A prospective, multicentric scoring system to predict mortality in febrile neutropenic children with cancer
Many studies have succeeded in identifying a subset of children with febrile neutropenia (FN) who are at lower risk of infectious complications and eventual death. Conversely, to the authors' knowledge, no scoring system has been published to date with which to assess the risk of mortality for the whole group of children with neutropenia and fever.
Between March 2000 and July 2004, 1520 episodes of FN in 981 children were included in a multicentric prospective study to evaluate a scoring system that was designed to identify high mortality risk at the onset of an FN episode in children with cancer.
In the derivation set (714 episodes), 18 patients died (2.5%). A multivariate analysis yielded the following significant mortality-related risk factors: advanced stage of underlying malignant disease (odds ratio [OR], 3122.1; 95% confidence interval [95% CI], 0.0001–5.2), associated comorbidity (OR, 25.3; 95% CI, 7.7–83.2), and bacteremia (OR, 7.2; 95% CI, 2.4–22.0). A mortality score could be built with 3 points scored for the presence of advanced-stage underlying malignant disease, 2 points scored for the presence of associated comorbidity, and 1 point scored for bacteremia. If patients collected 4 points of the risk score at onset, then their risk of mortality was 5.8%; if patients had a score of 5 points, then their risk of mortality was 15.4%; and, if they reached the maximum score of 6 points, then their risk of mortality was raised to 40%. The sensitivity of the scoring system was 100%, and it had a specificity of 84.2%. In the validation set (806 episodes), 19 children died (2.3%). For children with scores >3, the scoring system had a sensitivity of 84.2%, a specificity of 83.2%, and a negative predictive value of 99.54% for predicting mortality.
The use of a mortality score for high-risk patients was validated statistically by the current results. This is a major prognostic approach to categorize patients with high-risk FN at onset. A better initial predictive approach may allow better therapeutic decisions for these children, with an eventual impact on reducing mortality. Cancer 2007. © 2007 American Cancer Society.
Infections are a major cause of mortality in patients who display febrile neutropenia (FN) during anticancer chemotherapy.1 However, this group is heterogeneous and includes patients who have different risks of developing severe infections and related complications.2
The standard of care for patients with FN usually has been limited to hospitalization; the administration of intravenous, broad-spectrum antibiotics; and close observation.3 Several additional approaches have been attempted during the last decade, and many studies succeeded in identifying a subset of patients with lower risk of developing infectious complications and death.4 Bacteremia, other severe bacterial infections, and mortality were the most frequently assessed variable.5–9 However, the studies that intended to identify risk factors used different designs and definitions, hindering the development of a standardized risk profile. However, several authorities analyzed the performance of different mortality scores on adult patients with cancer, either with or without FN, who were admitted to intensive care units,10–12 including the Acute Physiology and Chronic Health Evaluation, the Organ Dysfunction and Infection Score, and the Simplified Acute Physiology Score. Results varied according to the method employed.12 Furthermore, a recent report from a multicenter European study of adults with FN used a scoring system that was designed to define low-risk patients.13 However, to the best of our knowledge, no scoring system to assess the risk of mortality in children with FN has been published to date. Indeed, a scoring system applied to the initial evaluation of children with FN would be useful for identifying patients with a high risk of mortality, because it would allow the selection of more aggressive initial therapy in the hope of improving their outcome. Likewise, the identification of low-risk patients could provide a basis for the development of less expensive and less aggressive therapies. In fact, it is becoming a frequent strategy to treat these kinds of patients on sequential parenteral-oral14–16 or oral17 schedules in an ambulatory setting.18–23
The objective of the current prospective study was to test a scoring system to assess mortality in children with FN. We identified the risk factors for mortality in children with cancer and FN in our center and, subsequently, designed a scoring system to predict mortality within the first hours of hospitalization. Thereafter, this scoring system was tested and was validated statistically at 8 pediatric institutions in Argentina.
