Risk and timing of hospitalization for febrile neutropenia in patients receiving CHOP, CHOP-R, or CNOP chemotherapy for intermediate-grade non-Hodgkin lymphoma

Abstract

BACKGROUND

Hospitalization for chemotherapy-induced febrile neutropenia is associated with substantial cost and may negatively impact clinical outcome due to associated dose attenuation.

METHODS

Medical records of 1355 patients with intermediate-grade non-Hodgkin lymphoma receiving cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP) or similar chemotherapy were reviewed. The potential risk factors associated with first hospitalization for febrile neutropenia were evaluated.

RESULTS

In the current study, 230 patients (17%) experienced 1 or more hospitalizations for febrile neutropenia and greater than one-half of all initial hospitalizations for febrile neutropenia occurred in Cycles 1 or 2. Increased risk of hospitalization for febrile neutropenia, based on Cox proportional hazards models, was significantly associated with the following characteristics: age 65 years or older (hazard ratio [HR] = 1.79; 95% confidence interval [95% CI], 1.35–2.37), serum albumin level at presentation less than or equal to 3.5 g/dL (HR = 1.34; 95% CI, 1.01–1.78), planned average relative dose intensity greater than or equal to 80% (HR = 2.70; 95% CI, 1.47–4.98), baseline absolute neutrophil count less than 1500/mm³ (HR = 1.98; 95% CI, 1.28–3.06), and the presence of hepatic disease (HR = 2.18; 95% CI, 1.11–4.28). Lack of early granulocyte colony-stimulating factor in Cycles 1 and 2 was also associated with increased risk of hospitalization for febrile neutropenia, but this did not reach statistical significance. A composite risk score based on these potential risk factors effectively distinguished patients at greater risk of hospitalization for febrile neutropenia (P < 0.001), the majority of which were observed during the first cycle of chemotherapy.

CONCLUSIONS

The data from the current study demonstrated that the risk of initial hospitalization for febrile neutropenia occured early in the course of CHOP-like chemotherapy. Identified risk factors for febrile neutropenia hospitalization may facilitate the use of targeted supportive care. Cancer 2003. © 2003 American Cancer Society.

Combination therapy with cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP) is considered to be the standard treatment for patients with intermediate-grade non-Hodgkin lymphoma (NHL).1 Variations of this chemotherapy regimen include the addition of rituximab (CHOP plus rituximab)² or the substitution of mitoxantrone for doxorubicin (CNOP).3 Although use of these regimens may improve tumor response rates, the prognosis for patients may be compromised by myelosuppression and its complications, including febrile neutropenia. Febrile neutropenia generally requires hospitalization, often resulting in reduced chemotherapy dose intensity due to dose delays or reductions.

Granulocyte colony-stimulating factor (G-CSF) has been shown to reduce the incidence of chemotherapy-induced febrile neutropenia in various malignancies and chemotherapy regimens.4–6 However, the identification of risk factors associated with neutropenia is important to ensure that the patients at greatest risk are considered for growth factor prophylaxis. Previous efforts to identify risk factors for severe and febrile neutropenia in patients with NHL being treated with CHOP and similar regimens have been limited by small sample size, missing data, and variable definitions of outcome. In addition, to our knowledge, none of these models have been validated.7–11 Furthermore, previous risk modeling efforts have provided limited information regarding the risk and timing of the first hospitalization for febrile neutropenia.

The current study was undertaken to identify potential risk factors associated with time to first hospitalization for febrile neutropenia in a large population of patients with NHL who were receiving CHOP or similar chemotherapy regimens. The results should aid clinicians in assessing the risk and timing of febrile neutropenia in these patients and in guiding the appropriate targeted use of the hematopoietic growth factors.

MATERIALS AND METHODS

Study Design and Patient Selection

A retrospective survey of community oncology practices was performed using a large, longitudinal database of patients receiving systemic chemotherapy in the U.S. The initiative, named Project ChemoInsight, was sponsored by Amgen, Inc. (Thousand Oaks, CA) and included healthcare professionals at medical practices throughout the U.S. Participating practices represented a sample of mostly community oncology practice physicians who each identified approximately 10 patients treated for NHL between March 1999 and March 2002. Data were abstracted from medical records onto case report forms by oncology nurses.

