Evaluating real-world overall survival (rwOS) across oncology data sources: Friends of Cancer Research non-small cell lung cancer pilot study.

e18749 Background: Friends of Cancer Research and five data partners used electronic health record-sourced datasets to conduct parallel analyses in real-world data (RWD). The study assessed estimates of rwOS among patients taking platinum doublet chemotherapy (chemo) or pembrolizumab in combination with chemo (IO+chemo) in first-line treatment for metastatic non-small cell lung cancer (mNSCLC). Methods: The KEYNOTE-189 clinical trial was selected as a reference for cohort selection in the RWD. We applied select available KEYNOTE-189 eligibility criteria to compare rwOS in a homogenous real-world population. Analyses were conducted following a mutually developed statistical analysis plan that was harmonized across data partners. Twelve-month rwOS, and unadjusted and adjusted hazard ratios (HR) for death were calculated using time-to-event methods with patients censored at last structured activity prior to the data cutoff. Results: The five datasets included patients treated with chemo (range of n=155 to 1501) and IO+chemo patients (n=36 to 405, Table). Twelve-month rwOS estimates ranged from 45% to 58% in the chemo arm and 44% to 68% in the IO +chemo arm. The HR for death comparing IO+chemo to chemo varied across groups, ranging from 0.79 (95% CI: 0.68, 0.92) to 1.10 (95% CI: 0.70, 1.72). The HR for death adjusted for age, gender, performance status, and smoking status did not significantly alter results, ranging from 0.80 (95% CI: 0.69, 0.93) to 1.15 (96% CI: 0.71, 1.85). Conclusions: The systematic evaluation of reproducibility and performance of rwOS across five real-world cohorts and the impact of implementing consistent eligibility criteria and methods was useful to understand the consistency of results in real-world populations and relationship between treatments and outcomes in the absence of randomization. A well-articulated research question and a pre-specified analytic plan that includes covariate definitions are necessary preliminary steps to guide data harmonization, standardization, and analysis. A range of observed results highlights the complexity of real-world data, including differences in sample size, underlying patient variability across datasets, and missing covariate data. The harmonization process and observed results underscore the importance of a well-articulated research question and a pre-specified analytic plan to guide data harmonization, standardization, and analysis. [Table: see text]