Causal Inference
Mediation analysis with time-varying mediators and time-to-event outcomes accounting for competing risks Arce Domingo* Arce Domingo Yuchen Zhang Ana Navas-Acien Kiros Berhane Brent Coull Linda Valeri
Background: In survival settings, competing events refer to any event that makes it impossible for the event of interest to occur. Not accounting for competing events by death can lead to biases caused by the fact that individuals that die do not have the opportunity to develop the event of interest. In this work, we propose a framework to deal with competing events in the causal mediation setting in presence of longitudinal mediators and time-to-event outcomes.
Methods: We used the path-specific effects framework to adapt the mediational g formula to consider the competing event as a nested mediator with our mediator of interest. Thus, we consider two indirect effects: the indirect effect through the history of the mediator of interest, and the indirect effect through the history of the competing event. We used additive hazards models to obtain effects estimates in an attributable risk scale. We applied our algorithm to real data from the Strong Heart Study, a prospective cohort of American Indians. We evaluated the potential mediating role of systolic blood pressure on the association between urinary cadmium and arsenic (in separate models) and cardiovascular disease (CVD), accounting for competing events by death. Metals were measured at visit 1 (1989-1991). Blood pressure was measured at three time points: visit 1, visit 2 (1993-1995) and visit 3 (1998-1999).
Results: 148.58 (-81.25, 376.41) CVD cases per 100,000 person-years were attributable to an interquartile range (IQR) increase in urinary cadmium. Of those, 28.16 (4.13, 56.72) cases were mediated by the blood pressure trajectory, and 28.87 (-69.59, 141.12) cases were avoided due to the individual dying before CVD happened. For urinary arsenic, 274.98 (8.58, 537.57) CVD cases were attributable to an IQR increase. Of those, 45.27 (21.68, 77.05) cases were mediated by the blood pressure trajectory, and 12.92 (-39.42, 65.28) cases were avoided by death.
Conclusion: We introduced, for the first time, a framework using path-specific effects to define causal mediated effects in presence of competing risks. Our algorithm estimates causal effects in terms of attributable cases per a number of person-years, which is a measure of public health impact. We found mediated effects of blood pressure on the association between urinary arsenic and cadmium and CVD.