Mortality, morbidity, and cost associated with febrile neutropenia in adult cancer patients


  • Nicole M. Kuderer M.D.,

    1. James P. Wilmot Cancer Center and the Department of Medicine, University of Rochester, Rochester, New York
    Search for more papers by this author
    • Drs. Kuderer and Cosler have no industry funding. Drs. Dale and Lyman have served on the speaker's bureau for Amgen. Drs. Lyman and Crawford have received research grant support from Amgen.

  • David C. Dale M.D.,

    1. Department of Medicine, University of Washington, Seattle, Washington
    Search for more papers by this author
    • Drs. Kuderer and Cosler have no industry funding. Drs. Dale and Lyman have served on the speaker's bureau for Amgen. Drs. Lyman and Crawford have received research grant support from Amgen.

  • Jeffrey Crawford M.D.,

    1. Duke Comprehensive Cancer Center and the Department of Medicine, Duke University, Durham, North Carolina
    Search for more papers by this author
  • Leon E. Cosler M.D., Ph.D.,

    1. Albany College of Pharmacy, Union University, Albany, New York
    Search for more papers by this author
    • Drs. Kuderer and Cosler have no industry funding. Drs. Dale and Lyman have served on the speaker's bureau for Amgen. Drs. Lyman and Crawford have received research grant support from Amgen.

  • Gary H. Lyman M.D., M.P.H.

    Corresponding author
    1. James P. Wilmot Cancer Center and the Department of Medicine, University of Rochester, Rochester, New York
    • University of Rochester School of Medicine and Dentistry, 601 Elmwood Ave., Box 704, Rochester, NY 14642
    Search for more papers by this author
    • Fax: (585) 276-1885

  • Presented in part at the 38th Annual Meeting of the American Society of Clinical Oncology, May 18–21, 2002, Orlando, FL.

  • The information contained in this article was based on the Clinical Data Products Data Base maintained by the University HealthSystem Consortium.



Hospitalization for febrile neutropenia (FN) in cancer patients is associated with considerable morbidity, mortality, and cost. The study was undertaken to better define mortality, length of stay (LOS), cost, and risk factors associated with mortality and prolonged hospitalization in cancer patients with FN.


The longitudinal discharge database derived from 115 US medical centers was used to study all adult cancer patients hospitalized with FN between 1995 and 2000, comprising a total of 41,779 patients. Primary outcomes included mortality, LOS, and cost per episode.


Overall, in-hospital mortality was 9.5%. Patients without any major comorbidities had a 2.6% risk of mortality, whereas 1 major comorbidity was associated with a 10.3% and more than 1 major comorbidity with a ≥21.4% risk of mortality, respectively. Mean (median) length of stay was 11.5 (6) days, and the mean (median) cost was $19,110 ($8,376) per episode of FN. Patients hospitalized for ≥10 days (35% of all patients) accounted for 78% of overall cost. Independent major risk factors for inpatient mortality included invasive fungal infections, Gram-negative sepsis, pneumonia and other lung disease, cerebrovascular, renal, and liver disease. Main predictors for LOS ≥10 days included leukemia, invasive fungal infections, other types of infection, and several comorbid conditions.


Factors associated with increased mortality, LOS, and cost in hospitalized adult cancer patients with FN include patient characteristics, type of malignancy, comorbidities, and infectious complications. These factors may be useful in identifying patients at increased risk of serious medical complications and mortality for more aggressive supportive care measures. Cancer 2006. © 2006 American Cancer Society.

