SEARCH

SEARCH BY CITATION

Keywords:

  • neutropenia;
  • infection;
  • mortality;
  • chemotherapy;
  • propensity score

Abstract

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

BACKGROUND:

Febrile neutropenia (FN) is a serious and potentially life-threatening condition that may develop in patients with cancer who receive myelosuppressive chemotherapy. The risk of mortality from FN is not well characterized in current clinical practice.

METHODS:

Patients with cancer who were receiving chemotherapy in clinical practice were identified from a large US healthcare claims database, and mortality was confirmed using the National Death Index. Patients with FN had their propensity scores matched within tumor types of interest (non-Hodgkin lymphoma and breast, lung, colorectal, and ovarian cancers) to patients who did not experience FN. Study endpoints of overall mortality (anytime during follow-up), early mortality (during the first 12 months of the first chemotherapy course), and hospitalization were examined using univariate and multivariate techniques.

RESULTS:

Matched FN and control groups each included 5990 patients, and the average follow-up for both groups was 17.6 months. Crude incidence rates of early mortality were significantly higher for patients with FN compared with controls for all tumor types. Proportional hazards regression demonstrated a significant increase in the risk of overall and early mortality in patients with FN compared with controls (hazard ratio [HR], 1.15 [95% confidence interval, 1.02-1.29] and HR, 1.35 [95% confidence interval, 1.09-1.67], respectively).

CONCLUSIONS:

The adjusted risk of mortality in patients who experienced FN was at least 15% higher than in comparably matched patients without FN, supporting the inference that infectious complications because of neutropenia resulting from myelosuppressive chemotherapy are clinically important. Cancer 2010. © 2010 American Cancer Society.

Febrile neutropenia (FN) is a common but potentially serious complication in patients with cancer who receive myelosuppressive chemotherapy.1-3

FN and Early Mortality

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

The risk of FN varies directly with the severity and duration of neutropenia and reportedly occurs most frequently early during the course of chemotherapy.4-6 Major risk factors for the development of FN include older age, comorbid conditions, the type of cancer, and the type and number of myelosuppressive chemotherapy agents used.5-7 FN represents a life-threatening complication that requires urgent evaluation for the identification and treatment of documented or presumed infection.8 Infection in the neutropenic cancer patient often can be difficult to identify because of the lack of neutrophils and associated clinical symptoms and signs. The febrile response also can be blunted in some immunosuppressed cancer patients because of reduced circulating leukocytes producing interleukin 1 (IL-1), IL-6, and tumor necrosis factor alpha.3, 8-10 FN is associated with significant morbidity and mortality, with estimates of 6.8% to 20% mortality among patients who are hospitalized for FN-related complications and with higher rates observed in patients who have major comorbidities and documented sepsis or shock.2, 10-12

FN and Late Mortality

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

Despite successful management, FN may have an equally important impact on the overall chemotherapy treatment plan, resulting in dose reductions, treatment delays, chemotherapy discontinuation, or a switch to alternative, less toxic, or potentially less effective regimens.11, 13 Myelosuppression continues to represent the major cause of reduced chemotherapy dose intensity.13 In vivo animal studies have demonstrated that relatively small decreases in dose can cut cure rates by 50% or more.14 Several retrospective and prospective randomized controlled trials (RCTs) support the importance of chemotherapy dose intensity in achieving optimal clinical outcomes, including long-term disease-free and overall survival.15-21

The major strategies for decreasing the risk of FN and its complications include chemotherapy dose reductions and delays, the prophylactic use of a myeloid growth factor, and the selective prophylactic use of antimicrobial agents. A recent meta-analysis of 17 RCTs demonstrated a significant reduction in FN and in infection-related and early all-cause mortality with primary granulocyte colony-stimulating factor (G-CSF) prophylaxis.22

FN and Hospitalization

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

FN is treated as a serious and life-threatening condition for which the standard of care for most patients includes hospitalization for further evaluation and empiric administration of broad-spectrum antibiotics.8, 12 The mean length of stay for hospitalized patients with FN ranges from 3 days to 15 days, and the mean hospitalization cost per FN episode is $19,110, representing a significant burden on the healthcare system.2, 23 Increasingly, low-risk patients are being considered for outpatient management of their FN. Careful patient selection and appropriate supportive care are necessary to safely and effectively implement an outpatient FN management system.24

To our knowledge, no study using a large private insurer health plan database has examined the impact of FN on early and overall mortality. In this analysis, we explored the current impact of FN on early and overall mortality as well as hospitalization in a large managed health plan database.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

For this retrospective cohort study, we used data from a group of managed care payers that were linked to the National Death Index (NDI). Unique cancer patients with and without FN were identified to generate comparative estimates of the risk of hospitalization and mortality overall and by tumor type.