The database includes data on 5659 patients with NHL treated at 567 medical oncology practices. Effective January 2001, a revised data collection form was implemented that augmented the data with the date of onset of hospitalization for febrile neutropenia. The current study was based on this subset of the data, which included 1450 patients with NHL from 147 sites.

Patients were eligible for the study if they were at least 16 years old, had intermediate-grade NHL (National Cancer Institute [NCI] working formulation D–H),12 and were receiving chemotherapy with CHOP, CHOP plus rituximab, or CNOP. Ninety-five patients were excluded from the analysis because they did not have intermediate-grade NHL treated with a CHOP, CHOP plus rituximab, or CNOP regimen. Therefore, our analyses were performed on data from the remaining 1355 patients across 147 sites.

Standard Chemotherapy Regimens

The standard dose of CHOP chemotherapy was comprised of cyclophosphamide (750 mg/m²), doxorubicin (50 mg/m²), and vincristine (1.4 mg/m²) on Day 1, and prednisone (100 mg) on Days 1–5. The standard dosing for CHOP plus rituximab chemotherapy was rituximab (375 mg/m²) on Day 1, followed by cyclophosphamide (750 mg/m²), doxorubicin (50 mg/m²), and vincristine (1.4 mg/m²) on Day 3, and prednisone (100 mg) on Days 3–7. CNOP standard dosing consisted of cyclophosphamide (750 mg/m²), mitoxantrone (10 mg/m²), and vincristine (1.4 mg/m²) on Day 1, and prednisone (50 mg/m²) on Days 1–5. Regimens were repeated every 21 days for six cycles.

Independent Variables and Operational Definitions

Pretreatment patient characteristics abstracted from medical records included the chemotherapy regimen, patient age, patient gender, body surface area (BSA), comorbid conditions (e.g., congestive heart failure, myocardial infarction, angina, hepatic disease, renal disease, diabetes, and pulmonary disease), disease stage, number of extranodal sites, lymphoma histology, serum lactate dehydrogenase (LDH) level, serum albumin level, bone marrow involvement, and absolute neutrophil concentration (ANC) at the time of presentation. Age was dichotomized as younger than 65 years or 65 years or older and BSA was categorized as less than 2.0 m² or greater than or equal to 2.0 m². In various analyses, comorbid conditions were classified as the presence or absence of any of the conditions specified above (i.e., the presence or absence of comorbidity overall) and as the presence or absence of each comorbidity. In addition, the presence or absence of any cardiac comorbidity (e.g., congestive heart failure, myocardial infarction, or angina) was considered. Disease stage was categorized as limited stage (Ann Arbor Stage I or II) or advanced stage (Ann Arbor Stage III or IV).13 The number of extranodal sites was classified as 0 to 1 or greater than or equal to 2. Lymphoma histology was categorized according to the NCI working formulation commonly utilized at the time patients were diagnosed and treated in the current study.12 The baseline serum LDH level was categorized as either greater than 300 U/L or less than or equal to 300 U/L and the baseline serum albumin level was classified as either less than or equal to 3.5 g/dL or greater than 3.5 g/dL. Bone marrow involvement was categorized as either present or absent and ANC at the time of presentation was dichotomized as either less than 1500/mm³ or greater than or equal to 1500/mm³. Based on previous research, the following characteristics were predicted to be associated with a higher risk of hospitalization for febrile neutropenia: age 65 years or older, female gender, BSA less than 2.0 m², advanced-stage disease, LDH level greater than 300 U/L, albumin level less than or equal to 3.5 g/dL, presence of bone marrow involvement, and ANC less than 1500/mm³.7, 8, 10

