Download Introduction - Open Research Exeter

Exfoliated colonocyte DNA levels and clinical features in the diagnosis
of colorectal cancer: a cohort study in patients referred for investigation.
Lalitha Mahadavan,1,3 Alexandre Loktionov,2 Ian R Daniels,1,2 Angela Shore,3 Diane
Cotter,1 Andrew H Llewelyn2 and William Hamilton.3
1. Royal Devon and Exeter NHS Foundation Trust, Barrack Road, Exeter
2. Colonix Medical Ltd.
3. Peninsula College of Medicine and Dentistry, University of Exeter and
Peninsula NIHR clinical research facility. Veysey Building, Salmonpool Lane,
Exeter, EX2 4SG.
Lalitha Mahadavan, research fellow
Alexandre Loktionov, laboratory researcher
Ian R. Daniels, consultant colorectal surgeon
Angela Shore, professor of cardiovascular science and Vice Dean research
Diane Cotter, researcher
Andrew H Llewelyn, medical researcher
William T. Hamilton, general practitioner and professor of primary care diagnostics
Original article, submitted to Colorectal Disease
Key words: colorectal cancer, diagnosis, proctoscopy
Word count: 2962
Corresponding author: Prof. Hamilton [email protected]. Phone number
(Secretary) 01392 726090. Prof. Hamilton is deaf, however, and e-mail
communication is the optimum.
1
Abstract
Aim. Selection of patients for investigation of suspected colorectal cancer is difficult.
One possible improvement may be to measure DNA isolated from exfoliated cells
collected from the rectum.
Method. This was a cohort study in a surgical clinic in the Royal Devon & Exeter
Hospital, Devon, UK. Participants were aged ≥40 years referred for investigation of
suspected colorectal cancer. Exclusion criteria were: inflammatory bowel disease,
previous gastro-intestinal malignancy, or recent investigation. A sample of the
mucocellular layer of the rectum was taken with an adapted proctoscope (the Colonix
system). Haemoglobin, mean cell volume, ferritin, carcino-embryonic antigen and
faecal occult bloods were tested. Analysis was by logistic regression.
Results. 828 patients were offered participation; 717 completed the investigations.
Three were lost to follow-up. 72 (10%) had colorectal cancer. Exfoliated cell DNA
was higher in cancer: median 5.4μg/ml (inter-quartile range 1.8,12), without cancer
2.0μg/ml (0.78,5.5), p<0.001.
Seven variables were independently associated with cancer: age, odds-ratio 1.05 per
year (95% confidence interval 1.02,1.08), p<0.001; DNA 1.05 per μg/ml (1.01,3.6),
p=0.01; mean cell volume per fl 0.93 (0.89,0.97), p=0.001; carcino-embryonic antigen
1.02 per μg/l (1.00,1.04), p=0.02; male sex 2.0 (1.1,3.6), p=0.02, rectal bleeding 2.4
(1.3,4.5), p=0.007; positive faecal occult blood 6.7 (3.4, 13), p<0.001.
The area under the receiver-operating characteristic curve for the DNA score was 0.65
(0.58 to 0.72)and for the seven-variable model 0.88 (CI 0.84 to 0.92).
2
Conclusions. Quantification of exfoliated DNA from rectal cellular material has
promise in the diagnosis of colorectal cancer, but this requires confirmation in a
second cohort.
What is new in this paper
Better selection of patients for investigation of suspected colorectal cancer is needed.
Rectal exfoliated DNA is associated with cancer, though not enough to use alone.
However, when combined with symptoms, blood tests, and faecal occult bloods, the
overall predictive power is much higher than for any previous system of triage.