MATERIALS AND METHODS
The identification and assessment of risk factors for mortality, along with the initial development of the scoring system (derivation set), were carried out at the Professor Dr. Juan P. Garrahan Pediatric Hospital in Buenos Aires, Argentina, a 550-bed tertiary-care pediatric center with special assistance for oncohematology patients and solid-organ recipients. The validation of the scoring system was performed at the same center and in 7 additional pediatric institutions from different regions of Argentina: Pedro de Elizalde Hospital (Buenos Aires), Professor Dr. Alejandro Posadas National Hospital (Buenos Aires), Dr. Humberto Notti Pediatric Hospital (Mendoza), Victor J. Vilela Hospital (Rosario), Children's Hospital (Rosario), Juan Pablo II Pediatric Hospital (Corrientes), and Castro Rendon Hospital (Neuquen).
All patients who were hospitalized between March 2000 and September 2002 potentially were eligible for the study. The study was conducted according to Declaration of Helsinki guidelines.
The inclusion criteria were: 1) children aged <18 years with neutropenia after chemotherapy for a primary malignant disease, 2) an absolute neutrophil count of either <500/mm3 or between 500/mm3 and 1000/mm3 with a predicted decline <500/mm3, and 3) 1 episode of a fever >38.5°C or 2 records >38°C within 24 hours. Patients who underwent bone marrow transplantation were not included. Patients who were included for the validation phase were those who were hospitalized between March 2003 and July 2004.
Peripheral blood cultures, urine cultures, and chest x-rays were obtained from all patients at the onset of the study. Blood samples from port catheters and from peripheral veins for quantitative differential cultures also were taken from all patients. In addition, when skin and soft tissue infections, diarrhea, phyarangitis, or any other localized infection were suspected, cultures of specimens from the involved sites were obtained.
High-risk patients were defined by the presence of any of the following risk factors: 1) severe comorbidities, eg, incoercible bleeding, refractory hypoglycemia, hypocalcemia, hypotension, altered mental status, renal insufficiency (estimated glomerular filtration rate <50% of normal levels for the patient's age), hepatic dysfunction (serum alanine aminotransferase levels >4 times normal or bilirubin levels >3 mg% at the time of the preceding chemotherapy cycle); 2) respiratory failure; 3) poor clinical condition; 4) fascial, perineal, or catheter-associated cellulitis; 5) evidence of enteritis or severe mucositis; 6) uncontrolled local infection; 7) positive blood cultures within the first 24 hours; 8) advanced stage of underlying diseases, ie, bone marrow involvement, recurrence, second tumor, highly myelotoxic chemotherapy, or genetic disease; and 9) parents or caregivers deemed absolutely reliable by the medical staff.
After samples were collected, empiric antibiotic treatment was started. Therapy was based on our previous experience,15, 16, 22 as follows: ceftriaxone (100 mg/kg per day intravenously [iv] as a single dose; maximum dose, 2 g plus amikacin (15 mg/kg per day iv as a single dose) the first day, followed by ciprofloxacin (20 mg/kg per day orally every 12 hours) for low-risk children and ceftazidime (100 mg/kg per day iv every 8 hours) or imipenem (50 mg/kg per day iv every 6 hours) plus amikacin for high-risk patients. Patients who had been receiving prophylactic acyclovir continued to receive this agent. The use of hematopoietic growth factors did not hinder the enrollment of patients, and their administration was at the discretion of the attending oncologist for each patient.
Low-risk patients were discharged after 24 hours of hospitalization and were subjected to follow-up every 24 hours as outpatients by clinical examination and a differential leukocyte counts every 48 hours when persistent neutropenia was observed. High-risk patients remained hospitalized and were discharged only when their clinical state was satisfactory, they remained afebrile for 72 hours, and had an absolute neutrophil count >100/mm3.