Treatment Characteristics and Outcomes

To determine how closely the patient's planned dose of chemotherapy compared with the standard dose regimen, the agent-specific relative dose intensity (RDI) was calculated as a ratio of the planned first-cycle dose intensity divided by the corresponding standard dose in mg/m² per week. If the patient's chart did not contain an explicit reference to planned amounts, the actual dose administered for Cycle 1 was used to infer the planned dose. The planned average relative dose intensity (ARDI)14 for each patient was obtained by averaging the RDIs of cyclophosphamide and doxorubicin for the CHOP and CHOP plus rituximab therapies and by averaging the RDIs of cyclophosphamide and mitoxantrone for CNOP therapy. ARDI was dichotomized as greater than or equal to 80% or less than 80%. The 80% value was chosen based on data from Epelbaum et al.,15 which showed a correlation between improved response and greater than 80% ARDI of cyclophosphamide and doxorubicin chemotherapy. In addition, previous risk modeling studies have suggested an association between greater than 80% ARDI and increased risk of a febrile neutropenic event.8, 9

Hospitalization for febrile neutropenia was determined from the patient's medical record. Time to first hospitalization for febrile neutropenia was defined as the number of days from the start of Cycle 1 chemotherapy to the day that the first hospitalization for febrile neutropenia occurred. Early G-CSF use was defined as at least 7 days of G-CSF therapy within the first 5 days of both chemotherapy Cycles 1 and 2.

Statistical Methods

In the exploratory, post-hoc analysis, the proportion of patients hospitalized for febrile neutropenia was evaluated for subgroups of each independent variable using the chi-square test. All P values should be considered a descriptive statistic. A correlational analysis of all independent variables was undertaken to assess multicollinearity. The risk of first hospitalization for febrile neutropenia was estimated using the Kaplan–Meier method and is presented both as the cumulative risk of occurrence and as the hazard rate. Hazard ratio (HR) estimates with 95% confidence intervals (95% CI) were generated for each covariate using the Cox proportional hazards regression method.16

Missing data were evaluated for each variable considered in the models. Less than 10% were found to be missing for all variables except LDH. To assess the joint impact of the potential predictors on each dependent variable, stepwise logistic regression analysis was employed using patients with complete data for all variables (815 patients from 134 sites). The stepping algorithm employed was the stepwise method of SAS Proc Logistic (SAS Inc., Cary, NC). The method began with a model containing only an intercept parameter. In the first step, the predictor with the largest associated chi-square value was entered, conditional on its P value being less than 0.05. This process continued until no predictors achieved a significance level of 0.05. After each step in which a variable was entered into the model, the model was recalculated with all variables entered up to that point. Had the P value for any effect in the model exceeded 0.05, it would be dropped and the model recalculated without that term. Iteration stopped when no more variables could be entered into (P < 0.05) or removed from (P ≥ 0.05) the model. This stepwise analysis identified a subset of six potential predictors that met the criteria specified earlier. A direct logistic regression analysis was performed, retaining only the significant predictors from the previous analysis and using patients with complete data for these six predictors (1186 patients from 142 sites).

A risk score was generated based on the number of high-risk characteristics included in the fixed model. Patients with zero to two risk factors were considered to have low risk, patients with three risk factors were considered to have intermediate risk, and patients with more than three risk factors were considered to have high risk.

RESULTS

Study Population

We studied 1355 patients who received CHOP, CHOP plus rituximab, or CNOP for intermediate-grade NHL in 2001 or 2002. The median patient age was 63 years (range, 16–96 years). Fifty-nine percent of the patients (804 of 1355 patients) were diagnosed with diffuse large-cell histology (NCI working formulation G) and 82% (1117 of 1355 patients) received CHOP chemotherapy.

Hospitalization for Febrile Neutropenia

Table 1 shows that 230 patients (17%) experienced one or more hospitalizations for febrile neutropenia. Patients who were hospitalized for febrile neutropenia were significantly more likely to be age 65 years or older, female, have a BSA less than 2.0 m², at least 1 comorbid condition, a baseline serum albumin level less than or equal to 3.5 g/dL, an ANC less than 1500/mm³ at the time of presentation, and a planned ARDI of greater than or equal to 80%. No significant correlation was observed between hospitalization for febrile neutropenia and tumor histology, type of chemotherapy regimen, disease stage, or number of extranodal sites. Similarly, no significant association was observed between hospitalization for febrile neutropenia and the presence of specific comorbidities including the three cardiac comorbidities. Greater than half (59%) of all initial hospitalizations for febrile neutropenia occurred during the first two chemotherapy cycles (Table 2). Among patients who were hospitalized for febrile neutropenia, the median time from the initiation of chemotherapy to first febrile neutropenic hospitalization was 34 days, approximately the expected ANC nadir for Cycle 2.