3
Introduction
Over 30,000 new colorectal cancers (CRCs) are diagnosed annually in the UK,
leading to over 16,000 deaths.1 Survival in the UK is poor in comparison with many
other European countries.2-3 It has recently been estimated that over 1000 lives are lost
annually from colorectal cancer in the UK, when compared with the European mean,
and over 1600 when compared with the best in Europe.4 A significant factor
contributing to this is late diagnosis.5 Currently, most patients are diagnosed after
symptoms have developed, and after presentation to primary care.6 This will remain
the case even after full introduction of the UK National Bowel Cancer Screening
Programme, as only around one cancer in ten is identified by screening.7
In primary care, CRC is difficult to diagnose, as most of the symptoms of cancer can
also be experienced with benign conditions, such as irritable bowel syndrome or
haemorrhoids. These are also more common than cancer. National Referral
Guidelines for Suspected Cancer have been introduced (NICE Criteria), although
they have a weak evidence base, and concentrate upon typical presentations of
cancer.8-9 These guidelines are used to select patients for rapid investigation in the
‘two-week’ clinics (so named, as specialist opinion is guaranteed within that time).1011
However, this system has led to the concern that a two tier system has been created,
with patients not qualifying for referral to a two-week clinic suffering undue delays.1214
Only half of patients with CRC report a symptom (or symptoms) that qualify for a
referral to a two-week clinic.15 Indeed, only a quarter of all CRCs are identified by
such a route.16
4
A particular problem in the diagnosis of CRC is the absence of an adequate test that
can be used to select patients at higher risk from those presenting with a low-risk
symptom, such as diarrhoea or abdominal pain.17 Primary care investigation generally
includes measurement of the haemoglobin, though at least half of patients with CRC
have a normal value.18 Similarly, faecal occult blood testing has inadequate sensitivity
and specificity for reliable use.19 A symptom scoring system, the CAPER score, has
been derived, but has not entered routine use.20 In secondary care, similar problems
exist, with increasing numbers of referrals for suspected colorectal cancer,
accompanied by a fall in the percentage of referred patients who have cancer. It would
be attractive to identify low-risk patients in this population, who could avoid a
colonoscopy. Another symptom scoring system has been derived, the SELVA score,
but this has also not entered routine clinical practice.21 In one clinic, use of this score
had considerable diagnostic accuracy, as shown by an area under the receiveroperating characteristic curve (AUC) of 0.86. However, a recent report from Welsh
colorectal clinics found a lower performance, with an AUC of 0.75.22 Even so, this
was regarded as being worthy of use for prioritisation of referred patients and acts as a
comparator for our data.23
Recent studies have suggested that measurement of DNA in exfoliated material
collected from the surface of the rectal mucosa may identify colorectal disease,
including cancer.24-26 In dysplasia and neoplasia, normal systems controlling cell
exfoliation and death break down, with exfoliated cells progressing distally in the
mucocellular layer of the colon, eventually reaching the rectum.24 These cells can be
collected from the surface of the rectal mucosa using a simple device that incorporates
an inflatable elastic membrane. The device is inserted into the rectum through a
5
proctoscope (the Colonix system). Earlier studies in 30 patients with colorectal cancer
(28) or large polyps (2) demonstrated a mean DNA score of 15.1μg/ml in rectal
samples, compared with 3.9μg/ml in 52 outpatients with a normal bowel.25 A second
study included 66 patients with cancer, plus 110 healthy controls aged 50-70,
reporting mean rectal DNA scores of 9.0 and 2.1μg/ml in cancer patients and controls
respectively.26 Given this initial promise, we performed a study to examine the
diagnostic accuracy of a developmental version of such a test in symptomatic patients
referred for investigation of possible colorectal cancer.
Patients and Methods
This was a cohort study performed at the Royal Devon & Exeter Hospital, in Exeter,
Devon, UK, between May, 2008 and May 2009. This hospital provides investigation
of suspected colorectal cancer for a population of around 400,000, receiving
approximately 125-140 such referrals monthly.
Participants
All new NHS patients aged ≥40 years were eligible if they had been referred for
investigation of suspected colorectal cancer through the two-week wait system.
Exclusion criteria were: previous confirmed inflammatory bowel disease, previous
gastro-intestinal malignancy, or relevant investigation of the bowel within the
previous six months. Patients were offered participation as soon as the referral letter
was received, and participants gave signed consent. Recruitment into the study was
not allowed to interfere with current clinical and management priorities, which
required patients to be offered an appointment within the next available clinic or to be
invited directly for colonoscopy. At times, this process was too efficient to allow
participation in the study to be offered.
6
Data collection
Consenting participants completed a questionnaire detailing their symptoms and
regular medications. The questionnaire was posted with the appointment letter along
with full study details, except when the clinic appointment date and study entry was
agreed by telephone, when the questionnaire and consent process was completed in
clinic. At the hospital appointment, the timings of their most recent alcohol and food
intake and defecation were recorded. Blood samples were taken for haemoglobin,
mean red cell volume (MCV), ferritin and carcino-embryonic antigen testing, and a
Haemoccult™ kit for faecal occult blood testing (FOBT) was given to the participant.
