Title: Handling missing scalar and functional data in integrative analysis with applications to mental health research
Authors: Adam Ciarleglio - George Washington University (United States) [presenting]
Abstract: In mental health research, the number of clinical trials and observational studies that include multimodal neuroimaging is growing rapidly. Often, the goal is to integrate both the clinical and imaging data in order to address a specific research question. Functional data analytic tools for analyzing such data do exist and can perform well, but most/all methods assume that the data being analyzed are complete. Analyses that use these tools can be undermined by the fact that some proportion of both the clinical and imaging data may be missing for some study participants, often leading to data sets where only a small number of subjects have data available for all variables. We present approaches for imputation of missing scalar and functional data when the goal is to fit a scalar-on-function regression model for the purpose of either (1) estimating the association between a scalar outcome and a scalar or functional predictor or (2) developing a predictive model. We present results from a simulation study showing the performance of various imputation approaches with respect to fidelity to the observed data, estimation of the parameters of interest, and prediction. The proposed approaches are also illustrated using data from a placebo-controlled clinical trial assessing the effect of SSRI in subjects with major depressive disorder.