Sepsis, septic shock, and organ dysfunction were defined as described previously.23 Renal impairment was defined as a glomerular filtration rate <50% of the normal rate for the patient's age, and hepatic impairment was defined by a serum alanine aminotransferase levels that was 4 times normal or by a bilirubin level that was >3 mg% at the time of the preceding chemotherapy cycle. Metabolic values were based on previous recommendations.24 Advanced stage of underlying disease implied bone marrow involvement, recurrence, second tumor, high-dose myelotoxic chemotherapy, or genetic disease.
Treatment success was defined as the resolution of the episode of fever and neutropenia without any readmission because of a new fever episode within 7 days of discharge or a new febrile episode during the same period of neutropenia. Death was the main variable in the assessment of final outcome.
The following variables were collected and registered for analysis: age, sex, underlying disease and staging, predicted period of neutropenia, presence of intravenous device (Hickman or Broviac catheters), previous administration of antibiotics or granulocyte colony-stimulating factor (G-CSF), and presence of a clinical site of infection. Facial, oral, anal, and catheter cellulitis as well as enteritis, sepsis, necrotizing gingivitis, and mucositis were considered high-risk clinical foci. Pneumonia was analyzed separately. Also analyzed separately were the results from microbiologic isolations and the presence of associated comorbidities (ie, incoercible bleeding, refractory hypoglycemia and hypocalcemia, hypotension, altered mental status, renal insufficiency, hepatic dysfunction, and respiratory failure).
Data analysis was performed using the Statistical Package for Social Sciences (SPSS) version 11.0. Descriptive data are expressed as percentages, medians, and ranges or as means and standard deviations. Associations between potential risk factors and mortality were analyzed through bivariate and multivariate methods. Statistical significance was calculated using chi-square or Fisher exact tests for nominal variables and Student t tests or Wilcoxon rank-sum tests for numerical variables. All independent variables that yielded P ≤ .10 in the bivariate analysis were included as covariables in the multivariate analysis (logistic regression, forward stepwise method; Pin = .05; Pout = .10). Risk is reported informed as the relative risk (RR) for bivariate analyses and the odds ratio (OR) for logistic regression analyses. Statistical significance is reported as P and 95% confidence interval (95% CI). Model and score performance is expressed as sensitivity, specificity, predictive values, and overall accuracy. All P values ≤.05 were regarded as significant.
In all, 714 episodes of FN in 458 patients who were admitted to the Professor Dr. Juan P. Garrahan Pediatric Hospital were registered prospectively for the derivation set. Fifty-three percent of patients were boys. The median age was 83.7 months (range, 1–215 months). Acute leukemia was the most frequent underlying disease. Twenty-five percent of patients had been treated with high-dose chemotherapy followed by long periods of neutropenia. The mean time to fever onset after the last chemotherapy course was 11.8 ± 42.3 days. Only 14% of children received G-CSF. Sixty percent of children had neutrophil counts <100/mm3 at registration (Table 1). Sixty-four percent of patients displayed an overt clinical site of infection at onset. Upper respiratory tract infections predominated. Thirteen percent of children showed some degree of mucositis. Bacteremia was detected in 97 patients (14%). Staphylococcus aureus, coagulase-negative staphylococci, and Escherichia coli were the most prevalent infectious organisms (Table 2). Eleven percent of patients had an associated, severe comorbidity.
