Targeted learning in observational studies with multi-level treatments: An evaluation of antipsychotic drug treatment safety for patients with serious mental illnesses

We investigate estimation of causal effects of multiple competing (multi-valued) treatments in the absence of randomization. Our work is motivated by an intention-to-treat study of the relative metabolic risk of assignment to one of six commonly prescribed antipsychotic drugs in a cohort of adults with serious mental illness. Doubly-robust estimators of multi-level treatment effects with observational data, such as targeted minimum loss-based estimation (TMLE), require that either the treatment model or outcome model is correctly specified to ensure consistent estimation. However, common TMLE implementations estimate treatment probabilities using multiple binomial regressions rather than a single multinomial regression. We implement a TMLE estimator that uses multinomial treatment assignment and ensemble machine learning to estimate average treatment effects. Our implementation achieves superior coverage probability relative to the binomial implementation in simulation experiments with varying treatment propensity overlap and event rates. An evaluation of the causal effects of six antipsychotic drugs on the risk of diabetes or death illustrates our approach. We find a relative safety benefit of moving from a second-generation antipyschotic thought to have more favorable metabolic risk profile relative to other second-generation drugs to a less commonly prescribed first-generation antipyschotic known for having a low rate of metabolic disturbance.