practice, prostate-specific antigen, prostate neoplasms
Additional Supporting Information may be found in the online
version of this article.
DOI: 10.1002/ijc.29008
Revised 29 Apr 2014; Accepted 20 May 2014; Online 6 June 2014
Correspondence to: Peter Hjertholm, Research Centre for Cancer
Diagnosis in Primary Care, Aarhus University, Research Unit for
General Practice, Aarhus University, Bartholins Alle 2, DK-8000
Aarhus, Denmark, Tel.: 145-8716-8043, Fax: 145-8612-4788,
E-mail: [email protected]
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
Study population
We conducted a population-based ecological study with a
well-defined cohort of all men aged 40 years or more listed
at a general practice in the Central Denmark Region (1.2 million inhabitants, corresponding to 20% of the Danish population) between January 1, 2004 and December 31, 2009. Only
general practices which remained active during the entire
study period were included.
Prostate cancer-free men were included from January 1,
2004, when turning 40 years old or when moving to the Central Denmark Region, whichever occurred last. Individuals
Epidemiology
Knowledge is sparse about the consequences of variation in prostate-specific antigen (PSA) testing rates in general practice.
This study investigated associations between PSA testing and prostate cancer- related outcomes in Danish general practice,
where screening for prostate cancer is not recommended. National registers were used to divide general practices into four
groups based on their adjusted PSA test rate 2004–2009. We analysed associations between PSA test rate and prostate
cancer-related outcomes using Poisson regression adjusted for potential confounders. We included 368 general practices,
303,098 men and 4,199 incident prostate cancers. Men in the highest testing quartile of practices compared to men in the
lowest quartile had increased risk of trans-rectal ultrasound (incidence rate ratio (IRR): 1.20, 95% CI, 0.95–1.51), biopsy (IRR:
1.76, 95% CI, 1.54–2.02), and getting a prostate cancer diagnosis (IRR: 1.37, 95% CI, 1.23–1.52). More were diagnosed with
local stage disease (IRR: 1.61, 95% CI, 1.37–1.89) with no differences regarding regional or distant stage. The IRR for prostatectomy was 2.25 (95% CI, 1.72–2.94) and 1.28 (95% CI, 1.02–1.62) for radiotherapy. No differences in prostate cancer or
overall mortality were found between the groups. These results show that the highest PSA testing general practices may not
reduce prostate cancer mortality but increase the downstream use of diagnostic and surgical procedures with potentially
harmful side effects.
436
PSA testing and prostate cancer outcomes
What’s new?
The impact of PSA testing on diagnosis and mortality of prostate cancer is not yet clear. In this study, the authors found that
patients of general practitioners (GPs) with high rates of PSA testing had a significantly increased incidence of prostate cancer, as well as greater use of diagnostic and surgical procedures. However, the mortality rate due to prostate cancer was similar to patients of GPs with low rates of PSA testing. This indicates that routine PSA testing may increase the use of diagnostic
and surgical procedures with potentially harmful side effects, without actually reducing prostate cancer mortality.
were censored at the end of the study period, upon death,
when moving out of Central Denmark Region or when leaving the list system, whichever occurred first.
The study cohort was established using the Danish Civil
Registration System where all Danish citizens are assigned a
unique personal identification number. The system also contains data on emigration, immigration and vital status.5 The
identification number was used as a key to allow accurate
linkage of information from different registries on the individual level. Information about any citizen’s general practice
is registered in the Patient Lists Register. All changes of practice are noted here with a specific date and any citizen’s general practice can be identified at any given time.
Epidemiology
Exposure
PSA tests rates.
PSA tests were identified in the clinical laboratory information system (“LABKA”) for all laboratories in
the Central Denmark Region.6 Results of biochemical analyses are electronically transferred directly to this database, and
it holds information on all PSA tests (including test date, test
requestor and test result). We calculated the overall test rate
for each general practice in the study period as the number
of PSA tests ordered in the practice divided by the sum of
person-years contributed by all eligible men listed in the
practice up to the point of prostate cancer diagnosis, leaving
the region, leaving the list system, death or end of study.
National Patient Register using a register-specific procedure
code.7
Prostate cancers (International Classification of Diseases,
ICD-10 code C61)8 were identified in the Danish Cancer
Registry.9 This registry holds information on all incident cancer cases in Denmark since 1943, including date of diagnosis,
age at diagnosis, cancer type and Tumor-Node-Metastasis
(TNM) stage.
Disease stage at diagnosis was categorised based on the
TNM classification using the clinical categories local,
regional, distant and unknown as proposed by NguyenNielsen et al. (Supporting Information Table 1).10
Radical prostatectomy and radiotherapy were identified in
the Danish National Patient Register. Radiotherapy was
counted as number of prostate cancer patients who received
radiotherapy within six months after diagnosis and who had
not been treated with prostatectomy.
Relative survival was calculated as the survival of prostate
cancer patients relative to the survival expected had they
been subject to the background mortality given the same
demographic factors (year of birth, calendar time). Information on the general population mortality was collected from
national life tables available from Statistics Denmark.
Mortality was calculated separately as prostate cancerspecific and all-cause deaths in the entire study population
per person-years. Information on date of death was retrieved
from the Civil Registration System and cause of death was
retrieved from the Danish Register of Causes of Death.11
Outcome measures
Trans-rectal ultrasound of the prostate and biopsy of the
prostate on all included men were identified in the Danish
Background variables
The Charlson Comorbidity Index was computed for each
man, and data were used in a comparison of four groups of
practices.12,13 All available primary and secondary diagnoses
in the Danish National Patient Register were included (both
inpatient and outpatient hospital diagnoses) from 1994 until
date of censoring. Included diagnoses are listed in Supporting
Information Table 2. The men were divided into three groups
based on their individual score (0, 1–2 and 3 or higher).
Information from January 2004 on education and marital status was obtained from Statistics Denmark.
Statistical methods
Figure 1. Danish guidelines for PSA testing, 2012.
The analyses consisted of two main steps; (i) categorisation
of general practices into four groups (quartiles) based on
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
437
Hjertholm et al.
Table 1. Characteristics of men aged 40 years or more according to PSA test rate in the four practice groups (group 1 lowest rate of PSA testing, group 4 highest rate of PSA testing)
Total
Group 1
Group 2
Group 3
Group 4
No. practices
368
92
92
92
92
No. men1
303,098
70,491
87,915
88,608
74,795
Person-years
1,471,869
317,601
400,504
410,397
343,368
Observation time per practice,
mean (sd), person-years
4,000 (2,683)
3,452 (2,350)
4,353 (2,989)
4,461 (3,053)
3,732 (2,130)
Number
106,882
11,401
21,511
29,984
43,986
Crude rate, mean tests/1,000
person-years
72.6
35.9
53.7
73.1
128.1
1 (Reference)
1.50 (1.46–1.53)
2.04 (1.99–2.08)
3.57 (3.50–3.64)
PSA tests
IRR, unadjusted (95% CI)
IRR, adjusted (95% CI)
1 (Reference)
1.52 (1.49–1.56)
2.02 (1.98–2.06)
3.56 (3.48–3.63)
PSA value, mean, mg/L2,3
5.37
5.17
4.53
3.97
Proportion of PSA tests >4 mg/L, %3
27.3
30.5
31.0
27.2
24.6
Age, mean (sd), yrs
57.9 (12.1)
58.2 (12.4)
57.6 (12.1)
58.0 (12.0)
58.0 (12.0)
<10 yrs (primary and lower
secondary school)
34.2
36.5
33.0
33.4
34.5
10–12 yrs (upper secondary school
or vocational training)
43.7
42.3
43.6
44.3
44.3
>12 yrs (higher education)
22.1
21.2
23.4
22.3
21.3
4
Educational level (%)
Marital status5 (%)
Married/cohabitating
76.7
75.0
77.2
77.5
76.7
Single
23.3
25.0
22.8
22.5
23.3
Danish
95.1
94.5
95.1
95.3
95.4
Western immigrant
1.8
1.8
1.8
1.8
1.7
Non-western immigrant
2.9
3.4
2.9
2.6
2.7
0
72.0
71.2
72.2
72.6
71.6
1–2
20.1
20.5
19.9
19.7
20.4
3 or more
8.0
8.3
7.9
7.7
8.0
Ethnicity5 (%)
1
Sum of groups adds to more than total because men can change practice
One-way analysis of variance shows that values are statistically significantly different between the groups, except groups 1 and 2
3
Only values less than 200 mg/L were included (99.1% of the tests)
4
Information missing in 4.7%. Men with missing educational level were included in the lowest educational level
5
Information missing in 0.4%. Persons were excluded from analyses.
Abbreviations: sd: standard deviation, IRR: incidence rate ratio, CI, confidence interval.
2
their adjusted PSA test rates, (ii) computation of associations
between PSA test rate groups and outcomes.
First step. PSA test rates at practice level were calculated
using a Poisson regression adjusting for calendar-year in 2year periods (2004–2005, 2006–2007, 2008–2009), age in
10-year age groups (40–49, 50–59. . . 90–99, >100), educational level (<10 years, 10–12 years, >12 years), ethnic
origin (Danish, western immigrant, and non-western immigrant) and marital status (married/cohabitating or living
alone).
