Combining experimental and observational data through a power likelihood

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Health Data Science

Background:

• Randomized controlled trials (RCTs) are crucial for causal inference in evidence-based medicine but often lack sufficient statistical power due to small sample sizes.
• Observational data offers large sample sizes but is susceptible to bias from unmeasured confounding, limiting its causal inference capabilities.
• Bridging the gap between RCTs and observational data is essential for robust treatment effect estimation.

Purpose of the Study:

• To propose and validate a power likelihood approach for augmenting RCTs with observational data.
• To enhance the efficiency and statistical power of treatment effect estimation by integrating complementary data sources.
• To provide a data-adaptive method for optimal information regulation from observational data.

Main Methods:

• Development of a power likelihood framework to fuse RCT and observational data.
• Implementation of a data-adaptive procedure to select the optimal learning rate by maximizing the expected log predictive density (ELPD).
• Validation through simulation studies and a real-world data fusion application.

Main Results:

• The proposed method demonstrated increased statistical power compared to using RCT data alone.
• The approach maintained approximate nominal coverage rates, ensuring reliable causal inference.
• A real-world application augmenting the PIONEER 6 trial with health claims data confirmed the method's effectiveness.

Conclusions:

• Augmenting RCTs with observational data using the power likelihood approach improves treatment effect estimation efficiency and power.
• The data-adaptive ELPD maximization provides a robust way to balance information from different data sources.
• This method offers a practical solution for leveraging large-scale observational data in clinical research while mitigating bias.