Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Hepatitis B wikipedia , lookup
Rocky Mountain spotted fever wikipedia , lookup
Hepatitis C wikipedia , lookup
Marburg virus disease wikipedia , lookup
Neonatal infection wikipedia , lookup
Coccidioidomycosis wikipedia , lookup
Oesophagostomum wikipedia , lookup
ORIGINAL STUDIES Risk Prediction in Pediatric Cancer Patients With Fever and Neutropenia Hana Hakim, MD,* Patricia M. Flynn, MD,*†‡ Deo Kumar Srivastava, PhD,§ Katherine M. Knapp, MD,*† Chenghong Li, PhD,§ James Okuma, MS,§ and Aditya H. Gaur, MD* Background: To identify predictors for 2 risk measures—“proven invasive bacterial infection or culture-negative sepsis (IBD)” and “clinical complications (CC)”—in pediatric cancer patients with fever and neutropenia (FN). Methods: Records of 390 patients with FN hospitalized over 2 years were reviewed. For the 332 who met inclusion criteria, one FN episode was randomly selected. Independent predictors at presentation were analyzed using multiple regression models. Optimal cut-off risk prediction scores were determined. These models were validated by bootstrap analysis. Results: Patients’ median age was 6.0 years; 66% had an underlying diagnosis of leukemia. Independent predictors of IBD (n ⫽ 56) were absolute neutrophil count ⬍100, temperature at presentation ⱖ39.0°C, “sick” clinical appearance, and underlying diagnosis of acute myeloid leukemia. A total weighted score ⬍24 reliably identified patients at low risk for IBD. Independent predictors of CC (n ⫽ 47) were relapse of malignancy, non-white race, “sick” clinical appearance, and underlying diagnosis of acute myeloid leukemia. A total weighted score ⬍19 predicted patients at low risk for CC. Of those misclassified as low risk, 11 of 12 with IBD and 3 of 9 with CC had the outcome within 24 hours of presentation. Of the remaining patients classified as low-risk for IBD and CC, 99.5% and 97.1%, respectively, remained outcome-free after 24 hours of observation. Conclusions: This study identifies predictors of infection/complications in pediatric patients with FN, establishes clinical cut-off scores and highlights the importance of the initial clinical impression and 24 hours of observation. These prediction models warrant prospective validation. Key Words: fever, neutropenia, children, cancer, risk, prediction (Pediatr Infect Dis J 2010;29: 53–59) F tient setting is the mainstay of therapy for FN in pediatric patients.1–3 However, this patient population is heterogeneous, and not all pediatric patients with FN are at the same risk for complications and bacterial infections. Models of risk prediction for FN-related morbidity and mortality and risk-based stratification of care including outpatient and oral antibiotics have been suggested4 –21 and are being increasingly used for adults. Risk prediction can result in tremendous cost saving, better resource utilization, and improved quality of care.22 The key factor in risk prediction is to consistently and reliably identify patients “at risk.” Of the several risk prediction models proposed,23–35 few apply to pediatric patients. While many previous studies focused on derivation datasets using retrospective study designs, the 2 conducted prospectively,26 –28 found different predictor variables, did not establish weighted risk prediction scores and, unlike the approach now commonly used in adult patients,23,24 had limited focus on clinical complications (CC) as an outcome measure. Clinical complications evaluated in previous models did not include adverse events occurring without associated hemodynamic instability or sepsis syndrome, such as respiratory failure and altered mental status. With the above summarized limitations and needs, as the first step of a planned 3-part evaluation, we designed a retrospective study to examine candidate predictor variables for 2 measures of risk for FN: proven invasive bacterial infection or culture-negative sepsis (IBD) and clinical complications (CC). The next steps of our planned study will be to use this information to build prediction models that could be further optimized and validated prospectively, and as a final step, to evaluate risk-stratified FN management in a randomized clinical trial. ever and neutropenia (FN) is a common complication in patients undergoing chemotherapy to treat cancer. Empiric treatment with intravenous (IV) broad-spectrum antibiotics in an inpa- Study Population and Design Accepted for publication October 1, 2009. From the *Department of Infectious Diseases, St. Jude Children’s Research Hospital, Memphis, TN; Departments of †Pediatrics and ‡Preventive Medicine, University of Tennessee Health Science Center, Memphis, TN; and §Department of