Matching Methods for Comparative Effectiveness Studies of Longitudinal Treatments

Duke University, Durham NC

Investigators

Linked publications & trials

Abstract

? DESCRIPTION (provided by applicant): Comparative effectiveness studies of medical treatments that are initiated during longitudinal, observational follow-up require advanced statistical methods to account for potential bias due to time-dependent confounding, where confounders may influence selection but also mediate treatment effects. Examples of time-dependent treatment include cardioversion at some point after diagnosis with atrial fibrillation, statin initiation in patients at risk for coronary artery disease, or bariatric surgery in an obese population, where these treatments are observed during longitudinal follow-up of an initially untreated population. As technological advances such as electronic health records make large registries and longitudinal data sources increasingly common, the need for valid, interpretable methods for studies of time-dependent treatment will also increase. Existing methods include marginal structural models, g-estimation and structural nested models, all of which involve parameters and assumptions that are hard to communicate to a non-statistical audience. Alternatively, matching methods create pairs of patients, one of whom initiates treatment at time T and another who doesn't, who are comparable: both eligible for treatment at time T and having a similar covariate history. This creates a pseudo-experiment for every treated patient. The pairs can be combined to create a matched sample and analyzed using standard methods for matched data. On the surface, this approach produces clinically intuitive results that look jus like studies of baseline treatment. That is a tremendous advantage; however, there are important complexities in how to define comparable and eligible patients over time. Thus, many variations have been proposed for specific applications, with unique terminology such as sequential stratification matching, propensity score matching with time-dependent covariates, and balanced risk set matching. Perhaps as a consequence of the diffuse literature in this area, these methods are underutilized in comparative effectiveness research (CER). This project will address these gaps, first by conducting a comprehensive review of longitudinal matching methods, and second by assessing the impact of these methods compared to naïve methods and alternative strategies for time-dependent confounding, when applied to evaluate the long-term efficacy and safety of statin use in the Framingham Heart Study. Together these aims will facilitate a shift in clinical research paradigms towards the use of appropriate methods for time-dependent confounding, and promote the correct and consistent application of longitudinal matching methods to reduce bias in CER studies of time-dependent treatments.