Use of Bevacizumab in Community Settings: Toxicity Profile and Risk of Hospitalization in Patients With Advanced Non–Small-Cell Lung Cancer

Findings here confirm the need for adherence to clinical recommendations for judicious use of carboplatin-paclitaxel-bevacizumab, but provide reassurance regarding the relative risk.

Abstract

Purpose:

Little is known regarding toxicities and hospitalizations in community-based settings for patients with advanced non–small-cell lung cancer (NSCLC) who received commonly prescribed carboplatin-paclitaxel (CP) or carboplatin-paclitaxel-bevacizumab (CPB) chemotherapy.

Methods:

Patients with stages IIIB-IV NSCLC age ≥ 21 years diagnosed between 2005 and 2010 who received first-line CP or CPB were identified at four health maintenance organizations (N = 1,109). Using patient and tumor characteristics and hospital and ambulatory encounters from automated data in the 180 days after chemotherapy initiation, the association between CP and CPB and toxicities and hospitalizations were evaluated with χ² tests and propensity score–adjusted regression models.

Results:

Patients who received CPB were significantly younger and had significantly more bleeding, proteinuria, and GI perforation events (all P < .05). For these patients, the unadjusted odds ratio associated with the likelihood of having a hospitalization was 0.46 (95% CI, 0.32 to 0.67). As shown by multivariable and propensity score–adjusted models, patients who received CPB were less likely to have been hospitalized (odds ratio, 0.48; 95% CI, 0.32 to 0.71) and had fewer total hospitalizations (rate ratio, 0.62; 95% CI, 0.47 to 0.82) and hospital days (rate ratio, 0.53; 95% CI, 0.47 to 0.60) than patients who received CP.

Conclusion:

Consistent with earlier randomized clinical trials, significantly more toxicity events were identified in patients treated with CPB. However, both unadjusted and adjusted models showed that patients who received CPB were less likely than patients who received CP to experience a hospital-related event after the initiation of chemotherapy. Findings here confirm the need for adherence to clinical recommendations for judicious use of CPB, but provide reassurance regarding the relative risk for hospitalizations.

Introduction

Usually diagnosed at an advanced noncurable stage, lung cancer is the leading cause of cancer-related death in the United States.^1–3 The antiangiogenic monoclonal antibody bevacizumab was approved by the US Food and Drug Administration FDA in 2004 for metastatic colorectal cancer. In 2006, a label extension was approved for its administration in combination with carboplatin-paclitaxel (CP) for first-line treatment of advanced non–small-cell lung cancer (NSCLC).^4–6 However, in a phase III trial, patients who received carboplatin-paclitaxel-bevacizumab (CPB) had significantly higher rates (P < .05) of toxicities, including hemorrhage, hyponatremia, hypertension, and neutropenia, than patients who received CP.⁷

To assess the toxicity concerns of these two commonly prescribed regimens, a phase III trial evaluated the safety of bevacizumab combined with several standard chemotherapy regimens. The three most common serious adverse events were thromboembolic (8%), hemorrhagic (6%), and GI perforation (4%).^8,9 Although bevacizumab toxicity has been studied in clinical trials, only a few studies have assessed bevacizumab treatment toxicity, including hospitalizations, in the community setting.^10–13

Hospitalization after chemotherapy initiation, whether related to toxicity or not, is prevalent. Yabroff et al reported that more than 60% of net medical costs for elderly patients with lung cancer were related to hospitalization.¹⁴ Vera-Llonch et al reported that 72% of community-based patients with advanced lung cancer had at least one hospitalization after chemotherapy initiation.¹⁵ These studies did not assess hospitalization specifically after bevacizumab treatment. Describing the patterns of toxicity profiles and hospitalizations associated with bevacizumab use in a community setting is important because it directly addresses the need for more clinically relevant information regarding the potential balance of benefits and harms for an expensive drug that may confer only limited improvements in survival.^16–18

In our previous comparative-effectiveness study, we confirmed a survival benefit associated with the addition of bevacizumab to CP in a cohort of patients with advanced NSCLC who received care in a community setting.¹⁹ Here, we use the same cohort and examine patterns of toxicity-related events for patients with NSCLC who received CP with or without bevacizumab in a community setting. In addition, we test the hypothesis that if the bevacizumab-related toxicities observed in the community setting are similar to those noted in the original clinical trial, then the likelihood of a hospitalization after the initiation of bevacizumab should increase with adjustment for observable factors, including age and comorbidity status.