Table 1. Demographic Characteristics of the Derivation and Validation Sets
| Median± SD||83.76± 59.5||85.82± 55.08|
|Boys||381 (53)||420 (52)|
|Underlying malignant disease|
| Acute leukemia and lymphomas||500 (68)||614 (76)|
| Solid tumors||125 (32)||192 (24)|
|Underlying malignant disease staging|
| High risk*||450 (63)||490 (61)|
| Low risk||265 (37)||316 (39)|
|Absolute neutrophil count at onset, /mm3|
| <100||427 (60)||533 (66)|
| 100–50||288 (40)||273 (34)|
|G-CSF therapy||98 (14)||239 (30)†|
|Central venous catheter||403 (56)||364 (45)†|
|Malnutrition||96 (13)||119 (15)|
|Associated comorbidity||81 (11)||70 (9)|
|Time to fever onset after the last chemotherapy course, mean±SD days||11.8± 42.3||12.6± 45.1|
Table 2. Clinical Sites of Infection and Blood Microbiologic Findings in the Derivation and Validation Sets
|Clinical site of infection||458 (64)||494 (61)|
|Type of infection|
| Upper respiratory tract infection*||225 (49)||136 (28)†|
| Celullitis||79 (17)||73 (15)†|
| Lower respiratory tract infection‡||67 (15)||113 (23)†|
| Gastroenteritis||47 (10)||74 (15)†|
| Other foci||40 (9)||98 (20)†|
|Mucositis||95 (13)||223 (27)†|
| Grade 1||45 (47)95 (43)||95 (43)|
| Grade 2||32 (34)93 (42)||93 (42)|
| Grade 3||10 (10)28 (12)||28 (12)|
| Grade 4||8 (9)||7 (3)†|
|Positive blood cultures||97 (14)§||135 (17)‖|
| Staphylococcus aureus||18 (18)||31 (22)|
| CNS||13 (14)||15 (10)|
| E. coli||12 (13)||8 (6)|
| Klebsiella spp.||12 (13)||12 (8)|
| Viridans streptococcus||9 (9)||19 (13)|
| Streptococcus pneumoniae||8 (8)||7 (5)|
| Pseudomonas aeruginosa||5 (5)||10 (7)|
| Candida spp.||8 (8)||12 (8)|
| Others||17 (17)||28 (20)|
Eighteen patients died (2.5%). Table 3 shows the bivariate analysis applied to the potential risk factors related to mortality. Advanced stage of underlying malignant disease (ie, bone marrow involvement), severe neutropenia (<100 neutrophils/mm3), the presence of an overt clinical site of infection, bacteremia, and associated signs of comorbidity proved to be significant risk factors of mortality (P < .05); whereas age, sex, time from the last chemotherapy course to fever onset, use of G-CSF, lower respiratory tract infection at onset, and mucositis were not associated with mortality (P > .05).
Table 3. Bivariate Analysis of Risk Factors of Mortality in 714 Episodes of Neutropenia and Fever in Children With Malignant Diseases in the Derivation Set
|Median age±SD, mo||67± 59.5||69.5± 63.2|| || ||.80|
|Median ± SD days from last chemotherapy course to onset of fever||10± 40||9± 43|| || ||.96|
|Advanced stage of underlying malignant disease|
| No||264||0|| || || |
| No||600||17|| || || |
| No||285||3|| || || |
|Clinical site of infection|
| No||612||12|| || || |
| No||631||17|| || || |
| No||620||15|| || || |
| No||605||6|| || || |
| No||629||4|| || || |
The multivariate analysis is shown in Table 4. Advanced stage of underlying malignant disease (OR, 3116; 95% CI, 0.0001–5.5102), associated comorbidity (OR, 25.5; 95% CI, 7.7–83.9), and bacteremia (OR, 7.2; 95% CI, 2.3–21.9) remained independent risk factors for mortality. This model had a specificity of 93.53% and a sensitivity of 77.78%.
Table 4. Multivariate Analysis of Risk Factors for Mortality in 714 Episodes of Neutropenia and Fever in Children With Malignant Diseases in the Derivation Set
|Advanced stage of underlying malignant disease||3116.3||0.0001–5.5102||.001|
|Associated comorbidity signs||25.552||7.774–83.980||.0001|
The scoring system to predict mortality was developed by assigning differential points to independent risk factors that were identified in the multivariate analysis as follows: the presence of an advanced stage of underlying malignant disease (ie, bone marrow involvement) was assigned 3 points, the presence of associated comorbidity was assigned 2 points, and bacteremia was assigned 1 point. The likelihood of death for patients who had totals 4, 5, and 6 points in the scoring system was 5.8%, 15.4%, and 40%, respectively. None of the patients who had a total of 3 points died (Table 5).
