Statistical Guardrails for Hybrid-Controlled Trials: Robust to Confounding and Between-Study Heterogeneity.

External evidence from prior trials, registries, and fit-for-purpose real-world data can improve drug development efficiency. Hybrid-controlled designs are particularly appealing for reducing concurrent control enrollment while simultaneously providing internal validity with a randomized control arm. Yet regulatory adoption is limited due to major concerns around bias due to possible differences in characteristics and outcomes between the external data and the trial. To realize the benefits of the hybrid approach without compromising credibility, methodological guardrails are crucial for mitigating bias and enabling valid inference. We assessed eight statistical methods which proactively address differences between external data and trial data. We apply these methods to both a large clinical trial as a case study, as well as within a comprehensive simulation study with continuous outcomes that varied the amount of measured versus unmeasured confounding, the severity of the between-data-source heterogeneity, and the number of external data sources. Results show that two-step strategy, propensity score-based balancing followed by Bayesian dynamic borrowing, consistently delivered the most favorable trade-off between precision gain and bias control. This approach when used with fit-for-purpose external data can provide a robust implementation of the hybrid trial design beyond the narrow set of conditions where there is currently precedent.