Biostatistics, St. Jude Children’s Research Hospital, Memphis, TN. Supported by a National Institutes of Health grant (CA21765) and the American Lebanese Syrian Associated Charities (ALSAC). Presented as a poster at the 45th Annual Meeting of Infectious Diseases Society of America (IDSA), San Diego, CA, October 2007. Another part was presented as a poster at the 48th Annual Interscience Conference on Antimicrobial Agents and Chemotherapy (ICAAC)/45th Annual Infectious Diseases Society of America (IDSA) Meeting, Washington, DC, October 2008. The authors report no conflicts of interest. Address for correspondence: Hana Hakim, MD, Department of Infectious Diseases, Mail Stop 600, St. Jude Children’s Research Hospital, 262 Danny Thomas Place, Memphis, TN 38105-3678. E-mail: [email protected]. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.pidj.com). Copyright © 2009 by Lippincott Williams & Wilkins ISSN: 0891-3668/10/2901-0053 DOI: 10.1097/INF.0b013e3181c3f6f0 We retrospectively reviewed the records of 390 pediatric (⬍22 years) patients with cancer who developed FN in an outpatient setting and were hospitalized for further management at St. Jude Children’s Research Hospital (St. Jude), Memphis, TN from January 1, 2004 to December 31, 2005. Data for patients who had already received a stem cell transplant or developed FN during hospitalization for other reasons were excluded from analysis. We developed a list of candidate variables (Table, Supplemental Digital Content 1, http://links.lww.com/INF/A314) on the basis of variables identified by previous risk-prediction studies23,24,26 –29 and by this retrospective medical record review. These included: demographics; underlying cancer; relapse; granulocyte-colony stimulating factor (G-CSF) use; steroid use within the past 14 days; duration of neutropenia; presence of indwelling central venous catheter (CVC); time since last chemotherapy; antifungal therapy for probable or proven fungal infection within the past 6 months; colonization by methicillin-resistant Staphylococcus aureus (MRSA), vancomycin-resistant Enterococcus (VRE), or Pseudomonas aeruginosa within the past 4 weeks; prophylactic or treatment antibiotics; vital signs; and laboratory tests. Blood cul- METHODS The Pediatric Infectious Disease Journal • Volume 29, Number 1, January 2010 www.pidj.com | 53 Hakim et al The Pediatric Infectious Disease Journal • Volume 29, Number 1, January 2010 tures were collected at admission in continuously-read BACTEC bottles (BD Diagnostics, Sparks, MD), and routinely included both AEROBIC/F bottles to detect aerobic organisms and MYCO/F Lytic bottles to detect fungi and mycobacteria. Data on comorbidities at the time of presentation were also analyzed. These included abdominal pain, altered mental status, blood loss requiring transfusion, clinical appearance, signs of viral infection, dehydration requiring IV fluids, mucositis, rectal pain, respiratory distress, vomiting, and suspected focus of infection. Definitions for the 2 study outcomes, (1) proven invasive bacterial infection or culture-negative sepsis and (2) clinical complications occurring after presentation until fever resolution for 5 consecutive days, were established a priori. This study was approved by the St. Jude Institutional Review Board. Children with FN were managed as inpatients as per institutional guidelines, which are consistent with the recommendations of the Infectious Diseases Society of America.3 Cefepime 1500 mg/m2/dose IV every 8 hours was empirically started at presentation. Vancomycin 400 mg/m2/dose IV every 8 hours was added for any of the following: obvious CVC or soft tissue infection, history of recent cytarabine treatment, quinolone prophylaxis or therapy at the time of fever, known colonization with MRSA, evidence of sepsis, suspected central nervous system shunt infection, and known or suspected infection with Bacillus cereus. Empiric antifungal therapy was added if the patient remained febrile on day 5 of antibiotics and if neutropenia was expected to last longer than 5 to 7 days. After fever resolution for ⬎24 hours, patients were discharged on oral cefpodoxime or IV cefepime if blood cultures remained sterile and there was no evidence of pneumonia, suspected bacterial infection, or sepsis during the episode, and no vomiting, stomatitis, or diarrhea at discharge. Antibiotics were continued until fever resolved for at least 24 hours and there was recovery from neutropenia; if neutropenia persisted, antibiotics were continued for at least 5 afebrile days and there was no evidence of documented infection or signs and symptoms of sepsis. The chemotherapeutic protocol for children with acute myeloid leukemia (AML) required administration