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
Second step. Age at prostate cancer diagnosis and mean
PSA values were compared among groups using one-way
analysis of variance. Incidence rate ratios in the period 2004
to 2009 of trans-rectal ultrasound and biopsy of the prostate,
prostate cancer, different disease stages at diagnosis, prostatectomy, radiotherapy and mortality among all included men
were calculated using Poisson regressions with male personyears in the four groups as the denominator. The practice
group with least testing (group 1) was chosen as the reference group. Adjusted models included calendar year, age
group, educational level, ethnic origin and marital status as
Epidemiology
Charlson Comorbidity Index (%)
2.46
n 5 3,801
71.5 (8.8)
IRR, only men aged 60 yrs
Age at diagnosis (sd), mean
1.19 (1.03–1.38)
18 (2.3)
Number (%)
723 (17.2)
137 (17.6)
1 (Reference)
Radiotherapy
IRR, adjusted
104 (13.3)
1 (Reference)
846 (20.1)
IRR, unadjusted
Number (% of cancer patients)
Prostatectomy
1 (Reference)
217 (27.8)
IRR unknown, adjusted
0.99 (0.80–1.23)
1 (Reference)
1,022 (24.3)
IRR distant, adjusted
Number with unknown
stage (%)
177 (16.8)
1.41 (1.08–1.83)
1.41 (1.09 (1.85)
186 (17.7)
1.05 (0.85–1.30)
263 (25.0)
174 (16.5)
150 (19.2)
651 (15.5)
Number with distant
disease (%)
1.70 (0.94–3.07)
1 (Reference)
IRR regional, adjusted
38 (3.6)
1 (Reference)
IRR local, adjusted
Number with regional
disease (%)
105 (2.5)
1.11 (1.00–1.24)
70.9 (8.6)
578 (54.9)
72.9 (9.2)
1.28 (0.93–1.78)
1.12 (1.02–1.25)
1.07 (0.96–1.19)
2.63
1,053
1.13 (1.00–1.28)
1.11 (0.98–1.27)
1,463
1.05 (0.83–1.33)
1.03 (0.81–1.30)
1,870
Group 2 (95% CI)
395 (50.6)
2,421 (57.7)
Number with local disease (%)
Stage at diagnosis
1 (Reference)
IRR, only men aged <60 yrs
1 (Reference)
1 (Reference)
n 5 398
IRR, adjusted
1 (Reference)
2.85
Crude rate, cancer/1,000 yrs
IRR, unadjusted
4,199
Number
780
1 (Reference)
Prostate cancer incidence
1 (Reference)
1,041
1 (Reference)
1 (Reference)
1,446
IRR, unadjusted
6,496
7,617
Group 1
IRR adjusted
Number
Biopsy
IRR, adjusted
IRR, unadjusted
Number
Trans-rectal ultrasound of prostate
Total
218 (17.7)
2.10 (1.64–2.68)
2.19 (1.72-2.81)
295 (23.9)
1.04 (0.84–1.28)
274 (22.2)
0.97 (0.77–1.21)
177 (14.4)
1.13 (0.56–2.24)
27 (2.2)
1.46 (1.26–1.69)
755 (61.2)
71.5 (8.8)
1.23 (1.11–1.36)
1.43 (1.04–1.96)
1.24 (1.13–1.37)
1.22 (1.10–1.37)
3.00
1,233
1.43 (1.26–1.62)
1.47 (1.29–1.67)
1,974
1.29 (1.03–1.61)
1.30 (1.03–1.63)
2,424
Group 3 (95% CI)
191 (16.9)
2.25 (1.72–2.94)
2.32 (1.78–3.02)
261 (23.0)
1.21 (0.99–1.48)
268 (23.7)
0.97 (0.78–1.21)
150 (13.2)
1.10 (0.58–2.10)
22 (1.9)
1.61 (1.37–1.89)
693 (61.2)
71.1 (8.5)
1.33 (1.19–1.49)
1.73 (1.27–2.35)
1.37 (1.23–1.52)
1.34 (1.20–1.50)
3.30
1,133
1.76 (1.54–2.02)
1.79 (1.56–2.06)
2,018
1.20 (0.95–1.51)
1.20 (0.95–1.52)
1,877
Group 4 (95% CI)
Table 2. Prostate cancer outcomes according to PSA test rate in the four practice groups (group 1 lowest rate of PSA testing, group 4 highest rate of PSA testing)
Epidemiology
p < 0.001
p < 0.001
p 5 0.10
p 5 0.75
p 5 0.70
p < 0.001
p < 0.001
p < 0.001
p < 0.001
p < 0.001
p < 0.001
p < 0.001
p < 0.001
p 5 0.035
p 5 0.027
Test for trend
438
PSA testing and prostate cancer outcomes
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
439
PSA test rates
Among the 498 general practices which ordered PSA tests in
the study period, 368 (74%) were active throughout the entire
study period and performed 10,882 PSA tests among 303,098
males (Fig. 2). The mean crude PSA test rate was 3.56-fold
(95% confidence interval, 3.48-3.63) higher among practices
which tested the most (128.1 per 1,000 person years, group
4) compared to practices which tested the least (35.9 tests per
1,000 person years, group 1) (Table 1). Mean PSA values varied between groups of practices with highest values in group
1 (5.37 mg/L) and lowest in group 4 (3.97 mg/L) (Table 1).
The four practice groups were comparable concerning sociodemographic composition and Charlson Comorbidity Indexes
of the men on their lists (Table 1).
Trans-rectal ultrasound and biopsy of prostate
Men in group 4 had a 20% non-statistically significantly
higher risk of having trans-rectal ultrasound examination of
the prostate (IRR: 1.20; 95% CI, 0.95–1.52) and a 76% higher
risk of having a prostate biopsy (IRR 1.76; 95% confidence
interval, 1.54–2.02) compared to men in group 1 (Table 2).
Cancer incidence
During the study period, 4,199 prostate cancers were diagnosed. Men in group 4 had a 37% higher prostate cancer
incidence (IRR: 1.37; 95% confidence interval, 1.23–1.52)
than men in group 1. This difference was more pronounced
among men younger than 60 years (IRR 1.73; 95% confidence interval, 1.27–2.35) (Table 2).
Age at diagnosis
Prostate cancer patients in group 1 were diagnosed at a statistically significantly older age (72.9 years) than in the three
other groups (Table 2).
Cancer stage at diagnosis
In total, 51% of the cancers in group 1 were diagnosed as local
stage disease compared to 61% in group 4 (Table 2). The IRRs
of localised disease were 1.19 (95% confidence interval, 1.03–
Epidemiology
1.01 (0.97–1.05)
1.00 (0.96–1.04)
1.01 (0.98–1.05)
8,379
Results
Abbreviations: sd: standard deviation, IRR: incidence rate ratio, CI, confidence interval.
8,630
9,999
9,734
IRR, adjusted
IRR, adjusted
Mortality, all cause
Number
36.742
1 (Reference)
1.11 (0.92–1.33)
0.94 (0.78–1.14)
0.97 (0.80–1.17)
1 (Reference)
237
237
234
206
914
Number
Mortality of prostate cancer
1 yr
5 yr
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
covariates. Robust variance estimation was performed to
account for possible cluster effects on practice level.14 Relative
survival was calculated by using the STATA procedure provided by Dickman (Ederer II method) and age standardized
using the International Cancer Survival Standard (ICSS)
weight distribution slightly modified to our age groups (40–
49 years: weight 0.09, 50–59 years: 0.17, 60–69 years: 0.27,
70–79 years: 0.29, 80 years1: 0.15).15,16 The relative survival
was modelled using excess mortality rates in a Poisson model
adjusting for calendar year, age group, educational level, ethnic origin and marital status. For survival and mortality men
were followed until January 1 2012.
All analyses were carried out using the statistical software
STATA 12.1 (StataCorp LP, Lakeway Drive, College Station, TX).
p 5 0.85
p 5 0.36
p 5 0.021
p 5 0.67
0.88 (0.45–1.72)
0.70 (0.50–0.97)
0.78 (0.39–1.54)
0.74 (0.54–1.01)
0.85 (0.42–1.71)
0.85 (0.62–1.16)
1 (Reference)
1 (Reference)
p 5 0.02
83.4 (79.3–86.7)
82.6 (78.8–85.8)
78.1 (73.3–82.2)
74.2 (65.8–80.8)
80.3 (77.9–82.4)
5 yr
Relative excess mortality rate ratios
96.3 (94.4–97.5)
97.7 (0.96–98.7)
96.6 (94.7–97.8)
96.3 (94.1–97.7)
Relative survival
IRR, unadjusted
IRR, adjusted
1 yr
96.8 (96.0–97.5)
p 5 0.61
p 5 0.02
1.28 (1.02–1.62)
1.22 (0.98–1.52)
1.05 (0.82–1.34)
p 5 0.01
1.29 (1.02–1.63)
1.23 (0.98–1.55)
1.02 (0.80–1.31)
1 (Reference)
1 (Reference)
Test for trend
Group 4 (95% CI)
Group 3 (95% CI)
Group 2 (95% CI)
Group 1
Total
Table 2. Prostate cancer outcomes according to PSA test rate in the four practice groups (group 1 lowest rate of PSA testing, group 4 highest rate of PSA testing) (Continued)
Hjertholm et al.
440
1.38) in group 2, 1.46 (95% confidence interval, 1.26–1.69) in
group 3, and 1.61 (95% confidence interval, 1.37–1.89) in
group 4. The IRRs of regional, distant and unknown disease
stages showed no differences across the four groups.
Radical prostatectomy
During the study period, 846 prostate cancer patients (20.1%)
were treated with a prostatectomy. The IRR of prostatectomy
was 2.25 (95% confidence interval, 1.72–2.94) in group 4
compared to group 1 (Table 2).
Radiotherapy
In total, 723 (17.2%) prostate cancer patients received radiotherapy within six months after diagnosis. The IRR of radiotherapy was 1.28 (95% confidence interval, 1.02-1.62) in
group 4 compared with group 1 (Table 2).
Relative survival and excess mortality rate ratios
Median follow-up on cancer patients was 3.43 years (range:
0–8 years). The 1-year relative survival showed no statistically
significant differences between the groups. The 5-year relative
survival was lower in group 1 than in group 4. The 5-year
relative excess mortality rate ratios between the two most
extreme groups showed a statistically significant difference of
0.70 (95% confidence interval, 0.50–0.97) in group 4 compared to group 1 (Table 2).
Mortality rates
Epidemiology
A total of 914 persons died from prostate cancer in the study
population, and 36,742 died from any cause during the study
period. The adjusted IRR for prostate cancer-specific mortality in group 4 was 1.11 (95% confidence interval, 0.92–1.33)
compared to group 1 (Table 2). No differences were found in
all-cause mortality between the four groups.
Discussion
Main findings
PSA testing rates varied with a factor of 3.6 between the
quartiles of general practices which tested the most and the
least and this variation was significantly associated with several clinical outcomes of the patients. Men assigned to general practices with the highest PSA test rates had a higher
incidence of trans-rectal ultrasound (albeit not statistically
significant), biopsies of the prostate, prostate cancer diagnoses, younger age at diagnosis, and higher incidences of local
disease at diagnosis, radical prostatectomy, and radiotherapy
compared to those assigned to practices with lower PSA test
rates. Furthermore, men assigned to practices with the highest PSA test rates had the same incidence of distant disease, a
higher 5-year relative survival, the same prostate cancerspecific and overall mortality.
