Background: Results from single-armed clinical trials can be put into context by comparing them against external controls (ECs) - cohorts consisting of patients derived from an external data source such as real-world data (RWD). Due to the lack of randomization and potential differences in how variables (incl. endpoints) are captured between data sources, EC analyses may yield biased results, leading to incorrect type I error probabilities. EC validation analyses feature both randomised and external comparators, providing an opportunity to assess the magnitude of this bias via comparison of the two control arms.
Objectives: To quantify EC bias in progression-free survival (PFS) outcomes (from all sources) and its between-study variation by conducting a series of EC validation analyses within a set of randomised non-small cell lung cancer (NSCLC) studies. This i) allows for bias adjustment in prospective EC analyses, and ii) yields empirical insights into whether clinical trial PFS outcomes can be replicated using real-world PFS in this setting.
Methods: PFS outcomes from chemotherapy control arms of 12 clinical trials in NSCLC were replicated in the US nationwide de-identified Flatiron Health electronic health record-derived database, with real-world PFS (rwPFS) as the endpoint. A meta-analysis of log-hazard ratios comparing clinical trial and RWD control arms was performed. A previously published method1 was employed to restore correct error probabilities for a prospective example study, using obtained meta-analysis estimates
Results: With the exception of one outlier, rwPFS outcomes in the RWD were on average similar to RCT PFS outcomes, with some between-study variation. RCT and RWD arms differed by a mean logHR of 0.04, with a standard deviation of 0.14 (outlier included). Applying these estimates to adjust error probabilities in a hypothetical prospective EC study revealed that bias in this setting should be adjusted for because it may lead to incorrect decision making.
Conclusions: The good alignment of RCT and RWD outcomes is reassuring, increasing confidence that i) we can use RWD ECs to provide context for single arm clinical trials in this setting, and ii) that rwPFS and PFS endpoints appear to give similar results despite differences in how the endpoints are assessed.
References:
[1] Incerti D, Bretscher MT, Lin R, Harbron C. A meta-analytic framework to adjust for bias in external control studies. Pharm Stat. 2023;22(1):162-180. doi:10.1002/pst.2266