of voriconazole for antifungal prophylaxis. Definitions Fever was defined as an oral temperature of ⱖ38.3°C or of ⱖ38.0°C persisting for ⱖ1 hour. Neutropenia was defined as an absolute neutrophil count of ⱕ500 cells/mm3. Hypotension was defined as a systolic blood pressure less than fifth percentile for age and sex, or need for vasopressor support. Respiratory failure was defined as an arterial oxygen pressure of ⬍60 mm Hg on room air or the need for mechanical ventilation or oxygen in a patient with no known supplemental oxygen requirement at baseline. Clinical appearance was classified as either well, sick, or toxic.28 A patient was classified as “well” if the admitting physician made no specific comment about the patient’s general appearance at the time of admission, or documented that the patient looked well, in no distress, or playful. A patient was classified as sick if he/she was noted to be irritable or looking ill, and classified as toxic if not breathing or noted to appear toxic, lethargic, or obtunded. Bacteremia was defined as a blood culture growing an organism determined not to be a contaminant. An organism was considered a contaminant if it was isolated from only one culture receptacle and was a member of the skin flora (diphtheroids except Corynebacterium jeikeium, Bacillus spp. except B. cereus, Propionibacterium spp., coagulase-negative staphylococcus or Micrococcus spp.). Proven invasive bacterial infection was defined as isolation of a pathogen from a sterile body site or as proven by histology. Culture-negative sepsis was defined, in the absence of a positive culture, as a systemic response to a possible infection because of 54 | www.pidj.com hemodynamic instability; focal or multiple organ involvement; or altered mental status or lethargy. CC from presentation until fever resolution for 5 consecutive days included altered mental status, arrhythmia or new electrocardiographic changes, bleeding requiring transfusion, congestive heart failure, disseminated intravascular coagulation (DIC), hypotension, admission to the intensive care unit (ICU), respiratory failure, renal failure, and other complications that the investigator judged as clinically significant. Statistical Analysis Derivation Only one episode per patient was randomly selected using a computer algorithm to avoid bias, given the possibility that multiple episodes in the same patient may not necessarily be independent. Frequency tables for categorical variables and descriptive parameters (median and range) for continuous variables were calculated for the predictor variables assessed at the time of presentation. For each outcome variable, univariate and multiple logistic regressions were used to determine the association between the variables and outcome. The multiple logistic regression model was fit to the data by using a stepwise selection method. Episodes that had the outcomes present at the time of presentation were excluded from the model building. A P value of less than 0.1 was used as the criterion for entry of the covariate into the model and a P value of less than 0.05 as that for staying in the model. The Hosmer-Lemeshow test was used to evaluate the model’s goodness-of-fit to the data. A P value of less than 0.05 was considered significant. All P values were 2-sided. The final multiple regression model was used to develop a scoring prediction rule by summation of weights assigned to each variable by multiplying the -regression coefficients by 10 and rounding to the nearest integer. Receiver operating characteristic (ROC) curves were used to determine the optimal cut-off score. Optimal cut-off points were determined by 3 methods36: (1) minimizing the absolute difference between sensitivity and specificity; (2) minimizing the distance to the point (0, 1), and; (3) maximizing the Youden index, J, the vertical distance to the chance line. Values equal to or greater than the cut-off point identified high-risk patients. Misclassification rate associated with the original predictive model was calculated by applying the ROC cut-off point to the original data that were used to develop the model. Predicted low-risk patients were re-evaluated within 24 hours of presentation for occurrence of the outcome. Validation The predictive model was validated using the bootstrap approach described by Efron and Chong.37 Within each bootstrap replication, a bootstrap sample was generated by drawing randomly with replacement from the original data and fitting a predictive logistic model as well as obtaining the ROC cut-off point using the same steps described above. The model was validated by applying the cut-off point to the episodes that were not selected in the bootstrap sample and calculating