Risk of developing chemotherapy-induced febrile neutropenia (FN) depends on patient-, treatment-, and disease-related characteristics.1 In our prior investigation, several chronic comorbidities were associated with significantly increased FN risk, including chronic obstructive pulmonary disease (COPD), congestive heart failure (CHF), autoimmune diseases, osteoarthritis, peptic ulcer disease, thyroid disorder, renal disease, liver disease, and HIV infection.2,3 We hypothesized that the increased FN risk associated with these comorbidities may be mediated at least in part through the following biological mechanisms: bone marrow suppression, impaired neutrophil and other immune cell function, disturbances of barrier function/increased availability of pathogenic microbes, and impaired bacterial flora.2 For example, renal disease and thyroid disorder have been associated with impaired neutrophil function.4–6 Peptic ulcer disease and COPD may be linked to FN via disturbances of barrier function, impaired bacterial flora, and, in the case of COPD, use of inhaled steroids.7–9
Based on these previously hypothesized underlying pathogenic mechanisms for FN, we further hypothesize that other clinical conditions that potentially lead to one or more of these pathogenic states may also increase risk of FN during chemotherapy. Specifically, we hypothesize that surgery, radiation therapy (RT), use of corticosteroids and antibiotics, and certain dermatologic/mucosal conditions might increase risk of chemotherapy-induced FN (hypothesized mechanisms and associated conditions are detailed in supplemental eTable 1, available with this article at JNCCN.org). Prior or concurrent RT may predispose patients with cancer to infections via bone marrow suppression and damage of mucosal barrier.10–12 Recent surgery and select dermatologic and mucosal conditions could indicate a breakdown in skin/mucosal integrity and increased exposure to pathogenic microbes.13,14 Antibiotic use prior to chemotherapy could alter bacterial flora.15,16 Corticosteroids are known to impair immune cell function.17
NCCN recommendations for prophylactic management of FN depends on the expected FN risk level the patient will experience, and accurate prediction of a given patient's FN risk depends on knowledge of key FN risk factors. The performance of currently available FN prediction models in the literature indicated room for improvement in clinical FN risk prediction.18,19 Thus, the objective of this study was to assess the effects of the aforementioned clinical conditions on FN risk during chemotherapy to shed light on novel FN risk factors that may be considered in clinical FN risk prediction.
Patients and Methods
Study Setting and Population
Kaiser Permanente Southern California (KPSC) is an integrated healthcare delivery system that provides comprehensive health services for 4 million enrollees who are representative of the racial/ethnic diversity in Southern California.20 KPSC maintains electronic medical records (EMRs) for virtually all aspects of care delivered, all linkable through a unique member identifier. The study cohort was composed of patients who were (1) diagnosed with non-Hodgkin's lymphoma (NHL) or breast, lung, colorectal, ovarian, or gastric cancer at KPSC between January 1, 2000, and December 31, 2009; (2) aged ≥18 years at diagnosis; (3) initiated myelosuppressive chemotherapy (defined per reference21) within 1 year of diagnosis; and (4) KPSC members for at least 12 months prior to diagnosis. These cancer types were chosen because they were relatively common and would be managed with myelosuppressive chemotherapy agents. Patients who met the following criteria were excluded: (1) had unknown cancer stage, chemotherapy agents, or cycle length; (2) received prophylactic granulocyte colony-stimulating factor (G-CSF) within 4 days of chemotherapy initiation and/or antibiotics prophylaxis dispensed between 3 days before and 10 days after first chemotherapy administration in the absence of fever/infection; (3) received dose-dense chemotherapy regimens or weekly regimens; (4) had bone marrow/stem cell transplantation; or (5) were patients of Orange County Medical Center if their cancer was diagnosed before 2007, for whom data on absolute neutrophil count (ANC) could not be readily accessed via EMR.
