|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
© 2000 American Society for Clinical Oncology The Multinational Association for Supportive Care in Cancer Risk Index: A Multinational Scoring System for Identifying Low-Risk Febrile Neutropenic Cancer PatientsFrom the Institut Jules Bordet, Brussels, Belgium; M.D. Anderson Cancer Center, Houston, TX; Royal Prince Alfred Hospital, Camperdown, Australia; Ontario Cancer Institute, Toronto, Canada; Geisinger Medical Center, Danville, PA; Herlev Hospital, University of Copenhagen, Copenhagen, Denmark; Medical Oncology Centre of Rosebank, Johannesburg, and Pretoria Academic Hospital, Pretoria, South Africa; and Dana Farber Cancer Institute, Boston, MA. Address reprint requests to J. Klastersky, MD, Service of Medicine, Institut Jules Bordet, 1 rue Héger-Bordet, B 1000 Brussels, Belgium; email jean.klastersky{at}bordet.be
PURPOSE: Febrile neutropenia remains a potentially life-threatening complication of anticancer chemotherapy, but some patients are at low risk for serious medical complications. The purpose of this study was to develop an internationally validated scoring system to identify these patients. MATERIALS AND METHODS: Febrile neutropenic cancer patients were observed in a prospective multinational study. Independent factors assessable at fever onset, predicting low risk of complications, on a randomly selected derivation set, were assigned integer weights to develop a risk-index score, which was subsequently tested on a validation set.
RESULTS: On the derivation set (756 patients), predictive factors were a burden of illness indicating absence of symptoms or mild symptoms (weight, 5; odds ratio [OR], 8.21; 95% confidence interval [CI], 4.15 to 16.38) or moderate symptoms (weight, 3; OR, 3.70; 95% CI, 2.18 to 6.29); absence of hypotension (weight, 5; OR, 7.62; 95% CI, 2.91 to 19.89); absence of chronic obstructive pulmonary disease (weight, 4; OR, 5.35; 95% CI, 1.86 to 15.46); presence of solid tumor or absence of previous fungal infection in patients with hematologic malignancies (weight, 4; OR, 5.07; 95% CI, 1.97 to 12.95); outpatient status (weight, 3; OR, 3.51; 95% CI, 2.02 to 6.04); absence of dehydration (weight, 3; OR, 3.81; 95% CI, 1.89 to 7.73); and age less than 60 years (weight, 2; OR, 2.45; 95% CI, 1.51 to 4.01). On the validation set, a Multinational Association for Supportive Care in Cancer risk-index score CONCLUSION: The risk index accurately identifies patients at low risk for complications and may be used to select patients for testing therapeutic strategies that may be more convenient or cost-effective.
FEBRILE EPISODES IN cancer patients with chemotherapy-induced neutropenia can be life-threatening and may require administration of empiric, broad-spectrum antibiotics.1 The accepted standard of care for such patients has been routine hospitalization for administration of parenteral antibiotics, close monitoring for development of complications, and evaluation for response to therapy.2 This practice, along with the use of substantially improved antimicrobial agents, has resulted in a dramatic decrease in mortality among febrile neutropenic patients. Many investigations have indicated that neutropenic patients with fever are a heterogeneous population, with subsets with varying risks regarding response to initial therapy, development of serious medical complications, and mortality.3-7 Over the past decade, several investigators have identified subsets of febrile neutropenic patients who are at low risk for the development of complications, including mortality.8-11 Several clinical studies involving neutropenic patients with predicted low risk have demonstrated the feasibility of newer approaches, such as outpatient therapy after early discharge from the hospital or outpatient therapy for the entire febrile episode, using parenteral, sequential (intravenous [IV] followed by oral), or oral antibiotic regimens.12-17 Because these studies were conducted at single centers, using various definitions of low-risk patients and without adequate controls, they cannot be considered as having demonstrated the safety and efficacy of these new approaches. However, the safety of an oral regimen was suggested in an early controlled study.18 Furthermore, two recent randomized large trials demonstrated the efficacy and safety, in clinically selected low-risk patients, of an oral empiric antibiotic treatment compared with a parenteral regimen19,20; however, in both studies, all patients were treated as inpatients. In published guidelines on the management of febrile neutropenia, oral therapy has been identified as a possible option for low-risk patients.21,22 Pivotal for the adoption of newer therapeutic strategies are a uniformly acceptable definition of low-risk febrile neutropenic patients and a simple but accurate clinical prediction rule that reliably identifies such patients at the onset of a febrile episode. Talcott et al23 developed a clinical prediction rule classifying patients into four risk groups. They suggested that neutropenic patients with controlled cancer and no serious comorbidity who developed fever in an outpatient setting would be at low risk, with an expected rate of serious medical complications of less than 5%. This