Methods

Study Design and Setting

This retrospective cohort analysis assessed toxicities and hospitalizations within 180 days after receipt of CP or CPB in patients with NSCLC diagnosed between January 1, 2005, and December 31, 2010. Follow-up was through December 31, 2011. Patients received their care at four nonprofit health maintenance organizations (HMOs; the Colorado, Northern California, and Northwest regions of Kaiser Permanente, and Group Health Cooperative) in which the majority of cancer care was delivered by salaried physicians in plan-owned facilities. This study was approved by the institutional review boards of the four participating HMOs.

Data Sources

The primary data source was the Cancer Research Network's (http://crn.cancer.gov/) Virtual Data Warehouse (VDW).^20–23 Briefly, the VDW is populated with patient-level data extracted from electronic health records and administrative and claims databases. The VDW contains tumor registry, procedure, diagnosis, and census data, including measures of socioeconomic status (eg, median neighborhood education), as well as infusion and pharmacy data associated with all hospital and ambulatory encounters (including claims both internal and external to the HMO). International Classification of Diseases, ninth Revision, Clinical Modification (ICD-9), Healthcare Common Procedure Coding System, the Fourth Edition of the Common Procedure Terminology codes, National Drug Code, and diagnosis-related-groups (DRG) codes were extracted from the VDW.

Study Sample

All patients in the VDW Virtual Tumor Registry age ≥ 21 years with a first cancer diagnosis of pathologically confirmed stage IIIB/IV nonsquamous NSCLC were identified. Patients were an HMO member at the time of NSCLC diagnosis, survived at least 1 month after NSCLC diagnosis, and received either CP or CPB as first-line chemotherapy treatment. Patients diagnosed with squamous cell NSCLC and those who had concurrent administration of chemotherapy and radiotherapy, defined as initiation dates within 14 days of each other, were excluded. Patients were monitored from chemotherapy initiation until death, health plan enrollment termination, or 180 days, whichever came first.

Identification of First-Line Treatment

Methods to identify first-line (defined as the first chemotherapy regimen received within 120 days of cancer diagnosis) CP or CPB treatment have been described elsewhere.^19,24 The date of the first chemotherapy administration after NSCLC diagnosis was considered the chemotherapy initiation date. The first-line regimen included all chemotherapy agents administered within 8 days of the chemotherapy initiation date.

Outcome Measures

The primary outcomes were CPB-related toxicities, all-cause hospitalization, number of hospitalizations, and total hospital days within 180 days of chemotherapy initiation. Although we did not conduct primary data collection to capture drug-related serious adverse events,^8,9,25,26 we did use methods consistent with those of other published studies. Using CPB-related toxicities noted in previous studies^5–8 that were translated to a comprehensive ICD-9 list with clinician review, the following toxicity classes were identified from hospital or ambulatory events: hemorrhage (overall as well as hemoptysis, epistaxis, and CNS hemorrhage [CNS bleed]), neutropenia, venous thromboembolism, anemia, thrombocytopenia, hyponatremia, proteinuria, GI perforation, or newly diagnosed hypertension (ICD-9 codes available upon request). The primary diagnosis for an inpatient event could not be identified during the study period, so all diagnoses and the DRG associated with the inpatient stay were collected. The DRGs were then mapped to a Major Diagnostic Category (MDC).