Results of these tests were accepted in lieu of repeat testing if samples had been taken
in the thirty days before study entry. Participants were not excluded from the study if
they declined blood or FOBT sampling. A sample of exfoliated material was collected
by LM or ID using the Colonix system. This sampling was performed before rectal
examination to minimise disturbance to the rectal contents. The sample was preserved
in cell lysis buffer and sent to an off-site laboratory for DNA isolation and
quantification using a Pico-Green assay and real-time polymerase chain reaction by
AL. We also measured faecal contamination by measuring the optical absorbance
value for the sample at 340nm, defining a value below 1.5 as low contamination and
≥1.5 as high contamination. Throughout the study the laboratory was blinded to the
clinical outcome and the clinicians to the laboratory findings.
Determination of the outcome
All patients were offered appropriate investigation of the bowel by colonoscopy or CT
as advised by the assessing specialist, generally within 2-3 weeks. However, the
7
outcome of cancer was defined as a colorectal cancer diagnosed within six months of
entry into the study – this allowed for any cancers missed at initial investigation, or
for cancers to become apparent in those who declined definitive investigation.
Cancers were confirmed histologically, and their site and staging collected from the
hospital records. Two groups required additional efforts to determine their outcome:
firstly, treatment of large rectal polyps is currently undertaken in Cheltenham,
Gloucestershire, UK, using specialist trans-anal endoscopic microsurgery. We
obtained copies of histology from these operations. Second, some patients declined
definitive investigation. Their six-month outcome was determined by checking both
their primary care and hospital records for colorectal cancer.
Analysis
The main method of determining association with cancer was logistic regression. All
putative explanatory variables with a p-value ≤0.1 in univariable analysis entered
multivariable logistic regression models. For the multivariable analyses, a p-value of
0.05 was chosen for significance. Data were incomplete for four variables: faecal
occult blood (116 missing values, 16% of the total), carcino-embryonic antigen (43,
6%), ferritin (34, 5%), and mean cell volume (5, 1%). For the multivariable analyses,
imputed values for missing data in these variables were created using multiple
imputation by chained equations.27 No important differences were observed in oddsratios between the imputed dataset and the original dataset (full details in Appendix
A): we provide the imputed regression results here. Plausible interaction terms
relating to sample contamination, recent alcohol use, recent defecation and use of
non-steroidal anti-inflammatory drugs were added to the final model, and tested with
likelihood ratio tests. We tested for evidence against linear relationships between the
8
continuous predictors and the log odds of cancer status using fractional polynomials.28
A receiver-operating characteristic (ROC) curve was compiled for the final model, in
both the original and imputed datasets. The areas under the curves in these were
nearly identical: however, calculation of confidence intervals in imputed datasets is
impossible so we report so we report ROC curve characteristics from the original
dataset. Analyses used Stata, version 10.29
The sample size calculation was based on providing an area under the ROC curve
with a margin of error of ± 0.05 (based on 95% confidence intervals) of our estimated
figure of 0.85.26 30 We assumed 9% of referrals would have cancer: this required 614
participants. As a quarter of patients in an earlier study had rectal samples that were
not considered suitable for reliable analysis, we aimed for 800 participants in total.
The study was approved by Devon and Torbay Research Ethics Committee
(08/HO202/14).
Results
The study outline is shown in Figure 1. This shows patient recruitment, eligibility,
withdrawals and losses to follow-up. The age and sex of participants, broken down by
outcome, is shown in Table 1. The two samples lost in transport had leaked
completely. 714 participants had both a DNA score and knowledge of their outcome.
Diagnostic outcomes
72 participants (10%) had a colorectal cancer: 32 (44%) of these cancers were situated
in the rectum, 18 of which were palpable, 17 (24%) in the sigmoid, 12 (17%) in the
9
caecum and 11 (15%) elsewhere in the bowel. All were adenocarcinomas, and only
one was identified belatedly – in a patient whose apparently benign rectal polyp was
found to contain a cancer after trans-anal endoscopic microsurgery. The median age
of those with cancer was 74 years (inter-quartile range 70, 80) and of those without
cancer 70 years (62, 80), those with cancer being significantly older, p=0.006, MannWhitney test, and more likely to be male, p=0.014, chi2 test.
Eighteen patients (2.7%) had other malignancies identified: four pancreatic, three
lymphomas, three neuroendocrine tumours, two squamous anal and two liver cancers,
plus one each of stomach, gastro-intestinal stromal tumour, kidney and lung.
Seventeen (2.4%) patients had new diagnoses of inflammatory bowel disease.
Adenomatous polyps at least 1cm in diameter were found in 102 (14.9%) of patients,
though in nine of these patients a colorectal cancer was also found.
