Propensity Score Methods in Rare Disease: A Demonstration Using Observational Data in Systemic Lupus Erythematosus

Area of Science:

• Rheumatology
• Epidemiology
• Biostatistics

Background:

• Observational studies are vital for understanding rheumatic disease natural history, risk factors, and treatment effects from real-world data.
• These studies are susceptible to confounding bias, which can distort findings.
• Propensity scores (PS) are statistical tools designed to balance covariates between treatment and control groups in observational research.

Purpose of the Study:

• To review propensity score methods for observational research.
• To demonstrate the application of propensity score methods for achieving study group balance in rare diseases.
• To illustrate the utility of propensity scores using an example of infection risk in patients with systemic lupus erythematosus (SLE) and hypogammaglobulinemia.

Main Methods:

• Review of propensity score methodology, including matching, stratification, adjustment, and inverse probability weighting.
• Application of propensity score methods to a rare disease cohort.
• Utilizing a case study focusing on the risk of infection in SLE patients with hypogammaglobulinemia.

Main Results:

• Propensity score methods can effectively balance study groups in observational research, even in rare diseases with small sample sizes or low event rates.
• Demonstrated successful application of PS methods in achieving balance for analyzing real-world data in SLE.
• The methods provide a robust approach to mitigate confounding bias in the context of rare rheumatic conditions.

Conclusions:

• Propensity score methods are valuable tools for strengthening the validity of observational studies, particularly in rare diseases like SLE.
• These methods facilitate the estimation of treatment effects and risk factors using real-world data by reducing confounding bias.
• The application in SLE highlights the practical utility of PS for addressing challenges posed by small sample sizes and low event rates.