the misclassification rate. A total of 500 bootstrap replications were performed; the average misclassification rate and frequency distribution of each candidate variable selected in 500 predictive models were obtained. SAS version 9.1.3 (SAS institute, Cary, NC) was used. RESULTS Overall Description of the FN Episodes A list of 885 FN episodes was obtained from the Medical Records Department. One episode was randomly selected per patient. Of the 390 medical records reviewed, 58 were excluded because of absence of: fever (23), neutropenia (23), both (9), or FN development © 2009 Lippincott Williams & Wilkins The Pediatric Infectious Disease Journal • Volume 29, Number 1, January 2010 while inpatient (3). A total of 332 FN episodes were analyzed. The median number of FN episodes per patient is 2 (range: 1–12). Table, Supplemental Digital Content 1, http://links.lww.com/INF/A314, shows patients’ baseline characteristics at the time of presentation. Fifty-six patients had IBD: bacteremia (41), urinary tract infection (2), and culture-negative sepsis (13). Forty-seven patients developed CC: hypotension (21), respiratory failure (14), altered mental status or seizure (6), congestive heart failure (1), DIC (1), arrhythmia (1), paraplegia (1), deep venous thrombosis (1), and persistent hypoglycemia (1). Twenty-five patients developed both IBD and CC. Prediction Rule for the Outcome “Proven Invasive Bacterial Infection or Culture-Negative Sepsis (IBD)” Table 1 shows the predictor variables at the time patients presented with FN that were analyzed. Using multiple regression analysis, independent predictors in the final model for IBD included an underlying diagnosis of AML (OR: 7.46; 95% CI: 2.76 –20.14), “sick” clinical appearance (OR: 4.23; 95% CI: 2.10 – 8.53), temperature of ⱖ39.0°C at presentation (OR: 2.91; 95% CI: 1.30 – 6.49), and an absolute neutrophil count ⬍100 (OR: 2.68; 95% CI: 1.25–5.76). Weighted scores were assigned (Table, Supplemental Digital Content 2, http://links.lww.com/INF/A315) and the ROC curve was plotted (Fig. 1A). A total weighted score of ⬍24 points identified patients at low risk for IBD, with a high negative predictive value (NPV) but a low positive predictive value (PPV) (75% sensitivity, 77% specificity; 95% NPV, 36% PPV). The 3 statistical methods agreed on the optimal cut-off score of 24 points. Prediction Rule for the Outcome “Clinical Complications (CC)” Table 1 summarizes the results for the outcome CC. Independent predictors of CC included “sick” clinical appearance (OR: 7.41; 95% CI: 3.50 –15.72), relapse of malignancy (OR: 3.14; 95% CI: 1.36 –7.25), nonwhite race (OR: 2.21; 95% CI: 1.00 – 4.90), and an underlying diagnosis of AML (OR: 3.15; 95% CI: 1.10 – 9.04). Weighted scores were assigned (Table, Supplemental Digital Content 2, http://links.lww.com/INF/A315) and the ROC curve was plotted (Fig. 1B). A total weighted score of ⬍19 points predicted patients at low risk for CC (78% sensitivity, 73% specificity; 29% PPV, 96% NPV). Bootstrap Validation for the Outcome “Proven Invasive Bacterial Infection or Culture-Negative Sepsis (IBD)” The 4 IBD predictors in the proposed predictive model were most frequently identified in 500 bootstrap replications, suggesting the proposed model was appropriate. The misclassification rate of the original model was 0.23, which is smaller than the average misclassification rate from the bootstrap replications (0.31). This agrees with Efron and Chong37 who showed that error rate obtained from the original model-building sample is an underestimate of the true error rate. Bootstrap Validation for the Outcome “Clinical Complications (CC)” The 4 predictors in the proposed predicting model for CC were most frequently selected in 500 bootstrap replications. The average misclassification rate from the bootstrap replications (0.28) was similar to that of the original predicting model (0.26). Inpatient Observation of the Low-Risk Group for 24 Hours The “serious error” or false-negative misclassification rate in the predicted low-risk group for IBD was 3.7%. Within the first 24 hours of presentation (Fig., Supplemental Digital Content 3, © 2009 Lippincott Williams & Wilkins Infection and Cancer http://links.lww.com/INF/A316), 91% of the IBD and 56% of the CC occurred. Re-evaluating low-risk patients 24 hours after presentation for IBD occurrence decreased the serious misclassification error to 0.3% (Fig. 2A). The one patient who developed IBD and was misclassified by the suggested prediction scheme did clinically well. This patient was a 7-year-old with T-cell lymphoma, febrile for ⬍1 day, and hospitalized for 2 days, whose admission blood culture yielded Capnocytophaga sputigena after 143.7 hours of