Introduction

Myelosuppressive chemotherapy is regularly used to treat different malignancies; however, it is often complicated by hematopoietic toxicity [1]. Chemotherapy-induced neutropenia (CIN) is the most common hematologic toxicity of myelosuppressive chemotherapy and can result in serious consequences [2]. Neutrophils are the most abundant leukocytes in circulation and play a crucial role in defending against infections [3]. Patients with CIN are thus at high risk for developing infection [2]. For patients with cancer, infection can be a life-threatening complication that is associated with suboptimal delivery of planned chemotherapy and significant increase in morbidity, mortality, and healthcare resource use [4–6].

Absolute neutrophil count (ANC) is a measure of the concentration of neutrophils in blood and is generally used to grade severity of neutropenia [7]. Myelosuppressive chemotherapy decreases ANC until it reaches its lowest point (the nadir), and ANC subsequently rises after bone marrow recovery. The shape of the ANC trajectory curve during chemotherapy varies based on the type of chemotherapy administered, patient characteristics, and use of granulocyte colony-stimulating factor (G-CSF) [8, 9].

G-CSF regulates production of neutrophils within the bone marrow and affects neutrophil progenitor proliferation and differentiation [10, 11]. Filgrastim (NEUPOGEN®, Amgen Inc., Thousand Oaks, CA, USA) [12] and pegfilgrastim (Neulasta®, Amgen Inc., Thousand Oaks, CA, USA) [13] are recombinant human G-CSFs indicated to decrease the incidence of infection, as manifested by febrile neutropenia (FN; neutropenia with fever), in patients with nonmyeloid malignancies receiving myelosuppressive anticancer drugs. Several randomized controlled trials have shown that chemotherapy-treated cancer patients who received prophylactic G-CSF experienced a substantially earlier and shallower ANC nadir and a more rapid recovery of ANC and lower incidence of infection (characterized by FN) compared with patients who did not receive G-CSF prophylaxis [14–16]. Prior studies provided some evidence that cancer patients with lower ANCs and longer duration of severe CIN during chemotherapy were at higher risk of developing infection [17, 18]. However, there is limited information on the quantitative relationship between ANC trajectory and infection risk.

The current study was conducted to quantify the relationship between severity and duration of CIN and risk of infection. We pooled individual patient data from several randomized controlled trials to estimate the hazard of first infection associated with different severities and durations of CIN among patients with nonmyeloid cancer who did not receive prophylactic G-CSF. An understanding of this relationship will facilitate clinical decision-making with respect to the need for preventing infections in cancer patients receiving chemotherapy.

Methods

Study design

The current study pooled individual patient data from six phase 2 or 3 randomized controlled trials sponsored by Amgen Inc. These trials were originally designed to evaluate the effectiveness of G-CSF (filgrastim or pegfilgrastim) in reducing CIN and infection in cancer patients who were receiving myelosuppressive chemotherapy. In the present study, we focused exclusively on the control/placebo arms in which no prophylactic G-CSF was administered to quantify the relationship between severity and duration of CIN with risk of infection-related hospitalization.

Study population

From the Amgen-sponsored phase 2 or 3 clinical trials in CIN, we included trials that had arms within which patients met the following criteria: adult patients with nonmyeloid malignancies who were treated with myelosuppressive chemotherapy; no prophylactic G-CSF was used; body temperature was measured on a daily basis; ANC was measured at least once at baseline of cycle 1 (days 1–4) and at least three times per week between day 4 and cycle end; and infection or FN was included as a study endpoint.

Patients in the selected trials were considered eligible for inclusion in the current analysis if they had ANC ≥ 1500/μL and normal body temperature before chemotherapy initiation. Patients were excluded if they had a recent infection before chemotherapy, had prior bone marrow or stem cell transplant, received prophylactic antibiotics, or received pelvic irradiation or radiation therapy extending beyond a single involved field within 4 weeks before chemotherapy initiation or during the first chemotherapy cycle.

Exposure and endpoint

Area over the curve (AOC) of ANC time-response curve, below different thresholds, was used to measure both the severity and duration of CIN. AOC was calculated as the area above the ANC time-response curve in the first chemotherapy cycle and below the threshold of 0.5 × 109/L or 1.0 × 109/L. The threshold is based on the Common Terminology Criteria for Adverse Events: ANC < 0.5 × 109/L is categorized as grade 4 neutropenia, and between 0.5 × 109/L and 1.0 × 109/L as grade 3 neutropenia [7].

We determined whether patients met our definition of infection-related hospitalization by reviewing reasons for hospitalization in patients’ case report forms. Patients were classified as having infection-related hospitalization if at least one reason for hospitalization was an infection-related condition (including FN).

Statistical analysis

Descriptive analyses were conducted to characterize study patients’ demographics, disease and treatment characteristics, and medical history. Body surface area (BSA) was calculated using the Mosteller formula [19]. Chemotherapy regimens’ risk categories for developing FN were classified based on the National Comprehensive Cancer Network (NCCN) guideline [20]. For regimens that remain unclassified, FN incidence among patients treated with the regimen but with no G-CSF prophylaxis reported either in the literature or in Amgen-sponsored clinical trials was used to determine FN risk category.

