Combining Initial Radiographs and Clinical Variables Improves Deep Learning Prognostication of Patients with COVID-19 from the Emergency Department

Kwon, Young Joon (Fred); Toussie, Danielle; Finkelstein, Mark; Cedillo, Mario A.; Maron, Samuel Z.; Manna, Sayan; Voutsinas, Nicholas; Eber, Corey; Jacobi, Adam; Bernheim, Adam; Gupta, Yogesh Sean; Chung, Michael S.; Fayad, Zahi A.; Glicksberg, Benjamin; Oermann, Eric K.; Costa, Anthony B.

“Just Accepted” papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. PURPOSE: To train a deep learning classification algorithm to predict chest radiography severity scores and clinical outcomes in patients with coronavirus disease 2019 (COVID-19). MATERIALS AND METHODS: In this retrospective cohort study, we identified patients of ages of 21 to 50 who presented to the emergency department (ED) of a multicenter urban health system from March 10–26, 2020 with COVID-19 confirmation on real-time reverse transcription polymerase chain reaction. We collected the initial chest radiographs (CXRs), clinical variables, and outcomes including admission, intubation, and survival within 30 days (n = 338; median age 39; 210 men). Two fellowship-trained cardiothoracic radiologists examined CXRs for opacities and assigned a clinically validated severity score. We trained a deep learning algorithm to predict outcomes on a holdout test set composed of confirmed COVID-19 patients who presented from March 27–29, 2020 (n = 161; median age 60; 98 men) for both younger (ages 21–50; n = 51) and older (ages > 50; n = 110) populations. Bootstrapping methods computed confidence intervals. RESULTS: The model trained on the CXR severity score produced the following areas under the receiver operating characteristic (AUCs): 0.80 (0.73,0.88) for the CXR severity score, 0.76 (0.68,0.84) for admission, 0.66 (0.56,0.75) for intubation, and 0.59 (0.49,0.69) for death. The model trained on clinical variables produced the following AUCs 0.64 (0.55,0.73) for intubation and 0.59 (0.50,0.68) for death. Combining CXR and clinical variables increased AUC of intubation and death to 0.86 (0.79,0.96) and 0.82 (0.72,0.91), respectively. CONCLUSION: The combination of imaging and clinical information improves outcome predictions.