A0930
Title: Joint estimation and false discovery control of causal effects in metabolomics randomized trials
Authors: Rebecca Deek - University of Pittsburgh (United States) [presenting]
Abstract: Randomized experiments are agreed to be the gold standard for assessing causality. Most often, their focus is on the relationship between a single outcome and exposure of interest. However, randomized trials with multiple outcomes are becoming increasingly common. The involvement of omics data, such as metabolomics, in randomized experiments, has seen expansion that is likely to grow. For example, the interest is in using randomized experiments to confirm causal relationships, previously identified in large-scale observational studies, between metabolite concentrations and modifiable risk factors or exposures such as diet and drug or antibiotic use. Accordingly, a covariate-adjusted multivariate regression model is proposed for multivariate estimation of the average treatment effects (ATEs) while also utilizing the correlation between metabolites to improve false discovery control. The utility of the procedure is demonstrated using simulation studies and data from a real randomized clinical trial of Crohn's patients. It is shown that the proposed estimator of the ATEs is more efficient than standard estimators, and the conditional calibration procedure has better false discovery control and higher power than traditional FDR correction methods.