Table 5. Statistically Significant Variables Related to High Mortality Rates and Point Scores After the Analysis of 714 Episodes of Fever and Neutropenia in Children With Malignant Diseases in the Derivation Set
|4||65 (94.2)||4 (5.8)||69|
|5||33 (85.6)||6 (15.4)||39|
|6||12 (60)||8 (40)||20|
|Total||696 (97.5)||18 (2.5)||714|
Using a cut-off score of 4 points to discriminate between high-risk and low-risk patients, the scoring system had a sensitivity of 85.94%, a specificity of 100%, a positive predictive value of 100%, and a negative predictive value of 84.2%. Patients with scores ≥4 points had an RR for mortality of 6.33 (95% CI, 5.33–7.51).
In all, 806 episodes of neutropenia and fever in 523 patients were included. The number of FN episodes that were included in each center were as follows: Professor Dr. Juan P. Garrahan Pediatric Hospital, 461 episodes; Pedro de Elizalde Hospital, 101 episodes; Professor Dr. Alejandro Posadas National Hospital,79 episodes; Dr. Humberto Notti Pediatric Hospital, 56 episodes; Victor J. Vilela Hospital, 45 episodes; Juan Pablo II Pediatric Hospital, 40 episodes; Children's Hospital, 16 episodes; and Castro Rendon Hospital, 8 episodes. No significant differences were observed in the demographic, clinical, or microbiologic characteristics at the study onset or in the final outcome among all registered patients (Tables 1, 2). In the validation set, there was a greater proportion of G-CSF therapy, a greater proportion of central venous catheters, more patients had lower respiratory tract infections, there was more mucositis, and there was a lower incidence of upper respiratory tract infections compared with the derivation set (P < .05).
The validation statistical analysis confirmed that the differences in score categories between children who died because of an infection and those who survived was statistically significant (P < .0001). The RR of mortality for children who accumulated scores >4 points was 23.71 (95% CI, 7–80.33) (Table 6). The sensitivity of the score was 84.21%, the specificity was 83.23%, the positive predictive power was 89.19%, and the negative predictive power was 99.54%. Less than 0.5% of registered children who had scores <4 points were at risk of death.
Table 6. Statistically Significant Variables Related to High Mortality Rates and Point Scores After the Analysis of 806 Episodes of Fever and Neutropenia in Children With Malignant Diseases in the Validation Set
|4||90 (93.8)||6 (6.2)||96|
|5||24 (89)||3 (11)||27|
|6||18 (72)||7 (28)||25|
|Total||787 (97.7)||19 (2.3)||806|
Patients who receive therapy for malignant disease and FN constitute a wide-spectrum group. The risk of severe infection and eventual death varies according to different patient characteristics.2 Risk analysis offers a more rational approach for these patients.
The onset and duration of neutropenia may be predicted for each patient, and the risk of a febrile episode at its onset may be evaluated accurately within the context of the patient.2 Many high-risk variables have been analyzed and reported, such as bacteremia,5 severe bacterial infections,6 and mortality.2, 7 The reported mortality rate for patients with FN is <5%.4, 5, 7 This low incidence is similar to the results from our current study and makes it necessary to gather a large numbers of patients to obtain valid conclusions. Other studies have analyzed variables related to FN in low-risk patients to test ambulatory and oral therapies.8, 9, 13–17
Several factors predict the onset of severe bacterial infections, bacteremia, or infection-related mortality, such as underlying disease characteristics (bone marrow involvement, recurrence, second tumor, highly myelotoxic chemotherapy, genetic diseases, delayed bone marrow recovery), severe neutropenia, predicted neutropenia for >7 days, absolute monocyte count <100/mm3, fever >39°C, thrombocytopenia <50.000/mm3, fever onset <7 days after the last chemotherapy course, signs of sepsis, arterial hypotension, and associated comorbid factors.3–10, 22, 25, 26 A score to predict mortality is a useful tool to evaluate and manage patients with severe infections. A major advantage is the objectivity of criteria to evaluate patients. Some risk profiles have been tested, but very few have proposed a well-designed and statistically supported score to predict mortality at onset in children who are treated for malignant diseases. Published studies with analyses that establish a risk score for mortality are scarce and mainly are based on adult and pediatric patients without neutropenia in the setting of intensive care unit admissions, adults hospitalized in oncology units, and low mortality risk scores in an adult population.11, 13 To our knowledge, there are no publications to date concerning the analysis of mortality-based scores in hospitalized pediatric patients with FN.
