Continuous-time Marginal Structural Models for Adverse Drug Effects in Pharmacoepidemiology

Analysis of new user cohort studies of adverse drug effects can be based on either intention-to-treat or as-treated paradigms. In the latter case, controlling for time-dependent confounding is necessary, and can be implemented through inverse probability of treatment (IPT) weighted estimation of marginal structural models (MSM). To develop the weights in the context of pharmacoepidemiological administrative data, besides a time-fixed initial dose model, the subsequent time-varying exposure can be decomposed into modeling the time of the dispensations and the dose at each time through a multinomial or multiordinal logistic model, as a marked point process. We compare different approaches for constructing the weights in a simulation study, simulating data from a multistate model. For estimation of the outcome MSM, we propose IPT weighted case-base sampling, which can reduce the size of long-format data sets needed, especially for rare outcomes. The methods are then demonstrated in a study of the effect of glucocorticoid use on fracture risk.M.Sc