Propensity Score: an Alternative Method of Analyzing Treatment Effects

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Randomized controlled trials (RCTs) ensure balanced patient characteristics for causal inference but may lack external validity.
• Non-randomized trials offer better external validity but risk confounding due to differing patient characteristics.
• Propensity score (PS) methods are increasingly used to analyze non-randomized intervention trials.

Purpose of the Study:

• To present, explain, and illustrate the propensity score method for analyzing non-randomized intervention trials.
• To highlight the advantages of propensity score methods over conventional regression modeling.

Main Methods:

• The propensity score (PS) is the probability of receiving an intervention.
• PS is estimated using logistic regression based on available patient data.
• Treatment effects are estimated using PS matching, inverse probability of treatment weighting (IPTW), stratification, or regression adjustment.

Main Results:

• The propensity score method provides a framework for estimating treatment effects in non-randomized studies.
• Four distinct analytical approaches leverage the propensity score for effect estimation.

Conclusions:

• The propensity score method is a valuable alternative for analyzing non-randomized trials, offering epistemological advantages.
• This method can adjust for known, measured confounding factors.
• Randomized controlled trials remain the gold standard for controlling unknown confounding factors.