Matching Weights to Simultaneously Compare Three Treatment Groups: Comparison to Three-way Matching

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Methods

Background:

• Propensity score matching is widely used but less explored for more than two treatment groups.
• A novel propensity score weighting method, matching weights, is evaluated for multi-group settings.

Purpose of the Study:

• To assess the performance of matching weights against traditional methods in three-treatment group scenarios.
• To compare matching weights, three-way propensity score matching, and inverse probability of treatment weighting (IPTW) via simulation and empirical data.

Main Methods:

• The matching weight method, an extension of IPTW, was simulated and compared to 1:1:1 propensity score matching and IPTW.
• An empirical study analyzing the safety of three analgesics was used to apply and compare the methods.

Main Results:

• Matching weights showed comparable bias but superior mean squared error (MSE) to three-way matching across all simulations.
• Performance benefits of matching weights were greatest with rare outcomes, unequal group sizes, or poor covariate overlap.
• Matching weights achieved better covariate balance and narrower confidence intervals in the empirical example compared to other methods.

Conclusions:

• Matching weights offer improved MSE over three-way matching, particularly in challenging scenarios like difficulty in finding matched subjects.
• The method's scalability to more than three groups makes it a recommended choice for multi-group outcome comparisons, especially with rare outcomes or uneven exposure distributions.