|
Abstract:
|
The analysis of intervention effects in longitudinal naturalistic data, such as data from observational studies and healthcare claims databases, presents multiple statistical challenges. To assess causal effects, researchers must address time-dependent confounders, treatment switching, and data censoring. Newer approaches to producing estimates of causal effects in such settings include marginal structural models (MSM), structural mean models, and longitudinal propensity score subclassification (PSSUB). Each method takes a different statistical approach to addressing these challenges. For instance, MSMs utilize inverse weighting to produce a confounder balanced pseudopopulation while PSSUB uses expected responses within propensity trajectory subgroups. A summary of the statistical challenges and differences in the various methodological solutions will be presented.
|