Patient Characteristics

Age at NSCLC diagnosis, year of NSCLC diagnosis, sex, race/ethnicity, tumor grade, morphology, and stage were collected from the Virtual Tumor Registry, and socioeconomic status measures were obtained from VDW Census files. The Quan adaptation of the Charlson Comorbidity Index, modified to exclude cancer diagnoses,²⁷ along with other pretreatment comorbidities used as exclusion criteria in the original trial⁷ were derived from diagnosis codes captured from all hospital and ambulatory encounters that occurred 13 months to 1 month before cancer diagnosis.

Statistical Analyses

Differences in baseline characteristics between CP and CPB were evaluated using the Wilcoxon rank-sum test for interval-level data and the χ² test for nominal- or ordinal-level data. Differences in toxicity proportions were tested using the χ² test. To address findings from the original trial data coupled with differences in baseline demographics, we conducted subanalyses by age and sex based on the same age categories as described in Ramalingam et al.^28,29 Analyses were performed using SAS 9.2 (SAS, Cary, NC).

The likelihood of hospitalization was evaluated using logistic regression models. Associations between the number of hospitalizations and total hospital days and treatment group were evaluated using Poisson regression modeling with the canonical log-link function and an offset parameter to accommodate the effect of unequal length of the individual follow-up periods.³⁰ To evaluate the effect of the addition of bevacizumab, unadjusted and multivariable-adjusted models (controlling for all demographic and clinical characteristics noted above) were estimated for the logistic and Poisson models.

We performed propensity score–adjusted analyses to balance characteristics of patients nonrandomly assigned to CP or CPB by using logistic regression models and controlling for all demographic and clinical characteristics described above.^17,30–36 Diagnostic tests, including checks on the balance of the scores within quintiles and standardized differences,^36,37 were performed to ensure that the propensity score was adequately specified. The propensity score was then used as a covariate adjustment in all models.

Results

We identified 1,109 patients with NSCLC who received CP or CPB as first-line chemotherapy treatment: 911 (82%) and 198 (18%), respectively. Patients who received CPB were younger and more likely to have had stage IV disease and well- or moderately-differentiated tumors (Table 1). Patients were more likely to receive CPB in the later years of the study (P < .001 for trend). Groups were balanced with respect to race/ethnicity, sex, comorbidity score, and the census tract proxy for education. However, patients receiving CPB were significantly (P < .05) less likely to have had a pretreatment diagnosis code for cardiovascular disease, diabetes, or a bleeding disorder.

Table 1.

Baseline Characteristics of Patients With Stages III-IV NSCLC Who Received Carboplatin-Paclitaxel With or Without Bevacizumab