Data Source

Data were obtained for this analysis from a large, geographically diverse US medical claims database from a group of 12 managed care plans with more than 20 million unique patients (HealthCore Integrated Research Database, Wilmington, Del). Claims data included both inpatient and outpatient diagnoses (in International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] format) and procedures (in Current Procedural Terminology, Fourth Edition [CPT-4] and Healthcare Common Procedure Coding System formats) as well as both standard and mail-order prescription records from the National Drug Code directory. Dates of service for all claims, additional patient demographic variables (eg, age, sex, geographic region), and types of healthcare coverage were available.

Mortality information was examined by linking each patient to the NDI using at least 2 items of unique private patient identifying information (eg, full name, Social Security number). Mortality data were accessed in a manner that complied with federal and state laws and regulations, including those related to privacy and security of individually identifiable health information pursuant to the Health Insurance Portability and Accountability Act of 1996. Institutional Review Board (IRB) approval was obtained through a central IRB (Quorum Review IRB) to access the NDI.

Selection of Participants

Source population creation

We identified all patients aged ≥18 years who had at least 1 diagnosis for a malignant neoplasm (ICD-9-CM code range 140.0-208.9) on or before the first chemotherapy claim that occurred between January 1, 2001, and December 31, 2006. The index date was defined as the first claim date for administration of a chemotherapeutic agent preceded by 6 months without any chemotherapy claims.

Patients were excluded from the source population if they 1) had claims indicating admission to a skilled nursing inpatient facility or hospice on or before the index date, 2) had a claim for hematopoietic stem cell transplantation, 3) had 1 or more chemotherapy courses that could not be classified (ie, specific chemotherapy agent could not be determined), or 4) were not continuously eligible in the managed care plan during the 6-month period preceding the first chemotherapy claim date.

Study cohort creation

Patients who were identified with the following malignancies were eligible for inclusion: non-Hodgkin lymphoma (NHL) and cancers of the breast, lung, colorectum, and ovary. Patients were classified as having a neutropenic event (ie, FN) using the following principal study definition: either a primary or secondary diagnosis code for neutropenia (ICD-9-CM code 288.0) or a diagnosis code for infection (bacterial or fungal) during the first chemotherapy course. By using both the primary and secondary diagnoses for the receipt of care for an infection, a greater proportion of those truly diagnosed with FN were identified.2

We used propensity score matching to maximize comparability between the study groups. Within each tumor type of interest, we used logistic regression to compute the propensity to develop FN according to preindex data from all available variables that were predictive of FN (Table 1). Once each patient was assigned a propensity score, we matched FN patients with and without FN using nearest neighbor matching (1:1 with a caliper of 25% standard deviation25) of the propensity score without replacement. For analyses regarding the overall cohort, we pooled over tumor type, creating 1 group of FN patients and 1 group of non-FN patients.

Table 1. Patient Characteristics for the Overall Study Cohort of Major Tumor Types
CharacteristicNo. of Patients (%)P
Patients With FNPatients Without FN
  1. FN indicates febrile neutropenia; SD, standard deviation; HMO, health maintenance organization; PPO, preferred provider organization; POS, point of service; FFS, fee for service; NHL, non-Hodgkin lymphoma; GERD, gastroesophageal reflux disease; COPD, chronic obstructive pulmonary disease; G-CSF, granulocyte colony-stimulating factor.

