Primary Submission Category: Design-Based Causal Inference
The test-negative design for the estimation of COVID-19 vaccine effectiveness: development of statistical methods in the evolving context
Authors: Helen Bian, Cong Jiang, Denis Talbot, Robert Platt, Mireille Schnitzer,
Presenting Author: Helen Bian*
The test-negative design (TND) has been widely used for the rapid estimation of vaccine effectiveness against infectious diseases. The TND typically includes individuals with a common symptom profile who are receiving a laboratory test for an infection of interest. Among them, participants who test positive for the target infection are “cases” and those who test negative are “controls”. Existing statistical approaches for the TND have certain limitations in a dynamic longitudinal setting, where data can be periodically collected from different individuals over the study period.
First, time-dependent confounders may be influenced by previous vaccination and health status, such as previous infection, while also affecting the subsequent vaccination decisions. Thus, the causal relation of interest cannot be properly estimated using traditional covariate-adjusted models. Secondly, since individuals can test positive multiple times over the study period, past infections may alter immunity and create non-positivity for vaccination for the following time-period. Therefore, we propose a causal framework that accounts for the time-varying effects as well as changing risk sets, particularly in the context of the dynamic nature of infectious diseases. We propose IPTW estimators of discrete-time hazard and hazard ratios, which can be identified from the TND samples. Simulation studies are used to show the performance of these estimators.