Exposures of interest were surgery prior to chemotherapy, RT before or concurrent with chemotherapy, corticosteroid and antibiotic use prior to chemotherapy, and selected dermatologic and mucosal conditions. Prior surgery was defined as cancer-related excisional treatment procedures received within 6 months prior to chemotherapy initiation. Prior and concurrent RT were defined as RT administered within 2 months prior to chemotherapy and within 5 days after chemotherapy initiation, respectively. A 2-month window was chosen for prior RT because previous studies showed that WBC counts did not return to normal until 6 to 8 weeks after radiation exposure.12,22 Corticosteroid use was identified within 3 months prior to chemotherapy initiation. Because most corticosteroid use by the study population was oral prescriptions (88%), only oral corticosteroid use was evaluated in this study. Antibiotic use was defined as at least 5 days of continuous outpatient antibiotic prescriptions, with the last day of use within 3 months prior to chemotherapy initiation. A 3-month window was chosen because previous studies showed that bacterial flora became normalized 3 months after withdrawal of antibiotics.23 Only outpatient oral antibiotic use was considered in the main analysis. A subanalysis evaluating both oral and any duration of intravenous antibiotic use from both outpatient and inpatient settings was limited to 2008 to 2009, when data on inpatient antibiotic use became assessable electronically. Selected dermatologic and mucosal conditions likely to damage the integrity of the barrier function were assessed using ICD-9 codes within 1 month prior to chemotherapy. The specific data source, ICD-9 diagnosis, and CPT codes to identify exposures of interest are shown in supplemental eTable 2.
The outcome of interest was the occurrence of FN within the first cycle of chemotherapy. FN was defined using one of the following methods: (1) neutropenia ICD-9 code 288.0 and fever ICD-9 code 780.6 within 7 days; (2) ANC <1,000/mcL and fever ICD-9 code 780.6 within 7 days; (3) hospitalization with neutropenia, ICD-9 code 288.0, as the primary diagnosis; (4) neutropenia ICD-9 code 288.0 or ANC <1,000/mcL and hospitalization with ICD-9 code of bacterial/fungal infection within 7 days; or (5) neutropenia ICD-9 code 288.0 or ANC <1,000/mcL and intravenous antibiotic use within 7 days. When not specified, both inpatient and outpatient corresponding ICD-9 codes were used to capture FN outcomes. FN outcomes were assessed only in the first chemotherapy cycle to allow the most unbiased assessment of the effect of new risk factors on FN. Subjects were followed from the first chemotherapy administration to the FN event, the beginning of the second chemotherapy cycle, termination of KSPC membership, death, or 28 days after the first chemotherapy administration, whichever came first.
Patient characteristics included age at diagnosis, sex, race/ethnicity, census block income and education level, Medicare or Medicaid enrollment, and length of KPSC membership. Cancer and treatment characteristics included stage at diagnosis and chemotherapy regimens, which were categorized based on its risk of inducing FN into highly myelosuppressive (>20% risk of FN), moderately myelosuppressive (10%–20% risk), or unclassified (<10% risk and all other regimens not in these lists).24 In addition, the number of myelosuppressive agents used with and without dose reduction (ie, received <90% of standard dose based on body surface area) was also considered. Laboratory measures of interest included ANC at the time of diagnosis. Comorbidities previously shown to increase the risk of FN were captured, including COPD, CHF, HIV infection, rheumatoid diseases, other autoimmune disorder, thyroid disorder, peptic ulcer disease, diabetes, renal failure, and liver disease. Recent infection history, defined as any bacterial or fungal infection within 3 months prior to chemotherapy initiation, was also captured. Both comorbidities and infection history were ascertained using ICD-9 codes.
Distributions of patient characteristics and occurrence of FN were described overall and by cancer type. Bivariable and multivariable Cox models were used to evaluate the association between each of the potential risk factors of interest and occurrence of FN in the first chemotherapy cycle. For the multivariable analysis, we used propensity score functions to adjust for potential confounders. The following variables were included in the propensity score function for each potential risk factor of interest: age, sex, race/ethnicity, length of KPSC membership, public insurance status, census level socioeconomic indicators, total number of comorbidities, recent infection, previous cancer, and cancer stage at diagnosis (Table 1 footnote). An additional 12 subjects were excluded from the multivariable analysis due to missing data on census level socioeconomic status. The distribution of these confounders by exposure status within propensity score quintiles was examined to determine if balance was achieved. Additional confounders that are specific to each potential risk factor of interest were also selected based on prior knowledge and adjusted in the propensity score function (see Table 1 footnote). The propensity score was included as a covariate (smoothed with natural cubic splines) in the Cox model to assess the effect of potential risk factors of interest on FN. The analyses were stratified by cancer type using the “strata” comment in SAS Phreg to account for potential confounding by cancer type.
