Predicting the Effectiveness of Covid-19 Vaccines from SARS-CoV-2 Variants Neutralisation Data

Abstract

AbstractRapid and accurate prediction of Covid-19 vaccine effectiveness is crucial to response against SARS-CoV-2 variants of concern. Despite intensive research, several prediction tasks are not well supported, such as predicting effectiveness of partial vaccination, of vaccine boosters and in vaccinated subpopulations. This paper introduces a novel predictive framework to accommodate such tasks and improve prediction accuracy. It was developed for predicting the symptomatic effectiveness of the BNT162b2 (Comirnaty) and ChAdOx1 nCoV-19 (Vaxzevria) vaccines but could apply to other vaccines and effectiveness types. Direct prediction within the framework uses levels of vaccine-induced neutralising antibodies against SARS-CoV-2 variants to fit efficacy and effectiveness estimates from studies with a given vaccine. Indirect prediction uses a model fitted for one vaccine to predict the effectiveness of another. The directly predicted effectiveness of Comirnaty against the Delta variant was 44.8% (22, 69) after one and 84.6% (64, 97) after two doses, which is close to 45.6% and 85.5%, the average estimates from effectiveness studies with the vaccine. The corresponding direct predictions for Vaxzevria were 41.6% (18, 68) and 63.2% (37, 86); and the indirect predictions, from the model fitted to Comirnaty data, were 45.5% (23, 70) and 61.2% (37, 83). Both sets of predictions are comparable to the average estimates, 42.5% and 66.3%, from effectiveness studies with Vaxzevria. Further results are presented on age subgroups; prediction biases and their mitigation; and implications for vaccination policies.