Comparison with existing literature
Many studies have documented variation in clinical activity
in general practice, but only few studies have looked at
PSA testing and prostate cancer outcomes
consequences of this variation and findings have been ambiguous. Franks et al. found no relationship between overall
referral rates and patient outcomes or costs. However, higher
referral rates were associated with higher patient satisfaction.17 O’Kane et al. found large variations in test rates for
HbA1c and thyroid functions, but no association with disease
prevalence or clinical outcome measures.18 Hippisley-Cox
found no association between overall referral rates and stages
of breast and colorectal cancer.19 A recent study by Shawihdi
et al. found an association between low gastroscopy referral
rate and poorer outcomes of oesophago-gastric cancer
regarding emergency admission, major surgery, and
mortality.20
To our knowledge, the present study is the first to investigate the associations between variation in the use of PSA
testing on general practice level and prostate cancer incidence
and outcomes. The overall findings of an association between
PSA testing intensity and downstream effects are in line with
previous studies.21,22
We found a strong link between PSA test rates in general
practice and prostate cancer incidence. The European
Randomized Study of Screening for Prostate Cancer
(ERSSPC) found an IRR of 1.63 for prostate cancer in the
screening group compared to the non-screening group.23
Outzen et al. found an increase in prostate cancer incidence
in Denmark after the introduction of PSA tests in general
practice in 1997,24 and a similar effect of PSA introduction
was seen in the United States.25
Age at diagnosis was significantly higher in the practices
ordering fewest PSA tests. Correspondingly, Outzen et al.
found that median age at diagnosis on a national level in
Denmark decreased from 75.1 years in 1988–1992 to 69.7 in
2008–2009; a period in which PSA test rates increased more
than 40-fold.24 This pattern is also seen in other countries.1,26
Outzen et al. found an increase in localised cancers after
the introduction of PSA tests, which is consistent with our
findings.24 Sandblom et al. also found an increased proportion of localised cancers in the screened group of men in the
Norrkoping screening trial from Sweden.26 This association
between higher PSA testing and more localised cancers, but
no change in distant cancers, is in accordance with a recent
comparison between Denmark, Iceland and Sweden.27 Thus,
general practices ordering more PSA tests tend to identify
more patients with localised prostate cancers, but not fewer
with distant cancers. The treatment of persons with localised
prostate cancers is continuously discussed and overdiagnosis
and overtreatment are well-known issues.28
Persons with localised prostate cancer may have a prostatectomy, which often has side-effects such as erectile dysfunction and urinary incontinence. The relation between PSA
testing and prostatectomy was highlighted by Welch et al.
who estimated that about one million additional men in the
US had been treated for prostate cancer and half a million
with surgery after the introduction of PSA testing.25 The
higher risk of prostatectomy of men in group 4 in our study
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
Figure 2. Flow-chart of men included in the study.
could be because the most testing practices tend to diagnose
patients with local cancers particularly suited for
prostatectomy.
Sandblom et al. found a tendency for better relative survival among the screened patients compared to the control
group.26 Bray et al. found an increasing five-year relative survival in the Nordic countries which intensified during the
period when PSA testing increased.2 The association between
PSA testing and survival was also found when comparing
countries.1 These survival differences and our findings are
probably caused by lead-time and length-time biases and
may therefore not reflect actual clinical benefits.27,29
The similarity in mortality (prostate cancer-specific and
overall) in the four groups is in line with the findings by several
studies and a recent Cochrane review on five screening trials.2,21,22,24,30 However, this contrasts the findings from the
ERSPC, which found a small decrease in prostate cancer-specific
mortality in the screened group but not before 9 years of followup.23 We cannot rule out that a longer follow-up could have
resulted in a mortality difference in our study, but subanalysis
on the 1,174 patients in our study diagnosed in 2004 and 2005
(median follow-up 6.0 years) also failed to show reductions in
prostate cancer mortality, IRR 5 1.15 (0.88–1.50) in group 4.
Strengths and weaknesses of the study
Strengths of the present study include the large number of
included men, PSA tests, and prostate cancers and also the
population-based nature of the study. We included men in a
well-defined geographical area and all had complete followup, which improves statistical precision. Data was collected
prospectively.
The Civil Registration Number assigned to each Danish
citizen allowed us to link high quality data from various
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
registers at the individual level.5,9 An additional strength was
that we were able to calculate accurate risk time as change of
general practice was noted with a specific date. Data for
included PSA tests came from a clinical database used in the
daily practice, which assures a high degree of completeness.
Information on disease stage was missing in nearly 20% of
the cancers, but missing information was almost equally distributed in the four groups and could not explain the findings (data not shown). Cause of death was unknown or
missing in 4.2% of the deaths. These were evenly distributed
across the four groups and do not affect our results.
Our study was limited by a lack of important clinical
information about the indication for testing but previous
studies find that opportunistic screening in Denmark is ubiquitous even though guidelines advice otherwise.31 Another
limitation was that we had only practice identification numbers. PSA test rates from partnership practices are a mix of
two or more general practitioners, but our results were consistent when looking only at single-handed practices. Discarding 130 general practices that were not active in the
entire study period also caused exclusion of 562 incident
prostate cancer patients. This was done in order to obtain a
more homogenous group of practices. Analyses including all
498 practices showed similar results.
Based on the included variables, the practice populations
were similar, but we cannot exclude confounding from factors
not accounted for. However, such factors should be associated
with both the PSA test rates in the practices and the included
person’s a priori risk of prostate cancer to explain our findings. This potential bias is minimised by our study design,
which exploits that men’s choice of practice is independent of
the practice’s PSA test rate. Associations between outcomes
and distance to a urology department performing prostatectomies could not explain our findings (data not shown).
The results could be influenced by outliers, but exclusion
of the 5% most and the 5% least testing practices did not
alter the results. Neither did analyses only based on first-testonly, not including repeated PSA tests.
The finding that mean PSA values varied among the four
groups of general practices and the 3.6-fold variation in testing rates indicate a variation in clinical practice, specifically
in the threshold for testing. Variations in PSA testing might
not be the only explanation of our findings, but may reflect
other differences in diagnostic reasoning. However, the similarity with previous epidemiologic findings indicates that variation in the practitioners’ propensity to use PSA testing
could explain our findings.
As this study was population-based and studied the variation in use of PSA testing which can be observed in most
health care systems, the findings are generalizable to other
healthcare settings
Conclusions
Our study is the first to show an association between general
practice variations in PSA testing and outcomes in relation to
Epidemiology
441
Hjertholm et al.
442
PSA testing and prostate cancer outcomes
prostate cancer. More PSA tests were associated with a higher
prostate cancer incidence and increased use of diagnostic and
surgical procedures, with no mortality reduction. Our findings add to the ongoing discussion about the appropriateness
of PSA testing in asymptomatic men and indicate that variation among general practices may have consequences for the
practice populations. To better guide the future diagnostic
activity in general practices, we need more studies on the
consequences of variation for cancer detection and outcomes.
Acknowledgements
The project was approved by the Danish Data Protection Agency (J.no.:
2009-41-3471). According to Danish law, approval from the National Committee on Health Research Ethics was not required as de-identified register
data were used for analyses. This work was supported by the Danish
National Research Foundation for Primary Care, the Novo Nordic Foundation and the Danish Cancer Society. The authors thank Dr. Thomas Mukai
(MD, PhD, Research Center for Cancer Diagnosis in Primary Care, Aarhus
University) for collecting data on blood samples. The authors thank the Statistics Denmark for providing data and Kaare Flarup (Data Manager,
Research Center for Cancer Diagnosis in Primary Care, Aarhus University)
for the initial data management.
Hjertholm had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Hjertholm and Fenger-Grïn are responsible for the data analysis. Study concept
and design: Hjertholm, Fenger-Grïn, Christensen, Vestergaard, Mïller
Vedsted. Acquisition of data: Hjertholm. Analysis and interpretation of data:
Hjertholm, Fenger-Grïn, Vestergaard, Borre, Mïller, Vedsted. Drafting of the
manuscript: Hjertholm, Fenger-Grïn, Vestergaard, Vedsted. Critical revision
of the manuscript for important intellectual content: Hjertholm, FengerGrïn, Vestergaard, Christensen, Borre, Mïller, Vedsted. Statistical analysis:
Hjertholm, Fenger-Grïn. Study supervision: Vestergaard, Vedsted
References
Epidemiology
1.
Baade PD, Youlden DR, Krnjacki LJ. International epidemiology of prostate cancer: geographical distribution and secular trends. Mol Nutr
Food Res 2009;53:171–84.
2. Bray F, Klint Å, Gislum M, et al. Trends in survival of patients diagnosed with male genital cancers in the Nordic countries 1964–2003 followed
up until the end of 2006. Acta Oncol 2010;49:
644–54.
3. Pedersen KM, Andersen JS, Sïndergaard J. General Practice and Primary Health Care in Denmark. J Am Board Fam Med 2012;25:S34–8.
4. DACROPA. Available at: http://ducg.dk/prostata/
retningslinjer. Accessed September 5, 2013.
5. Pedersen CB. The Danish Civil Registration System. Scand J Public Health 2011;39:22–5.
6. Grann AF, Erichsen R, Nielsen AG, et al. Existing
data sources for clinical epidemiology: The clinical laboratory information system (LABKA)
research database at Aarhus University, Denmark.
Clin Epidemiol 2011;3:133–8.
7. Lynge E, Sandegaard JL, Rebolj M. The Danish
National Patient Register. Scand J Public Health
2011;39:30–3.
8. .International Classification of Diseases. Available
at: http://www.who.int/classifications/icd/en/.
Accessed January 28, 2013.
9. Gjerstorff ML. The Danish Cancer Registry.
Scand J Public Health 2011;39:42–5.
10. Nguyen-Nielsen M, Froslev T, Friis S, et al. Completeness of prostate cancer staging in the Danish
Cancer Registry, 2004–2009. Clin Epidemiol 2012;
4(Suppl 2):17–23.
11. Helweg-Larsen K. The Danish register of causes
of death. Scand J Public Health 2011;39:26–9.
12. Sundararajan V, Quan H, Halfon P, et al., International Methodology Consortium for Coded
Health Information (IMECCHI). Cross-national
13.
14.
15.
16.
17.
18.
19.
20.
21.
comparative performance of three versions of the
ICD-10 Charlson index. Med Care 2007;45:1210–
15.
Thygesen S, Christiansen C, Christensen S, et al.
The predictive value of ICD-10 diagnostic coding
used to assess Charlson comorbidity index conditions in the population-based Danish National
Registry of Patients. BMC Med Res Methodol
2011;11:83.
Williams RL. A note on robust variance estimation for cluster-correlated data. Biometrics 2000;
56:645–6.
Corazziari I, Quinn M, Capocaccia R. Standard
cancer patient population for age standardising
survival ratios. Eur J Cancer 2004;40:2307–16.
