What is already known? •Basic lab results and other clinical and demographic attributes have predictive power in predicting risk of adverse events during COVID-19 hospitalization •These include things like age, BMI, general inflammatory markers, and others •Machine learning can be used to combine these attributes to make reliable predictions of COVID-19 prognosis

What does this study add? •Robust, validated, multi class machine learning risk prediction for 3 endpoints during COVID-19 hospitalization using only easily collectable attributes that are common in EHRs •Rigorous explainability analysis to describe which attributes contributed more to the machine learning algorithms’ assignment of risk not found in any similar work