Characteristic	CP (n = 911), No. (%)	CPB (n = 198), No. (%)	P
Age at diagnosis, years
< 60	281 (30.8)	89 (44.9)	< .01
60-64	163 (17.9)	37 (18.7)
65-69	170 (18.7)	37 (18.7)
70-74	143 (15.7)	20 (10.1)
≥ 75	154 (16.9)	15 (7.6)
Sex
Female	463 (50.8)	102 (51.5)	.86
Male	448 (49.2)	96 (48.5)
Race/ethnicity
White	691 (75.9)	155 (78.3)	.44
Hispanic	38 (4.2)	< 6
Black	69 (7.6)	11 (5.6)
Asian/Pacific Islander	94 (10.3)	22 (11.1)
Other	19 (2.1)	< 6
College degree or greater, census tract quintile
1 (lowest)	191 (21.0)	30 (15.2)	.22
2	173 (19.0)	49 (24.7)
3	184 (20.2)	39 (19.7)
4	179 (19.6)	42 (21.2)
5 (highest)	184 (20.2)	38 (19.2)
Median income, census tract quintile
1 (lowest)	185 (20.3)	36 (18.2)	.62
2	179 (19.6)	43 (21.7)
3	177 (19.4)	46 (23.2)
4	183 (20.1)	38 (19.2)
5 (highest)	187 (20.5)	35 (17.7)
Modified Charlson comorbidity score
0	505 (55.4)	121 (61.1)	.17
1	227 (24.9)	49 (24.7)
≥ 2	179 (19.6)	28 (14.1)
AJCC stage at diagnosis
IIIB	189 (20.7)	30 (15.2)	.07
IV	722 (79.3)	168 (84.8)
Level of differentiation (tumor grade)
Well-moderately	110 (12.1)	44 (22.2)	< .01
Poorly/undifferentiated	201 (22.1)	32 (16.2)
Unknown	600 (65.9)	122 (61.6)
Year of diagnosis
2005	164 (18.0)	14 (7.1)	< .01
2006	162 (17.8)	27 (13.6)
2007	136 (14.9)	40 (20.2)
2008	158 (17.3)	48 (24.2)
2009	141 (15.5)	35 (17.7)
2010	150 (16.5)	34 (17.2)
Other pretreatment comorbidities
Cardiovascular disease*	156 (17.1)	20 (10.1)	.01
COPD	216 (23.7)	53 (26.8)	.36
Diabetes	113 (12.4)	15 (7.6)	.05
Bleeding disorder†	99 (10.9)	11 (5.6)	.02
Other‡	156 (17.3)	25 (12.6)	.11
Had at least 1 hospitalization within 180 d after chemotherapy initiation	310 (34.0)	38 (19.2)	< .0001
Hospitalization ended in death	50 (5.5)	6 (3.0)	.15

NOTE. Any table results for which n = 1 to 5 were marked as “< 6.”

Abbreviations: COPD, chronic obstructive pulmonary disease; CP, carboplatin-paclitaxel; CPB, carboplatin-paclitaxel-bevacizumab; NSCLC, non–small-cell lung cancer.

^*Cardiovascular disease includes myocardial infarction, congestive heart failure, cardiovascular disease, ischemic heart disease, atrial fibrillation, and angina.

^†Bleeding disorder includes hemoptysis, hemorrhagic diathesis, anticoagulation disease, and stroke.

^‡Other comorbid conditions include liver disease, plegia, renal disease, rheumatologic disease, dementia, and pulmonary vascular disease.

Approximately 57% and 53% of CPB and CP patients, respectively, had evidence of any toxicity event (P = .40; Table 2). Patients receiving CPB were more likely to have had a hemorrhage, proteinuria, and GI perforation event (all P < .05). Among hemorrhage events, they were more likely to have had epistaxis (4.5% v 1.3%, P < .05). Patients receiving CPB had a higher percentage of hemoptysis and lower percentage of CNS bleeds, but neither reached statistical significance.

Table 2.

No. and Rate of Coded Toxicities Associated With Patients With Stages III-IV NSCLC Who Received Carboplatin-Paclitaxel With or Without Bevacizumab

Toxicity	P		CPB		χ2 P
	No. (%)	Rate per 1,000 Patients	No. (%)	Rate per 1,000 Patients
Any event	485 (53.2)	532.4	112 (56.6)	565.7	.39
Hemorrhage events (overall)	80 (8.8)	87.8	29 (14.6)	146.5	.01
Hemoptysis	25 (2.7)	27.4	8 (4.0)	40.4	.33
Epistaxis	12 (1.3)	13.2	9 (4.5)	45.5	< .01
CNS bleed	< 6	< 6	0 (0)	0	.59
Neutropenia	77 (8.5)	84.5	14 (7.1)	70.7	.52
Venous thromboembolic events	82 (9.0)	90.0	15 (7.6)	75.8	.52
Asthenia	0	0	0	0	NA
Anemia	200 (22.0)	219.5	36 (18.2)	181.8	.24
Thrombocytopenia	42 (4.6)	46.1	7 (3.5)	35.4	.50
Hyponatremia	68 (7.5)	74.6	9 (4.5)	45.5	.14
Proteinuria	< 6	< 6	< 6	< 6	.02
GI perforation	< 6	< 6	< 6	< 6	.02
Newly diagnosed hypertension*	77 (8.5)	84.5	22 (11)	111.1	.23

NOTE. Toxicities were identified via International Classification of Diseases (ed 9) codes that were associated with hospital or ambulatory events that occurred within 180 days after chemotherapy initiation. Any table results for which n = 1 to 5 were marked as “< 6.”