incubation. Similarly, the “serious error” or false-negative misclassification rate in the predicted low-risk group for CC was 2.7%. By re-evaluating low-risk patients 24 hours after presentation for CC occurrence, the serious misclassification error decreased to 1.8% (Fig. 2B). The 6 patients who were misclassified by the sequential prediction rule at presentation followed by a 24-hour observation had an average age of 9.6 years (range: 0.8 –18.0 years). Their complications were respiratory distress (4), seizure (1), and altered mental status (1). The median time to complications was 86.5 hours after presentation (range: 68.5–130.7 hours). The diagnoses of their FN episodes were Clostridium difficile colitis (1), pneumonia (2), rotavirus gastroenteritis (1), Streptococcus mitis bacteremia detected within 11 hours of presentation (1) and no identified focus (1). No deaths occurred. DISCUSSION This study identifies predictors of 2 measures of “risk” in pediatric and adolescent cancer patients with FN–invasive bacterial disease and clinical complications. The 2 outcome measures have some overlap but are complementary and not necessarily redundant when making treatment decisions for patients with FN. The risk of clinical complications influences the decision of whether the patient should receive care inpatient or outpatient, whereas the risk of IBD determines the route of antibiotic: intravenous or oral. Our study highlights the importance of having an initial period of inpatient observation, which minimizes the chance of a high-risk FN patient being inadvertently categorized as low-risk at admission and subsequently managed less aggressively using an outpatient approach. Risk assessment at admission combined with 24-hour observation shows a high negative predictive value. Despite the considerable progress made over the last decade in developing prediction rules for infections or clinical complications in patients with FN, there remain many limitations to the universal use of these rules, including lack of direct applicability of adult models to children, lack of multinational prospective validation of pediatric models, lack of pediatric prediction models with weighted scores for ease of use in clinical practice, and fundamental differences in definitions of “low-risk” and prediction end points such as bacterial infections, clinical complications or lack of response to therapy.22 To date, our model is the only one that evaluates clinical complications not occurring in association with sepsis syndrome, such as altered mental status or respiratory failure caused by a viral etiology. The presence of hemodynamic instability or sepsis syndrome is an integral part of the model developed by Santolaya et al.26 Table, Supplemental Digital Content 4, http://links.lww.com/INF/A317, compares the performance of various prediction models with that of our model. Our study addresses some of the above-mentioned limitations. Ours is the first study to examine the ability to predict the risk of invasive bacterial disease and clinical complications in pediatric and adolescent patients with FN, and provides an easily usable scoring tool for clinicians. Ours is also the only model that takes into account the value of including clinical appearance in risk prediction. The importance of “clinical impression,” which all clinicians consider in www.pidj.com | 55 The Pediatric Infectious Disease Journal • Volume 29, Number 1, January 2010 Hakim et al TABLE 1. Risk Predictors for “Proven Invasive Bacterial Infection or Culture-Negative Sepsis” and “Clinical Complications” Univariate Analysis Variables (n) Outcome Proven invasive bacterial infection or culture-negative sepsis† (n ⫽ 320) Cancer diagnosis (332) AML (16%) ALL/lymphoma (50%) Solid tumor (34%) Receiving G-CSF (332) Yes (29%) No (71%) Tmax at presentation (332) ⱖ39.0°C (18%) ⬍39.0°C (82%) ANC (330) ⬍100 (56%) ⱖ100 (44%) Plts (330) ⬍50,000 (44%) ⱖ50,000 (56%) Clinical appearance (331) Sick/toxic (25%) Well (75%) Prior relapse (332) Yes (18%) No (82%) Anticipated neutropenia ⬍7 d? (319) Yes (51%) No (49%) Outcome Clinical complications‡ (n ⫽ 324) Antifungal therapy within the past 6 mo? (313) Yes (8%) No (92%) Race (332) Nonwhite (25%) White (75%) Duration from neutropenia to fever onset (304) ⬎7 d (23%) ⱕ7 d (77%) Clinical appearance (331) Sick/toxic (25%) Well (75%) Prior relapse (332) Yes (18%) No (82%) At least one comorbidity at presentation (332) Yes (50%) No (50%) Cancer diagnosis (332) AML (16%) ALL/lymphoma (50%) Solid tumor (34%) OR 95% CI 6.91 1.93 1.00 2.77–17.24 0.83– 4.51 0.39 1.00 0.17– 0.90 2.05 1.00 Multiple Logistic Regression*  Regression Coefficient OR 95% CI P 2.0095 0.7218 7.46 2.06 1.00 2.76 –20.14 0.84 –5.02 ⬍0.0001 