Why Are We Weighting? Understanding the Estimates From Propensity Score Weighting and Matching Methods

Area of Science:

• Cardiovascular Surgery
• Biostatistics
• Epidemiology

Background:

• Observational studies use propensity scores to balance patient characteristics when randomization is not feasible.
• Understanding how different propensity score methods influence treatment effect estimates is crucial for accurate interpretation.

Purpose of the Study:

• To explain the treatment effect estimates derived from various propensity score methods in long-term mortality after coronary artery bypass grafting.
• To compare the effectiveness of single internal mammary artery versus bilateral internal mammary artery (BIMA) conduits.

Main Methods:

• Retrospective analysis of 47,984 coronary artery bypass grafting procedures (1992-2014).
• Utilized multivariable Cox regression, 1:1 propensity score matching, and inverse probability weighting (IPW) among treated and overall populations.
• Assessed long-term mortality following single versus BIMA conduit use.

Main Results:

• Multivariable Cox regression and propensity score matching showed BIMA significantly decreased mortality (HR 0.83 and 0.79, respectively).
• IPW among the treated also indicated a protective effect of BIMA (HR 0.83).
• IPW for the overall population suggested a non-significant increased risk of mortality with BIMA (HR 1.08).

Conclusions:

• While most methods show BIMA reduces mortality, IPW in the overall population yielded a non-significant increased risk.
• Discrepancies arise from the different populations weighted by IPW approaches (overall vs. treated).
• The choice of propensity score method significantly impacts treatment effect estimates and should align with the desired population inference.