Estimating the Causal Effect of Digoxin using Marginal Structural Models

Abstract

Digoxin is one of the oldest treatments for heart failure. Recent observational studies have shown strong associations with adverse events (AEs), while older randomized clinical trials showed it to be safe. We studied the effect of Digoxin on AEs using a longitudinal, observational cohort of 207 patients implanted with a Left Ventricular Assist Device (LVAD) at Columbia University Medical Center. Cox models with 1 to 1 propensity score matching at baseline showed the hazard ratio of AEs for patients taking Digoxin to be 1.8 (95% CI 1.4-26), similar to previous observational studies. However, Cox models are known to yield biased results in the presence of a time varying exposure and time dependent confounders. We adjusted for time dependent confounding through a marginal structural Cox model, using the time-varying probabilities of censoring and treatment to construct inverse probability weighted estimators. This model showed the effect of Digoxin on AEs to be non-significant 1.1 (0.7-1.7), consistent with older randomized trials. Our results suggest Digoxin is still a safe treatment for modern heart failure patients.

Date
Event
ASA Joint Statistical Meeting Poster Session
Location
Denver, CO
Avatar
Katherine Hoffman
Research Biostatistician I

I am passionate about meaningful, reproducible medical research.