Targeted Maximum Likelihood Based Estimation for Longitudinal Mediation Analysis

• DOI: 10.48550/arxiv.2304.04904
• Publication Date: 2023-01-01
• Authors: Zeyi Wang; Lars van der Laan; Maya Petersen; Thomas Alexander ...; Kajsa Kvist; Mark van der Laan
• Keywords: Censoring (clinical trials); Lasso

Causal mediation analysis with random interventions has become an area of significant interest for understanding time-varying effects longitudinal and survival outcomes. To tackle causal statistical challenges due to the complex data structure confounders, competing risks, informative censoring, there exists a general desire combine machine learning techniques semiparametric theory. In this manuscript, we focus on targeted maximum likelihood estimation (TMLE) natural direct indirect defined interventions. The proposed estimators are multiply robust, locally efficient, directly estimate update conditional densities that factorize likelihoods. We utilize highly adaptive lasso (HAL) projection representations derive new (HAL-EIC) efficient influence curves problems propose fast one-step TMLE algorithm using HAL-EIC while preserving asymptotic properties. method can be generalized other parameters smooth functions likelihoods, thereby provides novel flexible toolbox.