Leveraging external controls in clinical trials: estimands, estimation, assumptions

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Randomized controlled trials (RCTs) are often augmented with external controls from observational data to assess intervention effects.
• Traditional methods for treatment effect estimation rely on ambiguous causal estimands and strong assumptions like mean exchangeability.
• Implicit biases in causal inference arise from various sources, complicating accurate treatment effect evaluation.

Purpose of the Study:

• To introduce a transparent framework for defining causal estimands using double-indexed potential outcomes notation.
• To develop a novel statistical method for estimating treatment effects by integrating concurrent and external control data.
• To address limitations of existing methods by removing the need for mean exchangeability assumptions.

Main Methods:

• Developed a double-indexed notation for potential outcomes to clearly define causal estimands and identify sources of bias.
• Derived a consistent and locally efficient estimator for weighted average treatment effect (WATE) estimands.
• Proposed a Frisch-Waugh-Lovell style partial regression method to estimate systematic outcome differences between concurrent and external units.

Main Results:

• Demonstrated the critical role of the concurrent control arm in validating assumptions and enabling unbiased causal estimation.
• The proposed estimator successfully combines concurrent and external data without requiring mean exchangeability.
• Simulations and application to cardiovascular trials showed the proposed methods outperform existing approaches.

Conclusions:

• The novel approach offers a more robust and transparent method for causal inference in clinical trials by integrating diverse data sources.
• This framework enhances the reliability of treatment effect estimation when external data is used to augment RCTs.
• The proposed estimator provides a valuable tool for biostatisticians and clinical researchers seeking accurate causal effect evaluation.