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

Anshul Thakur, Bernardo Meza-Torres, Xuejuan Fan, Rachel Byford, Mark Joy, Wilhelmine Meeraus, Sudhir Venkatesan, Sylvia Taylor, Simon de Lusignan and David A Clifton

PLOS ONE, 2026, vol. 21, issue 2, 1-22

Abstract: Objectives: This study identifies predictors of severe COVID-19 following completion of two-dose primary series of the AZD1222 COVID-19 vaccine, employing eXtreme Gradient Boosting (XGBoost) and Shapely additive explanations (SHAP), as an explainable artificial intelligence (AI) approach. Method: A retrospective cohort study using linked primary care data from the Oxford-Royal College of General Practitioners Clinical Informatics Digital Hub (ORCHID), including computerised medical records of over 19 million people in England, for the period from 8th December 2020–31st December 2021, as part of the Real-world effectiveness of the AZD1222 COVID-19 vaccine in England (RAVEN) study. We evaluated a two-dose primary series of the AZD1222 vaccine on COVID-19 related hospitalisation, ICU admission or death. Results: A total of 4,515,280 individuals with a two-dose primary series of AZD1222 vaccine were analysed, where 7,171 individuals had a record of severe COVID-19. Variables with the greatest predictive weight for COVID-19 mortality in vaccinated individuals were age ≥ 85 years, high Cambridge Multi-Morbidity Score, and chronic heart, respiratory and kidney diseases; variables predicting COVID-19 hospitalisation following completed primary series included high CMMS, obesity, and being offered early COVID-19 vaccination in the national vaccine campaign (e.g., vaccinated during the first quarter of 2021); predictors of COVID-19 ICU admission included obesity, female sex, being offered early COVID-19 vaccination in the national vaccine campaign, chronic kidney disease and diabetes. Across models, age ≥ 85 years was highly predictive of mortality and moderately predictive of hospitalisation. However, for ICU admission it was reported as not predictive. Conclusion: Obesity, chronic heart, respiratory and kidney diseases were the main predictors across models, which is comparable to the scientific literature, validating the explainable AI approach. XGBoost can accurately predict severe outcomes in fully vaccinated individuals. Predictive models built on real-world primary care data can help to timely identify individuals to be prioritised for vaccination booster.

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0336449 (text/html)
https://journals.plos.org/plosone/article/file?id= ... 36449&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pone00:0336449

DOI: 10.1371/journal.pone.0336449

Access Statistics for this article

More articles in PLOS ONE from Public Library of Science
Bibliographic data for series maintained by plosone ().