Causal Inference for Estimation of Vaccine Effects from Time-to-Event Data

Xiong M, Li Z, Zhang K, Talebi Y, Wu H, Chan W and Boerwinkle E

Published on: 2023-10-20

Abstract

Vaccines are the most efficient method for controlling infectious diseases. Vaccine effectiveness estimation is extremely important in monitoring vaccine efficacy and controlling disease spread. We find that great challenges arise in vaccine effectiveness estimation due to various sources of bias, including confounding/treatment selection bias, censoring bias, and healthy vaccine bias, event-induced covariate shifts, competing events, prior immunity, the emergence of new virus variants, and healthcare-seeking behavior. A vital step in improving vaccine effectiveness studies is to develop novel methods to address bias. To achieve this goal, we adapt recently developed generalization bounds from domain adaptation and treatment effect estimation to time-to-event data and apply the counterfactual causal inference method with deep neural network-based balancing representations and the framework of empirical risk minimization (ERM), which is called SurvITE [10], to vaccine effectiveness estimation from the time-to-event data that are extracted from the Optum® de-identified COVID-19 Electronic Health Record dataset (2007-2021). The estimated vaccine effectiveness by SurvITE is compared with the Cox regression model and the random survival forest model. The preliminary results showed that the vaccine effectiveness curves estimated by the SurvITE method and RSF model were closed, but the vaccine effectiveness curve estimated by the Cox regression model was much higher than that estimated by the SurvITE method and RSF model.