Performance of propensity score methods when comparison groups originate from different data sources.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Estimating relative risks from disparate data sources presents methodological challenges.
• Propensity score methods are increasingly used in comparative effectiveness research.

Purpose of the Study:

• To evaluate the performance of propensity score methods for relative risk estimation when exposed and comparison subjects come from different data sources.
• To identify conditions under which these methods perform optimally or encounter bias.

Main Methods:

• Monte Carlo simulations were employed to assess propensity score method performance.
• Scenarios varied the degree of systematic differences between exposed and comparison groups from different data sources.

Main Results:

• Propensity score methods yielded accurate relative risk estimates in most simulated scenarios.
• Severe bias emerged when comparison groups systematically differed on outcome-associated factors, unless the exposed group was similarly restricted.
• Mean squared error was minimized with similarly restricted groups or when the comparison group was a random sample of the exposed group's source population.

Conclusions:

• Propensity score methods demonstrate good performance for relative risk estimation with different data sources in most tested situations.
• Careful consideration of group comparability and potential confounding is crucial for accurate results.