YSPH Biostatistics Seminar: “Causal inference with interdependent and multivariate, derived outcomes”

NOTE: BIS 525 students are required to attend in person. Others are invited to attend in person, but may also attend via Zoom.

SPEAKER: Youjin Lee, PhD, Manning Assistant Professor, Department of Biostatistics, Brown University

TITLE: “Causal inference with interdependent and multivariate, derived outcomes”

ABSTRACT: Many scientific studies seek to examine the causal effect of an intervention on complex, abstractly defined outcomes—such as personality, brain connectivity, or healthy aging—that are often derived from multiple correlated components. Assessing causal effects in these settings is particularly challenging when interest centers on specific components of the outcome, such as connectivity between brain regions. In such cases, it is necessary to address not only confounding between the intervention and the outcome but also dependencies among the outcome components that are unrelated to the intervention.

This study develops causal inference methods with interdependent and multivariate, derived outcomes by addressing each of the key challenges, motivated by brain effective connectivity research that involves time-series data to derive connectivity measures. We first address within-outcome dependencies in time-series data using sample splitting method to obtain derived outcomes. Then, under certain statistical and causal assumptions, we apply an inverse probability weighting method to the derived outcomes to adjust for observed confounding. We demonstrate that the proposed method is asymptotically valid under the established conditions, effectively controlling for type-I error rates and familywise error rates. We apply our proposed approach to the resting-state fMRI data from Alzheimer’s Disease Neuroimaging Initiative.