classification scheme was prospectively validated in a study conducted at two United States institutions.24 It was further tested clinically in a pilot study involving 30 low-risk patients who were discharged after an initial 48-hour hospital stay.25 Thirteen percent of these patients developed serious complications, and nine patients (30%) were readmitted. However, five of these readmissions were due only to prolonged fever that resolved immediately afterward, and no patient died. In other studies, patients selected using clinical criteria were treated successfully as outpatients with IV or orally administered antibiotics.14,26 The purpose of this study, which was designed as a multicenter prospective observational study involving febrile neutropenic cancer patients treated with empiric antibiotic regimens, was to develop an internationally validated scoring system to identify low-risk patients (those defined as having a high probability of fever resolution without development of serious medical complications or death) at the onset of their febrile episodes. The performance of this clinical prediction rule was compared with that of the Talcott model. The ultimate goal of the scoring system was selection of patients who might be candidates for new therapeutic strategies, including, but not limited to, outpatient management. This led us to register patients already hospitalized at fever onset, most of whom probably could not have been sent home even after initial hospital-based antimicrobial treatment. Inclusion of only outpatients in our study would have been detrimental for the generalizability of our model, given that the indication for hospitalization of cancer patients has changed during the last decade and will probably continue to change. The study was conducted by the Study Section on Infections of Multinational Association of Supportive Care in Cancer (MASCC).
Patients The participating institutions agreed to study prospectively all consecutive febrile episodes (or randomly selected febrile episodes) occurring in patients meeting the following eligibility criteria: diagnosis of malignancy treated by chemotherapy that was causative of or contributive to neutropenia (granulocyte count < 500/µL, including polymorphonuclear leukocytes and band forms), temperature greater than 38°C (measured orally and documented by the patient or the medical/nursing staff), and age greater than 16 years. In addition, patients’ febrile neutropenia had to be treated with an appropriate initial empiric antibiotic regimen based on known distribution of local pathogens and local susceptibilities, including an antipseudomonal beta-lactam in combination with an aminoglycoside, two beta-lactams in combination, or monotherapy with a third-generation cephalosporin or a carbapenem. Appropriateness of each administered empiric regimen was reviewed during a meeting of the authors of the present report and, if necessary, discussed with the local investigator. Because allowing patients to be enrolled several times during successive febrile episodes could induce a covariance structure between outcome data and invalidate statistical comparisons, only the first febrile episode occurring in a patient during the study period was considered. Finally, each participating institution contracted to enroll at least 20 patients.
Outcome
Data Collection All case report forms were sent to one of the institutions (Institut Jules Bordet), where a central review was performed to verify eligibility as well as homogeneity in outcome assessment and where the database was managed and checked for consistency.
Statistical Methodology For both samples, patient baseline characteristics were summarized using descriptive statistics: frequency tabulations for categorical variables and summary parameters (median and range) for continuous variables. Model derivation. Logistic regression analysis was applied to the derivation set to assess the association between the covariates and outcome. The covariates considered as potential prognostic factors are listed in Table 2; all were measured at the time of presentation with fever. All continuous variables were categorized on the basis of clinical arguments (the categories are then easy to use and to interpret) using a priori cut points, chosen according to physician judgment or to published reports.28, 29 Univariate analyses were performed to select a first set of covariates to be tested for inclusion in a multivariate model; all covariates with a P value less than .1 for the null hypothesis of no effect were considered. Estimates of the covariate coefficients were obtained using the maximum likelihood method. Estimated odds ratios (ORs) with confidence intervals (CIs) were calculated. A priori second-order interaction terms, selected using clinical judgment and expertise, were also considered, namely diagnosis (solid tumor or hematologic tumor) by age, shock (systolic blood pressure < 90 mmHg), and any comorbidity; control of cancer by any comorbidity; profound neutropenia (< 100/µL) by fever (temperature > 39°C); inpatient status by Acute Physiology and Chronic Health Evaluation (APACHE) II score; burden of illness by any comorbidity; and prophylaxis administration by growth factor administration.