Dickman PW, Sloggett A, Hills M, et al. Regression models for relative survival. Stat Med 2004;
23:51–64.
Franks P, Mooney C, Sorbero M. Physician referral rates: style without much substance? Med
Care 2000;38:836–46.
O’Kane MJ, Casey L, Lynch PLM, et al. Clinical
outcome indicators, disease prevalence and test
request variability in primary care. Ann Clin Biochem 2011;48:155–8.
Hippisley-Cox J, Hardy C, Pringle M, et al. Are
patients who present late with cancer registered
with low referring practices? Br J Gen Pract 1997;
47:731–2.
Shawihdi M, Thompson E, Kapoor N, et al. Variation in gastroscopy rate in English general practice and outcome for oesophagogastric cancer:
retrospective analysis of Hospital Episode Statistics. Gut 2014;63:250–61.
Lu-Yao G, Albertsen PC, Stanford JL, et al.
Screening, treatment, and prostate cancer mortality
in the Seattle area and Connecticut: fifteen-year
follow-up. J Gen Intern Med 2008;23:1809–14.
22. Shao YH, Albertsen PC, Shih W, et al. The
impact of PSA testing frequency on prostate cancer incidence and treatment in older men. Prostate Cancer Prostatic Dis 2011;14:332–9.
23. Schroder FH, Hugosson J, Roobol MJ, et al. Prostate-cancer mortality at 11 years of follow-up. N
Engl J Med 2012;366:981–90.
24. Outzen M, Brasso K, Martinussen N, et al. Prostate cancer in Denmark 1978–2009 - trends in
incidence and mortality. Acta Oncol 2013;52:831–
6.
25. Welch HG, Albertsen PC. Prostate cancer diagnosis and treatment after the introduction of
prostate-specific antigen screening: 1986–2005. J
Natl Cancer Instit 2009;101:1325–9.
26. Sandblom G, Varenhorst E, L€
ofman O, et al.
Clinical consequences of screening for
prostate cancer: 15 years follow-up of a randomised controlled trial in Sweden. Eur Urol 2004;
46:717–24.
27. Brasso K, Ingimarsdottir IJ, Rusch E, et al. Differences in survival from prostate cancer in Denmark, Iceland and Sweden. Eur J Cancer 2013;49:
1984–92.
28. Wilt TJ, Brawer MK, Jones KM, et al. Radical
prostatectomy versus observation for localized
prostate cancer. N Engl J Med 2012;367:203–13.
29. Finne P, Fallah M, Hakama M, et al. Lead-time
in the European randomised study of screening
for prostate cancer. Eur J Cancer 2010;46:3102–8.
30. Ilic D, Neuberger MM, Djulbegovic M, et al.
Screening for prostate cancer. Cochrane Database
Syst Rev 2013;1:CD004720.
31. Borre M. Screening by lower urinary tract symptoms vs asymptomatic prostate-specific antigen
levels leading to radical prostatectomy in Danish
men: tumour characteristics and treatment outcome. BJU Int 2009;104:205–8.
C 2014 UICC
Int. J. Cancer: 136, 435–442 (2015) V
Clinical Activity in General Practice and Cancer
Supporting information Table 1, Paper II
Algorithm for prostate cancer staging according to the TNM classification in the Danish
Cancer Registry1
Stage
TNM codes
Localized
T1–4,x N0 M0
T1–2 N0 Mx
T1–2 Nx M0,x
Regional
T1–4,x N1, M0
Distant
T1–4,x N1, M1
T1–4,x N0 M1
T1–4,x Nx M1
Unknown
T3–4,x Nx M0,x
T3–4,x N0 Mx
T1-4,x N1 Mx
Note: Tis, Ta, N2, and N3 are not used for prostate cancer, and such registrations were categorized as
“unknown.” Cases registered with “T0” were also categorized as “unknown”.
1
Nguyen‐Nielsen M, Froslev T, Friis S, Borre M, Harving N, Sogaard M. Completeness of prostate cancer staging in the Danish Cancer Registry, 2004‐2009. Clin Epidemiol. 2012;4 Suppl 2:17‐23. 148 Paper II
Supporting information Table 2, Paper II
ICD-10 codes used for calculation of Charlson Comorbidity Index (CCI). The CCI was
calculated using the Quan version as described by Sundararajan et al. modified
according to the ICD-10 codes listed in this table.1
Weights
1
Condition
Myocardial infarction
Congestive heart failure
ICD-10 codes
I21; I22; I23
I50; I11.0; I13.0; I13.2
Peripheral vascular disease
Cerebrovascular disease
I70; I71; I72; I73; I74; I77
Dementia
Chronic pulmonary disease
I60-I69; G45; G46
F00-F03; F05.1; G30
J40-J47; J60-J67; J68.4; J70.1;
J70.3; J84.1; J92.0; J96.1;
J98.2; J98.3
Connective tissue disease
M05; M06; M08; M09; M30;
M31; M32; M33; M34; M35;
M36; D86
Ulcer disease
K22.1; K25-K28
Mild liver disease
2
B18; K70.0-K70.3; K70.9; K71;
K73; K74; K76.0
Diabetes without end organ E10.0, E10.1; E10.9; E11.0;
damage
E11.1; E11.9
Diabetes with end organ damage E10.2-E10.8, E11.2-E11.8
Hemiplegia
Moderate to severe renal disease
G81; G82
Non-metastatic solid tumour
I12; I13; N00-N05; N07; N11;
N14; N17-N19; Q61
C00-C75
Leukaemia
C91-C95
C81-C85; C88; C90; C96
3
Lymphoma
Moderate to severe liver disease
6
Metastatic cancer
AIDS
B15.0; B16.0; B16.2;
K70.4; K72; K76.6; I85
B19.0;
C76-C80
B21-B24
1
Sundararajan V, Quan H, Halfon P, et al. Cross‐national comparative performance of three versions of the ICD‐10 Charlson Index. Med Care. 2007;45(12):1210‐1215. 149 Clinical Activity in General Practice and Cancer
150 Paper III
PAPER III 151 Variation among general practitioners in the referral propensity for lower endoscopy and outcome of colorectal cancer patients (submitted) Peter Hjertholm1,2,3, Research Fellow, MD [email protected] Morten Fenger‐Grøn1,2, Biostatistician, MSc [email protected] Mogens Vestergaard2,3, Professor, MD, PhD [email protected] Morten Bondo Christensen1,2, GP, Senior Researcher, PhD [email protected] Lene Hjerrild Iversen4,5, Consultant surgeon, DMSci, PhD [email protected] Peter Vedsted1,2, Professor, MD, PhD [email protected] Author affiliations: 1
Research Centre for Cancer Diagnosis in Primary Care, Department of Public Health, Aarhus University, Denmark 2
Research Unit for General Practice, Department of Public Health, Aarhus University, Denmark 3
Section for General Medical Practice, Department of Public Health, Aarhus University, Denmark 4
Department of Surgery, Aarhus University Hospital, Denmark 5
The Danish Colorectal Cancer Group Corresponding author: Peter Hjertholm, MD, PhD student Research Centre for Cancer Diagnosis in Primary Care, Aarhus University Research Unit for General Practice, Aarhus University Bartholins Alle 2 DK‐8000 Aarhus Denmark Email: [email protected] Phone (fixed, office): +45 8716 8043 Cell phone: +45 4041 4454 Keywords: Colorectal Neoplasms, Primary Health Care, General Practice, Endoscopy, Digestive System Short running title: Endoscopy variation and colorectal cancer outcome Abstract Objective Variation in the propensity to refer patients to lower endoscopy among general practitioners (GPs) may be related to differences in colorectal cancer (CRC) outcomes. We aimed to investigate possible associations in a Danish clinical setting. Design We performed a nationwide population‐based cohort study of persons aged ≥40 years in 2004‐2011 (N=3.172.469). Practice populations were categorised into quartiles (Q1‐Q4) for each year according to GP referral propensity, which was estimated as adjusted referral rates for lower endoscopies in the preceding two years. Quartiles were compared for CRC incidence, stage at diagnosis, and proportions of patients receiving elective surgery, treatment with curative intent, or a diagnosis of poor‐prognosis CRC. Results In total, 324,157 outpatient lower endoscopies were performed after a GP referral. Rates varied two‐fold between the quartile with the lowest (Q1) and the highest (Q4) referral propensity. In total, 29,985 patients were diagnosed with CRC. The lowest incidence was found for Q1 patients compared to Q2, Q3 and Q4; the incidence rate ratio in Q4 was 1.04 (1.00‐1.08) owing to a higher incidence of colon cancer. More Q4 patients with rectal cancer were diagnosed in an early stage (49.2%) than Q1 patients (44.8%); odds ratio: 1.19 (1.05‐1.34). More Q4 patients were treated electively and with curative intent and fewer had poor‐
prognosis CRC, albeit the differences were not statistically significant. Conclusions This study revealed a trend towards better outcomes for CRC patients in practices with the highest propensity to refer to endoscopies, but differences were limited and the clinical relevance debatable. SUMMARY BOX What is already known about this subject? ‐
Sigmoidoscopy use has shown to improve the prognosis for colorectal cancer in screening trials, but effects of lower endoscopies are unknown in a non‐screening setting. ‐
Healthcare systems with a strong gatekeeping function in general practice have impaired colorectal cancer survival. ‐
General practitioners have a pivotal role in the diagnosis of symptomatic colorectal cancer but are challenged by vague symptomatology, and variation in the referral propensity to lower endoscopy may consequently impact the prognosis of colorectal cancer patients. What are the new findings? ‐
We showed variation among Danish general practitioners in the propensity to refer patients for lower endoscopy. ‐
We found that more rectal cancer patients were diagnosed in early stage in the general practices with the highest referral propensity ‐
Overall, exposure to different referral propensities resulted in only small differences in outcomes for included practice populations regarding received treatment. How might it impact on clinical practice in the foreseeable future? ‐
Our study indicates a small positive effect of performing more lower endoscopies in symptomatic patients; however, the endoscopy rates are generally low in Denmark. ‐
Future studies should investigate whether decreasing the referral threshold for lower endoscopy from general practice could improve the prognosis for colorectal cancer. BACKGROUND Colorectal cancer (CRC) is the third most common cancer1 worldwide. It is also the third most common cause of global cancer‐related death.2 CRC survival has improved in Denmark over the last decade, but remains poor compared with many other countries.3 Earlier and more expedited diagnosis is important to improve survival as longer time to diagnosis is associated with increased mortality.4 Furthermore, survival is strongly dependent on stage at diagnosis; the 5‐year survival is approximately 80% for early disease stages (UICC5 I and II) and less than 20% for more developed stages (stage IV).6 Randomised controlled screening trials with faecal occult‐blood testing or flexible sigmoidoscopy have shown reductions in CRC mortality even after 30 years of follow‐up.7‐14 A national screening programme with an immunochemical Fecal Occult Blood Test (iFOBT) was introduced in Denmark in March 2014. Nevertheless, the role of the GP remains crucial in diagnosing symptomatic CRC and more knowledge is needed about the possible effects of increasing the number of lower endoscopies among patients with symptoms of possible CRC. Most CRCs are still diagnosed after symptomatic presentation to first‐line health‐care providers, even in settings with CRC screening.15 Danish general practitioners (GP) constitute the first line of health care and act as gatekeepers to specialist medical treatment in a publicly funded healthcare system. Cancer diagnosis in general practice is challenged by the fact that the prevalence of possible cancer symptoms is high, while the positive predictive values of these symptoms are low.16‐19 In addition, half of CRC patients present with symptoms that are interpreted by the GP as ‘unspecific’.20 No referral guidelines for CRC existed in Denmark until 2008. Before the introduction of these guidelines, endoscopy referrals were based on an individual assessment made by the GP in each separate case. However, the implementation of referral guidelines and diagnostic urgent referral pathways in 2008 allowed the Danish GPs to refer patients presenting with symptoms suspicious of CRC (50‐65%of all patients20, 21) directly to lower endoscopy (Figure 1). The GPs must still diagnose the remaining cancers among patients with low‐risk‐but‐not‐no‐risk symptoms, and