Trials augmented by external control data using balancing weights: A comparison of estimands and estimators

Area of Science:

• Biostatistics
• Clinical Trials
• Real-World Evidence

Background:

• Randomized controlled trials (RCTs) are the gold standard but can be infeasible for rare diseases.
• Real-world data (RWD) and external controls (ECs) offer alternatives for treatment effect estimation.
• Integrating ECs requires harmonizing data and employing robust statistical methods to address patient characteristic differences.

Purpose of the Study:

• To elucidate potential causal estimands when using propensity scores (PS) with external controls.
• To evaluate the performance of PS-weighted estimators in hybrid or single-arm trials incorporating EC data.
• To assess the feasibility of different estimands for small-scale studies, exemplified by the LIMIT-JIA trial.

Main Methods:

• Utilized propensity score (PS) methodology to summarize and account for differences between trial and EC patients.
• Explored various PS-based estimation methods, including balancing weights, augmented estimators, and Bayesian dynamic borrowing.
• Defined and evaluated different causal estimands in the context of integrating external control data.

Main Results:

• Demonstrated that the interpretability and feasibility of estimating causal estimands vary.
• Showcased that certain estimands are more amenable to estimation than others, particularly for smaller studies.
• Highlighted the importance of careful estimand selection for valid treatment effect estimation using ECs.

Conclusions:

• Propensity score methods are valuable tools for integrating external control data in clinical trials.
• The choice of causal estimand significantly impacts the feasibility and interpretation of results from hybrid trials.
• The findings support the use of ECs in specific contexts, provided appropriate statistical methodologies and estimand definitions are employed.