Development and Validation of an Electronic Health Record-Derived Prediction Model for Preventing COVID-19...

Hospitalization and death following COVID-19 infection continue to pose a major public health concern and place strain on health system resources. Outpatient antiviral medication can reduce the risk of COVID-19 hospitalization and death for those at risk of poor outcomes, but identifying high-risk populations who may benefit most from treatment is challenging. The objective of this study was to develop and validate a prediction model for the composite outcome of hospitalization or death in the 14 days following COVID-19 infection. Our sample included 67,530 COVID-19 infections documented in outpatient care and occurring between April 1, 2020, and November 1, 2022, for 64,529 Kaiser Permanente Washington patients who did not receive outpatient antiviral treatment; 1378 (2.0%) of these infections resulted in hospitalization or death. Our prediction model, estimated using logistic regression with LASSO variable selection and ridge penalization, included 19 risk factors and showed high performance, including an area under the curve of 0.825 (95% confidence interval 0.813-0.836). Among the 10% of infections with the highest risk predictions, the true positive rate was 48% (46-51%) and the positive predictive value was 9.9% (9.2-10.6%). Supplemental analyses confirmed strong model performance across racial and ethnic subgroups and over time. We also present our process for selecting a risk threshold above which to recommend antiviral treatment and discuss considerations for prospective clinical implementation. This project demonstrates that machine learning tools can be used by health systems to deliver timely, targeted secondary prevention to reduce the risk of serious illness or death.