An artificial intelligence system for predicting the deterioration of COVID-19 patients in the emergency department

This paper proposes a data-driven AI system for predicting COVID-19 patient deterioration in emergency departments. The system uses a deep neural network (DNN) analyzing chest X-ray images and a gradient boosting model (GBM) using clinical variables. Trained on data from 3661 patients, the system achieved an AUC of 0.786 (95% CI: 0.745–0.830) for predicting deterioration within 96 hours. The DNN's interpretability was demonstrated through saliency maps, and its performance was comparable to two radiologists in a reader study. A preliminary DNN version was successfully deployed at NYU Langone Health during the pandemic, showing real-time accuracy. The study demonstrates the potential of the system to assist clinicians in COVID-19 patient triage.

Farah E. Shamout, Yiqiu Shen, Nan Wu, Aakash Kaku, Jungkyu Park, Taro Makino, Stanisław Jastrzębski, Jan Witowski, Duo Wang, Ben Zhang, Siddhant Dogra, Meng Cao, Narges Razavian, David Kudlowitz, Lea Azour, William Moore, Yvonne W. Lui, Yindalon Aphinyanaphongs, Carlos Fernandez-Granda, Krzysztof J. Geras