Predicting immune protection against outcomes of infectious disease from population-level effectiveness data with application to COVID-19

Quantifying the extent to which previous infections and vaccinations confer protection against future infection or disease outcomes is critical to managing the transmission and consequences of infectious diseases.

We present a general statistical model for predicting the strength of protection conferred by different immunising exposures (numbers, types, and strains of both vaccines and infections), against multiple outcomes of interest, whilst accounting for immune waning. We predict immune protection against key clinical outcomes: developing symptoms, hospitalisation, and death. We also predict transmission-related outcomes: acquisition of infection and onward transmission in breakthrough infections. These enable quantification of the impact of immunity on population-level transmission dynamics. Our model calibrates the level of immune protection, drawing on both population-level data, such as vaccine effectiveness estimates, and neutralising antibody levels as a correlate of protection. This enables the model to learn realised immunity levels beyond those which can be predicted by antibody kinetics or other correlates alone.

We demonstrate an application of the model for SARS-CoV-2, and predict the individual-level protective effectiveness conferred by natural infections with the Delta and the Omicron B.1.1.529 variants, and by the BioNTech-Pfizer (BNT162b2), Oxford-AstraZeneca (ChAdOx1), and 3rd-dose mRNA booster vaccines, against outcomes for both Delta and Omicron. We also demonstrate a use case of the model in late 2021 during the emergence of Omicron, showing how the model can be rapidly updated with emerging epidemiological data on multiple variants in the same population, to infer key immunogenicity and intrinsic transmissibility characteristics of the new variant, before the former can be more directly observed via vaccine effectiveness data.

This model provided timely inference on rapidly evolving epidemic situations of significant concern during the early stages of the COVID-19 pandemic. The general nature of the model enables it to be used to support management of a range of infectious diseases.