After completion of the univariate analyses, a multiple logistic regression model was fit to the data. A backward and forward step-up procedure was used for selection of the covariates. A P value less than .01 was the criterion for entry of a covariate in the model, and a further P value greater than .05 led to removal of the variable from the model. All P values were two-sided. Determination of APACHE II score (APACHE II was developed and validated in intensive care settings) requires additional computation and the values of some variables not routinely assessed in patients with febrile neutropenia. For these reasons, the score was not included in further assessment. Based on the logistic model, a score for prediction was calculated for each patient. Each coefficient estimate was multiplied by the same factor and rounded to the nearest integer. The multiplicative factor was chosen to obtain the smallest individual weight equal to 2. The global score was the sum of these individual weights and was substituted to the exact value of the logistic equation for ease of clinical implementation. A higher global score indicated a greater likelihood of fever resolution without development of any serious medical complication. High values identified low-risk patients. The accuracy of the scoring system was assessed by plotting observed versus predicted outcomes. On the basis of this scoring system, several prediction rules were derived with the aim of selecting patients at low risk, using each possible value of the score as a threshold. Patients with scores higher than the thresholds constituted the group at low risk for complications and were predicted to recover favorably from their neutropenic febrile episodes. The sensitivity, specificity, and positive and negative predictive values of these rules were evaluated, with the convention that patients predicted by the rule to be at low risk and indeed recovering without complications represented true positives. Considering that our objective was to select, for therapeutic investigations, patients at low risk, we chose score thresholds to obtain safe positive predictive values with low rates of false identification of patients as being at low risk. Validation set. The validation set was analyzed to test the performance of the derived prediction rule and to compare that rule with the Talcott prediction rule. Receiver operating characteristic curves were calculated to compare the performance of the rule in the derivation and validation sets.
Missing data.
Missing data were handled in the following manner to avoid subsequent patient data exclusions from the analysis: a missing absolute neutrophil count was categorized, on the basis of the WBC count, into two categories (
Eligibility Between December 1994 and November 1997, 1,351 patients from 20 institutions (in 15 countries) were registered onto the study. Of these patients, 212 (16%) were considered ineligible, for the following reasons: undocumented neutropenia (72 patients), no or undocumented fever (45 patients), no chemotherapy or chemotherapy not contributive to neutropenia (38 patients), previous entry onto the study (30 patients), no empiric treatment or ineligible empiric antibiotic treatment (15 patients), no cancer (four patients), age less than 16 years (three patients), retrospective data collection (three patients), and incomplete (for key variables) case report form (two patients). Therefore, data relative to 1,139 patients were kept for the final analysis. Participating investigators and institutions are listed in the Appendix.
Patient Characteristics
Outcome Distribution Complete resolution of fever without complications occurred in 645 patients (85%; 95% CI, 82.7% to 87.9%) in the derivation set and 310 patients in the validation set (81%; 95% CI, 76.9% to 85.0%), whereas 111 in the derivation set (15%; 95% CI, 12.1% to 17.3%) and 73 in the validation set (19%; 95% CI, 15.0% to 23.1%) developed at least one serious medical complication, including death before complete resolution of fever (35 patients in the derivation set and 20 patients in the validation set). The most frequent complications were confusion, hypotension, and respiratory failure. A microbiologically documented infection with bacteremia was reported in 206 patients in the derivation set (27%; 95% CI, 24.0% to 30.5%) and 94 patients in the validation set (25%; 95% CI, 20.1% to 20.9%).
