Alliance for Pandemic Preparedness

May 22, 2020

Clinical Predictors of COVID-19 Mortality



  • [pre-print, not peer reviewed] Machine learning techniques were applied to clinical data from a cohort of 5,051 COVID-19 patients in New York City to predict mortality (training set of n=3,841). A mortality predictor based on five clinical features (age, minimum oxygen saturation, type of encounter, hydroxychloroquine use, and maximum body temperature) had good predictive performance (AUC=0.91) on a test set of retrospective (n=961) and prospective (n=249) patients.  

Yadaw et al. (May 22, 2020). Clinical Predictors of COVID-19 Mortality. Pre-print downloaded May 21 from