Total cohort5990 (100)5990 (100) 
Mean±SD follow-up, mo17.63 ± 14.3517.59 ± 15.39
Preindex variables   
 Demographics   
  Mean±SD age, y51.37 ± 8.5951.38 ± 8.38.633
  Male995 (16.61)1003 (16.74).845
  Female4995 (83.39)4987 (83.26) 
 Region   
  East402 (6.71)400 (6.68).994
  South1727 (28.83)1737 (29) 
  Central904 (15.09)895 (14.94) 
  West2957 (49.37)2958 (49.38) 
 Plan type   
  HMO1526 (25.48)1538 (25.68).997
  PPO3847 (64.22)3830 (63.94) 
  POS306 (5.11)304 (5.08) 
  FFS86 (1.44)88 (1.47) 
  Other225 (3.76)230 (3.84) 
 Cancer type   
  Breast3689 (61.59)3689 (61.59)1.000
  Lung783 (13.07)783 (13.07) 
  NHL361 (6.03)361 (6.03) 
  Colorectal757 (12.64)757 (12.64) 
  Ovarian400 (6.68)400 (6.68) 
 Comorbidities   
  Kidney disease62 (1.04)67 (1.12).658
  Liver disease91 (1.52)96 (1.6).713
  Cardiovascular disease1625 (27.13)1620 (27.05).918
  Connective tissue disease36 (0.6)38 (0.63).816
  Obesity161 (2.69)159 (2.65).910
  Diabetes359 (5.99)361 (6.03).939
  Depression260 (4.34)279 (4.66).402
  GERD476 (7.95)481 (8.03).866
  HIV16 (0.27)16 (0.27)1.000
  Rheumatoid arthritis48 (0.8)51 (0.85).762
  Anemia772 (12.89)800 (13.36).449
  Blood disorder159 (2.65)168 (2.8).614
  COPD240 (4.01)251 (4.19).612
 Charlson comorbidity index ± SD5.15 (3.14)5.2 (3.15).589
 History of infection5724 (95.56)5733 (95.71).687
 History of radiation486 (8.11)473 (7.9).662
 History of surgery3828 (63.91)3844 (64.17).761
 History of hospitalization3342 (55.79)3395 (56.68).329
 Myelosuppressive regimen (>2 in index course)5066 (84.57)5072 (84.67).879
 Metastatic cancer, preindex2769 (46.23)2804 (46.81).5333
Study variables, ie, postindex   
 Metastatic cancer, postindex2217 (37.01)2021 (33.74)<.001
 G-CSF use during 1st cycle of index course2095 (34.97)1260 (21.04)<.001
 No. of courses   
  14621 (77.15)4319 (72.1)<.001
  2733 (12.24)714 (11.92) 
  3224 (3.74)295 (4.92) 
  4176 (2.94)212 (3.54) 
  589 (1.49)135 (2.25) 
  650 (0.83)103 (1.72) 
  725 (0.42)88 (1.47) 
  ≥872 (1.2)124 (2.07) 
Mean±SD no. of cycles per course8.11 ± 8.386.37 ± 6.52<.001

For the outcome of overall mortality, patients were followed until death, disenrollment from the health plan, or study end (December 31, 2006); thus, patients potentially were followed across 1 or more courses of therapy for this outcome. For the outcome of early mortality, patients were followed until death, the end of the first chemotherapy course (truncated at 12 months), or disenrollment from the health plan. For the outcome of hospitalization, patients were followed until inpatient admission, the end of the first chemotherapy course (truncated at 12 months), or disenrollment from the health plan.

Study Variables

We used the period of 6 months before the initiation of chemotherapy to identify preindex characteristics (age, sex, region of the United States), comorbidities, and selected risk factors (history of infection, history of hospitalization, history of surgery, and receipt of radiation therapy; CPT codes 77401-77421). Comorbidities incorporated the major risk factors for developing FN from clinical practice guidelines and included kidney disease, connective tissue disease, obesity, diabetes, cardiovascular disease, and the Charlson comorbidity index.26, 27 The use of G-CSF as primary prophylaxis, which we defined as administration of G-CSF within 5 days after the first day of the first cycle of chemotherapy in the first course, was examined dichotomously.

Chemotherapy

A chemotherapy course was defined as 1 or more repetitive cycles of treatment with the same chemotherapy. Chemotherapy type was defined dichotomously as highly myelosuppressive or not based on the number of myelosuppressive drugs received during a chemotherapy cycle/course (≤2 drugs vs >2 drugs) consistent with National Comprehensive Cancer Network guidelines.27 Drugs that were considered myelosuppressive were based on previously published work.28

Endpoints

The primary outcome of this study was overall mortality, which was defined as all-cause mortality during study follow-up. Early mortality was defined as all-cause mortality that occurred during the first chemotherapy course (truncated at 12 months). To identify deaths, patient records were matched to the death file contained in the NDI dataset. NDI information included cause of death (ICD-10) as provided by the state vital statistics offices. All-cause hospitalization after an FN event and during the first chemotherapy course (truncated at 12 months) was identified using administrative claims. A total of 39 patients were excluded from the hospitalization outcome analyses as these patients did not experience their FN event within the first 12 months of the first chemotherapy course.