0.1100 1.00 – 4.20 1.0675 2.91 1.00 1.30 – 6.49 0.009 2.82 1.00 1.41–5.65 0.9864 2.68 1.00 1.25–5.76 0.012 2.33 1.00 1.25– 4.34 3.84 1.00 2.02–7.28 1.4423 4.23 1.00 2.10 – 8.53 ⬍0.0001 1.84 1.00 0.90 –3.74 0.43 1.00 0.23– 0.83 3.59 1.00 1.37–9.39 1.99 1.00 0.99 – 4.00 0.7950 2.21 1.00 1.00 – 4.90 0.0495 2.35 1.00 1.11– 4.99 6.39 1.00 3.18 –12.85 2.0032 7.41 1.00 3.50 –15.72 ⬍0.0001 2.87 1.00 1.39 –5.92 1.1457 3.14 1.00 1.36 –7.25 0.007 3.58 1.00 1.69 –7.59 2.14 0.96 1.00 0.87–5.23 0.44 –2.09 1.1484 ⫺0.0333 3.15 0.97 1.00 1.10 –9.04 0.40 –2.33 0.033 0.94 *The multiple logistic regression model was built by using stepwise selection method and included only the observations having complete data. The Hosmer-Lemeshow goodness-of-fit test for “proven invasive bacterial infection or culture-negative sepsis” did not indicate lack of fit (P ⫽ 0.14). ‡ The Hosmer-Lemeshow goodness-of-fit test for “clinical complications” model did not indicate lack of fit (P ⫽ 0.47). AML indicates acute myelogenous leukemia; ALL, acute lymphoblastic leukemia; G-CSF, granulocyte colony-stimulating factor; ANC, absolute neutrophil count; AMC, absolute monocyte count; Plts, platelets. † medical decision-making but find difficult to define, is a key finding of this study and perhaps underemphasized in previous studies in the FN literature.18,34 Consistent with 2 previous reports, nonwhite race (primarily African-American) was an independent predictor of CC in our 56 | www.pidj.com study.38,39 A longitudinal epidemiologic study that evaluated sepsis-related disparities in risk and outcome in a nationally representative sample of hospitalizations in 8.9 million patients with cancer in the United States has shown that white cancer patients had lower risk for sepsis than nonwhites.38 © 2009 Lippincott Williams & Wilkins The Pediatric Infectious Disease Journal • Volume 29, Number 1, January 2010 Infection and Cancer FIGURE 1. Receiver operating characteristic (ROC) curve for the outcome (A) “proven invasive bacterial infection or culture-negative sepsis,” using the scoring method, AUC (95% CI) ⫽ 0.776 (0.700 – 0.852) and (B) for the outcome “clinical complications,” using the scoring method, AUC (95% CI) ⫽ 0.803 (0.730 – 0.876). While our study has limitations that we recognized a priori and made attempts to minimize, the consistency of study results with 500 random bootstrap sample runs suggests robust findings. The retrospective study design introduces limitations because of the inability to capture clinical information in a standardized, consistent fashion. This limitation is especially pertinent to an important variable, “clinical appearance,” which came out as a strong predictor in both risk prediction models. Although we used a priori definitions while abstracting information on this variable, it will be imperative to examine it in a prospective study wherein the definition is available as a guide to the examining clinician. Overall, to minimize the limitations related to the retrospective nature of this study, we used a priori definitions, a standardized data abstraction form, and only 2 chart abstractors. Second, the exclusion of inflammatory markers in the model because they were © 2009 Lippincott Williams & Wilkins not routinely assessed in patients with FN has both advantages and disadvantages. While CRP is inexpensive and accessible, other promising inflammatory markers such as procalcitonin and IL6 remain expensive and not routinely used in patient care.40 – 44 Our current risk prediction model remains relevant to clinical settings where access to inflammatory markers is not easy and routine. That said, the consideration of inflammatory markers in further optimization of risk prediction is clear and is one of the objectives of our ongoing prospective validation study. In summary, the field of risk-stratified management of patients with FN has advanced and transitioned from research to practice much more for adult than for pediatric and adolescent patients. What is lacking is the systematic development of an easy-to-use risk prediction tool. Our study contributes to the existing data by providing a simple scoring tool for use by www.pidj.com | 57 The Pediatric Infectious Disease Journal • Volume 29, Number 1, January 2010 Hakim et al Anil Thridandapani, BS, Department of Infectious Diseases, St. Jude Children’s Research Hospital for constructing the study’s database and data quality assurance; Georgeta Vaidean, MD, PhD, Department of Preventive Medicine, University of Tennessee Health Science Center, Memphis, TN for her helpful comments about the manuscript; and Vani J. Shanker, PhD, ELS, Department of Scientific Editing, St. Jude Children’s Research Hospital for scientific