Symptom reporting and test results are shown in Table 2. Two symptoms were
associated with cancer in univariable analyses: rectal bleeding, odds-ratio (OR) 2.2
(95% confidence interval 1.3 to 3.7), p=0.002 and diarrhoea OR 0.63 (0.39 to 1.04),
p=0.07, though the association for diarrhoea was apparently protective. Three
investigations were associated with cancer: a positive FOBT, OR 11 (6.2 to 20)
p<0.001 and the mean red cell volume, OR 0.94 per fl (0.90 to 0.97), p=0.001, and
carcino-embryonic antigen, OR 1.03 per mcg/l (1.01 to 1.05), p=0.001. Neither the
haemoglobin nor ferritin value was significant in univariable analyses.
10
DNA scores
No adverse effects of the rectal sampling were noted, other than temporary discomfort
for some participants during proctoscope introduction. The DNA scores are shown in
Figure 2. The DNA score was significantly higher in those with cancer: median score
in those with cancer 5.4μg/ml (inter-quartile range 1.8, 12), and in those without
cancer 2.0μg/ml (0.78, 5.5), p<0.001, Mann-Whitney test. The mean (standard
deviation) values were 7.5 (7.3) and 4.4μg/ml (5.6) respectively. Similar results were
found for polymerase chain reaction amplification of DNA (results not shown).
Eighteen samples may be unreliable, and were identified as so before unblinding of
the data: four had problems with deflation or bursting of the membrane; one container
leaked and the sample could only be retrieved from the outer bag, one sample was so
heavily contaminated with faeces that pipetting was extremely difficult, and three had
been overfilled with buffer (one cancer), producing an unknown level of dilution. A
further nine (three cancers) had unrealistically low DNA scores in all three assays, so
it is probable they were inadequate samples. These were all included in the main
analysis, though subsidiary analyses (below) excluded them.
Faecal contamination of the rectal sample was more common in those with cancer; 31
(43%) having high contamination, compared with 131 (20%), for those without
cancer, p<0.001, chi2. No association was seen between the DNA score and the
presence of other cancers or of polyps. In contrast, the DNA score was higher in those
with newly diagnosed inflammatory bowel disease when compared to the remaining
‘healthy’ patients: odds-ratio 1.10 (1.04 to 1.16), p=0.001.
11
Multivariable analyses
Multivariable modelling using imputed values for the missing variables is shown in
Table 3. No interactions were identified for contamination, use of non-steroidal antiinflammatory drugs, or for recent food or alcohol intake, or for defecation within the
two hours before clinic attendance. The ROC curve from the seven-variable model in
Table 3 is shown in Figure 3. The area under the curve was 0.88 (CI 0.84 to 0.92). For
the DNA value alone the area under the curve was 0.65 (0.58 to 0.72).
Multivariable analysis excluding the 18 potentially unreliable samples from analysis
produced the same seven-variable model with similar odds-ratios. The area under the
ROC curve was 0.90 (0.86 to 0.93), and for the DNA value alone it was 0.65 (0.58 to
0.72).
Finally, in clinical practice tumours that were palpable rectally would not require
additional testing. Additionally, previously unrecognised inflammatory bowel disease
is clearly worth diagnosing. Thus a clinically-driven model was derived, excluding
the palpable tumours but with the outcome variable being cancer or new inflammatory
bowel disease. This model contained five variables: age, DNA value, a positive
FOBT, carcino-embryonic antigen and mean red cell volume. The AUC was 0.84
(0.78 to 0.90)
12
Discussion and Conclusions
This study shows that DNA scores measured in exfoliated material from the rectum of
patients at moderate risk are associated with colorectal cancer. However, the
diagnostic power of DNA concentration when used alone was modest, but when
added to other symptoms, blood test results and FOBTs – all of which are available in
primary or secondary care – it became considerably stronger.
Strengths and limitations
This study was conducted in a single hospital. The percentage of patients diagnosed
with cancer in our study is similar to previous reports, suggesting it was nationally
representative.16 Only a proportion of eligible patients were offered study
participation. This was simply because patients were given the next available
appointment – which was often too soon to allow recruitment into the study: it is
unlikely those offered participation were atypical. The two surgeons who collected
samples rapidly assimilated experience in use of the cell capturing device during the
study. Sampling was straightforward, though eighteen samples (2.5%) were probably
inadequate, some from faults with the device or the buffer fluid, and others with DNA
scores so low as to suggest the balloon may have failed to make contact with the
rectal mucosa. If the Colonix system were to enter clinical use, such samples would
require repeating, though this ‘failure rate’ compares favourably with tests such as
cervical cytology, with the traditional smear test having a failure rate of over 11%.31
Excluding these patients from analysis moderately improved the performance of the
test.