Statistical Analyses

Descriptive analyses were conducted to assess the adequacy of the propensity score matching both within specific tumor types and for the combined overall cohort. Comparisons of these variables between patients with FN and their matched controls were made using chi-square tests for categorical variables and 2-sample Wilcoxon- Mann-Whitney tests for continuous variables.

To adjust for differences in the length of follow-up between groups, person-time in the denominator of incidence rates was computed separately for each outcome (hospitalization or mortality) in months from the index date to the first occurrence of the relevant outcome (hospitalization or mortality), disenrollment, or study end. Unadjusted incidence rates per 1000 person-months with 95% confidence intervals (CIs) for all study endpoints were calculated. Mortality incidence rates for a subgroup of FN patients who had inpatient admissions that met the definition for FN were also calculated.

Cox regression models were used to examine the hazard for hospitalization, overall mortality, and early mortality. A saturated model with all covariates was created from which variables of a priori interest (eg, age, sex, geographic region, tumor types) and variables that demonstrated significance at the α = .05 level were chosen for inclusion in the final multivariate analyses. The assumption of proportional hazards was tested based on Schoenfeld residuals. Violation of the proportional hazards assumption (P < .05) indicated that the covariate had a time-varying effect on outcomes. Therefore, such variables were specified as time-varying covariates in the Cox models.

RESULTS

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

The source population contained 27,979 patients aged ≥18 years who had at least 1 diagnosis for a malignant neoplasm on or before the first chemotherapy claim date. Forty-one percent of patients with tumor types of interest were classified with FN; 5990 FN patients and 5990 non-FN patients were selected by propensity score matching within tumor type for the study cohort (see Fig. 1). Postmatching comparison of the FN groups versus the control groups within each tumor type demonstrated good balance between the groups (Table 1). In univariate analysis, primary prophylaxis with G-CSF was received with greater frequency by the patients with FN (P < .001).

thumbnail image

Figure 1. This is a patient attrition diagram that represents the distribution of patients with and without febrile neutropenia (FN) overall and after tumor type propensity score matching. NHL indicates non-Hodgkin lymphoma.

Download figure to PowerPoint

Crude Incidence Measures

The incidence of overall mortality, early mortality, and hospitalization by FN occurrence is listed in Table 2. Overall mortality was only significantly higher in FN patients compared with controls among those with NHL (Table 2). Rates of overall mortality for all tumor types trended higher in the FN group than in controls, although the trend was not significant (Fig. 2). Early mortality incidence rates were significantly higher in patients with FN compared with controls (Table 2, Fig. 3). The highest early mortality rates were associated with lung cancer, and the lowest rates were associated with breast cancer (Table 2). In all strata, the incidence of all-cause hospitalization was significantly greater for patients with FN versus controls. Hospitalization rates were highest for patients with lung cancer and lowest for patients with breast cancer.

Table 2. Incidence of Hospitalization and Mortality (Overall, Early) for Patients With Febrile Neutropenia (FN), Patients Without FN, and FN-Related Hospitalized Patients
VariablePatients With FNPatients Without FNFN-Related Hospitalized Patients
No. of EventsIncidence per 1000 Person- Months (95% CI)No. of EventsIncidence per 1000 Person- Months (95% CI)No. of EventsIncidence per 1000 Person- Months (95% CI)
  1. CI indicates confidence interval; NHL, non-Hodgkin lymphoma.

