Primary Submission Category: Randomized Studies
Weighting-Based Estimators for the Survivor Average Causal Effect in Cluster Randomized Trials
Authors: Dane Isenberg, Fan Li, Nandita Mitra, Michael Harhay,
Presenting Author: Dane Isenberg*
Studies often examine the effect of a binary treatment on a non-mortal outcome, where participants may have truncated outcomes if they do not survive to follow-up. One approach to address truncation by death is to target the survivor average causal effect (SACE), a casually interpretable and estimable conditional treatment effect defined via principal stratification. However, current uses for SACE either focus on individual randomized trials or require strong distribution assumptions when applied to multilevel data. We develop a SACE framework for cluster randomized trials (CRTs) relaxing restrictions on the distributions. We establish sets of assumptions that address latent confounding due to clustering to enable point identification of SACE for CRTs. We propose weighting-based estimators for SACE and provide asymptotic variance expressions when survival status is modeled using a GLMM. In simulations, we evaluate our estimators demonstrating that they account for latent confounding and are robust to certain departures from assumptions. We apply our methods to a CRT assessing the impact of a sedation protocol on mechanical ventilation among children with acute respiratory failure.