A tutorial on propensity score estimation for multiple treatments using generalized boosted models.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Propensity scores are widely used in observational studies to control for pretreatment imbalances.
• Inverse probability of treatment weighting (IPTW) with propensity scores from boosted models is effective for two treatments.
• Existing guidance and tools are limited for studies involving three or more treatment groups.

Purpose of the Study:

• To offer step-by-step guidance for implementing propensity score weighting in multiple-treatment scenarios.
• To propose generalized boosted models (GBM) for estimating propensity scores in complex treatment settings.
• To define relevant causal quantities and derive weighted estimators for multi-treatment studies.

Main Methods:

• Utilizing generalized boosted models (GBM) for propensity score estimation.
• Implementing inverse probability of treatment weighting (IPTW) for causal effect estimation.
• Developing tools for assessing covariate balance and treatment group overlap in multi-treatment studies.

Main Results:

• The paper defines causal quantities and derives weighted estimators for multiple treatments.
• A detailed plan is presented for using GBM to estimate propensity scores, weights, and causal effects.
• Methods are demonstrated through a case study on adolescent substance abuse treatment programs.

Conclusions:

• This work extends propensity score methods to handle multiple treatment groups effectively.
• Generalized boosted models (GBM) are proposed as a robust tool for propensity score estimation in complex settings.
• The provided guidance and tools facilitate rigorous causal inference in observational studies with multiple treatments.