Risk of chemotherapy-induced febrile neutropenia with early discontinuation of pegfilgrastim prophylaxis in US clinical practice

A retrospective cohort design and data from two large US healthcare claims repositories were employed. To minimize healthy survivor bias and other (e.g., selection) biases that would likely occur in such an observational study if it were designed to mimic the above-described trial by Aarts and colleagues, the evaluation was limited to receipt or no receipt of pegfilgrastim prophylaxis during the second chemotherapy cycle among a cohort of patients who all received first-cycle pegfilgrastim prophylaxis. A detailed description of study design and study methods may be found in the online supplement (Online Resource A).

Data from the two US healthcare claims repositories spanned January 1, 2006 through December 31, 2013, and were pooled for analyses. The two study repositories, the Truven Health Analytics MarketScan^® Commercial Claims and Encounters and Medicare Supplemental and Coordination of Benefits Databases (“MarketScan Database”), and the IMS LifeLink™ PharMetrics Plus Health Plan Claims Database (“LifeLink Database”), comprise medical (i.e., facility and professional service) and outpatient pharmacy claims from a large number of participating private US health plans. Formal approval for this study from an Institutional Review Board (IRB) was not required because the design was retrospective in nature, and subjects in the study databases could not be identified—directly or indirectly—through variables linked to their claims and/or enrollment records. Use of the study databases for health services research is fully compliant with the Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule and federal guidance on Public Welfare and the Protection of Human Subjects [23].

The source population comprised all patients aged ≥18 years who, from July 2006 to June 2013, received a course of myelosuppressive chemotherapy of at least two cycles duration for a single primary solid tumor or non-Hodgkin’s lymphoma (NHL). For each patient in the source population, the first observed course of chemotherapy and the first two cycles of chemotherapy within that course were characterized. Only patients who received first-cycle pegfilgrastim prophylaxis had continuous health benefits for ≥6 months prior to chemotherapy, did not have evidence of reactive CSF use or FN in cycle 1, did not receive prophylaxis with other CSF agents (filgrastim, sargramostim) or antimicrobials in either cycle 1 or cycle 2, and met all other selection criteria (as described in the online supplement) that were retained in the source population. Prophylactic use of pegfilgrastim was defined as receipt 1–3 days following completion of myelosuppressive chemotherapy administration in a given chemotherapy cycle, which is consistent with the indicated administration schedule and National Comprehensive Cancer Network (NCCN) guidelines [8, 9]. Use of pegfilgrastim was identified based on medical claims with corresponding Healthcare Common Procedure Coding System (HCPCS) Level II codes (C9119, S0135, J2505). From the source population, all patients who did not receive pegfilgrastim prophylaxis in their second cycle of chemotherapy (“comparison patients”) were matched to those who received it (“pegfilgrastim patients”).

Matching was implemented for each patient in the source population who did not receive second-cycle pegfilgrastim prophylaxis by first identifying all “candidate” patients who received second-cycle pegfilgrastim prophylaxis and had the same cancer type and chemotherapy regimen. From all such candidates for each patient, the candidate with the closest propensity score to the comparison patient was selected as the matched patient using a fixed 1:1 ratio and nearest-neighbor approach [24]. Propensity scores represent the conditional probability of assignment to the exposure group and may be used to control for multiple observed covariates that are associated with exposure and outcome [25, 26]. Propensity scores for receipt of second-cycle pegfilgrastim prophylaxis were estimated using multivariate logistic regression; independent variables included all patient, cancer, and treatment characteristics described below. The study population was limited to patients who received intermediate/high-risk chemotherapy regimens for non-metastatic breast cancer, non-metastatic colorectal cancer, non-metastatic lung cancer, or NHL, and for which the number of patients who discontinued pegfilgrastim prophylaxis in cycle 2 was ≥100 (Table 1).