Myelosuppression including neutropenia and its complications remain among the major dose-limiting toxicities of systemic cancer chemotherapy.1 Febrile neutropenia (FN) is considered a medical emergency and generally prompts immediate hospitalization for evaluation and administration of empiric broad-spectrum antibiotics.2–5 In addition, FN frequently leads to chemotherapy dose reductions and treatment delays that may compromise long-term clinical outcomes in responsive and potentially curable malignancies.6–8

Hospital care accounts for approximately 40% to 50% of the total costs of cancer care.9 Although neutropenic complications, including FN, unquestionably add to the toll of cancer treatment, there has been little systematic study of the actual morbidity, mortality, and costs. Clinical trials routinely underestimate the risk of FN and its complications in cancer patients, because of the highly selected nature of such patients and given the underreporting of hematologic toxicities.10 By using a large longitudinal nationwide database of all hospitalizations to 115 US academic health centers, the study reported here was undertaken to better define the inpatient mortality, length of stay (LOS), and costs associated with hospitalization of adult cancer patients with FN. We also sought to further investigate the demographic and clinical factors associated with mortality and prolonged hospitalization. Such information may potentially help guide the clinician for the selection of cancer patients at higher risk of death and complications secondary to FN and in need of more aggressive supportive management and prevention strategies.


Patient Population

The study population consisted of all adult (≥18 years of age) cancer patients hospitalized with FN between 1995 and 2000 at the 115 member institutions of the longitudinal University HealthSystem Consortium (UHC) hospitalization database. Clinical data from all discharge abstract summaries and cost data from all payers were analyzed. To identify the patient population of interest, inclusion criteria were developed based on the ninth revision of the International Classification of Diseases–Clinical Modification (ICD-9-CM). All claims had to contain at least 1 diagnosis for malignant neoplasm (ICD-9-CM code range 140.00 through 208.9) and each record also had to contain the ICD-9-CM code for agranulocytosis (ICD-9-CM 288), used by institutions to identify patients with FN. On the basis of standard coding practices, patients admitted with FN often do not receive a principal diagnosis of FN. Many cancer patients hospitalized for FN receive a principal diagnosis of FN only if there is no clear source of infection, no serious complication or comorbid illness identified.11, 12 To avoid an analysis based on a potentially biased patient subgroup, patients with either principal or secondary FN were included. Therefore, we analyzed patients who develop FN as either an outpatient or an inpatient, in concordance with pivotal studies of FN.18–20 We created 3 distinct FN patient groups based on different principal diagnoses to investigate the potential differences in clinical outcomes between these 3 groups: patients with FN as principal diagnosis (Group 1), patients with infection as principal diagnosis and FN as secondary diagnosis (Group 2), and patients with all other principal diagnoses and FN as secondary diagnosis (Group 3). Patients having undergone bone marrow or peripheral blood stem cell transplantation were excluded from the analysis.

Data Source

The UHC longitudinal hospitalization database is a unique set of prespecified clinical and cost variables derived from all discharge summaries and charges generated between 1995 and 2000 in 115 US health institutions throughout the country (66 university hospitals and 39 community teaching hospitals or affiliated institutions). Hospitals from most US states are represented except for Mississippi, Oklahoma, North and South Dakota, Idaho, Wyoming, Nevada, New Hampshire, Maine, Alaska, and Hawaii. This database is a collaboration among academic and affiliated health institutions, providing clinical and cost data for research and clinical practice analysis purposes. The data collection is closely monitored and undergoes a strict quarterly quality control, also ensuring data completeness. Information on laboratory studies, radiologic studies, pathologic reports, medication utilization, and cause of death is not recorded in this database. No patient or institutional identifiers were provided to the investigators.

Cost Data

Costs were derived from charges reported on patient records by multiplying the department charge (cost center) by a cost-to-charge ratio derived from the Hospital Cost Report Information System (HCRIS). Total costs per patient were computed by summing individual cost center estimates with subsequent appropriate adjustment for wage differences.