Table 1. Characteristics of 1355 Patients with Intermediate-Grade Non-Hodgkin Lymphoma Who Received Chemotherapy by Hospitalization for Febrile Neutropeniaa

Patient characteristics (n = 1355)	Patients with FNH (n = 230) No. (%)	Patients without FNH (n = 1125) No. (%)	P valueb
FNH: hospitalization for febrile neutropenia; BSA: body surface area; NOS: not otherwise specified; LDH: lactate dehydrogenase; ANC: absolute neutrophil count; ARDI: average relative dose intensity; G-CSF: granulocyte–colony-stimulating factor. a Patients received cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP); CHOP plus rituximab (CHOP-R); and cyclophosphamide, mitoxantrone, vincristine, and prednisone (CNOP). b Likelihood ratio chi-square value. c Percentages corrected when characteristic contained missing data. d Defined as GCSF factor use for at least 7 days within the first 5 days of Cycles 1 and 2. e Statistically significant.
Regimen (n = 1355)			0.856
CHOP (n = 1117)	187 (81)	930 (83)
CHOP-R (n = 123)	23 (10)	100 (9)
CNOP (n = 115)	20 (9)	95 (8)
Age (yrs) (n = 1355)
≥ 65 (n = 619)	132 (57)	487 (43)	< 0.001e
Gender (n = 1355)
Female (n = 696)	137 (60)	559 (50)	0.006e
BSA (n = 1355)
< 2.0 m^2 (n = 865)	164 (71)	701 (62)	0.009e
Comorbid condition (n = 1355)
Any present (n = 346)	74 (32)	272 (24)	0.013e
Cardiac comorbidity present (n = 113)	24 (10)	89 (8)	0.220
Stage of disease (n = 1201)c
Advanced-stage disease (n = 607)c	115 (55)	492 (50)	0.132
Extranodal sites (n = 1224)c
≥ 2 (n = 238)c	48 (23)	190 (19)	0.177
Histology (n = 1355)			0.619
D: follicular large cell (n = 155)	30 (13)	125 (11)
E: diffuse small cleaved cell (n = 93)	16 (7)	77 (7)
F: diffuse mixed small and large cell (n = 135)	24 (10)	111 (10)
G: diffuse large cell (n = 804)	128 (56)	676 (60)
H: large cell immunoblastic (n = 77)	18 (8)	59 (5)
Intermediate-grade NOS (n = 91)	14 (6)	77 (7)
LDH level (n = 1067)c
> 300 U/L (n = 362)c	59 (35)	303 (34)	0.862
Albumin level (n = 1191)c
≤ 3.5 g/dL (n = 394)c	84 (41)	310 (31)	0.008e
Bone marrow (n = 1355)
Involvement (n = 211)	37 (16)	174 (16)	0.814
Baseline ANC (n = 1347)c
< 1500/mm^3 (n = 98)c	28 (12)	70 (6)	0.003e
Planned ARDI (n = 1355)
≥ 80% (n = 1209)	214 (93)	995 (88)	0.031e
Early G-CSF use (n = 1355)d
None (n = 1246)	217 (94)	1029 (92)	0.127

Table 2. Distribution of the Initial Hospitalization for Febrile Neutropenia by Cycle and Regimen

Regimen	Cycle 1 No. (%)	Cycle 2 No. (%)	Cycle 3 No. (%)	Cycle 4 No. (%)	Cycle 5 No. (%)	Cycle 6+ No. (%)	Total No. (%)
CHOP: cyclophosphamide, doxorubicin, vincristine, and prednisone; CHOP-R: CHOP plus rituximab; CNOP: cyclophosphamide, mitoxantrone, vincristine, and prednisone.
CHOP	66 (35)	43 (23)	23 (12)	22 (12)	14 (8)	19 (10)	187 (100)
CHOP-R	6 (26)	8 (35)	1 (4)	3 (13)	3 (13)	2 (9)	23 (100)
CNOP	10 (50)	2 (10)	2 (10)	2 (10)	4 (20)	0 (0)	20 (100)
Total	82 (36)	53 (23)	26 (11)	27 (12)	21 (9)	21 (9)	230 (100)