13
Thirty seven (4.9%) of the 754 eligible participants declined the test, though we do
not know how many of these were specifically concerned about the technique for
sampling of the rectal contents – some may have disliked rectal examination per se, or
any type of proctoscopy. This percentage is much lower than the 16% who did not
submit a FOBT. As well as missing data for FOBTs, there were a small number of
patients without blood test results. The imputation methods to compensate for missing
data followed standard practice, and the results were very similar with and without
imputation, so it is unlikely this procedure introduced any important bias.27 Eighteen
of the cancer patients had a palpable rectal tumour, which could interfere with
sampling. In practice, some of these patients would be identified clinically and would
not require sampling.
One strength was the very thorough follow up, with only one patient wholly lost from
study. Using an endpoint of cancer at six months allowed us to identify the one cancer
missed on investigation, though no cancers were belatedly diagnosed in the few
patients who declined investigation. It is a reasonable assumption that in a
symptomatic population any colorectal cancer not identified because investigation
was declined would have come to light within six months.32
Comparison with previous literature
Two pilot studies reported mean DNA scores of 15.1 and 9μg/ml in those with cancer,
though in fewer patients.25-26 The mean score in this study was 7.5μg/ml, though
many fewer patients were excluded in the current study. DNA scores were skewed,
and had a median of 5.4μg/ml. The cancer and non-cancer populations in the present
study had overlapping DNA distributions, suggesting that using the DNA value on its
14
own would have insufficient discriminatory power to allow safe decision making on
the need for investigation. Although high DNA scores were observed in all heavily
contaminated samples, such contamination did not alter the association between the
DNA score and the presence of cancer. Non-compliance with the restriction of food
or alcohol did not alter the association either, implying that such restrictions may be
unnecessary for future use of the test, thus making this test easier to use.
Place in colorectal cancer diagnosis
This study was performed in the referred population. It was efficient to do so, as a
study in primary care would have had to be much larger to compensate for the relative
rarity of cancer in that setting. The performance of the seven-item model was
excellent. Few selection methods for colorectal cancer have been validated in either
the low-risk population of primary care, or the moderate-risk referred population. The
model in this study compares well with the CAPER and SELVA scores, which have
arguably the largest evidence base.20-22 When the CAPER score was tested in a second
primary care dataset the area under the ROC curve was 0.79 (data submitted for
publication): the SELVA score, when tested in a referred population (similar to ours)
in Wales the area was 0.76.22 In the study reported here, the area below the ROC
curve was 0.88, and 0.90 if unreliable samples are excluded. Furthermore, the ROC
curve is steep on the left, an area with high specificity and low sensitivity. This would
allow for rationalisation of further investigations. Arguably, many of the patients at
this point on the curve could be spared colonoscopy and could be re-assessed
clinically or investigated by minimal preparation computerised tomography or Barium
enema. This is particularly relevant, as colorectal cancer has been selected for
increased rapid access to primary care diagnostics.33 Inevitably, this policy will yield
15
more referrals, so any valid method of selecting those at very low risk could
considerably reduce the total costs of investigation, estimated at over £200 million
annually.34 This study was not designed to test the DNA score in inflammatory bowel
disease but values were significantly higher in patients who were diagnosed with IBD.
Such a finding was not unexpected as colonocyte shedding is increased in
inflammatory bowel disease. Our analysis labelled such patients correctly as without
cancer, though definitive investigation in these patients was clearly of diagnostic
value. Our pragmatic secondary analysis including patients with inflammatory bowel
disease as a ‘positive’ outcome suggested that selection of patients in using our model
would also correctly identify them for investigation.
Rectal sampling could take place in primary care if general practitioners were willing
to learn the technique, and transport of the samples could be organised. However, our
results do not fully translate to the low-risk population seen in primary care. It was
notable that the only symptom remaining independently significant in the final model
was rectal bleeding; furthermore diarrhoea was apparently protective in univariable
analysis. Both these findings illustrate the selection process undertaken by general
practitioners. When a symptom is used to select a patient for investigation it is
understandable that its predictive power is reduced, or eliminated – or even reversed –
in the selected population. Therefore, if rectal sampling were to be used in primary
care, symptoms would almost certainly be relevant. This would require a similar study
stationed in primary care.
16
Conclusion
Rectal sampling has promise in the diagnosis of colorectal cancer. When combined
with simple blood tests, age, sex and results of faecal occult blood tests the method
presented has considerable ability to predict cancer. This combination may be
particularly useful in triage of referrals for investigation, with the intention being to
select those at very low risk, who can then be observed. A second cohort study in the
referred population is ongoing. If its results are similar to those reported here, then
serious consideration should be given to implementation of this diagnostic system.