Study Outcomes and Independent Variables

The primary outcomes for analysis were inpatient mortality, hospital LOS, including the proportion of patients hospitalized ≥10 days, and total hospital cost per episode of FN and per day of hospitalization. In addition, multivariate analysis of clinical factors predicting mortality and long length of stay was conducted. Independent variables evaluated include year of hospitalization, patient demographics (age, gender, race/ethnicity, payer type), institution and major cancer diagnoses (colorectal cancer, lung cancer, breast cancer, ovarian cancer, testicular and bladder cancer, malignant lymphoma, multiple myeloma, leukemia, and sarcoma). Patients with multiple cancer diagnoses were classified as “multiple cancer” category and all remaining malignant neoplasm diagnoses as “other cancer” category. Patient comorbidities and infectious complications were also examined based on up to 16 available diagnostic codes, including congestive heart failure, all other heart diseases, cerebrovascular disease, lung disease, liver disease, renal disease, diabetes mellitus, peripheral vascular disease, deep venous thrombosis, pulmonary embolism, anemia and transfusions, Gram-negative and Gram-positive bacteremia, other bacterial infections, invasive candida and invasive aspergillus infections, other fungal infections, all other infections, pneumonia, intravenous site infection, hypotension, and hypovolemia. For further risk stratification, patients were also categorized based on the number of major comorbidities. Major comorbidities were defined as any major organ dysfunction requiring diagnostic or therapeutic intervention.

Statistical Methods

Cost and resource utilization estimates for the study period were based on all hospitalizations associated with FN. Patient demographic characteristics, comorbid conditions, and infectious complications, along with multivariate analyses for mortality and prolonged hospitalization, were based on a single admission per patient. To avoid bias, a single admission was randomly selected for each patient with multiple hospitalizations. Group comparisons for categorical variables were based on chi-square or Fisher exact test, whereas comparisons for continuous variables were based on Student t-test or analysis of variance for normalized variables and Wilcoxon rank sum test for other continuous variables. Multivariate regression models were based on block entry of variables significant in univariate analysis. Global tests for model significance were based on the chi-square test. For dichotomous variables, an adjusted odds ratio and 95% confidence intervals are derived from the exponential function of the logistic regression coefficient. The R2 represents the proportion of the variability in the outcome accounted for by the variables in the model. Model performance of discrimination was evaluated by the c-statistic representing the area under the receiver operating characteristic curve.


Patient Population

A total of 55,276 hospitalizations with FN were reported in 41,779 adult cancer patients between 1995 and 2000 in the 115 nationwide US health centers. During the period of observation, 15% of patients reported 2 or more admissions with FN. The number of admissions per hospital over the 6 years of study averaged 456.8 (range, 10-1849). The average age of the patient population was 53.6 years (range, 18-99). Table 1 presents demographic characteristics and cancer types. Table 2 summarizes the number and percentage of major comorbidities and infectious complications. Major comorbid illnesses in addition to cancer and FN were reported in 48.8% of patients, with 19.1% reporting 2 or more major comorbidities.

Table 1. Patient Characteristics and Univariate Analysis for Mortality, Length of Stay, and Cost
VariableNo. (%)% Mortality (± 95% CI)Length of Stay, dTotal Cost per Hospitalization, $
Mean (± 95% CI)MedianMean (± 95% CI)Median
  1. CI: confidence interval.

All subjects41,779 (100.0)9.5 ± 0.311.5 ± 0.1619,110 ± 3058376
Age, y      
 <6529,880 (71.5)8.2 ± 0.311.7 ± 0.2620,867 ± 3808789
 ≥6511,899 (28.5)12.7 ± 0.611.0 ± 0.2616,869 ± 4777782
 Male19,438 (46.5)10.6 ± 0.412.1 ± 0.2721,222 ± 4719289
 Female22,341 (53.5)8.5 ± 0.411.0 ± 0.2618,436 ± 3957820
 White30,595 (73.2)9.1 ± 0.311.1 ± 0.1619,156 ± 3398190
 Black4837 (11.6)11.2 ± 0.912.1 ± 0.4718,920 ± 9108481
 Hispanic1831 (4.4)8.9 ± 1.313.1 ± 0.7721,538 ± 170110,254
 Other/unknown4516 (10.8)10.6 ± 0.913.3 ± 0.7723,955 ± 112810,256
Cancer type      
 All solid tumors22,142 (53.0)8.0 ± 0.38.1 ± 0.1513,354 ± 2706593
 Colorectal957 (2.3)8.8 ± 1.89.3 ± 0.7612,850 ± 18687309
 Lung3339 (8.0)13.4 ± 1.28.3 ± 0.3510,976 ± 5156144
 Breast3077 (7.4)3.6 ± 0.78.0 ± 0.4512,372 ± 7075203
 Ovary1431 (3.4)6.1 ± 1.28.1 ± 0.5510,875 ± 8795,053
 Testis/bladder541 (1.3)5.0 ± 1.88.4 ± 0.9511,500 ± 14695760
 Sarcoma1115 (2.7)3.9 ± 1.18.0 ± 0.6511,341 ± 9776116
 Lymphoma8871 (21.2)8.9 ± 0.610.7 ± 0.3618,437 ± 6268599
 Multiple myeloma2025 (4.9)8.2 ± 1.212.1 ± 0.5823,143 ± 117413,563
 Leukemia8703 (20.8)14.3 ± 0.719.7 ± 0.41538,583 ± 96825,242
 Other11,244 (26.9)7.8 ± 0.58.6 ± 0.2512,512 ± 3766,621
 Multiple cancers476 (1.1)18.9 ± 3.516.1 ± 1.7929,646 ± 389112,882
Table 2. Univariate Analysis for Mortality, Length of Stay, and Cost for Comorbidities, Type of Infection, and Principal Diagnosis
VariableNo. (%)% Mortality (± 95% CI)Length of Stay, dTotal Cost per Hospitalization, $
Mean (± 95% CI)MedianMean (± 95% CI)Median
  1. CI: confidence interval.