Abbreviations: CP, carboplatin-paclitaxel; CPB, carboplatin-paclitaxel-bevacizumab; NA, not applicable; NSCLC, non–small-cell lung cancer.

^*Newly diagnosed hypertension was defined as any hypertension diagnosed within 180 days after first chemotherapy treatment that was not previously identified prior to the NSCLC diagnosis.

In subanalyses among patients less than 70 years, CPB patients had a significantly higher percentage of GI perforations (1.8% v 0.0%; P < .01). In patients age ≥ 70 years, percentages of patients with bleeding overall (and epistaxis in particular) or a proteinuria event were higher in CPB patients (all P < .05), but GI perforation events were not. In subanalyses by sex, percentages of patients with an epistaxis or proteinuria event were higher (both P < .05) in female CPB patients, whereas all other toxicity event comparisons were nonsignificant. Increased proteinuria events in the CPB group may be due to more frequent monitoring in patients receiving bevacizumab. No significant differences for any toxicity were identified between male CP and CPB patients.

Overall, 34% and 19% of CP and CPB patients, respectively, had at least one hospitalization (P < .001). Specifically, 310 CP patients had 438 hospitalizations for a total of 2,360 hospital days, and 38 CPB patients had 62 hospitalizations for a total of 282 total hospital days. Six percent and 3% of CP and CPB patients, respectively, had a hospitalization that ended in death. Among patients who had a hospitalization, no significant differences were found between the groups with respect to age, race/ethnicity, sex, comorbidity score, year of diagnosis, stage and tumor grade (all P > .05, data not shown).

With the exception of the respiratory system, the most frequent MDC for CP patients involved the nervous system (14.0%), and the most frequent MDC for CPB patients involved the digestive system (16.1%; Appendix Table A1, online only). Other common MDCs varied by CP versus CPB status.

Consistent with the descriptive statistics shown in Table 1, the unadjusted logistic and Poisson models noted that CPB patients were 54% less likely to experience a hospitalization within 180 days of initiating treatment (odds ratio [OR], 0.46; 95% CI, 0.32 to 0.67; Table 3). In addition, the unadjusted rate ratios (RRs) associated with the number of hospitalizations and the total hospital days, within 180 days of chemotherapy initiation, were also significantly lower for CPB patients (RR, 0.61; 95% CI, 0.47 to 0.80 and RR, 0.51; 95% CI, 0.45 to 0.58, respectively). However, the likelihood of a hospitalization, the RR of hospitalizations, and RR of total hospital days did not change in any meaningful way, after adjusting for observable characteristics (including age and disease-specific comorbidities) in both multivariable and propensity score models. Specifically, the propensity-adjusted logistic model identified that CPB patients were 52% (v 54% in the unadjusted model) less likely to have had a hospitalization and had a statistically lower number of hospitalizations and total hospital days. CPB patients had 38% (v 39% in the unadjusted model) and 47% (v 49% in the unadjusted model) lower risks of hospitalization and total hospital days, respectively.

Table 3.