The log interpolation technique was used to derive ANC on days without a measurement, using the two ANC measurements between which it was bounded. ANC nadir was the lowest ANC value that occurred over the chemotherapy cycle. Time to ANC nadir was calculated as the number of days for a patient’s ANC to reach the nadir. Study patients were censored from the analyses of ANC trajectory upon occurrence of infection-related hospitalization, since potential treatment changes after infection might affect ANC trajectory. AOC of ANC was calculated using the Riemann sum method assuming ANC values to be constant within each day [21].

Time-dependent Cox proportional hazards models were used to quantify the hazard of first infection associated with each additional day of grade 4 CIN (ANC < 0.5 × 109/L) or grade 3/4 CIN (ANC < 1.0 × 109/L) as well as the hazard associated with AOC, all in the first chemotherapy cycle. The CIN exposure variable was coded as 0 if the patient had not developed CIN at a specific time t and was coded as 1 if the patient had developed CIN prior to or at time t. Potential confounders adjusted for in the model included sex, age (per 10 years increase), Eastern Cooperative Oncology Group (ECOG) performance status (0, 1, 2–3; nominal scaled indicator variable with ECOG 0 as the reference category), body mass index (BMI) (per 5 kg/m2), data source (filgrastim or pegfilgrastim trial), tumor stage (advanced, non-advanced), comorbidities related to impaired neutrophil function (congestive heart failure, diabetes, renal disease, or thyroid disorder), and comorbidities related to disturbance of barrier function (chronic obstructive pulmonary disease) [22]. Standard disease definitions were created to identify patients with history of relevant comorbidities from the clinical trial case report forms. Missing ECOG status for five patients was imputed with the median value. Missing weight and/or height for three patients were imputed with their respective medians by sex to derive BMI and BSA.

Results

Pegfilgrastim and filgrastim CIN clinical trials conducted by Amgen Inc. and for which patient-level data were available in-house were identified. Of the 24 pegfilgrastim and 19 filgrastim phase 2 or 3 trials identified, 22 pegfilgrastim and 15 filgrastim trials were excluded based on the study population or design (Fig. 1). Data from patients who met the eligibility criteria from the remaining six studies (see Online Resource 1) were analyzed.

Fig 1
figure 1

Selection of studies included in the analysis

Full size image

A total of 271 patients were eligible for inclusion in the current study. Demographic and clinical characteristics of the study population are shown in Table 1. Of the eligible patients, 60.5 % were male, 95.2 % were white, and 94.1 % had ECOG performance status ≤2. Mean (± standard deviation (SD)) age of patients was 59.9 (±8.6) years. Of the patients, 56.1 % had small cell lung cancer, 24.4 % had non-Hodgkin’s lymphoma, 11.4 % had head and neck cancer, and 8.1 % had breast cancer. Most (63.8 %) patients had advanced cancer, and most (77.5 %) received chemotherapy regimens associated with greater than 20 % FN risk.

Table 1 Demographic and clinical characteristics of study population
Full size table

In the first chemotherapy cycle, 238 patients (87.8 %) developed grade 3/4 CIN, and 216 patients (79.7 %) developed grade 4 CIN. Median (Q (quartile) 1, Q3) baseline ANC was 5.24 (3.90, 6.90) ×109/L, median (Q1, Q3) ANC at nadir was 0.08 (0.03, 0.32) ×109/L, and median time for ANC to reach the nadir was 13 days (Table 2). Figure 2 presents the daily median ANC (Q1, Q3) during cycle 1 on a natural logarithmic scale.

Table 2 Description of ANC trajectory in the first chemotherapy cycle
Full size table
Fig 2
figure 2

ANC trajectory in the first chemotherapy cycle

Full size image

During the first chemotherapy cycle, 51 patients (18.8 %) were hospitalized for infection-related diseases. For each additional day that patients had grade 3/4 or grade 4 CIN, their risk of infection-related hospitalization increased by 28 % (hazard ratio (HR) = 1.28, 95 % confidence interval (CI) 1.07, 1.51) and 30 % (HR = 1.30, 95 % CI 1.10, 1.54), respectively (Table 3).

Table 3 Risk of infection-related hospitalization associated with each additional day of CIN adjusted for potential confounders
Full size table

Table 4 shows elevated risk of infection-related hospitalization associated with each unit (day × 109/L ANC) increase in AOC. Each unit increase in the AOC with threshold of ANC < 0.5 × 109/L (grade 4 neutropenia) was associated with an almost two–fold increased risk of infection-related hospitalization (HR = 1.98, 95 % CI 1.35, 2.90). With the threshold of ANC < 1.0 × 109/L (grade 3/4 neutropenia), each unit increase in AOC was also associated with an elevated risk of infection-related hospitalization (HR = 1.42, 95 % CI 1.17, 1.72).

