External Control Arm with Synthetic Real-world Data for Comparative Oncology using Single Trial Arm Evidence (ECLIPSE): A Case Study using Lung-MAP S1400I

Abstract

Single-arm trials supplemented with external comparator arm(s) (ECA) derived from real-world data are sometimes used when randomized trials are infeasible. However, due to data sharing restrictions, privacy/security concerns, or for logistical reasons, patient-level real-world data may not be available to researchers for analysis. Instead, it may be possible to use generative models to construct synthetic data from the real-world dataset that can then be freely shared with researchers. Although the use of generative models and synthetic data is gaining prominence, the extent to which a synthetic data ECA can replace original data while preserving patient privacy in small samples is unclear. Objective: To compare the efficacy of nivolumab + ipilimumab combination therapy ('experimental arm') versus nivolumab monotherapy ('control arm') in patients with metastatic non-small cell lung cancer (mNSCLC) using real-world data from two real-world databases ('original ECA'), and synthetic data versions of these datasets ('synthetic ECA'), with the aim of validating synthetic data for use in ECA analysis. Study design: Non-randomized analyses of treatment efficacy comparing the experimental arm to the (i) original ECA and (ii) synthetic ECA, with baseline confounding adjustment. Data sources: The experimental arm is from the Lung-MAP no-match substudy S1400I (NCT02785952) provided by National Clinical Trials Network (NCTN) in the United States. The real-world data source for the ECA is data from population-based oncology data from the Canadian province of Alberta, and from Nordic countries in Europe, specifically Denmark and Norway.