Early dissemination of bevacizumab for advanced colorectal cancer: a prospective cohort study

CanCORS is a prospective, observational, population- and healthcare systems-based cohort study designed to determine how characteristics of cancer patients, providers, and health care organizations influence treatments and outcomes in newly diagnosed lung or colorectal cancer patients [8]. The full CanCORS cohort appears representative of the US Surveillance, Epidemiology, and End Results (SEER) registry sample: CanCORS participants did not differ from their corresponding SEER population by more than 8 percentage points in any important demographic characteristics such as gender, race, age, and stage [9]. Patients with colorectal cancer were nationally enrolled from geographically diverse populations and health care systems, including five integrated health care systems in the NCI-funded Cancer Research Network, and fifteen Veterans Administration Hospitals [8]. Primary data were collected from patient surveys and medical records. Human subject committees approved the study protocol at each participating site.

Eligible patients were at least 21 years old and were identified within 3 months of diagnosis of colorectal cancer. Three thousand six hundred sixty-four incident cases of colorectal cancer (Figure 1) were enrolled in the CanCORS cohort between September, 2003 to January, 2006, including 742 patients (20%) with advanced colorectal cancer (677 with stage IV cancer at diagnosis, and 65 with a metastatic or regional recurrence within the 15-month post-diagnosis follow-up period of the study). Of note, this stage distribution (20% with advanced cancer) is consistent with that seen in the SEER registry, where patients with advanced colorectal cancer comprise 19% of all patients [10]. Of the 742 patients in our cohort with stage IV or recurrent disease, 355 initiated first-line treatment after the bevacizumab approval date in February, 2004 and were included in this analysis. Hence, the subcohort of interest for this analysis began treatment from March, 2004-January, 2006. All patients or their surrogates provided informed consent at enrollment. Recruitment materials and patient surveys were translated into Spanish and Chinese.

Medical records data were abstracted by trained abstractors at each of the data-collection sites. The time window for records abstraction was three months before diagnosis to fifteen months after diagnosis. Data abstractors had professional experience as cancer registrars or nurses and underwent an intensive four-day training course on data collection processes and standards in CanCORS. Quality assurance was monitored by completion of up to six gold-standard reference chart abstractions per abstractor.

Presence of comorbid illness was collected as part of the medical record abstraction. The degree of comorbidity was scored using the Adult Comorbidity Evaluation 27 (ACE-27), a 27-item index developed to provide prognostic information for cancer patients [11].

Primary data were collected from patients via computer-assisted telephone interviews 4-7 months after diagnosis. For patients who had died or were too ill to be interviewed, a surrogate (relative or household member) familiar with their cancer care was interviewed. The surveys (available at http://​www.​cancors.​org/​public) assessed patients' sociodemographic characteristics (age, race/ethnicity, annual income), insurance coverage, comorbid conditions, and beliefs about cancer care (see Additional file 1) [12].

Surveys were the primary source of insurance information; medical records provided a secondary source. We categorized patients by whether or not they received care in an integrated health system, defined as one of the 5 participating health maintenance organizations, a California Kaiser Permanente plan, or one of the participating Veterans Administration hospitals [8]. Patients were also classified by US region (South, including sites in Alabama, Mississippi, Tennessee, Texas, Georgia; Atlantic, including sites in Massachusetts, Maryland, New York, and North Carolina; and West/Midwest, including sites in Arizona, California, Oregon, Washington, Iowa, Illinois, Indiana, Michigan, Minneapolis, Indiana).

First-line therapy with bevacizumab was defined as receipt of bevacizumab within 8 weeks of starting any chemotherapy for stage IV or recurrent colorectal cancer. For instance, a patient who began metastatic colorectal cancer treatment with cytotoxic chemotherapy, then had bevacizumab added to the regimen 6 weeks later was categorized as receiving first-line bevacizumab. Rates of first-line bevacizumab use pertain to the period from March, 2004 (the month after FDA approval) to January, 2006. To evaluate the relationship between diffusion of bevacizumab and calendar time since FDA approval, we analyzed quarterly usage. Since CanCORS participants were enrolled as a fixed cohort, and the rate of enrollment in CanCORS did not remain constant over the study period, the denominator of eligible patients per quarter (any patients starting any first-line chemotherapy) decreased over time.

The FDA approved bevacizumab based on the results of the pivotal phase III study where bevacizumab was delivered in combination with 5-fluorouracil and irinotecan;[1] however, approval was for use "with intravenous 5-fluorouracil-based chemotherapy," without specifying use of additional agents, such as irinotecan or oxaliplatin [13]. To determine how strictly practitioners interpreted trial data and FDA indication, we described which other drugs were delivered in combination with first-line bevacizumab.