 Congestive heart failure2475 (5.9)23.9 ± 1.716.3 ± 0.71030,079 ± 163914,138
 Other heart diseases12,578 (30.1)15.9 ± 0.613.1 ± 0.3723,335 ± 62610,223
 Lung disease8236 (19.7)26.8 ± 1.016.7 ± 0.41132,932 ± 98416,305
 Liver disease1293 (3.1)28.2 ± 2.518.7 ± 1.11337,532 ± 284120,394
 Renal disease4398 (10.5)30.2 ± 1.418.5 ± 0.51338,669 ± 149822,004
 Diabetes mellitus3041 (7.3)11.7 ± 1.112.1 ± 0.5719,570 ± 10019192
 Cerebrovascular disease620 (1.5)30.2 ± 3.618.2 ± 1.41336,846 ± 380920,096
 Peripheral vascular disease396 (0.9)18.9 ± 3.918.2 ± 1.91138,956 ± 540116,542
 Deep venous thrombosis2,506 (6.0)10.8 ± 1.216.5 ± 0.71128,654 ± 168514,516
 Pulmonary embolism346 (0.8)24.6 ± 4.516.2 ± 1.81129,124 ± 441316,004
 Anemia13,908 (33.3)10.4 ± 0.513.5 ± 0.2823,244 ± 57110,598
 Transfusion requirement9036 (21.6)13.1 ± 0.715.6 ± 0.31026,876 ± 73214,903
 All infections18,325 (43.9)15.3 ± 0.515.1 ± 0.2927,189 ± 58612,617
 Gram-negative bacteremia3901 (9.3)33.9 ± 1.516.3 ± 0.61032,312 ± 14036631
 Gram-positive bacteremia2174 (5.2)21.2 ± 1.720.1 ± 0.81438,167 ± 209721,188
 Other bacterial infections3019 (7.2)8.0 ± 0.915.7 ± 0.6925,582 ± 121212,741
 Invasive candidiasis424 (0.8)36.7 ± 5.330.8 ± 3.02770,052 ± 806552,226
 Invasive aspergillosis639 (1.2)39.2 ± 4.827.3 ± 1.82463,229 ± 513652,260
 Other fungal infection4046 (9.7)12.0 ± 1.015.3 ± 0.5925,063 ± 106812,352
 Other infections1524 (3.6)13.2 ± 0.716.3 ± 0.41029,706 ± 87014,115
 Pneumonia4518 (10.8)26.5 ± 1.318.0 ± 0.51235,378 ± 148017,076
 Intravenous site infection2198 (5.3)10.1 ± 1.320.6 ± 0.81539,147 ± 215024,000
 Hypotension1867 (4.5)25.5 ± 2.013.2 ± 0.6825,097 ± 161812,688
 Hypovolemia6786 (16.2)14.4 ± 0.910.7 ± 0.3717,431 ± 6478512
No. of major comorbidities      
 021,386 (51.2)2.6 ± 0.29.2 ± 0.1514,344 ± 3126605
 112,398 (29.7)10.3 ± 0.512.3 ± 0.2720,545 ± 5269462
 25865 (14.0)21.4 ± 1.115.5 ± 0.4929,057 ± 108013,700
 31772 (4.2)38.6 ± 2.319.5 ± 1.01441,934 ± 262123,935
 4358 (0.9)50.6 ± 5.222.0 ± 1.91750,085 ± 536831,979
Principal diagnosis      
 Febrile neutropenia14,186 (34.0)2.8 ± 0.36.0 ± 0.148,892 ± 3345189
 Infection5031 (12.0)14.8 ± 1.09.5 ± 0.3714,763 ± 5518619
 All other diagnoses22,582 (54.0)12.3 ± 0.415.7 ± 0.21027,738 ± 51514,318

