Evaluating Spillover Effects in Network-Based Studies In the Presence of Missing Outcomes

Estimating causal effects in the presence of spillover among individuals embedded within a social network is often challenging with missing information. The spillover effect is the effect of an intervention if a participant is not exposed to the intervention themselves but is connected to intervention recipients in the network. In network-based studies, outcomes may be missing due to the administrative end of a study or participants being lost to follow-up due to study dropout, also known as censoring. We propose an inverse probability censoring weighted (IPCW) estimator, which is an extension of an IPW estimator for network-based observational studies to settings where the outcome is subject to possible censoring. We demonstrated that the proposed estimator was consistent and asymptotically normal. We also derived a closed-form estimator of the asymptotic variance estimator. We used the IPCW estimator to quantify the spillover effects in a network-based study of a nonrandomized intervention with censoring of the outcome. A simulation study was conducted to evaluate the finite-sample performance of the IPCW estimators. The simulation study demonstrated that the estimator performed well in finite samples when the sample size and number of connected subnetworks (components) were fairly large. We then employed the method to evaluate the spillover effects of community alerts on self-reported HIV risk behavior among people who inject drugs and their contacts in the Transmission Reduction Intervention Project (TRIP), 2013 to 2015, Athens, Greece. Community alerts were protective not only for the person who received the alert from the study but also among others in the network likely through information shared between participants. In this study, we found that the risk of HIV behavior was reduced by increasing the proportion of a participant's immediate contacts exposed to community alerts.