A validated, real-time prediction model for favorable outcomes in hospitalized COVID-19 patients

This paper presents a prospectively validated, real-time prediction model for favorable outcomes within 96 hours of prediction in hospitalized COVID-19 patients. Using 3345 retrospective and 474 prospective hospitalizations, a parsimonious model was developed and validated using real-time lab values, vital signs, and oxygen support. The model achieved high average precision (88.6% and 90.8%) and discrimination (95.1%–95.2% and 86.8%) in retrospective and prospective validation, respectively. Integrated into the EHR, the model showed a positive predictive value of 93.3% with 41% sensitivity. Preliminary results suggest clinical adoption.

Narges Razavian, Vincent J. Major, Mukund Sudarshan, Jesse Burk-Rafel, Peter Stella, Hardev Randhawa, Seda Bilaloglu, Ji Chen, Vuthy Nguy, Walter Wang, Hao Zhang, Ilan Reinstein, David Kudlowitz, Cameron Zenger, Meng Cao, Ruina Zhang, Siddhant Dogra, Keerthi B. Harish, Brian Bosworth, Fritz Francoisi, Leora I. Horwitz, Rajesh Ranganath, Jonathan Austrian, Yindalon Aphinyanaphongs