Using machine learning to predict COVID-19 infection and severity risk among 4,510 aged adults: a UK Biobank cohort study

• DOI: 10.1101/2020.06.09.20127092
• Publication Date: 2020-06-11T15:05:35Z
• Authors: Auriel A. Willette; Sara A. Willette; Qian Wang; Colleen Pappas; Brandon S. Klined...; Scott Le; Brittany Larsen; Amy Pollpeter; Tianqi Li; Nicole Brenner; Tim Waterboer
• Keywords: 0301 basic medicine; 03 medical and health sciences; 16. Peace & justice; Article; 3. Good health

Abstract Background Many risk factors have emerged for novel 2019 coronavirus disease (COVID-19). It is relatively unknown how these collectively predict COVID-19 infection risk, as well a severe (i.e., hospitalization). Methods Among aged adults (69.3 ± 8.6 years) in UK Biobank, data was downloaded 4,510 participants with 7,539 test cases. We baseline from 10-14 years ago, including demographics, biochemistry, body mass, and other factors, antibody titers 20 common to rare infectious diseases. Permutation-based linear discriminant analysis used hospitalization risk. Probability threshold metrics included receiver operating characteristic curves derive area under the curve (AUC), specificity, sensitivity, quadratic mean. Results The “best-fit” model predicting achieved excellent discrimination (AUC=0.969, 95% CI=0.934-1.000). Factors age, immune markers, lipids, serology pathogens like human cytomegalovirus. more modest (AUC=0.803, CI=0.663-0.943) only titers. Conclusions Accurate profiles can be created using standard self-report biomedical collected public health medical settings. also worthwhile further investigate if prior host immunity predicts current COVID-19.