the decision whether to refer for further investigation may vary considerably among GPs. Knowledge about the possible consequences of these variations may help optimise the diagnostic process for CRC patients, and thereby diagnose patients at earlier stages and improve their survival.22 The aim of this study was to describe variations in the propensity to refer to lower endoscopy among Danish general practices and to investigate whether such propensity was associated with CRC incidence, stage at diagnosis, and proportions of patients subjected to elective surgery, being treated with curative intent, or having a poor‐prognosis CRC. METHOD Study design We conducted a population‐based study consisting of two steps. Firstly (step 1), we estimated the propensity to refer to lower endoscopies for each participating general practice in each calendar year from 2004 to 2011. Estimates were based on their referral rate for lower endoscopies in the preceding two‐year periods (2002‐2003, 2003‐2004…2009‐2010) (see Figure 2). Secondly (step 2), we categorised the practice populations into quartiles (Q1‐Q4) according to this referral propensity, and analysed the outcomes of cancer patients in these four exposure groups. The study period was 2002‐2010 for step 1 and 2004‐2011 for step 2. The following sections will describe the applied method and highlight any significant differences between the two steps. Study population The study population comprised all Danish citizens aged 40 years or older who were listed with a Danish general practice (>98% of all persons). All persons must contact their GP for medical advice, including referrals for specialty care. All persons in Denmark are assigned a unique personal identification number, which can be used to link information from various national registers at the individual level.23 Eligible study participants should be living in Denmark on 1 January 2004 (2002 for step 1). We included eligible persons at the beginning of the study or when they turned 40 years, whichever came last. Risk time was calculated as the time from inclusion in the study until CRC diagnosis, leaving the list system, death, emigration, or end of study, whichever came first. We excluded persons with a diagnosis of any chronic inflammatory bowel disease from step 1 on the first date of the diagnosis. Likewise, we excluded persons with a CRC diagnosis made before the beginning of the study. In addition, we excluded sixteen persons registered with a cancer diagnosis after their death. The Patient List Register was used to follow persons through time and allowed identification of the exact general practice at which each person was listed at any given time. The Register is used for remuneration of GPs and contains specific dates for any person’s change of general practices. Only persons at general practices with a calculated propensity for referral were eligible. Therefore, the person‐years of the 181 practices which exist for less than two years could not be included. This left us with 3,172,469 persons and 20,409,841 person‐years for the main analyses (step 2). Lower endoscopies (step 1) The Danish National Patient Register24 was used to identify all sigmoidoscopies and colonoscopies performed at public and private hospitals. The Register holds information on the referring entity and only endoscopies performed after a referral from general practice were included in the study. However, the ultimate decision to perform the endoscopy also relies on the surgeon/gastroenterologist (Figure 1). The Danish National Health Service Register25 was used to identify all lower endoscopies conducted in private practice gastroenterology, which were all considered to be initiated by GP referral. All performed endoscopies are publicly financed; also procedures conducted at private hospitals and private clinics. No mass screening programme including lower endoscopy existed in Denmark in the study period. Colorectal cancer (step 2) All cases of CRC were identified in the Danish Cancer Registry which contains records of all cancers diagnosed in Denmark since 1943.26 CRCs were identified using the ICD‐7 codes 153 and 154 and the ICD‐10 codes C18‐20. From the Danish Cancer Registry we used information on date of diagnosis and patient age at diagnosis. Since 1 May 2001, the Danish Colorectal Cancer Group (DCCG) has collected information on all patients with a first‐time diagnosis of colorectal adenocarcinoma treated at one of the Danish surgical departments. The completeness of this database has been shown to be above 95% since 2002.27 From the DCCG data, detailed information was collected on cancer stage according to the Union for International Cancer Control (UICC)5, type of cancer (colon vs. rectal), priority of surgery (acute/elective), and intent of treatment (curative/palliative).28, 29 We analysed the stage at diagnosis as the proportions with stages I‐II combined and stages I‐III combined. To investigate the proportion of patients with regional disease (stage III) among the potentially curable patients (stages I, II or III), we performed an analysis excluding stage IV cancers. Information about intent of treatment was only available for cancers diagnosed after 31 December 2004. We made a composite measure representing poor‐prognosis colorectal cancer for patients who had experienced one of the following: stage IV disease at diagnosis, acute surgery, or no surgery at all. Characteristics of the study population Data on education, ethnic origin, marital status, and urbanisation were obtained from Statistics Denmark for January 2002 (step 1) and January 2004 (step 2). Urbanisation was defined as the size of the city in which each person was living. The Charlson Comorbidity Index (CCI) was calculated for each year for every person by using the 10‐year history of diagnoses.30‐32 All available primary and secondary diagnoses in the Danish National Patient Register were included (both inpatient and outpatient hospital diagnoses). CRC diagnoses were not included in the CCI. Included diagnoses are listed in the Appendix. Included persons were divided into three groups according to their individual score (0, 1‐2, and ≥3). Statistical methods We calculated the number of expected lower endoscopies during 2002‐2010 (step 1) and the incidence rate ratios of CRC diagnoses in the four quartiles during 2004‐2011 (step 2) using Poisson regression models with person‐years as the denominator. The proportions of specific disease stages at diagnosis, performed elective surgery, and initiated treatment with curative intent were compared among the four groups using a logistic regression model. Adjusted models included calendar year, patient age in each calendar year (restricted cubic splines with six knots according to Harrel’s recommended percentiles33), gender, educational level (<10 years, 10‐12 years, or > 12 years), ethnic origin (Danish, western immigrant, or non‐
western immigrant), marital status (married/cohabitating or living alone) and CCI (0, 1‐2, or ≥3). Very small general practices with less than 100 person‐years in total were excluded. Practices were ranked and divided into quartiles for each year based on the ratio between observed and expected number of endoscopies in the preceding two years. Persons listed with practices with the lowest propensity to refer patients for lower endoscopy formed the reference group (Q1). Age at diagnosis was compared using one‐way analysis of variance. Robust variance estimation was performed to account for possible cluster effects at practice level. All analyses were stratified according to type of cancer (rectal and colon). All analyses were carried out using the statistical software STATA 13.1 (StataCorp LP, Lakeway Drive, College Station, TX, USA). RESULTS Categorisation of general practices (step 1) In step 1, we made the categorisation of general practices into four quartiles (Q1‐Q4) according to the preceding referral rates for lower endoscopy. We included a total of 3,171,781 persons (22,909,182 person years) who were registered with 2564 different general practices (see the flow diagram in Figure 1 in the Appendix). A total of 581,373 lower endoscopies were identified during 2002‐2010; 324,157 (56%) of these were outpatient endoscopies performed after referral from general practice. The exposure groups, Q1‐Q4, were similar with respect to patient gender, age, marital status, ethnicity and comorbidity, but differed slightly in educational level and urbanisation as more persons from the bigger cities formed part of Q4 (Table 1). Endoscopy rates showed a two‐fold variation in 2004‐2011 between the least and the most referred populations according to the categorisation in step 1, i.e. a crude endoscopy rate in Q1 on 9.9/1000 persons‐years and 20.3/1000 person‐years in Q4 (Table 1). The proportion of endoscopies performed at private clinics was 13.7% in Q1 and 29.8% in Q4. Colorectal cancer diagnoses Among the 3,172,469 persons registered with the general practices included in the main analyses, 29,985 were diagnosed with CRC (Table 2). Patients in Q2‐Q4 had a 3‐4% higher incidence of CRC (Table 2). The DCCG held information on 28,270 (95%) of the CRC patients; 18,666 were colon cancers and 9604 were rectal cancers. Stratification revealed that the higher incidences in Q2‐Q4 were statistically significant for colon cancer, whereas no differences were seen for rectal cancer (Table 3). Tumour characteristics The overall mean age at diagnosis was 71.2 years, with non‐significant differences between the quartiles. Information on stage at diagnosis was available for 26,524 cancer patients. The proportion of cancers diagnosed in stages I‐II was 44.0% in Q1 and 45.9% in Q4, corresponding to an OR in Q4 of 1.08 (1.01‐1.16) (Table 2). This difference was biggest for rectal cancer, with an OR in Q4 of 1.19 (1.05‐1.34) (Table 3). Overall, the proportions of stages I‐III cancers were almost similar, with an OR of 1.04 (0.96‐1.13) in Q4. The proportion of stage III cancers among the potentially curable disease stages (stages I‐III) was 37.2% in Q1 and 35.2% in Q4 (OR: 0.92 (0.84‐1.01)). This difference was mainly caused by rectal cancer, where 37.0% in Q1 and 32.8% in Q4 were diagnosed in stage III (OR: 0.92 (0.84‐1.01)). Treatment Among the 27,456 patients with available information on performed surgery, we found that the proportion of patients receiving elective surgery was marginally higher in Q3 and Q4. However, this was only statistically significant for colon cancer patients in Q3 (OR: 1.12 (1.01‐1.24). Information on curative/palliative intent was available for 24,163 patients (only after 2004), and the proportions of patients treated with curative intent were similar across the quartiles (Table 2). The proportion of patients with a poor‐prognosis CRC was lowest in Q3 and Q4, especially for rectal cancer patients, although not statistically significant (Tables 2 and 3). Sensitivity analyses When we included all endoscopies (not only those requested by GP referral) in the calculation of the exposure, the analyses showed similar results. When we included only single‐handed practices, the analyses produced comparable results as did analyses in which we had entirely excluded patients with inflammatory bowel disease (data not shown). To reduce the possible effects of urbanisation, we excluded practices and persons from cities with >50.000 inhabitants; this revealed similar results. In two areas, screening trials with FOBT were carried out in August 2005‐November 2006. This led to an increased use of endoscopies, but exclusion of general practices and patients from these areas did not affect the results.