Hospitalizationn=5951 n=5990  
 Breast95532.99 (30.96-35.15)52516.90 (15.52-18.41) 
 Lung38395.34 (86.25-105.38)16734.06 (29.26-39.63) 
 NHL12651.52 (43.27-61.35)6624.12 (18.95-30.70) 
 Colorectal29856.80 (50.70-63.63)15828.08 (24.03-32.82) 
 Ovarian16860.02 (51.60-69.82)9933.01 (27.11-40.19) 
 All tumor types1,93044.41 (42.47-46.44)101521.45 (20.17-22.81) 
Mortalityn=5990 n=5990 n=1223 
 Overall mortality      
  Breast1572.19 (1.88-2.57)1812.52 (2.18-2.92)474.84 (3.64-6.45)
  Lung28735.61 (31.72-39.98)23330.61 (26.92-34.81)14061.64 (52.23-72.74)
  NHL498.22 (6.21-10.87)284.29 (2.96-6.21)2820.25 (13.98-29.33)
  Colorectal1129.00 (7.48-10.83)988.01 (6.57-9.76)5518.45 (14.16-24.03)
  Ovarian415.40 (3.98-7.33)375.12 (3.71-7.06)168.52 (5.22-13.90)
  All tumor types6466.12 (5.66-6.61)5775.48 (5.05-5.94)28615.70 (13.98-17.63)
 Early mortality      
  Breast593.31 (2.56-4.27)422.69 (1.99-3.64)3211.18 (7.90-15.80)
  Lung11635.65 (29.72-42.77)5328.83 (22.03-37.74)7360.03 (47.72-75.51)
  NHL1911.47 (7.32-17.99)96.67 (3.47-12.82)1432.28 (19.12-54.50)
  Colorectal489.84 (7.41-13.05)236.34 (4.21-9.53)2720.45 (14.03-29.83)
  Ovarian166.69 (4.10-10.92)94.43 (2.31-8.53)1117.98 (9.96-32.46)
  All tumor types2588.59 (7.61-9.71)1365.56 (4.70-6.57)15724.36 (20.83-28.49)
thumbnail image

Figure 2. This Kaplan-Meier plot illustrates overall mortality for patients with and without febrile neutropenia (FN) based on propensity score matching.

Download figure to PowerPoint

thumbnail image

Figure 3. This Kaplan-Meier plot illustrates early mortality as defined in the text (see Materials and Methods) for patients with and without febrile neutropenia (FN) based on propensity score matching.

Download figure to PowerPoint

Multivariate Analysis

The occurrence of FN in the Cox regression analysis was a significant predictor of overall mortality when controlling for all other variables in the model and was associated with a 15% greater relative risk of dying (Table 3). Consistent with crude incidence, patients who had lung cancer had the greatest hazard for overall mortality (hazard ratio [HR], 5.26; 95% CI, 4.42-6.26) compared with patients who had breast cancer. Geographic region was associated with a significant difference: Patients in the East had the lowest hazard. Use of prophylactic G-CSF significantly decreased the hazard of overall mortality by 35%.

Table 3. Cox Regression Results
VariableHospitalization, n=11,941Overall Mortality, n=11,980Early Mortality, n=11,980
HR (95% CI)PHR (95% CI)PHR (95% CI)P
  • HR indicates hazard ratio; CI, confidence interval; FN, febrile neutropenia; NHL, non-Hodgkin lymphoma; HIV, human immunodeficiency virus; G-CSF, granulocyte colony stimulating factor.

  • a

    Indicates time-varying covariate.

Febrile neutropenia2.04 (1.89-2.20)<.0011.15 (1.02-1.29).0201.35 (1.09-1.67).006
Age. y1.00 (1.00-1.00).6581.02 (1.01-1.02)<.0011.03 (1.01-1.04)<.001
Women0.91 (0.81-1.02).0920.75 (0.65-0.86)<.0010.67 (0.53-0.86).002
Geographic region      
 EastReference Reference Reference 
 West1.15 (0.99-1.35).0724.85 (2.90-8.10)<.001
 Midwest1.15 (0.97-1.37).1174.46 (2.62-7.60)<.001
 South1.04 (0.88-1.22).6404.28 (2.55-7.19)<.001
Cancer type      
 BreastReferenceReferenceReference   
 Lung cancer1.63 (1.44-1.84)<.0015.26 (4.42-6.26)<.0012.72 (1.96-3.78)<.001
 NHL1.00 (0.85-1.18).9992.68 (2.03-3.53)<.0011.24 (0.77-1.99).368
 Colorectal cancer1.49 (1.32-1.69)<.0011.84 (1.50-2.25)<.0011.58 (1.10-2.29).014
 Ovarian cancer1.70 (1.49-1.94)<.0011.50 (1.16-1.93).0021.08 (0.69-1.69).737
HIV2.92 (1.49-5.73).002
Anemia1.43 (1.24-1.65)<.0011.09 (1.03-1.14).002
Radiation0.92 (0.88-0.96)a<.0011.49 (1.28-1.75)<.0011.52 (1.14-2.01).004
Prophylactic G-CSF0.65 (0.53-0.79)<.0010.55 (0.40-0.76)<.001
History of surgery, preindex0.81 (0.80-0.84)a<.0010.97 (0.96-0.98)a<.0010.25 (0.19-0.33)<.001
History of hospitalization, preindex1.18 (1.08-1.28)<.0011.00 (1.00-1.01)a.4061.68 (1.32-2.14)<.001
Metastatic cancer1.20 (1.17-1.22)a<.0011.81 (1.43-2.29)<.001
No. of chemotherapy courses1.00 (1.00-1.00)a.398
Average no. of cycles per course1.00 (1.00-1.00)a.542
No. of cycles in the index chemotherapy course0.98 (0.98-0.98)a<.001