For the future, there may also be a role for use in primary care, but this will require
further investigation.
Acknowledgements. We wish to thank all participating patients, plus staff of the Royal
Devon and Exeter Hospital. Tim Peters and Nicola Wiles were very helpful with the
imputation analyses. Thanks also to Obi Ukoumunne for comments on the regression
analyses.
Ethical review: The study was approved by Devon and Torbay Research Ethics
Committee (08/HO202/14).
Contributions. LM was responsible for all aspects of the study, as was WH (other than
data collection). AHL and AL provided the laboratory facilities, and contributed to
design. ID contributed to design and data collection. AS assisted in conduct of the
study and analysis. DC supervised the logistics of patient recruitment and data
collection. All authors have full access to the final data - though only LM, WH, ID
and DC had access to the clinical dataset during data collection, while AHL and AL
17
had sole access to the laboratory data. All revised the paper; the initial draft being
written by WH, who is the guarantor. The study sponsor was the University of Bristol,
who had no role in study design and the collection, analysis, and interpretation of data
and the writing of the article and the decision to submit it for publication, though did
audit conduct of the study. WH, AS, LM, and DC are independent from the funders
other than detailed in the competing interests; WH is employed by the sponsor.
Funding. The study was funded by a project grant from Colonix Medical Ltd.
Competing interests. All authors have completed the Unified Competing Interest form
at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding
author) and declare that (1) all authors have grant support from Colonix Medical Ltd
for the submitted work; (2) LM, ID and DC have received travel expenses for
conferences relating to this work from Colonix Medical Ltd.; ID is a medical advisor
to Colonix Medical Ltd. AL and AL were directors and remain shareholders of
Colonix Medical Ltd.; (3) all authors’ spouses, partners, or children have no financial
relationships that may be relevant to the submitted work; and (4) all authors have no
non-financial interests that may be relevant to the submitted work.
18
References
1. Office for National Statistics. MB1 No 37 - Cancer Registration Statistics 2006.
Cancer Statistics: Registrations Series MB1. London: ONS, 2009.
2. Verdecchia A, Francisci S, Brenner H, Gatta G, Micheli A, Mangone L, et al.
Recent cancer survival in Europe: a 2000-02 period analysis of EUROCARE4 data. The Lancet Oncology 2007;8(9):784-96.
3. Berrino F, Verdecchia A, Lutz JM, Lombardo C, Micheli A, Capocaccia R.
Comparative cancer survival information in Europe. European Journal of
Cancer 2009;45(6):901-08.
4. Abdel-Rahman M, Stockton D, Rachet B, Hakulinen T, Coleman MP. What if
cancer survival in Britain were the same as in Europe: how many deaths are
avoidable? Br J Cancer 2009;101(S2):S115-S24.
5. Richards MA. The size of the prize for earlier diagnosis of cancer in England. Br J
Cancer 2009;101(S2):S125-S29.
6. Barrett J, Jiwa M, Rose P, Hamilton W. Pathways to the diagnosis of colorectal
cancer: an observational study in three UK cities. Fam. Pract. 2006;23:15-19.
7. Goodyear S, Stallard N, Gaunt A, Parker R, Williams N, Wong LS. Local impact of
the English arm of the UK Bowel Cancer Screening Pilot study. British
Journal of Surgery 2008;95(9):1172-79.
8. NICE. Referral Guidelines for suspected cancer. London: NICE, 2005.
9. Hamilton W, Sharp D. Diagnosis of colorectal cancer in primary care: the evidence
base for guidelines. Family Practice 2004;21:99-106.
10. Ford AC, Veldhuyzen van Zanten SJO, Rodgers CC, Talley NJ, Vakil N,
Moayyedi P. Diagnostic Utility of Alarm Features for Colorectal Cancer:
Systematic Review and Meta-analysis
10.1136/gut.2008.159723. Gut 2008:gut.2008.159723.
11. Jellema P, van der Windt DAWM, Bruinvels DJ, Mallen CD, van Weyenberg
SJB, Mulder CJ, et al. Value of symptoms and additional diagnostic tests for
colorectal cancer in primary care: systematic review and meta-analysis. Bmj
2010;340(mar31_3):c1269-.
12. Hamilton W. Five misconceptions in cancer diagnosis. British Journal General
Practice 2009;59:441-47.
13. Neal R, Allgar V. Socio-demographic factors and delays in the diagnosis of six
cancers: analysis of data from the 'National Survey of NHS Patients: Cancer'.
