Improving the validity of nonexperimental comparative effectiveness research: The impact of calendar time on prescribing of novel chemotherapeutic therapies for stage III colon cancer

Oxaliplatin was rapidly adopted for stage III colon cancer treatment after FDA approval in 2004. Uncertainty remains regarding oxaliplatin's superiority to the former chemotherapeutic standard in older patients, the most affected population. The relationship between calendar time and treatment receipt during oxaliplatin's dissemination presents a challenging yet rich methodological research opportunity for comparative effectiveness research (CER). Stage III colon cancer patients aged 65+ initiating chemotherapy from 2003-2008 were studied using U.S. population-based cancer registry data linked with Medicare claims. We examine changes in treatment receipt using a novel calendar time-specific (CTS) propensity score (PS), which allows covariate predictive values to change over time. We compare this method and a calendar time instrumental variable (IV) with traditional adjustment to enhance understanding of oxaliplatin effectiveness for reducing cancer mortality, a strong driver of all-cause mortality among stage III patients. PSs for treatment receipt were constructed using logistic models with key components of demographics, tumor substage, grade, and comorbidities. The CTS PS was used to match oxaliplatin-treated and untreated patients within 1-year intervals. The two-level calendar time instrument was anchored at oxaliplatin's approval and based on IV strength and plausibility of assumptions. PS-matched hazard ratios (HR) were estimated using Cox models. Risk differences (RD) were derived from Kaplan-Meier survival curves. CTS PS and IV results were compared with conventional PS-matched estimates. Oxaliplatin use increased considerably during the study timeframe, with 8% receipt in the first time period vs. 52% in the last (N=2800). Channeling by comorbidities, income, and age appeared to change over time. The CTS PS improved covariate balance within calendar time strata and yielded an attenuated estimated benefit of oxaliplatin (HR=0.75) compared with the conventional PS (HR=0.69). The calendar time instrument resulted in 54% compliance (N=2881). The 3-year IV RD (95% confidence interval) was -0.09 (-0.15,-0.03) favoring oxaliplatin; PS-adjusted RD was -0.04 (-0.08,-0.01). All analyses indicated better survival among oxaliplatin-treated patients. These consistent results based on differing assumptions lend plausibility to the conclusion that oxaliplatin retains effectiveness among older stage III patients. In nonexperimental CER of emerging therapies, calendar time's role as a confounder or instrument should be carefully considered.Doctor of Philosoph