OBJECTIVES: This study aimed to augment a synthetic control arm (SCA) constructed from real-world data (RWD) using historical clinical trial data and Bayesian borrowing (BB) methods in first-line (1L) non-small cell lung cancer (NSCLC).

METHODS: ConcertAI Patient360™ was used to construct an SCA for a randomized controlled trial (RCT) in 1L NSCLC comparing chemotherapy with or without cetuximab (NCT00946712; individual patient data [IPD] provided by the National Cancer Institute via Project Data Sphere). For the without-bevacizumab subpopulation, patient characteristics were matched between the treatment arm (cetuximab+chemotherapy) and SCA using cardinality matching. Overall survival (OS) was evaluated using Cox proportional hazards (PH) models. The SCA was augmented through BB from historical controls (NCT00540514; pseudo-IPD obtained via digitized published Kaplan-Meier curves). BB was conducted using a static power prior under a Weibull PH parameterization with borrowing weights from 0.0-1.0.

RESULTS: The SCA analysis yielded much higher OS HR estimates (HR [95% CI]: 1.53 [1.21, 1.93]) than those obtained in the matched population of the RCT (HR [95% CI]: 0.91 [0.73,1.13]). BB reduced the estimates from the SCA (HR [95% CrI]: 1.30 [1.08,1.54], borrowing weight=1.0), providing a useful characterization of heterogeneity between SCA and historical control. An exploratory analysis where BB was used to augment the RCT control arm via historical controls yielded greater precision in estimates (HR [95% CrI]: 1.03 [0.86,1.22], borrowing weight=1.0) than those obtained in the matched population of the RCT.

CONCLUSIONS: The SCA using RWD was unable to replicate the OS estimates from the matched population of the RCT, possibly due to unmeasured confounding, differences in time periods and follow-up, and differences in subsequent therapy. BB from historical clinical trials showed potential as a bias assessment tool in the presence of heterogeneity between trials and to increase the precision of treatment effect estimates when minimal heterogeneity is present.