Hospitalizations and Total Hospital Days for Patients With Stages III/IV NSCLC Who Received Carboplatin-Paclitaxel With or Without Bevacizumab

Measure	Percentage of Patients With a Hospitalization, CP and CPB*	Likelihood of Hospitalization, Odds Ratio (95% CI)	Count of Hospitalizations, Rate Ratio (95% CI)	Total Hospital Days, Rate Ratio (95% CI)
Unadjusted model	34.0 and 19.2	0.46 (0.32 to 0.67)	0.61 (0.47 to 0.80)	0.51 (0.45 to 0.58)
Multivariable model†		0.46 (0.30 to 0.69)	0.63 (0.48 to 0.83)	0.54 (0.47 to 0.61)
Propensity score adjusted model‡		0.48 (0.32 to 0.71)	0.62 (0.47 to 0.82)	0.53 (0.47 to 0.60)
Subgroup analyses
Males
Unadjusted model	33.3 and 21.9	0.56 (0.33 to 0.94)	0.74 (0.52 to 1.06)	0.60 (0.51 to 0.70)
Propensity score‡		0.61 (0.35 to 1.06)	0.78 (0.53 to 1.13)	0.65 (0.55 to 0.78)
Females
Unadjusted model	34.8 and 16.7	0.38 (0.21 to 0.65)	0.49 (0.33 to 0.74)	0.44 (0.36 to 0.53)
Propensity score‡		0.38 (0.21 to 0.67)	0.50 (0.32 to 0.75)	0.43 (0.36 to 0.52)
< 70 yr
Unadjusted model	34.9 and 17.8	0.41 (0.26 to 0.62)	0.56 (0.42 to 0.76)	0.48 (0.42 to 0.56)
Propensity score‡		0.44 (0.28 to 0.69)	0.60 (0.44 to 0.82)	0.53 (0.46 to 0.61)
≥ 70 yr
Unadjusted model	32.3 and 25.7	0.72 (0.33 to 1.60)	0.74 (0.40 to 1.37)	0.63 (0.48 to 0.83)
Propensity score†		0.66 (0.29 to 1.50)	0.69 (0.37 to 1.30)	0.56 (0.42 to 0.74)

Abbreviations: AJCC, American Joint Committee on Cancer; COPD, chronic obstructive pulmonary disease; CP, carboplatin-paclitaxel; CPB, carboplatin-paclitaxel-bevacizumab; NSCLC, non–small-cell lung cancer.

^*Referent group was set to CP.

^†Multivariable model adjusted for age group, sex, race/ethnicity, health plan, tumor grade, census education proxy, AJCC stage, cardiovascular disease, COPD, diabetes, bleeding disorders, other comorbid conditions, and year of diagnosis.

^‡Propensity score added as a covariate in the model. The propensity of receiving CPB was estimated using a multivariable logistic regression model that included age group, sex, race/ethnicity, health plan, tumor grade, census education proxy, AJCC stage, cardiovascular disease, COPD, diabetes, bleeding disorders, other comorbid conditions, and year of diagnosis.

Similar results were found in the subanalyses by sex and within age groups. For females, and in both age groups (< 70 years v ≥ 70 years), CPB patients had statistically significant lower likelihoods of hospitalization (RR, 0.38; 95% CI, 0.21 to 0.67 and RR, 0.44; 95% CI, 0.28 to 0.69, respectively). However, for females, no differences were found between the unadjusted and the adjusted models, and for males, the likelihood of hospitalization increased slightly but was no longer statistically significantly different from the CP patients. For patients age < 70 years, relative to the unadjusted model, the odds of a hospitalization for CPB patients increased slightly in the multivariable and propensity-adjusted models (OR, 0.41; 95% CI, 0.25 to 0.62 v OR, 0.44; 95% CI, 0.28 to 0.69) but for patients age ≥ 70, no statistically significant differences were found.

Discussion

Our study examined the impact of adding bevacizumab to CP on toxicity profiles and hospital-related events in adult patients with NSCLC who received care in a community-based setting. Consistent with previous clinical trial findings,^6,8,9 we found that CPB patients were more likely to have experienced a toxicity event. However, CPB patients were less likely to experience a hospitalization within 180 days after initiation of first-line chemotherapy. More importantly, we rejected the null hypothesis that in the presence of a higher toxicity burden, the likelihood of a hospitalization after the initiation of bevacizumab would increase after the adjustment for observable factors, including age and comorbidity status. To our knowledge, this is the first study outside of a clinical trial setting to examine these outcomes in patients with NSCLC.

