Estimating the Causal Effect of Chloroquine Treatment on Mortality in Malaria Patients Using Marginal Structural Models

Abstract

Traditional methods of survival analysis to ascertain causal effects from RCT data are frequently biased. Our research explores the implementation of more robust statistical methods, namely marginal structural models (MSMs) with inverse probability weighting (IPW), to recover the true causal effect in trials with imperfect treatment adherence, losses to follow up, and complex time-varying confounding. We first simulate data for a 2,000-patient, 60-time point RCT assessing the causal effect of chloroquine treatment on the mortality of malaria patients. We then fit a special class of MSMs called weighted Cox proportional hazards models to show that when a patient's entire treatment, parasitemia, and chloroquine resistance history are known, MSM analysis with IPW successfully recovers the known causal effect. The key result, however, is that when only a subset of patient data is known, we show that even MSM analysis with IPW estimation fails to recover the true causal effect. In fact, we reveal a situation in which such analysis results in conclusions with significant clinical implications. Ultimately, such findings will be pivotal in informing the design and implementation of future RCTs to ensure accurate and robust causal claims.