FN episodes were ascertained beginning 4 days after completion of myelosuppressive chemotherapy administration in the second cycle of chemotherapy and ending on the last day of that cycle, and were identified using a “broad” definition, as follows [ 27]. FN episodes requiring inpatient care were identified based on hospital admissions with a principal or secondary diagnosis of neutropenia (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] 288.0), or fever (780.6), or infection (codes in Online Resource A). FN episodes requiring outpatient care only were identified based on ambulatory encounters (e.g., those in a physician’s office, emergency department, or home) with a diagnosis of neutropenia, or fever, or infection and—on the same date—a HCPCS Level I (i.e., Current Procedural Terminology [CPT]) code for IV administration of antimicrobial therapy. Such encounters that preceded or followed an FN-related hospitalization during the same cycle of chemotherapy were not considered as a separate outpatient episode (i.e., they were classified as part of the episode of FN requiring inpatient care). An alternative (“narrow”) definition for FN comprising inpatient encounters with a principal or secondary diagnosis of neutropenia, and outpatient encounters with a diagnosis of neutropenia and evidence of IV antimicrobial therapy, was also evaluated [27].

Patient characteristics included many of those listed by the American Society of Clinical Oncology (ASCO) and NCCN as important risk factors for FN and thus those that could confound the estimated relationship between prophylaxis discontinuation and FN risk. Patient characteristics were evaluated based on evidence during the period beginning up to 12 months prior to the date of chemotherapy initiation and ending 3 days after completion of chemotherapy in the second cycle (unless otherwise noted in the online supplement).

Characteristics included the following: age; sex; presence of selected chronic comorbidities (cardiovascular disease, diabetes, liver disease, lung disease, renal disease, osteoarthritis, rheumatoid disease, thyroid disorder); body weight/nutritional status (obesity, underweight, malnutrition); proxies for health status (hospice/skilled nursing facility [SNF] care) and physical function (use of hospital bed, supplemental oxygen, walking aid, wheelchair); use of immunosuppressive therapy; history of blood disorders (anemia, neutropenia, other), infection, recent surgery (i.e., ≤90 days prechemotherapy), hospitalization (all-cause and FN-related, respectively), chemotherapy, and radiation therapy; total healthcare expenditures in the baseline period; presence of metastatic disease; and calendar year of chemotherapy initiation.

The adequacy of the matching procedure in terms of patients’ baseline characteristics was evaluated using standardized differences; a value <0.1 was assumed to indicate a negligible difference in the characteristic between comparison patients and pegfilgrastim patients [28, 29]. Comparisons of second-cycle FN odds between comparison patients and pegfilgrastim patients were evaluated on an overall basis and within cancer- and regimen-specific subgroups using generalized estimating equation (GEE) regression models; a binomial distribution and logistic link function were specified for all GEE models, and the models were fitted using an exchangeable correlation structure. GEE models were used to account for the matched-pairs design; additional covariates were not included in the models (since groups were well-balanced on their baseline characteristics).

All statistical tests were two-sided and were performed at a significance level of α = 0.05. Assuming an approximate 50 % increase in second-cycle FN risk among patients not receiving pegfilgrastim prophylaxis in that cycle (6.4 vs. 4.2 % for those receiving pegfilgrastim prophylaxis) and assuming further that the minimum sample size for each group would be at least 2000 (4000 in total), we calculated that the beta (β) for this evaluation would be <20 % (two-sided α = 0.05) and thus the study should have adequate power (>80 %) to evaluate the primary objective [13].

The sensitivity of study results to alternative methods for confounding adjustment (i.e., using all patients qualifying for inclusion in the source population and multivariate regression) and alternative methods for propensity-score matching (i.e., 1:3 ratio and sequential) were evaluated. In multivariate regression analyses, all patients in the source population with a qualifying cancer-regimen combination were included, and the second-cycle FN odds ratio for comparison patients versus pegfilgrastim patients was estimated using a logistic model including all potential confounders as independent variables.