Predictive Factors for Favorable Outcome
Among the interaction tests investigated using logistic regression models with two main effects and one second-order interaction term, one was selected to be tested in the multivariate analysis: a term including underlying disease and previous fungal infection status (previous fungal infection is associated with a poorer prognosis in patients with hematologic tumors but not in patients with solid tumor [OR, 0.24; 95% CI, 0.11 to 0.55; P < .001]). Based on multivariate analysis of a set of covariates selected as described earlier, the characteristics included in the final model, which could be considered independent predictive factors of favorable outcome, were (in order of decreasing magnitude of the corresponding ORs) burden of illness with no or mild symptoms or moderate symptoms, absence of hypotension, absence of chronic pulmonary obstructive disease, solid tumor or no previous fungal infection in patients with hematologic tumors, absence of dehydration, burden of illness with moderate symptoms, outpatient status, and age less than 60 years. The coefficient estimates of the covariates included in our model are listed in Table 5 with the corresponding ORs and their CIs and a goodness-of-fit statistic for the model. The integer score derived from the logistic equation of this model gave the following weights to those characteristics: burden of illness with no or mild symptoms, five points; no hypotension, five points; no chronic obstructive pulmonary disease, four points; solid tumor or no previous fungal infection in the case of a hematologic tumor, four points; no dehydration, three points; burden of illness with moderate symptoms, three points; outpatient status, three points; and age less than 60 years, two points. The whole score was obtained by summing the individual weights (Table 6). The maximum score is 26, because the points given for the two favorable categories of burden of illness are not cumulative. Plots of the observed and predicted outcomes against score values are presented in Fig 1.
The prediction rules derived using this score, on the derivation set, are listed in Table 7 for different thresholds, with the associated sensitivities, specificities, positive and negative predictive values, and overall misclassification rates. The same parameters were calculated for the Talcott clinical prediction rule. The best overall misclassification rate that can be achieved is 13%. At that level, 616 of 756 patients are identified as low-risk patients; however, at that threshold, the proportion of patients falsely identified as being at low risk (12%) is too high for a safe clinical prediction rule. We propose to use the threshold of 21 because it corresponds to a still relatively low misclassification rate (21%), to a large proportion of patients identified as being at low risk (551 of 756, 73%), and an increased positive predictive value (94%).
Of the 551 patients identified as being at low risk using the threshold of 21, a total of 32 (6%) developed a serious medical complication, including death (six patients [1%]). In the group considered by the rule to be at high risk for complications (205 patients), 79 patients (39%) had a serious medical complication, including death (29 patients [14%]). The most frequent complications among the patients falsely identified as members of the good-prognosis group were confusion (nine patients), cardiac problems (ECG changes in seven and arrhythmia in seven), respiratory failure (six patients), hypotension (four patients), renal failure (four patients), and bleeding (four patients). Of the six patients who died before fever resolution, three had hematologic malignancies, four had advanced disease treated with first-line chemotherapy, and two had advanced disease treated with second (subsequent)-line chemotherapy; a microbiologic diagnosis was made in four patients (with bacteremia in three). The deaths occurred quite late after fever onset in most of the six patients (two died on day 8 and three died after > 2 weeks); one patient died on day 3. Three of these patients had been allocated to Talcott group IV. The application of the Talcott model on our derivation set led to the identification of 218 patients (29%) with predicted low risk. The positive predictive value is high (96%), but sensitivity is low (32%) and the misclassification rate is high (59%). The death rate in this Talcott group IV was 1% (three of 218 patients). Validation set. Of the 383 patients in the validation set, 243 (63%) were predicted, using the chosen prediction rule with a threshold of 21, to be at low risk, with a 91% rate of resolution without complications; whereas 140 were classified as being at high risk, with a rate of resolution without serious complications of 64%. The sensitivity, specificity, negative predictive value, and misclassification rate of the prediction rule were 71%, 68%, 36%, and 30%, respectively. By comparison, the Talcott prediction rule identified 99 patients (26%) as being at low risk, with a 7% false-positive rate (93% rate of resolution without occurrence of a serious medical problem). Among the other patients, identified as being at high risk, the rate of resolution without complication was 77% (218 of 284 patients). Four (1.6%) of the 243 patients in the low-risk group died. One of these patients had solid tumor, two had advanced disease and received first-line chemotherapy, one had advanced disease treated with second (subsequent)-line chemotherapy, one underwent bone marrow transplantation, two developed bacteremia, a microbiologic diagnosis was made without bacteremia documentation in one patient, and three patients were allocated to Talcott group I. As in the derivation set, these deaths did not occur immediately after the beginning of the febrile episode (one patient died on day 5 and three died after > 2 weeks). The most frequent complications in the false-positive patients were bleeding (three patients) and hypotension (three patients). Renal failure and respiratory failure occurred in one patient each. The sensitivity, specificity, negative predictive value, and misclassification rate of the Talcott prediction rule were 30%, 90%, 23%, and 59%, respectively. Three deaths occurred in Talcott group IV (on days 5, 6, and 8). Two of these patients had an MASCC risk-index score of less than 21 and advanced disease treated with first-line chemotherapy (one had solid tumor and one had non-Hodgkin’s lymphoma); both patients developed bacteremia. In Table 8, we provide the parameters of the MASCC clinical prediction rule at several thresholds, as well as those of the Talcott model, on the validation set. In Fig 2, we present, for both sets, the receiver operating characteristic curves assessing the discriminating ability of the MASCC scoring system.
