Plan Per-protocol (PP) Causal Inference Analysis Addressing Intercurrent Events Following the Targeted Learning Roadmap

Intercurrent events, such as deviations from ideal study protocols involving treatment non-adherence, changes or missing data may be common in longitudinal trials. This is particularly pronounced in pragmatic trial due to the flexibility of the design. Metformin for Overweight and Obese Children and Adolescents with Bipolar Spectrum Disorders Treated with Second Generation Antipsychotics (MOBILITY, ClinicalTrials.gov Identifier NCT02515773) was an open-label, multi-site, randomized pragmatic trial (pRCT) designed to assess the effectiveness of metformin at preventing or reversing weight gain in overweight or obese youth with bipolar spectrum disorders treated with second-generation antipsychotics (SGA). Participants were randomized to receiving metformin in addition to brief healthy lifestyle education (MET+LIFE) or to LIFE alone. Participants could also switch between treatment groups, sometimes repeatedly, and these changes were often a direct result of intermmitant weight gain. Following the targeted learning road map and using MOBILITY study as an example, we presented detailed considerations in planning statistical analyses targeting the per-protocol (PP) causal effect, imposing few and transparent modeling assumptions. In addition, we demonstrate the use of the R package ltmle for implementing the analysis plan, and conclude with a discussion of outstanding challenges and future directions.