Table 4 Risk of infection-related hospitalization associated with each unit increase in AOC of ANC (day × 109/L ANC) adjusted for potential confounders
Full size table

Discussion

The results of the current study add further evidence to earlier findings that prolonged exposure to severe neutropenia results in an increased risk of infection. Increase in AOC of ANC below given thresholds, a composite measurement for both severity and duration of CIN, is associated with a higher risk of infection in cancer patients receiving chemotherapy. Infection risk increased about 30 % with each additional day of exposure to grade 3 or grade 4 CIN.

Infection has significant clinical consequences and poses a substantial financial cost for cancer patients receiving chemotherapy. The inpatient case fatality rate with FN was reported to be 2.6–10.6 % [4, 23–25]. FN may also result in suboptimal delivery of planned chemotherapy, including reduction or delay of planned doses of chemotherapy or chemotherapy discontinuation [26–29]. Chemotherapy dose delays and dose reductions or discontinuations may lead to poorer disease-free survival, progression-free survival, and overall survival [27, 28, 30–39]. In addition, FN places substantial economic burden on the healthcare system. In the US, the mean (median) hospitalization cost of FN management ranged from $18,880 to $22,086 ($8376 to $10,396) per episode [4, 23, 40].

Guidelines recommend prophylactic use of G-CSF in patients with a risk of FN greater than 20 % and suggest consideration of G-CSF prophylaxis when the risk is 10–20 % [41–43]. The strong, positive association observed between severity and duration of CIN with risk of infection in the current study provides a more scientific explanation for findings from prior randomized controlled trials, which reported that patients with cancer who received prophylactic G-CSF had significantly lower FN incidence and different ANC trajectories (earlier and shallower ANC nadir and more rapid recovery of ANC) compared with those who did not receive prophylactic G-CSF [14–16].

Our findings are consistent with those of previous studies. Bodey et al. [18] found that risk of infection was higher at lower concentrations of granulocytes (a term that typically includes neutrophils, basophils, and eosinophils) and the risk increased with longer duration (in weeks) of granulocytopenia among 52 leukemia patients receiving chemotherapy. In that study population, any episode of granulocytopenia, regardless of duration, had a 39 % chance of resulting in identified infection. Six weeks of severe granulocytopenia (<100/mm3) or 12 weeks of persistent granulocytopenia (<1000/mm3) resulted in 100 % identified infection [18]. However, the extent of myelosuppression experienced by patients with acute leukemia may be different in nature from that experienced by patients with nonmyeloid malignancies who are receiving chemotherapy. Another analysis of two randomized phase 3 trials comparing pegfilgrastim to filgrastim reported that risk of FN increased with duration (1, 2, 3, and ≥4 days) of severe neutropenia (ANC < 0.5 × 109/L), with an odds ratio of 2.28 per day increase in duration of severe neutropenia using logistic regression analysis [17]. In a previously conducted simulation study, the Cox proportional hazards model with time-updated exposure was shown to provide the least biased estimates compared to logistic regression or Cox proportional hazards model with constant exposure, when studying the relationship between a biomarker and a binary outcome when duration of that biomarker stays beyond a threshold that is the predictor of the event of interest [44]. Further, the use of FN (which includes neutropenia in its definition) as an outcome in a model with neutropenia as an exposure may overestimate the effect estimate.

In the current study, we focused on cancer patients receiving no G-CSF prophylaxis to enable us to get an accurate estimate of the relationship of interest. Moreover, the current study used more quantitative methods to estimate the effects of severity and duration of CIN on risk of infection compared to the previous studies. Specifically, we used a composite variable, AOC of ANC, to measure severity and duration of CIN simultaneously and quantified the risk of infection with each additional day increase of CIN at different severities. Further, we adjusted for potential confounding by controlling for a number of covariates in multivariate regression models to better estimate these relationships (Tables 3 and 4). Where possible, definitions for endpoints and all the covariates were standardized across different trials, and we also used standardized inclusion and exclusion criteria for patient selection into this pooled analysis.

Despite the demonstrated improvements we made to the study methodology, several limitations of the current study should be noted. First, patients enrolled in filgrastim trials conducted in the 1990s as well as those enrolled in more recent pegfilgrastim trials were included in this pooled analysis. Clinical practice patterns and data collection and reporting methods are likely to have changed over this time period. To account for the temporal changes, data source (filgrastim or pegfilgrastim trial) was adjusted for in the analysis. Another limitation is that the analysis relied on existing data collected in the original clinical trials, and there is a possibility of differences in definitions used for evaluated outcomes across studies. Wherever possible, we have standardized to common definitions for this study. Lastly, this study analyzed data from patients originally recruited for clinical trials in which individuals with poor performance status or serious medical illnesses were likely excluded from enrollment.

Conclusions

In this study, we observed that severity and duration of CIN increase the risk of infection in cancer patients receiving chemotherapy. Interventions that limit the extent and duration of CIN are of critical importance in preventing infection and further improving subsequent treatment outcomes in this patient population.