DISCUSSION Main findings This large population‐based cohort study showed that in the quartile of general practices with the highest propensity to refer to lower endoscopies patients received twice as many endoscopies than patients in the quartile with the lowest propensity between 2004 and 2011. Overall, patients registered within practices with a higher referral propensity experienced a higher incidence of colon cancer and had a higher proportion of rectal cancers diagnosed in early stages. However, as the absolute differences were small, the clinical relevance is debatable. We found no statistically significant differences regarding the proportion of patients receiving elective surgery or treatment with curative intent or patients with a poor‐prognosis CRC (no surgery, acute surgery, or stage IV disease). Strengths and limitations A major strength of the present study is the use of high‐quality nationwide population‐based registers. The Danish Civil Registration System keeps records of deaths and emigration, which enabled us to establish a cohort of all Danish citizens of 40 years and older listed at a GP (>98%) and follow it for up to ten years without loss to follow‐up. The national registers are public and the Danish system is based on a principle of free and equal access to all citizens to healthcare services. Selection bias is, therefore, an unlikely explanation of our results. The inclusion of data from more than 320,000 lower endoscopies and almost 30,000 CRC patients gave the study a high statistical precision, which is also reflected in the narrow confidence intervals for most estimates. We had information on all lower endoscopies performed by public hospitals, private practicing specialists, and private hospitals in Denmark. The registrations of lower endoscopies are considered to be complete as they form the basis for remuneration of health contractors (i.e. hospitals and private practicing specialists). Only 1715 (5.7%) patients were not recorded in DCCG, but the missing information was equally distributed across the four investigated exposure quartiles. Furthermore, the quality of the registrations in DCCG is considered high.27‐29 Therefore, it is implausible that information biases explain our findings as any inconsistencies in registry recordings are unlikely to be related to both the exposure and the outcomes investigated in our study. Some non‐differential misclassification may have occurred as we analysed general practices and not the individual GPs. This may have caused an underestimation of the variation and bias our results towards the null as variations within a general practice could be masked when aggregated at practice level. Nevertheless, the analyses of single‐handed practices did not noticeably change the results. The observational design may present a limitation of the study as the use of endoscopy may not be the sole explanation of the findings. This implies that our findings may suffer from ecological fallacy, meaning that any identified associations might not be causal in nature, but may rather be confounded. We aimed to investigate an effect of the propensity to refer among GPs and not their overall use of referrals during the study period as this would be more influenced by confounding by indication. Furthermore, investigating the overall use could lead to reversed causality if GPs encountering patients with late‐stage cancer subsequently increased their referral rates. This may lead to the conclusion that more endoscopies were associated with poorer outcomes. Our approach showed that the rates of performed endoscopies in the preceding two‐year period did actually predict differences in the current referral rate as the variation between Q1 and Q4 still remained two‐fold (Table 1). However, if the indications were known and stratified into ‘symptoms indicating CRC’ and ‘non‐specific symptoms’, the study could have added even further valuable information. In line with other health services research, we investigated the effects of different sectors combined. The included endoscopies were performed after GP referral to a hospital department or private clinic performing lower endoscopies. Therefore, the decision to perform an endoscopy was also influenced by others than the GPs (Figure 1). The analyses were adjusted for plausible confounders driving the demand for lower endoscopies, but we cannot exclude residual confounding. The differences in Table 1 regarding urbanisation indicate some geographical variations in utilisation and perhaps access. The confounding effect of e.g. urbanisation on mortality was strong, and attempts to investigate mortality differences were discarded as the results were clearly influenced by residual confounding. Instead, we chose to investigate other factors such as treatment received that are known to be associated with prognosis and less likely to be confounded.34, 35 Comparison with other studies To our knowledge, this is the first study to investigate the effect of variations in GP propensity to refer to lower endoscopy among general practice populations. Other studies on the effects of variation in general practice have found diverse effects depending on the chosen subject matter. Hippisley‐Cox et al36 found no relation between cancer stage at CRC diagnosis and the overall referral rate in general practice. In contrast, Shawihdi et al37 found that low rates of gastroscopy were associated with suboptimal outcomes for oesophago‐gastric cancer patients. In the tertile of general practices in England ordering fewest gastroscopies, the registered patients had less major surgery, more emergency admissions, and higher mortality.37 This was contrasted by our findings as we saw no statistically significant differences in the proportion receiving treatment with curative intent. Acute surgery is related to later cancer stages and substantially lower survival than elective surgery.34 Importantly, we found only small differences in the proportions of patients receiving elective surgery or treatment with curative intent between the groups. One explanation of the discrepancies between the former studies may be differences in the magnitude of the variation, in the clinical settings, and/or in symptoms. The two‐fold variation in the present study is modest compared to other studies on variation and may not be sufficient enough to cause significant impact on outcomes. Our approach allowed for general practices to change behaviour, and this may be important as lower endoscopies are generally performed only rarely. Sigmoidoscopy screening trials and observational studies on colonoscopy use have formerly shown associations between increased number of endoscopies and reduced cancer incidence, more favourable stage distributions at diagnosis, and ultimately lower mortality.12, 38‐43 These studies are not entirely comparable to our study as we investigated the use of endoscopies in symptomatic patients, but they still suggest a potential for earlier diagnosis and improved prognosis in CRC patients. Screening with lower endoscopy has never been encouraged in Denmark, which may explain the low overall endoscopy rates. In our study period, the average rate was approximately 30 endoscopies per 1000 person‐years (including all endoscopies) compared to 60 per 1000 person‐years and above in studies from the US and Canada.38, 39 The substantially lower endoscopy rates in all four quartiles of our study may explain the limited effect of the relative differences in endoscopy rates, both for cancer incidence and stage distribution. Regarding the stage distribution among curable cancers (stages I‐III), we found fewer rectal cancers with regional disease (stage III) in Q4. However, this finding should be interpreted with caution as proportion differences in stage IV cancers were not considered in the calculations. Generalisation of our results to other healthcare settings demand the presence of similar contextual conditions and healthcare structures. Differences in access to endoscopy services and overall endoscopy rates may influence the possible associations between endoscopy rates and outcomes. Denmark and the UK are both lacking behind other European countries in terms of CRC survival.3 Different strategies, including public information campaigns and iFOBT screening (introduced in Denmark in March 2014), have been used to improve outcomes. The current study could not clearly confirm whether a higher referral rate to lower endoscopy for symptomatic patients could lead to a more favourable stage distribution and better outcomes. However, our study suggests that there may be a positive effect of performing more endoscopies, especially for rectal cancer, which could be investigated in a future intervention study offering some GPs easier access to endoscopic facilities. Differences in endoscopy rates across urban‐rural areas indicate that different access levels to healthcare facilities may cause inequality in health. Yet, our results suggest that the identified disparities only have limited impact on the prognostic factors measured in this study. CONCLUSION This is the first study to investigate possible consequences of variations in the referral propensity among GPs in a healthcare setting without CRC screening. We found that a higher propensity to refer to lower endoscopies was associated with a higher incidence of colon cancer and a higher proportion of rectal cancers diagnosed in early stage (UICC I+II). The differences were small, and the clinical relevance is debatable. The limited effect could reflect the modest variations among Danish GPs in their propensity to refer to lower endoscopies. A higher overall referral rate for endoscopies in Danish general practice could be beneficial for future CRC patients, and this issue should be investigated in an intervention study. Ethical approval The project was approved by the Danish Data Protection Agency (J.no.: 2009‐41‐3471). According to Danish law, approval from the Danish National Committee on Health Research Ethics was not required as de‐