British Journal of Cancer. 2005;92:1971-75.
14. Jones R, Rubin G, Hungin P. Is the two week rule for cancer referrals working?
BMJ 2001;322:1555-56.
15. Hamilton W, Round A, Sharp D, Peters T. Clinical features of colorectal cancer
before diagnosis: a population-based case-control study. British Journal of
Cancer 2005;93:399-405.
16. Rai S, Kelly MJ. Prioritization of colorectal referrals: a review of the 2-week wait
referral system. Colorectal Dis. 2007;9:195-202.
17. Hamilton W, Lancashire R, Sharp D, Peters T, Cheng K, 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.
18. Hamilton W, Lancashire R, Sharp D, Peters TJ, Cheng K, Marshall T. The
importance of anaemia in diagnosing colorectal cancer: a case-control study
using electronic primary care records. British Journal of Cancer 2008;98:32327.
19
19. Weller D, Coleman D, Robertson R, Butler P, Melia J, Campbell C, et al. The UK
colorectal cancer screening pilot: results of the second round of screening in
England. British Journal of Cancer 2007;97:1601-05.
20. Khan N, NCRI Colorectal Clinical Studies Group. Implementation of a diagnostic
tool for symptomatic colorectal cancer in primary care: a feasibility study.
doi:10.1017/S1463423608000996. Primary Health Care Research &
Development 2009;10:54-64.
21. Selvachandran S, Hodder R, Ballal M, Jones P, Cade D. Prediction of colorectal
cancer by a patient consultation questionnaire and scoring system: a
prospective study. Lancet 2002;360:278-83.
22. Ballal MS, Selvachandran SN, Maw A. Use of a patient consultation questionnaire
and weighted numerical scoring system for the prediction of colorectal cancer
and other colorectal pathology in symptomatic patients: A prospective cohort
validation study of a Welsh population. Colorectal Disease 2010;12:407-14.
23. Kelly MJ. Commentary: Use of the WNS Patient Questionnaire. Colorectal
Disease 2010;12:414-15.
24. Loktionov A. Cell exfoliation in the human colon: Myth, reality and implications
for colorectal cancer screening. International Journal of Cancer
2007;120:2281-89.
25. Loktionov A, Bandaletova T, Llewelyn A, Dion C, Lywood H, Lywood R, et al.
Colorectal cancer detection by measuring DNA from exfoliated colonocytes
obtained by direct contact with rectal mucosa. International Journal of
Oncology 2009;34:301-12.
26. Loktionov A, Ferrett CG, Gibson JJS, Bandaletova T, Dion C, Llewelyn AH, et al.
A case-control study of colorectal cancer detection by quantification of DNA
isolated from directly collected exfoliated colonocytes. International Journal
of Cancer 2010;126:1910-19.
27. Sterne JAC, White IR, Carlin JB, Spratt M, Royston P, Kenward MG, et al.
Multiple imputation for missing data in epidemiological and clinical research:
potential and pitfalls. Bmj 2009;338(jun29_1):b2393-.
28. Royston P, Ambler G, Sauerbrei W. The use of fractional polynomials to model
continuous risk variables in epidemiology. International Journal of
Epidemiology 1999;28:964-74.
29. StataCorp. Stata Statistical Software: Release 10. College Station, TX: Stata
Corporation, 2008.
30. Zhou X-H, Obuchowski N, McClish D. Statistical methods in Diagnostic
medicine. New York: Wiley, 2002.
31. Harrison W, Teale A, Jones S, Mohammed M. The impact of the introduction of
liquid based cytology on the variation in the proportion of inadequate samples
between GP practices. BMC Public Health 2007 2007;7:191.
32. Stapley S, Peters TJ, Sharp D, Hamilton W. The mortality of colorectal cancer in
relation to the initial symptom and to the duration of symptoms: a cohort study
in primary care. British Journal of Cancer 2006;95:1321-25.
33. Department of Health. Cancer Reform Strategy. Achieving local implementation second annual report. . London: Departmen of Health, 2009:81.
34. York Health Economics Consortium. Bowel Cancer Services: Costs and Benefits.
London: Department of Health, 2007.