Alternative models that additionally adjusted for potential confounders that may be on the causal pathway, such as baseline ANC, baseline anemia (defined as prior diagnosis of anemia or baseline hemoglobin <10 g/dL), chemotherapy regimen FN risk group, number of myelosuppressive agents with dose reduced and not reduced, were also evaluated in sensitivity
HRs for Estimating Risk of FN Associated With Potential New Risk Factors
We also evaluated the effect of corticosteroid by duration of use (no use, <15 days, 15–29 days, 30–44 days, and 45–90 days) and recentness of use (no use, <15 days, 15–29 days, 30–44, and 45–90 days prior to chemotherapy). Alternative recentness of use was also assessed for oral antibiotic use (≤30 days and <15 days prior to chemotherapy), as well as for prior surgery (≤30 days prior to chemotherapy). All analyses were performed with SAS 9.3 (SAS Institute Inc.).
A total of 62,714 adult patients were diagnosed with 1 of 6 cancer types of interest between 2000 and 2009 at KPSC. Of these, 15,971 were included in the analysis after applying the eligibility criteria. Figure 1 shows population dispositions after applying each inclusion and exclusion criterion. Table 2 presents the distribution of demographic and clinical characteristics of the study population. Prior surgery (60.2%) and oral antibiotic use (44.0%) were common in the study population. RT was received by 6.6% and 2.6% of the study population in the period prior to and concurrent with chemotherapy, respectively. Oral corticosteroid use was found in 5.6% of patients prior to chemotherapy. Selected dermatologic and mucosal conditions were found in 3.0% of patients.
Overall, 4.3% of the study patients developed FN in the first chemotherapy cycle. Table 1 presents the incidence proportions of FN in the first chemotherapy cycle for patients with and without each of the potential risk factors. The propensity score–adjusted hazard ratio (aHR) estimates for FN risk associated with each of the potential risk factors are also shown in Table 1. Selected dermatologic/mucosal conditions were marginally associated with increased risk of FN with statistical significance (aHR, 1.40; 95% CI, 0.98–1.93). Corticosteroid use overall was associated with a statistically significant increase in FN risk (aHR, 1.53; 95% CI, 1.17–1.98). A positive relationship was observed for the durationof corticosteroid use and FN risk, with the aHR increasing from 1.78 (95% CI, 1.50–2.11) for use of <15 days to 2.86 (95% CI, 1.74–4.47) for use of 45 to 90 days (Table 3). However, no clear pattern was observed with the recentness of corticosteroid use. For antibiotic use, no association was observed for oral antibiotic use in the main analysis nor in the sensitivity analysis restricting the use to ≤30 days or <15 days prior to chemotherapy (data not shown). However, in the subanalysis including data on inpatient antibiotic use, intravenous antibiotic use (aHR, 1.35; 95% CI, 0.97–1.87) but not oral antibiotic use (aHR, 1.07; 95% CI, 0.77–1.48) was marginally associated with increased risk of FN relative to no antibiotic use (Table 4).
No other potential risk factors examined were associated with risk of FN (Table 1). For prior surgery, no clear association was observed when exposure was restricted to ≤30 days prior to chemotherapy (aHR, 1.03; 95% CI, 0.82–1.28; data not shown). Similar results were obtained from all the sensitivity analyses additionally adjusted for potential confounders that could be on the causal pathways (Table 1).
Balanced distribution was achieved for most but not all measured confounders for each exposure of interest on examination within the propensity score quintiles (supplemental eTables 3–8).
We found that corticosteroid use, selected dermatologic/mucosal conditions, and intravenous antibiotic use were associated with increased FN risk. Our findings on corticosteroid use are consistent with those reported by Lyman et al,18 who found that use of immunosuppressive drugs was a risk factor for FN. We further found that a duration response relationship might exist for corticosteroid use, and even a short period of use (ie, <15 days) was significantly associated with increased FN risk. These findings have particular relevance for treatment regimens in which corticosteroids are often used prior to the start of chemotherapy. Our findings on the selected dermatologic/mucosal conditions and intravenous antibiotic use are novel, and to our knowledge, no prior study has examined the effect of these factors on FN risk. The results of this study suggest that corticosteroid use, intravenous antibiotics, and selected dermatologic and mucosal conditions should be taken into consideration when evaluating the need for prophylactic G-CSF or dose reduction, and during patient monitoring.