Inpatient Mortality

Over the 6 years of observation with 100% of mortality data availability, a total of 11% (4587) patients died after 1 or several hospitalizations with FN. Among 1 randomly selected admission for each patient, 9.5% (3967) led to death and among all FN admissions 8.3% had a fatal outcome, as some patients were admitted more than once for FN. The mortality rate for cancer patients with FN averaged 9.3% per hospital (range, 0% to 50%), with 35.5% of institutions reporting mortality rates of 10% or greater. Table 1 summarizes the risk of in-hospital mortality associated with different cancer types. The average mortality rates were 8.0%, 8.9%, and 14.3% for patients with solid tumors, lymphoma, and leukemia, respectively. Among the different solid tumor cancer types, breast cancer patients had the lowest mortality rate at 3.6% and lung cancer patients experienced the highest rate at 13.4%. Leukemia patients displayed the highest death rate among all cancer types. In univariate analysis, the highest mortality rates in cancer patients hospitalized with FN were observed in those with documented infections such as invasive aspergillosis (39.2%) and invasive candidiasis (36.7%), Gram-negative sepsis (33.9%), pneumonia (26.5%) or Gram-positive sepsis (21.2%), and those patients with major comorbidities (Table 2).

The number of major comorbid conditions is associated with a significant increase in the risk of mortality. Patients without any major comorbidities had on average a 2.6% risk of dying, whereas 1 major comorbidity was associated with a 10.3% and more than 1 with a ≥21.4% average risk of mortality (Table 2). Patients with documented infections experienced substantially higher mortality rates than patients without documented infections, independent of the number of comorbidities (Fig. 1). In addition, mortality was significantly associated with the patient's principal diagnosis (Table 2). Patients with FN as principal diagnosis had a mean mortality risk of 2.8%, whereas patients with any other principal diagnosis and concurrent FN as secondary diagnosis experienced an average risk of dying ≥12.3%.

Figure 1.

Impact of major comorbidities and infections on mortality (95% confidence intervals).

Cost and Length of Stay (LOS)

LOS data were available on all hospitalizations, whereas cost data were available on 95% of admissions. The mean and median LOS for all hospitalizations were 11.5 and 6 days, respectively. The mean (median) lengths of stay were 8.13 days (5 days) in patients with solid tumors, 10.7 days (6 days) in malignant lymphoma patients, and 19.7 days (15 days) in leukemia patients (Table 1). Other factors associated with prolonged hospitalization in univariate analysis included most types of infection and comorbid conditions (Table 2). Hospitalization for 10 days or longer occurred in 35% of cases. The 35% of hospitalizations for ≥10 days accounted for 74% of overall hospital days, 78% of overall hospital costs, and 65% of all inpatient deaths.

Overall, the mean and median costs of hospitalization were $19,110 and $8,376, respectively (Table 1). The total cost of hospitalization varied considerably among the various types of cancer as well as reported comorbidities and infections (Tables 1, 2). Estimated mean (median) hospital costs per day and costs per hospitalization adjusted for inflation to the year 2000 were $1,691 ($1,432) and $20,290 ($8,882), respectively, whereas the total estimated inflation-adjusted costs across all institutions for the 6 years of observation were $1.06 billion.