There is general agreement that febrile neutropenic cancer patients constitute a heterogeneous population, with a subgroup having a low risk of serious complications or death. This recognition of risk stratification has led to changes in the treatment of these patients, including changes in antimicrobial therapy (combinations v monotherapy), mode of administration (IV v orally), and treatment setting (hospital, outpatient center, physician’s office, home). Two recently published clinical trials19,20 demonstrated the safety and efficacy of oral antibiotics for low-risk patients in the inpatient setting. These studies used different definitions to select patients for the trials, which underscores the need for an internationally validated system to identify this low-risk group accurately so that clinical research into new treatment strategies, including outpatient management, can be conducted safely. In this multinational, multicenter study of more than 1,100 patients with fever and neutropenia, we demonstrated that certain characteristics, easily identifiable at the onset of the febrile episode, predict low risk of medical complications. Using these factors, we developed the simple and easy-to-use MASCC scoring system and its clinical prediction rule for identification of low-risk patients. When we were developing our rule, we examined the trade-offs between sensitivity, specificity, positive and negative predictive values, and overall misclassification rate. The rule that we are proposing, with a threshold of 21, has a low, if not minimal, misclassification rate, corresponding to a nearly optimal use of the logistic regression model, and a better sensitivity than rules using higher thresholds would have. If we apply our rule to the subgroup of outpatients, its positive predictive value, on the validation set, is improved (94% [133 of 142 patients]) and is comparable to that of the Talcott model (93%) when that model is applied to our patients. With our model, 142 (76%) of 186 outpatients are identified as being at low risk, compared with 99 Talcott group IV patients (53%). Each user of our model still has the opportunity to choose the threshold defining the clinical prediction rule. However, if we select a clinical prediction rule at a higher threshold (eg, 22), the overall performance of the rule is worse (at a threshold of 22, the misclassification rate is increased to 45%). Our model has several advantages over prior methods for identifying low-risk patients. Our prospective study involved worldwide participation (investigators were affiliated with 20 institutions in 15 countries); the external validity of the results was increased with a large sample size, ensuring internal validity. As far as external validity is concerned, our patient population is more representative of the varying patterns of clinical practice in the management of these patients. Our sample included patients who were treated with different antibiotic regimens after having undergone a variety of myelosuppressive treatments in various types of institutions. We developed our risk-index score on a derivation set and further tested it on a validation set. The two sets were obtained by randomizing institutions rather than patients, to better assess the validity of our model, even in the presence of varying types of patient management. By this method, we could also address the issue of consideration of subjectively assessed covariates, such as burden of illness, which was included in our final model but was perhaps not consistently measured by the different investigators. The generalizability of our model, in terms of reproducibility and transportability as defined by Justice et al,30 is confirmed by the fact that the positive predictive value in the validation set (94%) was close to that in the derivation set (91%). A similar decrease in positive predictive value also occurred when the Talcott model was applied to our patients; this was unexpected, because neither the derivation set nor the validation set was used for development of the Talcott rule. This suggests that the small loss in the positive predictive value of our rule is due more to a random variation than to a lack of reproducibility. We acknowledge that the validation of our scoring system is not a static process. Our model will need to be tested repeatedly over time and will probably need to be updated to reflect changes in the management of febrile neutropenic patients. Data were missing for some of the biologic variables (at rates from < 1% to up to 20%) but not for the covariates included in the final model. The reproducibility of the performance of our model should therefore not be jeopardized by that problem. Our model represents an improvement over the Talcott classification in that our model has a lower misclassification rate (30% v 59%, on the validation set) and a better sensitivity (71% v 30%). The positive predictive value was comparable (91% v 93%), but our model has a lower specificity (68% v 90%). The major advantage