The occurrence of FN was associated with a significant 35% increase in the hazard of early mortality in the Cox regression analysis (Table 3). Among tumor types, patients who had lung cancer had the greatest hazard of early mortality (HR, 2.72; 95% CI, 1.96-3.78) compared with patients who had breast cancer. Both a history of surgery and the use of prophylactic G-CSF significantly decreased the hazard of early mortality by >75% and by 45%, respectively.

Cox regression analysis on the occurrence of hospitalization in the first chemotherapy course revealed twice the hazard for patients with FN versus patients without FN (Table 3). Patients who had ovarian and lung cancers had the greatest hazard of hospitalization—a roughly 60% increase over that for patients who had breast cancer. Both a history of surgery and a history of radiation demonstrated a slightly decreased hazard of hospitalization by 19% and 8%, respectively.

DISCUSSION

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

The current results demonstrate an increased risk of overall mortality, early mortality, and hospitalization in patients who experience FN compared with patients who do not experience FN. A central consideration of this investigation was the possibility that FN may have both early and late effects on mortality in cancer patients matched on key predictive characteristics because of an increased risk of early FN-related deaths and late disease-related deaths potentially caused by suboptimal relative dose intensity of chemotherapy. The occurrence of FN in the initial chemotherapy course was associated with a statistically significant difference in both the unadjusted risk of early mortality and the adjusted hazard of early mortality based on multivariate analyses. Although overall mortality rates for the propensity-matched cohorts were not significantly different in the unadjusted comparison, the Cox regression models did result in a significant HR for FN on overall mortality after multivariate adjustment. Thus, the primary impact of FN appeared to be on early mortality, ie, within the first chemotherapy course and no later than 12 months after the start of chemotherapy. Because the average follow-up was 17.6 months for both groups, extended evaluation of the association between FN and mortality was limited. Further evaluation with longer average follow-up is warranted to fully assess the impact of FN on long-term outcomes. The results of the current study suggest that hospitalized patients with FN are more likely to experience poorer outcomes, as evidenced by the higher rate of overall mortality and particularly the higher rate of early mortality.

We conducted sensitivity analyses on the primary results given the limitations and assumptions described above (see Materials and Methods). We evaluated the range of patients that would be categorized as experiencing FN among patients who had the tumor types of interest (n = 17,349) using various claims definitions as stated below; the number of patients with FN ranged from 4.5% (narrow) to 64.7% (broad). The broader FN definition included any diagnosis with neutropenia (ICD-9 code 288.0), or a diagnosis of fever (780.6), or a diagnosis of infection or procedure code for infection treatment; the narrower definition of FN included a diagnosis with neutropenia (ICD-9 code 288.0) and diagnosis of either fever (780.6) or infection. In an exploratory analysis that allowed the occurrence of FN at any time during the entire study period, the unadjusted results demonstrated significant differences in favor of survival for various groups (Table 4). The adjusted HR for FN on early mortality was 1.54 (95% CI, 1.29-1.85), and the adjusted HR for FN on overall mortality was 1.53 (95% CI, 1.35-1.72).

Table 4. Incidence of Overall and Early Mortality for Patients With and Without Febrile Neutropenia
VariableIncidence per 1000 Person-Months (95% CI)
Patients With FNPatients Without FN
  1. CI indicates confidence interval; FN, febrile neutropenia.