20
21
Figure 1. Participation, eligibility, withdrawals and losses in the study
828 patients were offered
participation and gave initial consent
27 were ineligible for study
37 declined
rectal sampling
36 cancelled or
failed to attend
11 were
admitted as an
emergency
84 withdrew from study
10 previous
bowel cancer
5 recent bowel
investigation
6 inflammatory
bowel disease
1 under 40
5 other reasons
717 patients had rectal sampling
3 patients no result available
714 analysed
2 samples lost
in transit
1 declined
investigation
and moved
away
Note. The 828 initial participants were from approximately 1500 referrals: it was not possible
to keep accurate figures for those in whom the logistics of offering a rapid appointment
overrode the offer of study entry, as opposed to those who were offered study entry but
declined. However, it was clear, that the second of these groups was much smaller than the
former.
22
Table 1. Age and sex of participants
Colorectal cancer
Without colorectal
cancer
Age
Male
Female Total
Male
Female Total
band
40-49
1
0
1
18
27
45
50-59
1
3
4
37
59
96
60-69
11
4
15
82
101
183
70-79
19
16
35
77
81
158
80-89
9
4
13
54
86
140
90+
1
3
4
9
11
20
Total
42
30
72
277
365
642
23
Table 2. Symptoms and investigation results
Colorectal cancer
(n=72)
Present
Absent
Number (%) reporting a symptom
Without colorectal
Total
cancer
(n=642)
(n=714)
Present
Absent
Present
Absent
Rectal bleeding 44 (61)
28 (39)
266 (41)
376 (59)
310 (43)
404 (57)
Weight loss 27 (38)
45 (62)
192 (30)
450 (70)
219 (31)
495 (69)
Abdominal pain 30 (42)
42 (58)
286 (45)
356 (55)
316 (44)
398 (56)
Constipation 23 (32)
49 (68)
217 (34)
425 (66)
240 (34)
474 (66)
Diarrhoea 33 (46)
39 (54)
367 (57)
275 (43)
400 (56)
314 (44)
24 (41)
63 (12)
476 (88)
98 (16)
500 (84)
S.D
9.5
Mean
13.3
S.D
1.8
Mean
13.3
S.D
3.4
7.5
87.8
5.9
87.5
6.1
351
128
120
130
159
178
3.2
13
8.1
59
Investigations *
Positive faecal 35 (59)
occult blood
Blood tests *
Mean
Haemoglobin (g/dl) 13.4
Mean cell volume 85.1
(fl)
Ferritin (pmol/l) 147
Carcino-embryonic 53
antigen (μg/L)
* Not all participants agreed to investigation, with faecal occult blood samples
submitted by 598 (84% of total), mean cell volume 708 (99%), ferritin 680 (95%) and
carcino-embryonic antigen 671 (94%). S.D. = standard deviation.
24
Figure 2. Box plot of DNA scores in participants with and without cancer.
Colorectal cancer
0
10
dnascore
20
30
No colorectal cancer
25
Table 3. Multivariable analysis of diagnostic features for colorectal cancer
Feature
Odds ratio
95% confidence
interval
P-value
Age (years)
1.05
1.02 to 1.08
<0.001
DNA (μg/ml)
1.05
1.01 to 3.6
0.01
Mean red cell
volume (fl)
Carcino-embryonic
antigen (μg/L)
Binary variables
0.93
0.89 to 0.97
0.001
1.02
1.00 to 1.04
0.02
Male sex
2.0
1.1 to 3.6
0.02
Rectal bleeding
2.4
1.3 to 4.5
0.007
Positive faecal
occult blood test
6.7
3.4 to 13
<0.001
Continuous variables
Note: for continuous variables the odds ratio is for each unit of measure, such as each year for
age. For binary variables, it is used if the variable is present.
26
Figure 3. Receiver-operating characteristic curve for variables independently
0.50
0.25
0.00
Sensitivity
0.75
1.00
associated with cancer.
0.00
0.25
0.50
1 - Specificity
0.75
1.00
Area under ROC curve = 0.8776
27
Appendix A. Technique for generation of imputed values.
This used multiple imputation by chained equations.27 No associations were observed
between missing data for faecal occult bloods (the variable with the greatest number
of missing values) and possible explanatory variables, such as age or sex, so we
assumed the values were missing at random. Two variables required transformation
before generation of imputed values: ferritin by log transformation, and carcinoembryonic antigen by using the inverse. All variables that were significantly
associated with cancer in logistic regression in the dataset with full values (n=570),
with the addition of male sex, were included in the regression equations for
imputation, as was the outcome variable, cancer.29 25 cycles of imputation were run,
and the two transformed variables back-transformed.
Logistic regression models were performed across the 25 imputed datasets. The
combined odds ratios and confidence intervals were compared between those derived
in the dataset with full values (n=570) and those from datasets with imputed values
(n=714). No important differences were found between the two, other than the p-value
for male sex (not an imputed value) falling to 0.02.
28