Common methods for missing data in marginal structural models: What works and why
Electronic health records (EHR) are useful for addressing health-related questions, such as estimating the marginal effect of treatment over a long period of time. In practice, a patient’s treatment exposure may not be constant over time but it get updated as their medical history evolves. In turn, the new treatment may affect future health events and individual factors, potentially associated with the outcome of interest.
Marginal structural models (MSMs) have been proposed to estimate marginal effects in this type of settings. The parameters of MSMs are often estimated using inverse-probability-of-treatment-weighting (IPTW) with weights accounting for time-varying confounding through the modelling of the treatment assignment mechanism. A major issue when applying this method is missing data among confounders, where a poorly informed analysis method will lead to biased estimates of treatment effects. Despite several approaches described in the literature for handling missing data in MSMs, there is little guidance on what works in practice and why.
In this presentation, we will review existing missing-data methods for MSMs and discuss the plausibility of their underlying assumptions. In particular, we will focus on the complete case analysis, multiple imputations, the last observation carried forward, the missingness pattern approach and inverse probability of missingness weighting, under three mechanisms for nonmonotone missing data encountered in research based on electronic health record data.
This session is relevant to statisticians and epidemiologists interested in long-term causal treatment effect estimation using electronic health records.
Clemence Leyrat, Assistant Professor, Medical Statistics Department, London School of Hygiene & Tropical Medicine (LSHTM)
Please note that the recording link will be listed on this page when available.