Improving and Externally Validating Mortality Prediction Models for COVID-19 Using Publicly Available Data

We conducted a systematic survey of COVID-19 endpoint prediction literature to: (a) identify publications that include data adhere to FAIR (findability, accessibility, interoperability, and reusability) principles (b) develop reuse mortality models best generalize these datasets. The largest such cohort we knew was used for model development. associated published subjected recursive feature elimination find minimal logistic regression which had statistically clinically indistinguishable predictive performance. This could still not be applied the four external validation sets were identified, due complete absence needed features in some sets. Thus, generalizable (GM) built all An age-only as benchmark, it is simplest, effective, robust predictor currently known literature. While GM surpassed three cohorts, fourth cohort, there no significant difference. study underscores: (1) paucity being shared by researchers despite glut (2) difficulty creating any consistently outperforms an diversity available