Causal Mediation Analysis With Survival Data

摘要: Causal mediation analysis is considered for time-to-event outcomes and survival models. Different possible effect decompositions are discussed the function, hazard, mean time median scales. Approaches to in social sciences related counterfactual approaches using additive proportional hazard accelerated failure The product-coefficient method from gives mediated effects on difference scale models, log hazards model but only if outcome rare. With a common outcome, can provide valid test presence of mediator does not measure. When time, or rare-outcome models employed combined with approach, exposure-mediator interactions be accommodated relatively straightforward manner.