 This retrospective study of COVID-19 patients treated at NYU Langone Health used XG Boost model to predict mortality, ventilation, or ICU admission based on clinical data during different parts of their hospital stay. The most important feature was respiration rate, followed by SPO2, and age 75 years and over. The model's performance extended five days prior to death with AUC equals 0.81, specificity equals 70%, and sensitivity equals 75%. Other canonical markers such as diabetic history, age, and temperature offered minimal gain. Lab values such as blood urea nitrogen and LDH were most beneficial in predicting mortality. Features that were predictive of morbidity included LDH, calcium, glucose, and seriactive protein. This article was authored by Joshua M. Wong, Wen Klu, Xiaoshan Chen, and others.