of our model is that the rate of identification of patients as being at low risk is substantially increased (63% v 26%). The use of a higher threshold to define the clinical prediction rule, a threshold of 22 for instance, would result in a higher positive predictive value (94%) on the validation set and a specificity comparable to that of the Talcott model (86%). However, although this choice might be safer, especially in patients not rigorously monitored, it corresponds to a loss of 22% of low-risk patients. Furthermore, as already mentioned, the positive predictive value of our model (94%), in the outpatients included in the validation set, is comparable to that of the Talcott model (93%) when that model is applied to the same patient population. Another advantage of our model over the Talcott classification system is that we replaced the uncontrolled cancer variable, not selected in our multivariate model, with measures more specifically associated with the clinical severity of the febrile episode (rather than with the underlying cancer), such as burden of illness, hypotension, and dehydration. The general comorbidity variable was replaced with specific conditions (chronic obstructive pulmonary disease and age). It is not surprising that comorbidities such as diabetes, cardiac disease, and confusion, which were relatively uncommon in our series, are not included in our final model; they were probably replaced by other variables such as burden of illness. Our group of patients predicted as being at high risk remains heterogeneous; a substantial number of patients in that group recover from fever without complications. This is a common problem that has been encountered by other investigators attempting to predict the occurrence of complicated febrile neutropenia.31-34 There is a need to evaluate new covariates including variables related to the treatment of the underlying disease, which might be particularly important11, 35 but were not fully assessed in our study. Among the factors that we expected to be predictive of development of a serious complication were the underlying disease (hematologic malignancy v solid tumor) and the duration of neutropenia. Neither was included in our model. Nevertheless, the role of the underlying disease appears in our model as an interaction between prior fungal infections and hematologic malignancy. Its meaning is not completely clear; it may be a surrogate marker for refractory or relapsed leukemia or prolonged prior episodes of neutropenia. The expected duration of neutropenia, often used by other authors for selecting low-risk patients,19,20 although not prognostic of final outcome, predicted a higher probability of responding to empiric antimicrobial therapy without need for modification. However, the expected duration of neutropenia did not accurately predict the actual duration of neutropenia, which might be an important factor. Accurate prediction of the severity and duration of neutropenia is an important goal. These aspects should be taken into consideration during the development of future risk models, which should allow us to provide optimal care to febrile neutropenic patients with new therapeutic strategies focusing on patient comfort and quality of life as well as cost efficacy.
The following are participating investigators and institutions, with numbers of eligible patients given in parentheses: L. Elting, K. Rolston, E. Rubenstein, M.D. Anderson Cancer Center, Houston, TX (172 patients); T. Berghmans, J. Klastersky, P. Mommen, M. Paesmans, Institut Jules Bordet, Brussels, Belgium (159 patients); B. De Pauw, J.P. Donnelly, Algemeen Universitair Ziekenhuis, Nijmegen, the Netherlands (91 patients); R. Feld, A. McGeer, Ontario Cancer Institute, Toronto, Canada (83 patients); A. Cometta, G. Zanetti, Centre Hospitalier Universitaire Vaudois, Lausanne, Switzerland (73 patients); W. Feremans, A. Kentos, Hôpital Erasme, Brussels, Belgium (53 patients); I. Malik, National Cancer Institute, Karachi, Pakistan (52 patients); J.L. Michaux, S. Neumeister, Cliniques Saint-Luc, Brussels, Belgium (50 patients); J. Herrstedt, K. Wedervang, Herlev Hospital, University of Copenhagen, Copenhagen, Denmark (49 patients); W. Kern, Medizinisches Universitätsklinik, Ulm, Germany (46 patients); I. Bover, Hospital Sant Joan, Reus, Spain (43 patients); M. Boyer, Royal Prince Alfred Hospital, Camperdown, Australia (41 patients); P. Ljungman, Huddinge University Hospital, Huddinge, Sweden (40 patients); J. Vorlicek, Masaryk University Hospital, Brno, Czech Republic (35 patients); Z. Aziz, Jinnah Hospital, Lahore, Pakistan (27 patients); B.L. Rapoport, Medical Oncology Centre of Rosebank, Johannesburg, and Pretoria Academic Hospital, Pretoria, South Africa (27 patients); J. Gallagher, Geisinger Medical Center, Danville, PA (26 patients); J. Talcott, Dana Farber Cancer Institute, Boston, MA (25 patients); M. Viot, Centre Antoine Lacassagne, Nice, France (25 patients); R. de Bock, Algemeen Ziekenhuis Middelheim, Antwerp, Belgium (22 patients).
