Prediction of COVID-19 hospitalisation, ICU admission or death following ChAdOx1 vaccination using artificial intelligence: A clinical predictive model from the English RAVEN study

Comprehensive Summary

This retrospective cohort study used linked primary care data from over 19 million individuals in England to identify predictors of severe COVID-19 outcomes (hospitalization, ICU admission, death) following completion of a two-dose vaccine series. The study employed explainable AI (XGBoost and SHAP) to evaluate predictive factors. Key predictors of mortality were age, high multi-morbidity scores, and chronic heart, respiratory, and kidney diseases. Predictors for hospitalization included high multi-morbidity, obesity, and early vaccination timing, while ICU admission was associated with obesity, female sex, early vaccination, chronic kidney disease, and diabetes. Age above 85 was strongly predictive for mortality but not for ICU admission.

Outcomes and Implications

The study demonstrates that explainable AI can accurately predict severe COVID-19 outcomes in fully vaccinated individuals using real-world data. This approach could help prioritize high-risk individuals for booster vaccinations and targeted public health interventions. It also validates the use of AI-based predictive models in public health planning.