Get the most from your data: a propensity score model comparison on real-life data.

Area of Science:

• Medical Statistics
• Bariatric Surgery Research

Background:

• Propensity score analysis is debated regarding its capabilities and limitations.
• Real-world data analysis requires robust statistical methods to manage confounding variables.

Purpose of the Study:

• To examine the effect of propensity score analysis on real-life data.
• To introduce a simple and effective clinical approach using propensity scores.
• To assess weight loss outcomes in bariatric surgery patients.

Main Methods:

• Review of current publications on propensity score analysis.
• Application to a nonrandomized clinical trial of 173 bariatric surgery patients (127 Roux-en-Y gastric bypass, 46 sleeve gastrectomy).
• Analysis of covariate distribution and excess weight loss using Mann-Whitney U, chi-squared, and Student's t-tests.
• Development and application of three propensity score models with matching.
• Post-matching assessment of covariate distribution and weight loss.

Main Results:

• Initial analysis showed significant differences in excess weight loss.
• Propensity score models did not confirm these differences, indicating procedural equivalence in weight loss.
• Matching improved covariate distribution, ensuring greater balance between groups.
• All propensity score models demonstrated that both surgical procedures were comparable for weight loss.

Conclusions:

• A clinical approach using outcome-related covariates as a propensity score base balances covariate distribution and data loss.
• Propensity score models designed with clinical intent are effective in overcoming covariate heterogeneity.
• Propensity score analysis provides a valuable tool for analyzing real-world bariatric surgery data.