We thank R. Sylvester, PhD (European Organization for Research and Treatment of Cancer Data Center, Brussels, Belgium), for his expert statistical review of the analysis and the manuscript.
1. Klastersky J, Zinner SH, Calandra T, et al: Empiric antimicrobial therapy for febrile granulocytopenic cancer patients: Lessons from four EORTC trials. Eur J Cancer Clin Oncol 24:S35–S45, 1988 (suppl 1) 2. Hughes WT, Armstrong D, Bodey GP, et al: From the Infectious Diseases Society of America: Guidelines for the use of antimicrobial agents in neutropenic patients with unexplained fever. J Infect Dis 161:381–396, 1990[Medline] 3. Bodey GP, Buckley M, Sathe YS, et al: Quantitative relationships between circulating leukocytes and infections in patients with acute myeloid leukemias. Ann Intern Med 64:328–340, 1966
4.
Rolston KVI, Berkey P, Bodey GP, et al: A comparison of imipenem to ceftazidime with or without amikacin as empiric therapy in febrile neutropenic patients. Arch Intern Med 152:283–291, 1992
5.
Cometta A, Zinner SH, de Bock R, et al: Piperacillin/Tazobactam plus amikacin versus ceftazidime plus amikacin as empiric therapy for fever in granulocytopenic patients with cancer. Antimicrob Agents Chemother 39:445–452, 1995 6. Elting LS, Rubenstein EB, Rolston KVI, et al: Outcomes of bacteremia in patients with cancer and neutropenia: Observations from two decades of epidemiological and clinical trials. Clin Infect Dis 25:247–259, 1997[Medline] 7. Rossini F, Pioltelli P, Mingozzi S, et al: Amikacin and ceftazidime as empirical therapy in severely neutropenic patients: Analysis of prognostic factors. Support Care Cancer 2:259–265, 1994[Medline] 8. 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 77:791–798, 1996[Medline] 9. Viscoli C, Bruzzi P, Castagnola E, et al: Factors associated with bacteremia in febrile granulocytopenic cancer patients. Eur J Cancer 30:430–437, 1994 10. Jones GR, Konsler GK, Pusek SN: Infection risk factors in febrile neutropenic children and adolescents. Pediatr Hematol Oncol 13:217–229, 1996[Medline] 11. Engels EA, Ellis CA, Supran SE, et al: Early infection in bone marrow transplantation: Quantitative study of clinical factors that affect risk. Clin Infect Dis 28:256–266, 1999[Medline] 12. Mullen CA, Buchanan GR: Early hospital discharge of children with cancer treated for fever and neutropenia: Identification and management of the low-risk patient. J Clin Oncol 8:1198–2004, 1990
13.
Gardembas-Pain M, Desablens B, Sensebe L, et al: Home treatment of febrile neutropenia: An empirical oral antibiotic regimen. Ann Oncol 2:485–487, 1991 14. Rubenstein EB, Rolston KVI, Benjamin RS, et al: Outpatient treatment of febrile episodes in low-risk neutropenic patients with cancer. Cancer 71:3640–3646, 1993[Medline] 15. 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 74:189–196, 1994[Medline] 16. Meropol NJ, Fox KR, Vaughn DJ, et al: A pilot study of early hospital discharge in adult patients with fever and neutropenia. Eur J Cancer 30A:1595–1596, 1994 17. Horowitz HW, Holmgren D, Seiter K: Stepdown single agent antibiotic therapy for the management of the high-risk neutropenic adult with hematologic malignancies. Leuk Lymphoma 23:159–163, 1996[Medline] 18. Malik A, Abbas Z, Karim M: Randomised comparison of oral ofloxacin alone with combination of parenteral antibiotics in neutropenic febrile patients. Lancet 339:1092–1096, 1992[Medline]
19.