editing of the manuscript. REFERENCES FIGURE 2. Effect of the scoring prediction model application at presentation followed by a 24-hour period of inpatient observation on the ability to detect patients with outcome events (A) “proven invasive bacterial infection or culture-negative sepsis” and (B) “clinical complications.” clinicians that predicts pediatric and adolescent patients with FN at risk for invasive bacterial disease and clinical complications. ACKNOWLEDGMENTS The authors thank Jawwad Yusuf, MBBS, and Jenine Worley for chart review and data abstraction; Wally Bitar, MS, and 58 | www.pidj.com 1. Pizzo PA, Robichaud KJ, Gill FA, et al. Duration of empiric antibiotic therapy in granulocytopenic patients with cancer. Am J Med. 1979;67:194 – 200. 2. Pizzo PA, Rubin M, Freifeld A, et al. The child with cancer and infection. I. Empiric therapy for fever and neutropenia, and preventive strategies. J Pediatr. 1991;119:679 – 694. 3. Hughes WT, Armstrong D, Bodey GP, et al. 2002 guidelines for the use of antimicrobial agents in neutropenic patients with cancer. Clin Infect Dis. 2002;34:730 –751. 4. Gardembas-Pain M, Desablens B, Sensebe L, et al. Home treatment of febrile neutropenia: an empirical oral antibiotic regimen. Ann Oncol. 1991;2:485– 487. 5. Kern WV. Risk assessment and risk-based therapeutic strategies in febrile neutropenia. Curr Opin Infect Dis. 2001;14:415– 422. 6. Malik IA, Abbas Z, Karim M. Randomised comparison of oral ofloxacin alone with combination of parenteral antibiotics in neutropenic febrile patients. Lancet. 1992;339:1092–1096. 7. Malik IA, Khan WA, Karim M, et al. Feasibility of outpatient management of fever in cancer patients with low-risk neutropenia: results of a prospective randomized trial. Am J Med. 1995;98:224 –231. 8. Aquino VM, Buchanan GR, Tkaczewski I, et al. Safety of early hospital discharge of selected febrile children and adolescents with cancer with prolonged neutropenia. Med Pediatr Oncol. 1997;28:191–195. 9. Aquino VM, Herrera L, Sandler ES, et al. Feasibility of oral ciprofloxacin for the outpatient management of febrile neutropenia in selected children with cancer. Cancer. 2000;88:1710 –1714. 10. Aquino VM, Tkaczewski I, Buchanan GR. Early discharge of low-risk febrile neutropenic children and adolescents with cancer. Clin Infect Dis. 1997;25:74 –78. 11. Bash RO, Katz JA, Cash JV, et al. Safety and cost effectiveness of early hospital discharge of lower risk children with cancer admitted for fever and neutropenia. Cancer. 1994;74:189 –196. 12. Klaassen RJ, Allen U, Doyle JJ. Randomized placebo-controlled trial of oral antibiotics in pediatric oncology patients at low-risk with fever and neutropenia. J Pediatr Hematol Oncol. 2000;22:405– 411. 13. Mustafa MM, Aquino VM, Pappo A, et al. A pilot study of outpatient management of febrile neutropenic children with cancer at low risk of bacteremia. J Pediatr. 1996;128:847– 849. 14. Paganini HR, Sarkis CM, De Martino MG, et al. Oral administration of cefixime to lower risk febrile neutropenic children with cancer. Cancer. 2000;88:2848 –2852. 15. Park JR, Coughlin J, Hawkins D, et al. Ciprofloxacin and amoxicillin as continuation treatment of febrile neutropenia in pediatric cancer patients. Med Pediatr Oncol. 2003;40:93–98. 16. Petrilli A, Altruda Carlesse F, Alberto Pires Pereira C. Oral gatifloxacin in the outpatient treatment of children with cancer fever and neutropenia. Pediatr Blood Cancer. 2007;49:682– 686. 17. Petrilli AS, Dantas LS, Campos MC, et al. Oral ciprofloxacin vs. intravenous ceftriaxone administered in an outpatient setting for fever and neutropenia in low-risk pediatric oncology patients: randomized prospective trial. Med Pediatr Oncol. 2000;34:87–91. 18. Santolaya ME, Alvarez AM, Aviles CL, et al. Early hospital discharge followed by outpatient management versus continued hospitalization of children with cancer, fever, and neutropenia at low risk for invasive bacterial infection. J Clin Oncol. 2004;22:3784 –3789. 19. Shenep JL, Flynn PM, Baker DK, et al. Oral cefixime is similar to continued intravenous antibiotics in the empirical treatment of febrile neutropenic children with cancer. Clin Infect Dis. 2001;32:36 – 43. 20. Sung L, Feldman BM, Schwamborn G, et al. Inpatient versus outpatient management of low-risk pediatric febrile neutropenia: measuring parents’ © 2009 Lippincott Williams & Wilkins The Pediatric Infectious Disease Journal • Volume 29, Number 1, January 2010 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. and healthcare professionals’ preferences. J Clin Oncol. 2004;22:3922– 3929. Vidal L, Paul M, Ben dor I, et al. Oral versus intravenous antibiotic treatment for febrile neutropenia in cancer patients: a systematic review and metaanalysis of randomized trials. J Antimicrob Chemother. 