We found that CPB patients were more likely to experience hemorrhagic events (epistaxis in particular), proteinuria, and GI perforation. Sandler et al reported higher rates of hypertension, proteinuria, bleeding, neutropenia, thrombocytopenia, hyponatremia, rash, and headache in CPB patients.⁷ Our toxicity percentages appeared higher than those reported in the phase III Eastern Cooperative Oncology Group trial as we identified numerically higher percentages of CPB patients with newly identified hypertension (11% v 8%) and venous thromboembolism (8% v 5%).⁷ Compared with our results, Cohen et al identified numerically higher percentages of CPB patients who experienced proteinuria (3% v 2%) and neutropenia (27% v 7%),⁵ and Crino et al⁸ reported numerically lower percentages of both epistaxis (0.8% v 4.5%) and neutropenia (3.8% v 7.1%). These discrepancies, although not necessarily statistically significant, may be explained by the trial setting of their studies in which direct physician reports and/or chart reviews were used to identify toxicity events, and the trial inclusion/exclusion criteria limited participation of patients with toxicity risk factors, such as hypertension, thromboembolic disease, and coagulopathy.^5,7

Although our focus was not specific to end-of-life care, our estimates of hospitalization use are consistent with Warren et al,³⁸ who found that SEER-Medicare longer term survivors (> 6 months after diagnosis) of NSCLC (with and without receipt of chemotherapy) had at least one hospitalization in the last month of life. In addition, Paessens et al studied 273 patients with either NSCLC or lymphoproliferative disorder and found that 50% of chemotherapy cycles were associated with a significant toxicity and 37% of cycles were associated with a hospitalization.³⁹ Another retrospective study of utilization and costs associated with adult patients with metastatic lung cancer (not specific to NSCLC) found that 72% of patients had a hospitalization after the initiation of chemotherapy.¹⁵ Although the characteristics and inclusion/exclusion criteria used in our study are different from these studies, our findings do confirm that a large proportion of patients with advanced lung cancer are hospitalized after initiating a variety of chemotherapy regimens. Our study adds to the literature by providing the distribution of outcomes by chemotherapy regimen.⁴⁰

Although somewhat counterintuitive relative to our toxicity findings, our study found that patients who received CPB had significantly lower risks of hospitalizations and total hospital days in the 180 days after starting chemotherapy, in both the unadjusted and adjusted models. This is supported by the Hassett et al finding that the majority of hospitalizations experienced by patients with cancer after the initiation of chemotherapy were not chemotherapy related.⁴¹ In addition, our previous study that used the same cohort of patients with NSCLC found an overall survival benefit for patients receiving CPB (adjusted hazard ratio, 0.79; 95% CI, 0.66 to 0.94) relative to CP alone.¹⁹ It is possible that the CP patients in our cohort experienced a higher number of hospitalizations that were associated with progression of disease and end-of-life care.³⁸ Nonetheless, with respect to the risk of hospitalization, our findings provide reassurance to community oncologists that the use of bevacizumab for advanced NSCLC is not associated with a higher risk of hospitalization or unexpected toxicity.

Our study has many strengths: use of tumor registries to characterize NSCLC stage and histology, a wealth of demographic and clinical electronic health record data to describe patient characteristics, validated chemotherapy treatment-related data, and complete capture of both ambulatory and inpatient events. However, our findings should be interpreted with the limitations inherent to observational studies.⁴² We used a variety of propensity score–adjusted models specifically designed to address selection bias in observational oncology studies,³¹ and our findings were robust to all model specifications, including those limited to adults < 70 years old and women. Despite our efforts, the role of unaccounted selection factors such as performance status or level and location of disease burden in determining who received CPB is unknowable. Although we did capture and adjust for noncancer comorbidities and other factors that may be related to clinician and patient selection of therapy, they are not a perfect proxy for performance status.⁴³ However, our results and their role in better understanding the potential magnitude of selection bias on outcomes such as hospitalization will be critically important as greater emphasis is placed on comparative-effectiveness research related to alternative cancer treatments.^16,44–46