There is biological plausibility to support the associations observed in our study. Corticosteroid use suppresses immune cell function, and hence increases host susceptibility to pathogenic microbes. Several studies have shown that corticosteroids reduce neutrophils' adherence to endothelium, inhibiting migration to inflammatory sites.25 In hematologic malignancies such as NHL, corticosteroids are often used in high doses, increasing the risk of infection with a broad spectrum of pathogens.11 In addition to NHL, corticosteroids were also commonly used prior to chemotherapy among patients with lung cancer (14%) in our study. Therefore, our findings may have important implications for FN prevention in these cancers. Although our examination was limited to use prior to chemotherapy, our results suggest that
Distribution of Patient Demographics, Cancer and Treatment Characteristics, and History of Potential New Risk Factors
Skin and mucosal injuries that breakdown barrier integrity create a point of entry for microorganisms that cause blood stream infections and directly contribute to fever.26 In our study cohort, the most common dermatologic and mucosal conditions were gastritis/gastroduodenitis, contact dermatitis, and psoriasis. It is a limitation of this study that we were unable to determine the severity of these conditions using ICD-9 codes. However, it is possible that skin/mucosal injuries that are more severe or delayed in healing are more likely to be noted and coded by the physician. It has been suggested that neutrophils play an important role in the biological processes of wound healing by serving as first responders in clearing microbes and cellular debris in the wound area.27,28 Therefore, in addition to a direct causal role of skin injuries on risk of FN, it is also possible that patients with cancer and impaired neutrophil function are more likely to experience both delayed healing of barrier injuries and FN.
Although we did not find FN risk to have a clear association with oral antibiotic use, we did observe a positive association with intravenous antibiotic use, suggesting that intravenous antibiotics may have a
Adjusted HRs for Oral Corticosteroid Use on Risk for FN
Contrary to our hypotheses, we did not find any association between surgery and RT and risk of FN. It is possible that infection risk associated with these treatments was effectively managed with good wound care or prophylactic antibiotic use. Alternatively, for concurrent RT, it is possible that complications may not have occurred until cycle 2. It is also possible that those who received surgery and/or RT were inherently different from those who received chemotherapy alone. For example, those who
HRs for Effect of Antibiotic Use on Risk of FN in the Restricted Analysis (2008–2009)
Several other potential limitations should be considered. First, our study was limited by the use of an electronic algorithm to capture FN, rather than through chart review.29 However, our prior chart review validation demonstrated a positive predictive value of >90% of this algorithm (unpublished data). Second, the completeness of ICD-9 coding for the selected dermatologic/mucosal conditions is unknown, and undercoding may be likely due to the temporary and often ill-defined nature of these conditions. Our study may be subject to potential selection bias due to the exclusion of patients who received primary G-CSF prophylaxis, which may also explain the relatively lower incidence proportion of FN in our study. Furthermore, we did not examine the dose–response relationship for cumulative doses of corticosteroids or antibiotics. Duration and recentness of use are not entirely independent measures and should not be interpreted as such. We cannot completely exclude the possibility of residual confounding. We also only focused on the first cycle of chemotherapy. However, this restriction was used to avoid potential biases in assessing causal estimates that may occur in subsequent cycles of chemotherapy, such as drop out, switching chemotherapy regimens, or dose reduction due to other related treatment complications. Finally, the study period was not recent. However, this study assessed the etiologic risk factors for FN. The biology underlying this potential association should not change over time.
Our study examined several potential novel FN risk factors based on hypothesized underlying pathogenic mechanisms. Our findings suggest conditions that affect the skin/mucosal barrier, corticosteroid use, and intravenous antibiotic use may increase risk of FN, and should be considered in making prophylaxis and patient monitoring decisions. Because patients usually present with multiple FN risk factors, individual patient-level risk factors need to be considered simultaneously to determine each patient's FN risk level. Future research is warranted to develop FN prediction algorithms incorporating established and newly identified patient-level risk factors to help clinicians account for multiple risk factors when making prophylactic decisions.
Drs. Li, Page, and Klippel have disclosed that they are employees and stock holders of Amgen, Inc. Drs. Family, Chen, and Chao have disclosed that they received research funding from Amgen, Inc. to perform this study.
This work is supported by Amgen, Inc. (#135884). All procedures performed in studies involving human participants were in accordance with the ethical standards of KPSC's Institutional Review Board and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. For this type of study, formal consent is not required.
See JNCCN.org for supplemental online content.
HosmerWMalinJWongM. Development and validation of a prediction model for the risk of developing febrile neutropenia in the first cycle of chemotherapy among elderly patients with breast, lung, colorectal, and prostate cancer. Support Care Cancer2011;19:333–341.