identified register data were used for the performed analyses. Author contributions Hjertholm had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Hjertholm and Fenger‐Grøn are responsible for the data analysis. Study concept and design: Hjertholm, Fenger‐Grøn, Christensen, Vestergaard, Iversen, Vedsted Acquisition of data: Hjertholm Analysis and interpretation of data: Hjertholm, Fenger‐Grøn, Christensen, Vestergaard, Iversen, Vedsted Drafting of the manuscript: Hjertholm, Fenger‐Grøn, Christensen, Vestergaard, Vedsted Critical revision of the manuscript for important intellectual content: Hjertholm, Fenger‐Grøn, Vestergaard, Christensen, Iversen, Vedsted Statistical analysis: Hjertholm, Fenger‐Grøn Study supervision: Christensen, Vestergaard, Vedsted Conflicts of interest The authors report no conflict of interest. The authors alone are responsible for the content and writing of the paper. Funding/support This work was supported by Aarhus University, the Danish National Research Foundation for Primary Care, the Novo Nordic Foundation and the Danish Cancer Society. The funding sources had no role in the design and conduct of the study (data collection, analysis, and interpretation) or in preparation or approval of the manuscript. Acknowledgements The authors would like to thank Kaare Flarup at the Research Centre for Cancer Diagnosis in Primary Care at Aarhus University for the initial data management. REFERENCES 1. Ferlay J, Shin HR, Bray F, Forman D, Mathers C, Parkin DM. Estimates of worldwide burden of cancer in 2008: GLOBOCAN 2008. Int J Cancer 2010; Dec 15;127(12):2893‐917. 2. The Danish National Board of Health. Cause of Death Register 2012 [report in Danish]. ; 2013. 3. Coleman MP, Forman D, Bryant H, Butler J, Rachet B, Maringe C, et al. Cancer survival in Australia, Canada, Denmark, Norway, Sweden, and the UK, 1995‐2007 (the International Cancer Benchmarking Partnership): an analysis of population‐based cancer registry data. Lancet 2011; Jan 8;377(9760):127‐38. 4. Torring ML, Frydenberg M, Hamilton W, Hansen RP, Lautrup MD, Vedsted P. Diagnostic interval and mortality in colorectal cancer: U‐shaped association demonstrated for three different datasets. J Clin Epidemiol 2012; Jun;65(6):669‐78. 5. Sobin LH, Gospodarowicz MK, Wittekind C, International Union against Cancer. TNM classification of malignant tumours. 7. ed. ed. Chichester, West Sussex, UK: Wiley‐Blackwell; 2010. 6. The Danish Colorectal Cancer Group. Annual report 2013 [Danish]. Available at: http://www.dccg.dk/03_Publikation/2013.pdf. Accessed 11/03, 2014. 7. Shaukat A, Mongin SJ, Geisser MS, Lederle FA, Bond JH, Mandel JS, et al. Long‐term mortality after screening for colorectal cancer. N Engl J Med 2013; Sep 19;369(12):1106‐14. 8. Mandel JS, Church TR, Ederer F, Bond JH. Colorectal cancer mortality: effectiveness of biennial screening for fecal occult blood. J Natl Cancer Inst 1999; Mar 3;91(5):434‐7. 9. Kewenter J, Brevinge H, Engaras B, Haglind E, Ahren C. Results of screening, rescreening, and follow‐up in a prospective randomized study for detection of colorectal cancer by fecal occult blood testing. Results for 68,308 subjects. Scand J Gastroenterol 1994; May;29(5):468‐73. 10. Kronborg O, Fenger C, Olsen J, Jorgensen OD, Sondergaard O. Randomised study of screening for colorectal cancer with faecal‐occult‐blood test. Lancet 1996; Nov 30;348(9040):1467‐71. 11. Hoff G, Grotmol T, Skovlund E, Bretthauer M, Norwegian Colorectal Cancer Prevention Study Group. Risk of colorectal cancer seven years after flexible sigmoidoscopy screening: randomised controlled trial. BMJ 2009; May 29;338:b1846. 12. Atkin WS, Edwards R, Kralj‐Hans I, Wooldrage K, Hart AR, Northover JM, et al. Once‐only flexible sigmoidoscopy screening in prevention of colorectal cancer: a multicentre randomised controlled trial. The Lancet 2010;375(9726):1624‐33. 13. Segnan N, Armaroli P, Bonelli L, Risio M, Sciallero S, Zappa M, et al. Once‐only sigmoidoscopy in colorectal cancer screening: Follow‐up findings of the italian randomized controlled trial ‐ SCORE. J Natl Cancer Inst 2011;103(17):1310‐22. 14. Schoen RE, Pinsky PF, Weissfeld JL, Yokochi LA, Church T, Laiyemo AO, et al. Colorectal‐cancer incidence and mortality with screening flexible sigmoidoscopy. N Engl J Med 2012;366(25):2345‐57. 15. Hamilton W. Five misconceptions in cancer diagnosis. Br J Gen Pract 2009; Jun;59(563):441,5, 447; discussion 446. 16. Hamilton W, Round A, Sharp D, Peters T. Clinical features of colorectal cancer before diagnosis: a population‐based case‐control study. Br J Cancer 2005;93:399‐405. 17. Svendsen RP, Stovring H, Hansen BL, Kragstrup J, Sondergaard J, Jarbol DE. Prevalence of cancer alarm symptoms: a population‐based cross‐sectional study. Scand J Prim Health Care 2010; Sep;28(3):132‐7. 18. Shapley M, Mansell G, Jordan JL, Jordan KP. Positive predictive values of >/=5% in primary care for cancer: systematic review. Br J Gen Pract 2010; Sep;60(578):e366‐77. 19. Astin M, Griffin T, Neal RD, Rose P, Hamilton W. The diagnostic value of symptoms for colorectal cancer in primary care: a systematic review. Br J Gen Pract 2011; May;61(586):231‐43. 20. Nielsen TN, Hansen RP, Vedsted P. Symptom presentation in cancer patients in general practice]. Ugeskr Laeger 2010; Oct 11;172(41):2827‐31. 21. Neal RD, Din NU, Hamilton W, Ukoumunne OC, Carter B, Stapley S, et al. Comparison of cancer diagnostic intervals before and after implementation of NICE guidelines: analysis of data from the UK General Practice Research Database. Br J Cancer 2014; 02/04;110(3):584‐92. 22. Folkesson J, Engholm G, Ehrnrooth E, Kejs AM, Pahlman L, Harling H, et al. Rectal cancer survival in the Nordic countries and Scotland. Int J Cancer 2009; Nov 15;125(10):2406‐12. 23. Pedersen CB. The Danish Civil Registration System. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):22‐5. 24. Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient Register. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):30‐3. 25. Sahl Andersen J, De Fine Olivarius N, Krasnik A. The Danish National Health Service Register. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):34‐7. 26. Gjerstorff ML. The Danish Cancer Registry. Scandinavian Journal of Public Health 2011; July 01;39(7 suppl):42‐5. 27. Iversen LH. Aspects of survival from colorectal cancer in Denmark. Dan Med J 2012; Apr;59(4):B4428. 28. Available at: http://www.dccg.dk/. Accessed 05/05, 2014. 29. Gogenur I, Ingeholm P, Iversen LH. Danish Colorectal Cancer Database. Ugeskr Laeger 2012; Oct 15;174(42):2525. 30. Sundararajan V, Quan H, Halfon P, Fushimi K, Luthi JC, Burnand B, et al. Cross‐national comparative performance of three versions of the ICD‐10 Charlson index. Med Care 2007; Dec;45(12):1210‐5. 31. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40(5):373‐83. 32. Thygesen S, Christiansen C, Christensen S, Lash T, Sorensen H. The predictive value of ICD‐10 diagnostic coding used to assess Charlson comorbidity index conditions in the population‐based Danish National Registry of Patients. BMC Medical Research Methodology 2011;11(1):83. 33. Frank E. Harrell J. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. 1st ed. New York: Springer Science & Business Media; 2001. 34. McArdle CS, Hole DJ. Emergency presentation of colorectal cancer is associated with poor 5‐year survival. Br J Surg 2004; May;91(5):605‐9. 35. The Danish Colorectal Cancer Group. Annual report 2012 [Danish]. Available at: http://www.dccg.dk/pdf/Aarsrapport_2012_dccg.pdf. Accessed 09/04, 2014. 36. Hippisley‐Cox J, Hardy C, Pringle M, Fielding K, Carlisle R, Chilvers C, et al. Are patients who present late with cancer registered with low referring practices?. Br J Gen Pract 1997; Nov;47(424):731‐2. 37. Shawihdi M, Thompson E, Kapoor N, Powell G, Sturgess RP, Stern N, et al. Variation in gastroscopy rate in English general practice and outcome for oesophagogastric cancer: retrospective analysis of Hospital Episode Statistics. Gut 2014; Feb;63(2):250‐61. 38. Rabeneck L, Paszat LF, Saskin R, Stukel TA. Association between colonoscopy rates and colorectal cancer mortality. Am J Gastroenterol 2010; Jul;105(7):1627‐32. 39. Gross CP, Andersen MS, Krumholz HM, McAvay GJ, Proctor D, Tinetti ME. Relation between Medicare screening reimbursement and stage at diagnosis for older patients with colon cancer. JAMA 2006; Dec 20;296(23):2815‐22. 40. Siegel RL, Ward EM, Jemal A. Trends in colorectal cancer incidence rates in the United States by tumor location and stage, 1992‐2008. Cancer Epidemiology Biomarkers and Prevention 2012;21(3):411‐6. 41. Baxter NN, Goldwasser MA, Paszat LF, Saskin R, Urbach DR, Rabeneck L. Association of colonoscopy and death from colorectal cancer. Ann Intern Med 2009;150(1):1‐8. 42. Holme O, Loberg M, Kalager M, Bretthauer M, Hernan MA, Aas E, et al. Effect of flexible sigmoidoscopy screening on colorectal cancer incidence and mortality: a randomized clinical trial. JAMA 2014; Aug 13;312(6):606‐15. 43. Hermann Brenner, Christian Stock, Michael Hoffmeister. Effect of screening sigmoidoscopy and screening colonoscopy on colorectal cancer incidence and mortality: systematic review and meta‐analysis of randomised controlled trials and observational studies. BMJ 2014; BMJ Publishing Group Ltd;348. 44. Referral guidelines for patients with symptoms suspicious of colorectal cancer [in Danish]. Available at: http://sundhedsstyrelsen.dk/publ/Publ2012/06juni/KraeftPkforl/TykogEndetarm3udg.pdf. Accessed 12/4, 2014. Table 1: Characteristics of persons aged 40 years or more according to practice quartiles and lower endoscopy rates (Quartile 1 (Q1): lowest propensity to refer, quartile 4 (Q4): highest propensity to refer) Total Q1