Change over Time

Whereas no change in inpatient mortality over time was observed between 1995 and 2000, the average LOS decreased by 10% (P-trend<0.0001), the average cost per FN episode increased by 12% (P-trend <0.0001), resulting in an increase in the average cost per day of 28% (P-trend<0.0001). After adjustment for inflation, the mean cost per FN episode did not change significantly over the 6-year period. Nevertheless, the average cost per day increased by 13% (P-trend <0.0001).

Multivariate Analysis

Table 3 presents the results of a multivariate logistic regression analysis for inpatient mortality. Adjusted odds ratios in this model were greatest for the following independent risk factors: Gram-negative sepsis 4.92, invasive aspergillosis 3.48, invasive candidiasis 2.55, lung disease 3.94, cerebrovascular disease 3.26, renal disease 3.16, hepatic disease 2.89, pneumonia 2.23, Gram-positive sepsis 2.29, hypotension 2.12, pulmonary embolism 1.94, patients with leukemia, and heart disease 1.58. This multivariate mortality risk model was found to have an R2 of 0.805 and a c-statistic of 0.857 (95% CI: 0.850-0.863).

Table 3. Multivariate Logistic Regression Analysis: Inpatient Mortality*
CategoryOdds Ratio (95% CI)P
  • *

    Adjusted for gender, year, and institution.

  • R2 is 0.805, c-statistic is 0.857 (95% confidence interval [CI]: 0.850–0.863; P <.0001).

 Gram-negative bacteremia4.92 (4.50–5.39)<.0001
 Gram-positive bacteremia2.29 (2.01–2.60)<.0001
 Invasive candidiasis2.55 (1.94–3.34)<.0001
 Invasive aspergillosis3.48 (2.70–4.48)<.0001
 Other fungal infection1.12 (1.00–1.26)0.051
 Hypotension2.12 (1.85–2.42)<.0001
 Hypovolemia1.52 (1.38–1.66)<.0001
 Pneumonia2.23 (2.04–2.45)<.0001
 Congestive heart failure1.27 (1.12–1.45)<.0001
 Other heart disease1.58 (1.45–1.72)<.0001
 Lung disease3.94 (3.64–4.26)<.0001
 Liver disease2.89 (2.48–3.37)<.0001
 Renal disease3.16 (2.89–3.46)<.0001
 Cerebrovascular disease3.26 (2.64–4.02)<.0001
 Pulmonary embolism1.94 (1.44–2.60)<.0001
Cancer type  
 Leukemia1.47 (1.34–1.61)<.0001
 Lung cancer1.18 (1.04–1.34)0.010
 Age ≥65 y1.12 (1.04–1.22)0.006

In a multivariate logistic regression analysis, adjusted odds ratios for prolonged length of stay (≥10 days) were greatest for invasive candidiasis 5.72, invasive aspergillosis 2.74, Gram-positive sepsis 2.45, and those with leukemia 3.33, and several comorbidities (Table 4).

Table 4. Multivariate Logistic Regression Analysis: Length of Stay Greater than 10 Days*
CategoryOdds Ratio (95% CI)P
  • *

    Adjusted for age, gender, ethnicity, year, and institution.

  • R2 is 0.332; c-statistic is 0.761 (95% confidence interval [CI]: 0.756-0.766; P <.0001).

