Synthesis estimators for transportability with positivity violations by a continuous covariate

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Randomized trials may not reflect the target population, limiting generalizability.
• Positivity assumption is crucial for transporting treatment effect estimates between populations.
• Trial eligibility criteria can violate positivity, creating challenges for external validity.

Purpose of the Study:

• To extend a synthesis approach for handling positivity violations with continuous covariates.
• To develop novel statistical methods for estimating treatment effects in external populations.
• To compare the performance of new estimators against existing methods for nonpositivity.

Main Methods:

• Utilized a synthesis of statistical and mathematical models integrating multiple data sources.
• Proposed two novel augmented inverse probability weighting estimators.
• Extended previous work on binary covariates to continuous covariates for positivity violations.

Main Results:

• The proposed synthesis approach effectively handles positivity violations with continuous covariates.
• Novel estimators demonstrated competitive or superior performance in Monte Carlo simulations.
• The approach was illustrated using antiretroviral therapy data for HIV-positive women.

Conclusions:

• The extended synthesis approach provides a robust framework for addressing nonpositivity in treatment effect estimation.
• The novel estimators offer improved methods for generalizing findings from clinical trials to target populations.
• This work has significant implications for real-world application of evidence-based medicine.