Causal Inference Methods for Combining Randomized Trials and Observational Studies: A Review.

Area of Science:

• Biostatistics
• Epidemiology
• Health Data Science

Background:

• Randomized controlled trials (RCTs) offer high internal validity but limited generalizability.
• Observational studies provide representative data but are susceptible to confounding.
• Integrating both data types is crucial for robust causal inference.

Purpose of the Study:

• To review methods for causal inference using combined RCT and observational data.
• To enhance the generalizability of RCT findings with observational data.
• To improve the unconfoundedness and precision of treatment effect estimates.

Main Methods:

• Review of identification and estimation strategies for combined data analysis.
• Discussion of weighting, conditional outcome models, and doubly robust estimators.
• Comparison of potential outcomes and structural causal models frameworks.

Main Results:

• Methods exist to leverage observational data for RCT generalizability.
• Techniques can improve unconfoundedness and average treatment effect estimation.
• Simulation and real-world data analysis (tranexamic acid in trauma) demonstrate method performance.

Conclusions:

• Combining RCTs and observational studies offers a powerful approach to causal inference.
• The reviewed methods provide a framework for robustly evaluating treatment effects.
• Practical guidance on code and implementations is provided for researchers.