Overall mortality  
 Breast2.9 (2.5-3.4)2.3 (1.9-2.8)
 Lung44.3 (39.6-49.6)29.6 (26.0-33.8)
 Non-Hodgkin lymphoma7.2 (5.3-9.7)3.3 (2.1-5.1)
 Colorectal8.4 (7.1-9.9)6.4 (5.3-7.8)
 Ovarian5.6 (3.9-8.2)4.4 (2.9-6.7)
 All tumor types7.9 (7.3-8.5)5.6 (5.1-6.1)
Early mortality  
 Breast1.3 (1.0-1.6)1.1 (0.9-1.5)
 Lung20.1 (17.0-23.8)11.4 (9.2-14.1)
 Non-Hodgkin lymphoma3.0 (1.8-4.8)1.5 (0.8-3.0)
 Colorectal3.4 (2.6-4.5)2.9 (2.2-3.9)
 Ovarian2.1 (1.1-3.9)1.6 (0.8-3.2)
 All tumor types3.4 (3.1-3.9)2.4 (2.1-2.8)

These findings support previous studies demonstrating that FN in patients who received myelosuppressive chemotherapy had an impact on early mortality.2 It has long been acknowledged that FN represents the major dose-limiting toxicity of myelosuppressive chemotherapy and often results in prompt hospitalization for evaluation and treatment with empiric, broad-spectrum antimicrobial therapy. A systematic review of RCTs demonstrated that primary prophylaxis with a myeloid growth factor may reduce the risk of infection-related and all-cause early mortality.22 Preclinical studies as well as prospective and retrospective clinical studies have suggested that delivered chemotherapy dose intensity often is associated with long-term disease-free survival and overall survival.15-21 In the current study, we did not observe a significant unadjusted difference in the impact of FN on overall mortality when defining FN as occurring within the first chemotherapy course. When this criterion was relaxed to allow FN to occur at any time during the study period, statistically significant differences were observed in favor of the non-FN group for both early mortality and overall mortality.

The limitations of the current study were similar to those of other observational studies using data from healthcare claims sources. The use of a large managed care administrative claims database provided real-world demographics, complete medical care information (inpatient, outpatient, and pharmacy data), and follow-up for a large cohort of patients. However, only limited information involving clinical management, medications administered during hospitalization, and laboratory results was available. The absence of detailed clinical records and dependence on ICD-9-CM coding to define FN was a recognized limitation that could have resulted in misclassification of some cases of FN. Previous studies have reported >80% sensitivity for identifying FN with these algorithms.29 Nevertheless, such potential bias likely would result, if anything, in reduced differences in outcomes between the FN and non-FN groups. Finally, data on cancer diagnosis were validated in the HealthCore Integrated Research Database, and mortality was validated through the NDI database. Detailed cancer staging was not accessible in this study and may have had have an unobserved impact on the comparability of FN patients and non-FN patients. However, adjustment for the presence of metastatic disease (stage IV) was possible and may have diminished the potential for confounding by stage. Despite propensity score matching, prophylactic G-CSF was received by 14% more patients with FN. This probably reflects clinicians' judgment to administer prophylaxis to patients who had an inherently greater risk of FN. Although G-CSF may have been used therapeutically in patients who developed severe FN in Cycle 1, data on such use were not available and, thus, were not addressed further in this study.

Overall, this retrospective cohort study is 1 of the largest such investigations reported to date on the impact of neutropenic complications on both early and overall mortality in “real-world” clinical practice settings. The results reported here are consistent with other observational and clinical trial results demonstrating a significant impact of FN on cancer patient mortality.

CONFLICT OF INTEREST DISCLOSURES

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES

Karen Smoyer Tomic and Jingbo Yu are employees of HealthCore Inc. and received salary for work on this research. Rich Barron is an employee of Amgen and holds stock in Amgen. Gary Lyman is the principal investigator on a grant for research from Amgen to Duke University. Shannon Michaels and Matthew Reynolds received support from Amgen.

REFERENCES

  1. Top of page
  2. Abstract
  3. FN and Early Mortality
  4. FN and Late Mortality
  5. FN and Hospitalization
  6. MATERIALS AND METHODS
  7. RESULTS
  8. DISCUSSION
  9. CONFLICT OF INTEREST DISCLOSURES
  10. REFERENCES