Debiasing hazard-based, time-varying vaccine effects using vaccine-irrelevant infections: An observational extension of a pivotal Phase 3 COVID-19 vaccine efficacy trial

内容摘要

Understanding how vaccine effectiveness (VE) changes over time can provide evidence-based guidance for public health decision making. While commonly reported by practitioners, time-varying VE estimates obtained using Cox regression are vul- nerable to hidden biases. To address these limitations, we describe how to leverage vaccine-irrelevant infections to identify hazard-based, time-varying VE in the pres- ence of unmeasured confounding and selection bias. We articulate assumptions under which our approach identifies a causal effect of an intervention deferring vaccination and interaction with the community in which infections circulate. We develop sieve and efficient influence curve-based estimators and discuss imposing monotone shape constraints and estimating VE against multiple variants. As a case study, we examine the observational booster phase of the Coronavirus Vaccine Efficacy (COVE) trial of the Moderna mRNA-1273 COVID-19 vaccine which used symptom-triggered multi- plex PCR testing to identify acute respiratory illnesses (ARIs) caused by SARS-CoV-2 and 20 off-target pathogens previously identified as compelling negative controls for COVID-19. Accounting for vaccine-irrele