Transportability Without Positivity: A Synthesis of Statistical and Simulation Modeling

Area of Science:

• Causal inference and transportability in observational and randomized studies.
• Statistical modeling and simulation for addressing real-world data limitations.

Background:

• Estimates from studies often need to be generalized to a target population not fully represented in the sample.
• Transportability methods typically assume positivity, meaning all target population covariate patterns are present in the study sample.
• Strict eligibility criteria in trials can violate the positivity assumption, limiting generalizability.

Purpose of the Study:

• To address violations of the positivity assumption in causal inference when generalizing study estimates to a target population.
• To propose and evaluate a novel synthesis approach combining statistical and simulation models as an alternative to population or covariate restriction.

Main Methods:

• Developed a synthesis approach integrating statistical and simulation models to overcome positivity violations.
• Proposed g-computation and inverse probability weighting estimators within the synthesis framework.
• Compared the synthesis approach with traditional restriction methods using a simulation experiment and a real-world example.

Main Results:

• The proposed synthesis approach accurately addressed the research question in both simulation and the illustrative example.
• Traditional methods of restricting the target population or covariate set failed to accurately address the motivating research question.
• The synthesis approach demonstrated superior performance in handling positivity violations compared to restriction methods.

Conclusions:

• Model synthesis offers a viable strategy for causal inference when faced with imperfect target population information and positivity violations.
• Combining empirical data with external knowledge through thoughtfully selected simulation models is crucial for successful transportability.
• This approach enhances the reliability of public health decisions based on study findings.