Augmenting external control arms using Bayesian borrowing: a case study in first-line non-small cell lung cancer.

Area of Science:

• Oncology
• Biostatistics
• Health Economics

Background:

• Real-world data (RWD) offers potential for external control arms (ECAs) in clinical trials.
• Augmenting ECAs with RWD presents challenges, including unmeasured confounding and data heterogeneity.
• Bayesian borrowing (BB) is a statistical method to incorporate external information into analyses.

Purpose of the Study:

• To evaluate the application of Bayesian borrowing (BB) methods for augmenting an external control arm (ECA) constructed from real-world data (RWD).
• To improve comparative effectiveness estimates in first-line non-small-cell lung cancer (NSCLC) by addressing challenges with RWD-based ECAs.
• To assess the utility of BB in mitigating biases and enhancing precision when using historical clinical trial data.

Main Methods:

• Constructed an ECA for a first-line NSCLC randomized controlled trial (RCT) using RWD (ConcertAI Patient360™).
• Employed cardinality matching to align patient characteristics between the treatment arm and the ECA.
• Applied Bayesian borrowing (BB) with a static power prior under a Weibull proportional hazards model, varying borrowing weights from 0.0 to 1.0.

Main Results:

• The RWD-derived ECA did not replicate the overall survival (OS) estimates from the matched RCT population.
• Incorporating BB reduced the OS hazard ratio (HR) in the ECA, moving it closer to the RCT's findings.
• Augmenting the RCT control arm with a historical control via BB improved the precision of the HR estimate.

Conclusions:

• RWD-based ECAs may not fully replicate RCT outcomes due to unmeasured confounding and data variations.
• Bayesian borrowing (BB) can enhance the precision of comparative effectiveness estimates.
• BB shows potential as a bias assessment tool and can address limitations of traditional methods when suitable external data is available.