We calculated descriptive statistics summarizing sociodemographic, comorbidity data, and survey-based patient preferences and beliefs. We evaluated factors associated with treatment using logistic regression. Results of multivariable modeling are presented as odds ratios (OR), two-sided p-values, and 95% confidence intervals (with associated confidence interval plots) [14]. Variables were selected for inclusion in multivariable models using a combination of statistical selection criteria (e.g., sufficient variation across values of the outcome, p < 0.25 in unadjusted analyses) and a priori scientific interest (race/ethnicity and comorbidity). Multiple imputation was used to address item nonresponse for survey-based variables and was performed centrally by the CanCORS Statistical Coordinating Center [15]. The presented results from multivariable models incorporate formal imputation adjustments [16]. Data analysis was conducted at the Durham Veterans Administration Medical Center, the coordinating site for Veterans Administration hospitals participating in CanCORS. This analysis used CanCORS core data (version 1.9), medical record data (version 1.9) and patient survey data (version 1.8). Statistical analyses were performed using SAS for Windows Version 9.2 (SAS Institute, Cary, NC).

Three hundred fifty-five patients met inclusion criteria for this analysis (Figure 1). Their characteristics are detailed in Table 1; 66% percent were male, 18% were ≥ 75 years old, and 62% were white. Seventy-three percent had no or mild comorbidity (ACE-27 score of 0-1). Since the most common contraindications to bevacizumab are related to cardiovascular disease, we described incidence of cardiovascular disease based on data from the ACE-27 (Table 1). Forty-nine percent of patients had at least one cardiovascular risk factor. Thirty-six percent had private insurance, 13% had public insurance, 27% had Medicare plus supplemental insurance, and 19% had Veterans Administration healthcare. Of those with only public insurance, 64% had Medicare, 62% had Medicaid, and 33% had both.

Of the 355 patients in this cohort eligible to receive bevacizumab from March, 2004 (after FDA approval) to January, 2006, 31% received the drug as a component of their first-line systemic treatment for stage IV or recurrent disease (Figure 1). Figure 2 displays the quarterly use of bevacizumab 9 months before FDA approval to 12 months after FDA approval of bevacizumab. Only 20% of patients starting first line therapy for metastatic colorectal cancer received first-line bevacizumab in March 2004, the month after FDA approval. In the year following approval, the highest percentage of use was seen during June-August 2004, specifically in July 2004 (48%), a month after publication of the pivotal phase III study which reported the drug's efficacy [1].

The FDA approved bevacizumab for use with intravenous 5-fluorouracil-based chemotherapy. Seventy-five percent of patients received bevacizumab with intravenous 5-fluorouracil (with or without other chemotherapy), in accordance with the FDA indication (Table 2). Thirteen percent received bevacizumab with capecitabine (with or without other chemotherapy).

Only 20% of those who received first-line bevacizumab received it with irinotecan--19% with 5-fluorouracil, as described in the pivotal phase III study, and 1% with capecitabine (Table 2). Fifty-three percent received bevacizumab in combination with oxaliplatin. Eight percent received bevacizumab with 5-fluorouracil alone; 5% with capecitabine alone; and 14% with "other" agents. No patients received first-line bevacizumab as a single agent.

Twenty-six percent of men received first-line bevacizumab (Table 3). In the multivariable model (Figure 3), men were significantly less likely to receive first-line bevacizumab than women (adjusted OR 0.55; 95% CI 0.32-0.93; p = 0.026). Bevacizumab use declined with age: first-line bevacizumab was used by 41% of those < 55 years of age, 32% of those 55-64 years of age, 30% of those 65-74 years of age, and 16% for those ≥ 75 years of age. Patients ≥ 75 years old were significantly less likely to receive first-line bevacizumab than younger (< 55 years) patients (adjusted OR 0.13; 95% CI 0.04-0.46; p = 0.001). Patients in the Atlantic region were more likely to receive bevacizumab than those in the West-Midwest (adjusted OR 2.51; 95% CI 1.37-4.59; p = 0.003), and patients in the South region also tended to receive bevacizumab more often than those in the West-Midwest (adjusted OR 2.16; 95% CI 1.17-4.00; p = 0.014).

Patients' race was not significantly associated with receipt of bevacizumab, nor was insurance status. To assess the effect of organization of care, apart from insurance type, we substituted health system for insurance in the adjusted analyses. No significant association was seen between health system type and receipt of first-line bevacizumab, and none of the other associations were substantially altered. The relationship between patients' income and receipt of bevacizumab was not significant.

ACE-27 comorbidity score was not significantly associated with receipt of first-line bevacizumab. Since cardiovascular comorbidity presents a potential contraindication, we also examined the specific impact of cardiovascular comorbidity on bevacizumab use. In the unadjusted analysis, we found no statistically significant association between bevacizumab use and the presence of cardiovascular disease. Patients' knowledge, belief, and attitudes were also not statistically significantly associated with receipt of bevacizumab (Table 3). However, due to missingness, these variables were not included in the final analytic model.