Genetic matching for time-dependent treatments: a longitudinal extension and simulation study

Area of Science:

• Observational studies
• Health research methodology
• Machine learning in healthcare

Background:

• Longitudinal matching is crucial for mitigating confounding in real-world studies of time-dependent treatments.
• Current methods often require manual, iterative adjustments for covariate balance.
• A novel longitudinal extension of genetic matching automates balancing of covariate histories.

Purpose of the Study:

• To introduce and evaluate a longitudinal extension of genetic matching for observational studies.
• To compare its performance against baseline propensity score matching and time-dependent propensity score matching.
• To assess the automated balancing of covariate histories in time-dependent treatment analyses.

Main Methods:

• A Monte Carlo simulation framework was developed to evaluate comparative performance.
• Simulations involved 1,000 datasets with 1,000 subjects each.
• Three matching methods were applied: baseline propensity score matching, time-dependent propensity score matching, and longitudinal genetic matching.

Main Results:

• Baseline propensity score matching showed significant bias (29.7%-37.2%) with time-dependent confounding.
• Time-dependent propensity score matching and longitudinal genetic matching demonstrated reduced bias (0.7%-13.7%) and improved covariate balance.
• Longitudinal genetic matching performed comparably or better than time-dependent propensity score matching without requiring manual re-specifications or covariate normality assumptions.

Conclusions:

• Longitudinal genetic matching offers a valid and automated approach for analyzing time-dependent treatments.
• The method enhances covariate balance and reduces bias in observational studies.
• This approach supports future real-world assessments of treatments administered at multiple time points.