|
Abstract:
|
This papers presents how causal inference analysis of observational study data subject to time-dependent confounding may be enhanced when members of the cohort under study can be linked to an individual electronic health record (EHR). Given recent support for EHRs among policymakers, we illustrate this potential role using a planned Department of Defense / Veterans Affairs trauma casualty cohort, as an initial example of the broad impact that EHRs might have on clinical research. Applying marginal structural models to time-to-event outcomes can yield unbiased estimates under certain restrictive assumptions, such as no unmeasured confounders. Such assumptions may be more tenable, however, when investigators can access appropriately linked and de-identified data derived from EHRs. The paper concludes with an example application to U.S. Military HIV Natural History Study cohort data.
|