2004;54:29–37. Kern WV. Risk assessment and treatment of low-risk patients with febrile neutropenia. Clin Infect Dis. 2006;42:533–540. Klastersky J, Paesmans M, Rubenstein EB, et al. The Multinational Association for Supportive Care in Cancer risk index: a multinational scoring system for identifying low-risk febrile neutropenic cancer patients. J Clin Oncol. 2000;18:3038 –3051. Talcott JA, Siegel RD, Finberg R, et al. Risk assessment in cancer patients with fever and neutropenia: a prospective, two-center validation of a prediction rule. J Clin Oncol. 1992;10:316 –322. Talcott JA, Whalen A, Clark J, et al. Home antibiotic therapy for low-risk cancer patients with fever and neutropenia: a pilot study of 30 patients based on a validated prediction rule. J Clin Oncol. 1994;12:107–114. Santolaya ME, Alvarez AM, Aviles CL, et al. Prospective evaluation of a model of prediction of invasive bacterial infection risk among children with cancer, fever, and neutropenia. Clin Infect Dis. 2002;35:678 – 683. Santolaya ME, Alvarez AM, Becker A, et al. Prospective, multicenter evaluation of risk factors associated with invasive bacterial infection in children with cancer, neutropenia, and fever. J Clin Oncol. 2001;19:3415– 3421. Klaassen RJ, Goodman TR, Pham B, et al. “Low-risk” prediction rule for pediatric oncology patients presenting with fever and neutropenia. J Clin Oncol. 2000;18:1012–1019. Rackoff WR, Gonin R, Robinson C, et al. Predicting the risk of bacteremia in childen with fever and neutropenia. J Clin Oncol. 1996;14:919 –924. Lucas KG, Brown AE, Armstrong D, et al. The identification of febrile, neutropenic children with neoplastic disease at low risk for bacteremia and complications of sepsis. Cancer. 1996;77:791–798. Alexander SW, Wade KC, Hibberd PL, et al. Evaluation of risk prediction criteria for episodes of febrile neutropenia in children with cancer. J Pediatr Hematol Oncol. 2002;24:38 – 42. Paganini HR, Aguirre C, Puppa G, et al. A prospective, multicentric scoring system to predict mortality in febrile neutropenic children with cancer. Cancer. 2007;109:2572–2579. © 2009 Lippincott Williams & Wilkins Infection and Cancer 33. Ammann RA, Hirt A, Luthy AR, et al. Identification of children presenting with fever in chemotherapy-induced neutropenia at low risk for severe bacterial infection. Med Pediatr Oncol. 2003;41:436 – 443. 34. Rondinelli PI, Ribeiro Kde C, de Camargo B. A proposed score for predicting severe infection complications in children with chemotherapyinduced febrile neutropenia. J Pediatr Hematol Oncol. 2006;28:665– 670. 35. Wicki S, Keisker A, Aebi C, et al. Risk prediction of fever in neutropenia in children with cancer: a step towards individually tailored supportive therapy? Pediatr Blood Cancer. 2008;51:778 –783. 36. Perkins NJ, Schisterman EF. The inconsistency of “optimal” cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am J Epidemiol. 2006;163:670 – 675. 37. Efron B, Chong G. A leisurely look at the bootstrap, the jackknife, and cross-validation. Am Stat. 1983;37:36 – 48. 38. Danai PA, Moss M, Mannino DM, et al. The epidemiology of sepsis in patients with malignancy. Chest. 2006;129:1432–1440. 39. Metzger ML, Castellino SM, Hudson MM, et al. Effect of race on the outcome of pediatric patients with Hodgkin’s lymphoma. J Clin Oncol. 2008;26:1282–1288. 40. Fleischhack G, Kambeck I, Cipic D, et al. Procalcitonin in paediatric cancer patients: its diagnostic relevance is superior to that of C-reactive protein, interleukin 6, interleukin 8, soluble interleukin 2 receptor and soluble tumour necrosis factor receptor II. Br J Haematol. 2000;111:1093–1102. 41. Lehrnbecher T, Venzon D, de Haas M, et al. Assessment of measuring circulating levels of interleukin-6, interleukin-8, C-reactive protein, soluble Fc gamma receptor type III, and mannose-binding protein in febrile children with cancer and neutropenia. Clin Infect Dis. 1999;29:414 – 419. 42. Secmeer G, Devrim I, Kara A, et al. Role of procalcitonin and CRP in differentiating a stable from a deteriorating clinical course in pediatric febrile neutropenia. J Pediatr Hematol Oncol. 2007;29:107–111. 43. Stryjewski GR, Nylen ES, Bell MJ, et al. Interleukin-6, interleukin-8, and a rapid and sensitive assay for calcitonin precursors for the determination of bacterial sepsis in febrile neutropenic children. Pediatr Crit Care Med. 2005;6:129 –135. 44. Santolaya ME, Alvarez AM, Aviles CL, et al. Predictors of severe sepsis not clinically apparent during the first twenty-four hours of hospitalization in children with cancer, neutropenia, and fever: a prospective, multicenter trial. Pediatr Infect Dis J. 2008;27:538 –543. www.pidj.com | 59