Veridical Causal Inference using Propensity Score Methods for Comparative Effectiveness Research with Medical Claims

Area of Science:

• Biomedical research
• Health data science
• Observational studies

Background:

• Medical insurance claims are increasingly used for biomedical research due to their longitudinal nature and large patient populations.
• Claims data present unique challenges, including selection bias, missing data, and the observational nature of studies, hindering reproducible comparative findings.
• These limitations contribute to a crisis in the reproducibility and replication of comparative treatment effect findings derived from medical claims data.

Purpose of the Study:

• To offer practical guidance for the analytical process of using medical insurance claims data.
• To demonstrate methods for estimating causal treatment effects using propensity score methods for various outcome types.
• To enhance transparency and reproducibility in reporting results from claims-based investigations.

Main Methods:

• Utilizing propensity score methods to estimate causal treatment effects.
• Applying methods to handle binary, count, time-to-event, and longitudinally-varying outcome measures.
• Illustrating the analytic pipeline with a prostate cancer patient sub-cohort from the Clinformatics Data Mart Database.

Main Results:

• The study provides a practical framework for causal inference from medical claims data.
• Propensity score methods are demonstrated to effectively estimate treatment effects across different outcome types.
• The developed analytic pipeline enhances transparency and reproducibility in claims-based research.

Conclusions:

• Medical insurance claims data can be leveraged for robust causal inference with appropriate analytical strategies.
• The proposed methods and readily implementable code facilitate reproducible research using large-scale claims datasets.
• This work aims to address the reproducibility crisis in comparative findings from medical claims research.