Kern WV, Cometta A, de Bock R, et al: Oral versus intravenous empiric therapy for fever in patients with granulocytopenia who are receiving cancer chemotherapy. N Engl J Med 341:312–318, 1999
20.
Freifeld A, Marchigiani D, Walsh T, et al: A double-blind comparison of empirical oral and intravenous antibiotic therapy for low-risk febrile patients with neutropenia during cancer chemotherapy. N Engl J Med 341:305–311, 1999 21. Biron P, Fuhrmann C, Escande MC, et al: Standards, options et recommendations pour la prise en charge des neutropénies courtes. Bull Cancer 85:695–711, 1998[Medline] 22. National Comprehensive Cancer Network: NCCN practice guidelines for fever and neutropenia. Oncology (Huntingt) 13:197–257, 1999[Medline]
23.
Talcott JA, Finberg R, Mayer RJ, et al: The medical course of cancer patients with fever and neutropenia: Clinical identification of a low-risk subgroup at presentation. Arch Intern Med 148:2561–2568, 1988 24. 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 10:316–322, 1992[Abstract] 25. Talcott JA, Whalen A, Clark J, et al: Home antibiotic therapy for low-risk patients with fever and neutropenia: A pilot study of 30 patients based on a validated prediction rule. J Clin Oncol 12:107–114, 1994[Abstract] 26. 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 98:224–231, 1995[Medline] 27. Pompei P, Charlson ME, Ales K, et al: Relating patient characteristics at the time of admission to outcomes of hospitalization. J Clin Epidemiol 44:1063–1069, 1991[Medline] 28. Knaus WA, Draper EA, Wagner DP, et al: APACHE II: A severity of disease classification system. Crit Care Med 13:818–829, 1985[Medline] 29. Groeger JS, Lemeshow S, Price K, et al: Multicenter outcome study of cancer patients admitted to the intensive care unit: A probability of mortality model. J Clin Oncol 16:761–770, 1998[Abstract]
30.
Justice AC, Kovinsky KE, Berlin JA: Assessing the generalizability of prognostic information. Ann Intern Med 130:515–524, 1999 31. Bates DW, Cook EF, Goldman L, et al: Predicting bacteremia in hospitalized patients: A prospectively validated model. Ann Intern Med 113:495–500, 1990
32.
Rackoff WR, Gonin R, Robinson C, et al: Predicting the risk of bacteremia in children with fever and neutropenia. J Clin Oncol 14:919–924, 1996 33. Hann I, Viscoli C, Paesmans M, et al: A comparison of outcome from febrile neutropenic episodes in children compared with adults: Results from four EORTC studies—International Antimicrobial Therapy Cooperative Group (IATCG) of the European Organization for Research and Treatment of Cancer (EORTC). Br J Haematol 99:580–588, 1997[Medline] 34. Bates DW, Sands K, Miller E, et al: Predicting bacteremia in patients with sepsis syndrome: Academic Medical Center Consortium Sepsis Project Working Group. J Infect Dis 176:1538–1551, 1997[Medline]
35.
Blay JY, Chauvin F, Le Cesne A, et al: Early lymphopenia after cytotoxic chemotherapy as a risk factor for febrile neutropenia. J Clin Oncol 14:636–643, 1996 Submitted January 4, 2000; accepted April 26, 2000.
This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2000 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
21 identified low-risk patients with a positive predictive value of 91%, specificity of 68%, and sensitivity of 71%. 
100/µL or > 100/µL). In case of a WBC count greater than 100/µL, this categorization was impossible and a random assignment was performed, with stratification by institution and, within each institution, with respect to the existing ratio between the two considered categories. Other missing biologic values were assumed to be in the normal defined range. 