Q2
Q3 Q4
Risk time 20,409,841 4,358,504 5,521,772 5,679,677 4,849,888 Gender (% of men) 48.1 48.8 48.4 47.9 47.3 Age, mean (sd) 59.2 (12.9) 59.5 (13.0) 59.3 (12.9) 59.1 (12.8) 59.0 (12.9) Educationa < 10 years, % 37.5 40.1 38.4 37.0 35.0 10‐12 years, % 39.9 39.3 40.0 40.3 39.7 > 12 years, % 22.6 20.6 21.6 22.7 25.4 Marital statusb (% of single) 29.3 29.6 28.3 28.5 30.9 Ethnicityc Danish, % 93.9 94.4 94.3 94.0 93.0 Western immigrant, % 2.6 2.4 2.4 2.5 2.9 Non‐western immigrant, % 3.5 3.2 3.3 3.4 4.1 Charlson index 0 79.9 80.1 80.2 79.9 79.3 1 to 2 16.3 16.1 16.1 16.2 16.6 ≥ 3 3.9 3.7 3.8 3.9 4.1 Urbanisation Capital city (Copenhagen) 17.9 13.3 14.6 17.0 26.8 > 50,000 12.5 11.3 10.2 11.5 17.3 10,000‐49,999 22.2 20.4 22.7 24.1 20.9 1000‐9999 24.4 25.8 27.6 25.1 18.9 200‐999 8.5 10.5 9.1 8.4 6.1 < 200 and rural 9.8 14.4 18.4 15.8 13.9 Lower endoscopiesd Number 301,792 42,511 72,885 89,885 96,511 Crude rate (scopies/1000 years) 9.9 13.4 16.1 20.3 Unadjusted IRR Ref 1.35 (1.34‐1.37) 1.62 (1.61‐1.64) 2.05 (2.02‐2.07) Adjustede IRR Ref 1.34 (1.33‐1.36) 1.61 (1.59‐1.63) 2.02 (2.00‐2.04) Proportion of endoscopies 22.1 13.7
17.3
23.3 29.8
performed at private clinics, % a
Unknown in 6.0% of the persons, mainly among the elderly. They are included in the < 10 years group b
Missing in 65 persons excluded from the adjusted analyses c
Missing in 4 persons excluded from the adjusted analyses d
Scopies performed after referral from general practice from 2004‐2011 in the quartiles defined by the referral rate the preceding two years. e
Adjusted for gender, age, educational level, marital status, ethnicity, and Charlson Comorbidity Index IRR: incidence rate ratio OR: odds ratio Table 2: Colorectal cancer incidence and outcomes according to exposure quartile at diagnosis for all colorectal cancers combined. Proportions in % where not otherwise stated (bold numbers indicate statistically significance) Total
Cancer incidence Number IRR, unadjusted IRR, adjusted Age at diagnosis, mean Q1
Q2
29,985
6386
8254
Ref 1.02 (0.99‐1.06)
Ref 1.04 (1.00‐1.07)
71.2
71.5
71.2
Q3 Q4
8293 7052
1.00 (0.96‐1.03) 0.99 (0.96‐1.03)
1.03 (1.00‐1.07) 1.04 (1.00‐1.08)
71.1 71.3
Stage at diagnosis (N=26,524) Stage I‐II (n= 11,928) 45.0
44.0
44.4
45.5 45.9
25.7
26.0
26.5
25.4 24.9
Stage III (n= 6821) Stage IV (n=7775) 29.3
30.0
29.1
29.1 29.3
Stage I‐II, adjusted OR Ref 1.02 (0.95‐1.09) 1.07 (1.00‐1.14) 1.08 (1.01‐1.16)
Stage I‐III, adjusted OR Ref 1.04 (0.97‐1.13) 1.05 (0.98‐1.14) 1.04 (0.96‐1.13)
Elective surgery (N= 27,456, n=21,111) 76.9
76.2
76.4
77.5 77.4
Elective surgery, adjusted OR Ref 1.00 (0.93‐1.09) 1.08 (0.99‐1.18) 1.08 (0.99‐1.18)
Curative intent (N=24,163, n=18,490) 76.5
76.0
76.7
76.8 76.5
Curative intent, adjusted OR Ref 1.03 (0.94‐1.12) 1.05 (0.96‐1.15) 1.03 (0.94‐1.13)
b
Poor‐prognosis CRC (N=27,456, n=9980) 36.4
37.1
36.6
35.9 36.0
Poor‐prognosis CRC, adjusted OR Ref 0.99 (0.92‐1.06) 0.95 (0.88‐1.02) 0.95 (0.88‐1.03)
a
Information on intent of treatment was not available for cancers diagnosed in 2004 b Poor‐prognosis CRC: stage IV disease, no surgery, or acute surgery Adjusted analyses include gender, age, educational level, marital status, ethnicity, and Charlson Comorbidity Index IRR: incidence rate ratio OR: odds ratio Table 3: Colorectal cancer incidence and outcomes according to exposure quartile at diagnosis stratified into colon and rectal cancers. Proportions in % were not otherwise stated (bold numbers indicate statistically significance) Q2 Q3 Q4 Ref
1.05 (1.01‐1.10)
1.04 (1.00‐1.09) 1.08 (1.03‐1.13)
44.1
25.8
30.1
43.6
25.9
30.6
Ref
Ref
43.9
26.7
29.3
1.02 (0.94‐1.11)
1.07 (0.97‐1.17)
44.5 25.4 30.2 1.05 (0.97‐1.14) 1.04 (0.94‐1.14) 44.3
25.3
30.4
1.04 (0.95‐1.13)
1.02 (0.92‐1.12)
Elective surgery (N= 18,141, n= 13,358) Elective surgery, adjusted OR 73.6
72.6
Ref
73.1
1.02 (0.93‐1.12)
74.6 1.12 (1.01‐1.24) 74
1.08 (0.98‐1.20)
Curative intenta (N= 16,004, n= 12,303) Curative intent, adjusted OR 76.9
76.3
Ref
77
1.03 (0.92‐1.14)
77.5 1.08 (0.97‐1.20) 76.5
1.02 (0.91‐1.14)
Poor‐prognosis CRCb (N=18,141, n= 7222) 39.8
Poor‐prognosis CRC, adjusted OR 40.6
Ref
39.9
0.98 (0.90‐1.06)
39.2 0.94 (0.86‐1.03) 39.7
0.96 (0.88‐1.06)
Ref
1.02 (0.96‐1.07)
1.02 (0.96‐1.07) 0.97 (0.92‐1.03)
46.8
25.5
27.8
44.8
26.3
28.9
Ref
Ref
45.4
26
28.6
1.01 (0.90‐1.15)
1.01 (0.88‐1.16)
47.7 25.6 26.8 1.10 (0.98‐1.24) 1.10 (0.96‐1.25) 49.2
24.0
26.8
1.19 (1.05‐1.34)
1.12 (0.96‐1.29)
Elective surgery (N= 9315, n= 7753) Elective surgery, adjusted OR 83.2
83
Ref
82.6
0.98 (0.83‐1.16)
83.1 1.02 (0.86‐1.20) 84.4
1.13 (0.95‐1.34)
Curative intenta (N= 8159, n= 6187) Curative intent, adjusted OR 75.8
75.3
Ref
76.0
1.03 (0.88‐1.21)
75.6 1.02 (0.88‐1.19) 76.4
1.06 (0.90‐1.25)
Total
Colon (N=18,666) Cancer incidence, adjusted IRR Stage at diagnosis (N= 17,783) Stage I‐II (n= 7842) Stage III (n= 4592) Stage IV (n=5349) Stage I‐II, adjusted OR Stage I‐III, adjusted OR Rectal (N=9,604) Cancer incidence, adjusted IRR Stage at diagnosis (N= 8741) Stages I‐II (n= 4086) Stage III (n= 2229) Stage IV (n=2426) Stages I‐II, adjusted OR Stages I‐III, adjusted OR Q1 30.4
30.3
29.6 28.2
Poor‐prognosis CRCb (N= 9315, n= 2758) 29.6
Poor‐prognosis CRC, adjusted OR Ref 0.99 (0.87‐1.14) 0.96 (0.85‐1.09) 0.89 (0.78‐1.03)
a
Information on intent of treatment was not available for cancers diagnosed in 2004 b Poor‐prognosis CRC: stage IV disease, no surgery, or acute surgery Adjusted analyses include gender, age, educational level, marital status, ethnicity, and Charlson Comorbidity Index IRR: incidence rate ratio OR: odds ratio Figure 1: Guidelines introduced in 2008 for urgent referral of patients from general practice with symptoms indicating possible colorectal cancer according to the Danish Medicines and Health Authority44 Patients ≥40 years with one of the following symptoms should be referred through the urgent referral pathway ‐
‐
‐
‐
Visible rectal bleeding Change in bowel habits for four weeks or more Iron deficiency anaemia Significant general symptoms (e.g. unexplained weight loss, abdominal pain) A diagnostic strategy for surgeons was published in 2001 recommending lower endoscopy for the above symptoms. Surgical departments have used this strategy since 2002 in their decision‐making when assesing referrals from general practice (DCCG guidelines)28 Figure 2: Schematic presentation of the exposure calculation. Lower endoscopy rates are calculated in the two‐year period (grey dotted line) as a measure of the propensity to refer to lower endoscopy; the quartile of each practice is used as the exposure in the following year (black solid line) Paper III
Supporting information, Table 1 Study III
Webappendix for “Variations among general practitioners in the referral propensity for lower endoscopy and outcome of colorectal cancer patients”, Hjertholm et al. Table 1. ICD‐10 codes used for calculation of Charlson Comorbidity Index (CCI). The CCI was calculated using the Quan version as described by Sundararajan et al.1 modified according to the ICD‐10 codes listed in this table. Weights Condition ICD‐10 codes 1 Myocardial infarction I21; I22; I23 Congestive heart failure I50; I11.0; I13.0; I13.2 Peripheral vascular disease I70; I71; I72; I73; I74; I77 Cerebrovascular disease I60‐I69; G45; G46 Dementia F00‐F03; F05.1; G30 Chronic pulmonary disease J40‐J47; J60‐J67; J68.4; J70.1; J70.3; J84.1; J92.0; J96.1; J98.2; J98.3 Connective tissue disease M05; M06; M08; M09; M30; M31; M32; M33; M34; M35; M36; D86 Ulcer disease K22.1; K25‐K28 Mild liver disease B18; K70.0‐K70.3; K70.9; K71; K73; K74; K76.0 Diabetes without end organ E10.0, E10.1; E10.9; E11.0; E11.1; damage 2 E11.9 Diabetes with end organ E10.2‐E10.8, E11.2‐E11.8 damage Hemiplegia G81; G82 Moderate to severe renal I12; I13; N00‐N05; N07; N11; disease N14; N17‐N19; Q61 Non‐metastatic solid tumour C00‐C75 Leukaemia C91‐C95 Lymphoma C81‐C85; C88; C90; C96 3 Moderate to severe disease liver B15.0; B16.0; B16.2; B19.0; K70.4; K72; K76.6; I85 6 Metastatic cancer C76‐C80 AIDS B21‐B24 1. Sundararajan V, Quan H, Halfon P, et al. Cross‐national comparative performance of three versions of the ICD‐10 charlson index. Med Care. 2007;45(12):1210‐1215.
183 Clinical Activity in General Practice and Cancer
Supporting information, Figure 1 Study III
Figure 1: Flow of included persons and practices for the calculation of rates of lower endoscopies General practices: 2756 Persons: 3,174,737 Observation time (years): 23,029,860 Providers: 2575 Persons: 3,171,894 Observation time (years): 22,913,508 General practices: 2572 Persons: 3,171,785 Observation time (years): 22,909,245 184 Practices existing for less than two years and persons only registered with these: General practices: 181 Persons: 2843 Observation time (years): 116,352 General practices: 2564 Persons: 3,171,781 Observation time (years): 22,909,182 Scopies after GP referral: 324,157 All scopies: 581.373 Practices with non‐coherent observation time and persons only registered with these: General practices: 3 Persons: 109 Observation time (years): 4263 Practices with less than 100 observation years in total and persons only registered with these: General practices: 8 Persons: 4 Observation time (years): 63