A Two-Step Framework for Validating Causal Effect Estimates.

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Comparing causal effects from observational data to randomized controlled trials (RCTs) is crucial for validity assessment.
• Challenges arise from differing data generation mechanisms and unknown treatment assignment in observational studies.
• Confounding and sampling bias can compromise causal inference from observational data.

Purpose of the Study:

• To propose a novel two-step framework for validating causal effect estimates derived from observational data.
• To adjust for both the unknown treatment assignment mechanism and varying sampling mechanisms.
• To enhance the reliability of causal inference in real-world health research.

Main Methods:

• Developed an estimator for causal effects accounting for treatment assignment and sampling mechanisms.
• Constructed a two-step framework for comparing causal effect estimates.
• Implemented the framework in an R package named RCTrep for practical application.

Main Results:

• Simulation studies demonstrated that the proposed framework yields causal effect estimates from observational data similar to those from RCTs.
• A real-world application successfully validated adjuvant chemotherapy treatment effects using registry data.
• The framework effectively addresses biases inherent in comparing observational and RCT data.

Conclusions:

• The developed framework facilitates robust comparison of causal effect estimates between observational and RCT data.
• This approach significantly aids in assessing the validity of causal inference derived from observational studies.
• The RCTrep package provides a practical tool for researchers to implement this validation methodology.