Evaluating and improving a matched comparison of antidepressants and bone density

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Matching is a key technique for covariate adjustment in observational studies to estimate causal effects.
• Current informal methods for assessing covariate balance in matched samples have limitations, including multiplicity issues and ignoring joint covariate distributions.
• The effectiveness of existing diagnostics in identifying significant problems remains unclear.

Purpose of the Study:

• To develop a formal statistical method for assessing covariate balance in matched samples.
• To address the limitations of informal diagnostics by considering both marginal and joint covariate distributions.
• To provide a method that controls false positives while maintaining high power for detecting true imbalances.

Main Methods:

• A novel formal assessment of covariate balance is proposed.
• The method compares marginal and joint distributions of the matched sample against benchmark complete randomizations.
• Statistical control for the probability of falsely identifying imbalance is incorporated.

Main Results:

• The proposed method formally assesses covariate balance, considering both marginal and joint distributions.
• It offers control over the probability of Type I errors (falsely identifying imbalance).
• The method demonstrates a high probability of correctly detecting and identifying significant covariate imbalances.

Conclusions:

• The developed formal assessment provides a robust approach to evaluating covariate balance in matched observational studies.
• This method enhances the reliability of causal effect estimation by addressing limitations of informal diagnostics.
• An R package, 'met', is available to implement this formal covariate balance assessment.