Consistent with a retrospective observational study by Hershman et al³¹ that used SEER-Medicare data to examine use and toxicity of bevacizumab, we identified bevacizumab-related toxicity events via ICD-9 codes extracted from administrative databases, thus potentially limiting our ability to describe the details or severity of these events. We did not perform chart reviews to determine the severity or grade of an identified toxicity event. We also did not examine variation in the length of stay or costs associated with the hospitalizations.

Hospitalization is the largest single component of spending for cancer care, accounting for the majority of all spending both at the time of cancer diagnosis and in the end-of-life period, and chemotherapy toxicity management is often related to these hospitalizations.¹⁴ As noted repeatedly in a 2013 Institute of Medicine report, Delivering Affordable Cancer Care in the 21st Century–Workshop Summary,¹⁶ as additional novel and expensive chemotherapies are diffused beyond clinical trials and into community practices, patients and providers need pertinent, real-world outcome data, including data on complications and differences in the risk of hospitalization, in order to improve their understanding of the effects of their treatment decisions. Although our study begins to address these gaps with respect to the use of the chemotherapy regimens described here, future research should address differences in regimen-specific costs with respect to downstream outcomes, toxicities and complications, including hospitalizations. In addition, the inclusion of patient-reported preferences and outcomes that can be linked to treatment and cost data may be critical to understanding patient and physician decision making when balancing effectiveness, cost, and toxicity.

Appendix

Table A1.

Hospitalization Rates by Major Diagnostic Categories in Patients With Stage III/IV NSCLC Who Received CP With or Without Bevacizumab

MDC Description	CP (n = 438 hospitalizations) Rate per 100 Patients	CPB (n = 62 hospitalizations) Rate per 100 Patients
Respiratory system	16.4	9.1
Nervous system	6.8	3.5
Digestive system	4.6	5.1
Circulatory system	3.1	1.0
Musculoskeletal system and connective tissue	1.5	3.5
Infectious and parasitic DDs	2.9	1.0
Blood, blood-forming organ, immunology system	2.2	1.5
Endocrine, nutritional, metabolic	2.1
Unknown	2.5
Hepatobiliary-pancreas system	1.2	1.5
Kidney, urinary tract	1.0
Ears, nose, mouth, throat		1.0
Burns		1.0
Female reproductive system		1.0

Abbreviations: CP, carboplatin-paclitaxel; CPB, carboplatin-paclitaxel-bevacizumab; NSCLC, non–small-cell lung cancer.

Authors' Disclosures of Potential Conflicts of Interest

Disclosures provided by the authors are available with this article at jop.ascopubs.org.

Author Contributions

Conception and design: Nikki M. Carroll, Thomas Delate, Mark C. Hornbrook, Lawrence Kushi, Debra P. Ritzwoller

Provision of study materials or patients: Thomas Delate

Collection and assembly of data: Nikki M. Carroll, Thomas Delate, Mark C. Hornbrook, Lawrence Kushi, Debra P. Ritzwoller

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Use of Bevacizumab in Community Settings: Toxicity Profile and Risk of Hospitalization in Patients With Advanced Non-Small-Cell Lung Cancer

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jop.ascopubs.org/site/misc/ifc.xhtml.

Nikki M. Carroll

No relationship to disclose

Thomas Delate

No relationship to disclose

Alex Menter

No relationship to disclose

Mark C. Hornbrook

No relationship to disclose

Lawrence Kushi

No relationship to disclose

Erin J. Aiello Bowles

No relationship to disclose

Elizabeth T. Loggers

No relationship to disclose

Debra P. Ritzwoller

No relationship to disclose