Robust weights that optimally balance confounders for estimating marginal hazard ratios

Area of Science:

• Epidemiology
• Biostatistics
• Observational Studies

Background:

• Covariate balance is essential for unbiased treatment effect estimation in observational studies.
• Existing methods primarily focus on continuous outcomes, leaving a gap for time-to-event data.
• Estimating marginal hazard ratios is critical for time-to-event outcomes in medical research.

Purpose of the Study:

• To introduce robust orthogonality weights for improved covariate balance.
• To evaluate the performance of these weights in estimating marginal hazard ratios for time-to-event outcomes.
• To apply the method to real-world data from the Women's Health Initiative study.

Main Methods:

• Developed robust orthogonality weights via quadratic constrained optimization.
• Maximized precision while constraining covariate balance (correlation between confounders and treatment).
• Evaluated performance in a simulation study for binary and continuous treatments with time-to-event outcomes.

Main Results:

• Robust orthogonality weights demonstrated effectiveness in balancing covariates.
• The proposed weights improved the estimation of marginal hazard ratios in simulations.
• Successful application to hormone therapy and red meat consumption data.

Conclusions:

• Robust orthogonality weights offer a powerful tool for analyzing time-to-event data in observational studies.
• This method enhances the reliability of treatment effect estimates, particularly marginal hazard ratios.
• The approach is applicable to both binary and continuous treatments and has been validated in a large cohort study.