Predicting Severe COVID-19 Infection on Initial Diagnosis: Comparison and Validation of CT Imaging Triage Tools.

Abstract

Developing a reliable predictive tool of disease severity in COVID-19 infection is important to help triage patients and ensure the appropriate utilization of health-care resources. To develop, validate, and compare three CT scoring systems (CTSS) to predict severe disease on initial diagnosis of COVID-19 infection. One hundred and twenty and 80 symptomatic adults with confirmed COVID-19 infection who presented to emergency department were evaluated retrospectively in the primary and validation groups, respectively. All patients had non-contrast CT chest within 48 hours of admission. Three lobarbased CTSS were assessed and compared. The simple lobar system was based on the extent of pulmonary infiltration. Attenuation corrected lobar system (ACL) assigned further weighting factor based on attenuation of pulmonary infiltrates. Attenuation and volume-corrected lobar system incorporated further weighting factor based on proportional lobar volume. The total CT severity score (TSS) was calculated by adding individual lobar scores. The disease severity assessment was based on Chinese National Health Commission guidelines. Disease severity discrimination was assessed by the area under the receiver operating characteristic curve (AUC). The ACL CTSS demonstrated the best predictive and consistent accuracy of disease severity with an AUC of 0.93(95%CI:0.88-0.97) in the primary cohort and 0.97 (95%CI:0.91.5-1) in the validation group. Applying a TSS cut-off value of 9.25, the sensitivities were 96.4% and 100% and the specificities were 75% and 91% in the primary and validation groups, respectively. The ACL CTSS showed the highest accuracy and consistency in predicting severe disease on initial diagnosis of COVID-19. This scoring system may provide frontline physicians with a triage tool to guide admission, discharge, and early detection of severe illness.