CFE 2019: Start Registration
View Submission - CMStatistics
Title: Mediation analysis when outcome and mediator are semi-competing events with application in health disparities research Authors:  Linda Valeri - Columbia University (United States) [presenting]
Cecile Proust-Lima - Universite de Bordeaux (France)
Helene Jacqmin-Gadda - Universite de Bordeaux (France)
Abstract: Novel methodology for mediation analysis is proposed to explain how much of the effect of the exposure on a terminal time-to-event outcome, say death, is attributed to the non-terminal potential intermediate time-to-event. Addressing this question is important in health disparities research when we seek to quantify inequities in access to high quality and timely treatment and their impact on patients survival time. We formalize a type of direct and indirect effects using the potential outcome framework in the presence of semi-competing risks. Mediation is studied in a multistate model in continuous time. Monte Carlo simulation based as well as closed form estimators of the causal contrasts are developed. We show via simulations that mediation analysis ignoring censoring in mediator and outcome time-to event-processes and/or ignoring competing risks may give misleading results. Rigorous definition of the direct and indirect effects and estimation of the joint outcome and mediator distributions in the presence of semi-competing risks is crucial for valid investigation of mechanisms in continuous time. We employ this novel methodology to investigate the role of delaying treatment uptake in explaining racial disparities in cancer survival in a multi-center cohort study of colorectal cancer patients.