The better evaluation of low-risk children with FN has allowed the development of new therapeutic strategies, like the sequential parenteral-oral approach,16 oral-only treatments,17 and ambulatory treatments.18–21, 25, 27, 28 Our current results support the use of these new strategies for the low-risk group. Conversely, the use of a mortality score for high-risk patients allows a better initial approach for such children (ie, precocious admission for upper respiratory tract infections, antibiotic therapy, and aggressive clinical support for selected patients). Thus, it is possible to predict mortality in these patients.
Klastersky et al. carried out a multicentric study in several European countries and analyzed the use of a score to identify low-risk adult cancer patients with FN.13 That score yielded a sensitivity of 71% and a specificity of 68%. We cannot compare those results with the results from pediatric patients, because the diseases and therapies vary widely between the groups, and the complications also differ. Adults usually present a higher rate of associated comorbidity than children. In the current study, we also demonstrated that the presence of associated comorbidity is a high risk factor for mortality. Another study reported by Santolaya et al.25 relied on a risk profile tested in Chile to predict severe bacterial infections in children. That risk profile had a higher sensitivity (90%) than ours (78%) but had lower specificity (65%) compared with what we obtained in our study (93%). Santolaya et al. also established that disease recurrence and arterial hypotension were predictive factors for severe bacterial infection. Likewise, we were able to demonstrate that the presence of an advanced stage of underlying malignant disease (including disease recurrence) was related significantly to a higher risk of mortality together with the presence of associated comorbidity.
Bacteremia has been defined by some authors as a risk factor for cancer patients with FN.5, 29 A previous study at our institution of 863 FN episodes in pediatric cancer patients corroborated those results.7 In our analysis that defined risk factors for our scoring system, bacteremia was associated significantly with mortality. In 2 earlier former studies in which risk factors in severe bacterial infections were analyzed in children with thrombocytopenia, fever onset <7 days after the last chemotherapy course, C-reactive protein levels >90 mg/dL, the need of resuscitation, severe neutropenia (<200 neutrophils/mm3), and fever >39°C, were associated significantly with a high mortality rate.6, 25 With the exception of C-reactive protein levels, the same variables were analyzed in our study and did not yield any statistical significance.
Our scoring system was capable of indicating that a child with FN had a high risk of death if 1 or more of the following factors were present at onset: advanced stage of malignant disease (bone marrow involvement, recurrence, second tumor, highly myelotoxic therapy, genetic disease, bone marrow involvement), severe associated comorbidity, and bacteremia. The assessment for bacteremia requires ≥24 hours after sampling to have blood culture results. The concomitant use of scores to define the risk of bacteremia in these children may enhance the value of the current mortality score further without requiring blood culture results to complete the assessment of all mortality risk factors at onset.
The variables that are assessed with this score are simple, easy to determine, and easy to apply to patients with NF. The outstanding validation results of the multicentric study have provided solid statistical support for the judicious clinical application of this scoring system.
We thank Graciela Demirdjian, MD, and Carlos Bantar, PhD, for statistical assistance.