Potential Risk Factors Associated with Initial Hospitalization for Febrile Neutropenia

An increased risk of first hospitalization for febrile neutropenia was associated with age 65 years or older (P < 0.001) (Fig. 1), baseline serum albumin level less than or equal to 3.5 g/dL (P = 0.010) (Fig. 2), ANC less than 1500/mm³ at the time of presentation (P = 0.001) (Fig. 3), and the presence of hepatic disease (P = 0.047) (data not shown). A decreased risk of hospitalization for febrile neutropenia was associated with planned ARDI less than 80% (P = 0.020) (Fig. 4).

Figure 1. Cumulative probability of initial hospitalization for febrile neutropenia by age. The risk of first hospitalization for febrile neutropenia was higher among patients age 65 years or older (P < 0.001). o: censored observation.

Figure 2. Cumulative probability of initial hospitalization for febrile neutropenia by serum albumin level. The risk of first hospitalization for febrile neutropenia was higher among patients with serum albumin levels less than or equal to 3.5 g/dL (P = 0.010). o: censored observation.

Figure 3. Cumulative probability of initial hospitalization for febrile neutropenia by baseline absolute neutrophil count (ANC). The risk of first hospitalization for febrile neutropenia was higher among patients with an ANC less than 1500/mm³ at the time of presentation (P = 0.001). o: censored observation.

Figure 4. Cumulative probability of initial hospitalization for febrile neutropenia by the planned average relative dose intensity (ARDI). The risk of first hospitalization for febrile neutropenia was lower among patients with planned ARDI less than 80% (P = 0.020). o: censored observation.

Table 3 presents the results of a multivariate proportional hazards regression analysis of risk of first hospitalization for febrile neutropenia for both stepwise and fixed models. No significant interactions were observed for the major covariates. In the fixed model, the risk of first hospitalization for febrile neutropenia was significantly greater in the multivariate analysis for patients with the following characteristics: age 65 years or older, serum albumin level at the time of presentation less than or equal to 3.5 g/dL, planned ARDI greater than or equal to 80%, baseline ANC level less than 1500/mm³, and the presence of hepatic disease. The area under the curve in receiver operating characteristic analysis as a measure of predictive model performance (c statistic) was 0.65 (95% CI, 0.61–0.69; P < 0.0001).

Table 3. Impact of Patient Characteristics on the Hazard Ratio of Risk of Initial Hospitalization for Febrile Neutropenia during Chemotherapy

Patient characteristics	Stepwise model (n = 815)a			Fixed model (n = 1186)b
	Hazard ratio	95% CI	P valuec	Hazard ratio	95% CI	P valuec
95% CI: confidence interval; ANC: absolute neutrophil count; ARDI: average relative dose intensity; G-CSF: granulocyte–colony-stimulating factor. a Includes patients with complete data for all variables. b Includes patients with complete data for specified variables. c Cox proportional hazards regression. d Statistically significant. e Defined as G-CSF use for 7 days within the first 5 days of Cycles 1 and 2.
Age ≥ 65 yrs	1.70	1.19–2.42	0.004d	1.79	1.35–2.37	< 0.001d
Albumin level ≤ 3.5 g/dL	1.62	1.14–2.30	0.008d	1.34	1.01–1.78	0.041d
Baseline ANC < 1500/mm^3	2.56	1.55–4.23	< 0.001d	1.98	1.28–3.06	0.002d
Planned ARDI ≥ 80%	3.02	1.23–7.43	0.016d	2.70	1.47–4.98	0.001d
No early G-CSF usee	2.05	1.00–4.21	< 0.050d	1.64	0.93–2.88	0.086
Hepatic comorbidity	2.83	1.24–6.49	0.014d	2.18	1.11–4.28	0.024d

The cumulative risk and hazard rate of initial hospitalization for febrile neutropenia are shown by risk group in Figures 5 and 6. The risk of first hospitalization for febrile neutropenia differed significantly between these risk groups (P < 0.001). Again, the greater risk observed in intermediate-risk and high-risk patients was primarily the result of greater risk during the first two cycles of chemotherapy. High-risk patients exhibited a peak hazard rate during the first cycle that was than threefold greater than intermediate-risk patients and nearly five times greater than low-risk individuals.

