Causal Inference in Psychiatry Research

Abstract

In Chapter 1, we show how to emulate a hypothetical randomized trial ("target trial") that would answer our questions about the comparative effectiveness of complex dynamic treatment strategies in major depression using inverse probability weighting. We finally show that using the strategy proposed by most clinical guidelines was associated with an increased probability of remission and a lower risk of serious adverse events. In chapter 2, we underscore the importance of incorporating the per-protocol effect as a causal estimand of interest in randomized trials in psychiatry. We used data from a recent randomized trial to estimate the per-protocol effect of lithium on suicidality. Our analyses showed that the 12-month risk of suicidality under the lithium strategy was 6.7 percentage points lower than for placebo, with values between 19.2 percentage points lower and 5.9 percentage points higher were very compatible with our data. Our results suggest that lithium might be effective in preventing suicidality after adjusting for nonadherence, but our confidence intervals were wide. In Chapter 3, we introduce the FEP-CAUSAL Collaboration, an international consortium of observational cohorts from patients with a First Episode of Psychosis (FEP). We used data from this consortium to compare the effect estimates from a well-known randomized trial (the EUFEST trial) against the emulation of a target trial with the same protocol using the observational data, a process that is called benchmarking. From this experience, we report the strengths and limitations of our dataset to emulate target trials in early psychosis.