 Gram-negative bacteremia1.49 (1.38–1.61)<.0001
 Gram-positive bacteremia2.45 (2.21–2.71)<.0001
 Other bacterial infection1.61 (1.50–1.74)<.0001
 Invasive candidiasis5.72 (4.07–8.04)<.0001
 Invasive aspergillosis2.74 (2.09–3.60)<.0001
 Other fungal infection1.74 (1.62–1.87)<.0001
 Pneumonia1.72 (1.60–1.85)<.0001
 Intravenous site infection2.00 (1.80–2.21)<.0001
 Congestive heart failure1.28 (1.16–1.42)<.0001
 Other heart disease1.14 (1.08–1.20)<.0001
 Lung disease2.03 (1.92–2.15)<.0001
 Liver disease2.02 (1.78–2.30)<.0001
 Renal disease2.15 (2.00–2.31)<.0001
 Cerebrovascular disease2.04 (1.70–2.45)<.0001
 Peripheral vascular disease1.65 (1.32–2.07)<.0001
 Pulmonary embolism1.74 (1.37–2.22)<.0001
 Deep venous thrombosis2.17 (1.98–2.38)<.0001
 Anemia1.30 (1.24–1.37)<.0001
 Transfusion requirement1.87 (1.77–1.98)<.0001
Cancer type  
 Leukemia3.33 (3.16–3.52)<.0001
 Multiple cancers1.92 (1.58–2.34)<.0001


Among all cancer patients hospitalized with FN between 1995 and 2000 at 115 US academic health centers and community teaching hospitals comprising over 41,000 patients, the overall inpatient mortality was 9.5%. In addition to the type of cancer, several comorbid conditions and infectious complications were significantly associated with increased mortality. The most important independent risk factors for inpatient mortality were invasive aspergillosis and invasive candidiasis, Gram-negative and Gram-positive sepsis, pneumonia, and several comorbid illnesses. Mean LOS was 8.1 days for solid tumor patients, 10.7 days for lymphoma, and 19.7 days for leukemia patients. Mortality, cost, and LOS increased in proportion to the number of comorbidities and complications present. Cost associated with hospitalization for FN closely correlated with length of stay.

The limitations of this study are inherent to studies based on administrative data. Despite the longitudinal data collection at the time of discharge, our analysis should be considered retrospective in design. Detailed clinical information is limited, including laboratory results, cause of death, timing of events, duration and severity of neutropenia, and the utilization of specific medications. Like previous FN studies based on administrative datasets, we are limited by the nonspecific nature of the ICD-9-DM diagnostic code of 288.0 for agranulocytosis.13, 14 Whereas the majority of patients hospitalized with this ICD-9-CM code have fever and neutropenia, a small number of patients may have neutropenia without fever. As neutropenia without fever is significantly less life-threatening and has much lower complication rates, this contributes to a more conservative estimate of the economic and clinical impact of FN. In fact, this study most likely provides a more accurate estimate of mortality and major complications than most clinical trials with their highly selected healthy patient populations. Given the inclusion of all consecutive neutropenic cancer patients hospitalized to a large number of US health institutions of all sizes nationwide, our patient population is more likely to reflect a general cancer population compared with previous trials. In addition, the large sample size of this study allowed us to clarify the role of major comorbid conditions as predictors of mortality and prolonged LOS, which has been difficult to accomplish in previous studies.

Risk estimates of neutropenic complications, including FN and mortality, associated with specific chemotherapy regimens have been based largely on the published results of randomized controlled trials. However, a recent systematic review of reported rates of hematologic toxicity and delivered dose intensity has demonstrated inconsistent reporting of these outcomes in the medical literature.10 Half of the published studies failed to report rates of toxicity, whereas almost half failed to report delivered chemotherapy dose intensity. In a recent Cochrane metaanalysis looking at the effects of therapeutic colony-stimulating factors in established FN, the combined control groups were found to have a mean inpatient mortality rate of 7.1%.15 Whereas the inpatient mortality rate of 9.5% reported in the current study appears higher than most clinicians associate with FN, it is consistent with mortality rates of 5% to 11% reported in smaller series of consecutive FN cancer patients, including a mix of solid tumor and hematologic malignancy patients.16–18 Several studies have demonstrated that hypotension, septic shock, and bacteremia in the setting of neutropenia are significant risk factors for increased mortality.19–24 The high death rates that we found in certain high-risk patients confirms these studies, which report mortality rates from 24% to 82%.19–25 Interestingly, the subgroup of patients with a principal diagnosis of FN corresponds closely to those patients without major comorbidities. This confirms the low-risk FN patient population artificially created by only including patients with FN as principal diagnosis.1 This practice inadvertently excludes many FN patients admitted with concomitant serious comorbidities or complications, which often relegate FN to a secondary diagnosis.11, 12