Figure 5. Cumulative probability of initial hospitalization for febrile neutropenia by risk composite score. The risk of first hospitalization for febrile neutropenia differed significantly among the three risk groups (P < 0.001). o: censored observation.

Figure 6. Hazard of initial hospitalization for febrile neutropenia by risk composite score. High-risk patients exhibited a peak hazard rate during the first cycle that was more than threefold greater than that for intermediate-risk patients and nearly five times greater than that for low-risk individuals.

Granulocyte Colony-Stimulating Factor Use

One hundred nine patients (8%) received early G-CSF (Cycles 1 and 2). In both Cycles 1 and 2, the mean duration of G-CSF use was 9.2 days (range, 7–19 days). Although the use of early G-CSF was associated with a significant reduction in risk of hospitalization for febrile neutropenia in stepwise regression analysis, the results were not statistically significant in univariate analysis (P = 0.127) (Table 1) or in the fixed model (P = 0.08).

DISCUSSION

In the current retrospective study of patient medical records, several risk factors were found to be associated with hospitalization for febrile neutropenia among patients with intermediate-grade NHL who were treated with CHOP, CHOP plus rituximab, or CNOP chemotherapy. The greatest risk of initial hospitalization for febrile neutropenia occurred during chemotherapy Cycles 1 and 2. An increased risk was significantly associated with the following factors: age 65 years or older, baseline serum albumin level less than or equal to 3.5 g/dL, ANC less than 1500/mm³ at the time of presentation, planned ARDI greater than or equal to 80%, and the presence of hepatic disease. The use of early G-CSF was associated with a reduction in the risk of hospitalization for febrile neutropenia but only reached statistical significance in the stepwise multivariate model.

The finding that the majority of initial hospitalizations for febrile neutropenia occur within the first 2 cycles of chemotherapy is consistent with a previous study in which 83% of these events occurred by Cycle 3.8 The early time of onset is important because of the potential impact of resulting treatment delays or dose reductions on the overall dose intensity in patients with responsive and potentially curable malignancies such as NHL.15, 17, 18 These concerns are further confirmed by recent data, suggesting that patients hospitalized for febrile neutropenia in Cycle 1 were 4.4 times more likely to prematurely discontinue CHOP chemotherapy.19 Recent risk modeling efforts have identified factors associated with neutropenic events in patients with NHL receiving CHOP chemotherapy. Abnormal bone marrow, BSA less than 1.9 m², serum albumin level less than 3.5 g/dL, and LDH level greater than 1 × normal have all been identified as predictors of an increased risk for neutropenia in this patient population.7, 10 The characteristics associated with a significantly increased risk of hospitalization resulting from febrile neutropenia in these patients included age 65 years or older, a greater than 80% planned ARDI, renal disease, and no early G-CSF administration.8 In the current study, age 65 years or older was also associated with an increased risk of hospitalization for febrile neutropenia. Longer lengths of stay and increased mortality rates have been associated with hospitalizations for older patients, possibly because of comorbidities that are more common among the elderly.11, 20, 21

Significant reductions in the incidence of febrile neutropenia have been reported in randomized clinical trials in which G-CSF was administered prophylactically.4–6 Although not significant in the fixed multivariate model, early G-CSF use was associated with a trend toward less risk of hospitalization for febrile neutropenia after adjustment for the other covariates considered. Because patients selected to receive G-CSF may have been at higher risk, we believe the impact of the hematopoietic growth factors can only be addressed fully in prospective clinical trials.

Limitations are inherent in a retrospective chart analysis, such as missing data and variability in definitions. Performance status, a potential risk factor for febrile neutropenia, was not collected as a variable in our study and, therefore, could not be included in our analysis. Because the analyses performed in the current study were exploratory, validation of the risk model should be prospectively confirmed.

The current study suggests that patients with NHL receiving CHOP or similar chemotherapy regimens who are at increased risk of hospitalization for febrile neutropenia can be identified. This may permit a more rational-targeted application of supportive care measures, such as the hematopoietic growth factors, toward the patients at greatest risk and most likely to benefit.