Attempts to develop clinical prediction rules to identify high- or low-risk patients presenting with FN have focused on those at low risk for developing serious medical complications, as potential candidates for outpatient therapy. When 1 early model was validated in 444 patients in 2 US hospitals, main predictors for mortality in multivariate analysis were found to be patients having developed FN as an inpatient, patients with serious comorbidities, patients with uncontrolled cancer, rapid onset of FN since initiation of chemotherapy, and age greater than 40. Unfortunately, the size of the study did not allow determining the role of the specific comorbidities predisposing to death.17 The Multinational Association of Supportive Care in Cancer has developed a model based on low-risk characteristics identified in 756 patients.18 Independent predictive factors of favorable outcome in an episode of FN were low burden of illness, solid tumor, no previous fungal infection in patients with hematologic malignancies, age <60 years, outpatient status, and absence of hypotension, of chronic pulmonary obstructive disease, and of dehydration. In our analysis, we confirmed most of the previously found independent risk factors, including hypotension, diagnosis of leukemia, fungal infections, elderly patients, and hypovolemia. Furthermore, we found additional comorbid illnesses and infectious complications that are independent predictors for mortality, including Gram-negative and Gram-positive bacteremia, invasive aspergillosis and invasive candidiasis, pneumonia, lung disease, renal disease, liver disease, cerebrovascular disease, congestive heart failure, other cardiac diseases, pulmonary embolism, and lung cancer. In addition, we confirm that patients without comorbid illnesses, complications, or documented infection have a low mortality risk.

Corresponding with our reported median LOS of 6 days, most of the randomized-controlled trials of hospitalized FN patients report a median LOS for FN of 6-8 days among their patients in the standard control arm.26–29 Although the major economic impact of neutropenic complications is the cost associated with hospitalization mainly determined by the LOS, there has been little systematic study of the actual costs involved. A report on hospitalization costs associated with 794 admissions with FN to a single institution during 1993 and 1994, based on detailed cost accounting and allocation, published a mean cost per episode of FN of $35,814 with a mean cost per day of $1,892,1 compared with $19,110 and $1,598 found in the current study of over 41,000 patients. Among solid tumor patients in the previous study, mean costs per episode and per day were $16,500 and $1,488, respectively, compared with $13,354 and $1,466 in the current analysis. Recently, a retrospective multiinstitution cost analysis for FN was published, with a mean cost per FN episode for all patients of $13,372 and among solid tumor patients with a range of $7,100 to $9,200. This analysis reported a significantly lower proportion of leukemia patients. Furthermore, they included cancer patients with infection, but without requiring a diagnosis of neutropenia in all patients, resulting in underestimations of cost.13 Previous studies reported substantially higher costs among leukemia patients.13

In addition, we found that the 35% of patients with prolonged hospitalization (>10 days) accounted for the majority of overall cost, overall days spent in the hospital, and deaths. This provides additional evidence that the financial relief of treating low-risk patients with short hospitalizations as outpatients is substantially less than the financial gain from preventing long and complicated hospitalizations.30

In conclusion, the results of this study confirm that administrative data can add valuable information to help clarify complex clinical questions. Furthermore, it can remove the population bias of studies confined to only a single or a few medical institutions. The mortality, LOS, and cost estimates presented in the current analysis strengthen and substantially extend those reported previously in smaller series. Despite improved medical management, FN continues to be associated with substantial morbidity, mortality, and cost, not only placing a significant burden on the individual patient but on the healthcare system as a whole. The multivariate model presented in this study clarifies and further elucidates additional risk factors, in particular, specific comorbid conditions, predicting death and prolonged LOS in patients hospitalized with FN. A large, prospective, multiinstitutional patient cohort would be needed to determine in more clinical detail the role of specific medical conditions and factors associated with mortality in FN. Until then, we anticipate that the validation of our mortality risk model in an independent population, which is currently under way, will aid clinicians in identifying high-risk patients for improved targeting of preventive and supportive care measures in an effort to further reduce the risk of both mortality and major complications associated with FN.