External validation of CT-based severity scoring systems to determine prognosis of pneumonia caused by COVID-19 virus: a multicentric cohort study

Abstract

Background: Prompt identification of COVID-19 patients through an advanced and dependable prognostic model can alleviate pressure on healthcare facilities by enabling safe home recovery for individuals who do not require hospitalization. In this study, we ascertained the validity of three different CT scoring systems as the main component of prognostic models. Methods: This retrospective cohort studied 566 COVID-19 patients hospitalized in three tertiary centers. In addition to demographic and clinical data collection, all the patients’ chest CT images were evaluated by standard 15-, 20-, and 24-scoring systems introduced earlier for the prognosis of COVID-19. The outcome studied was mortality status. Using logistic regression analysis, four variables of age, sex, lymphocyte count, and CT severity score were investigated according to each model setup. To assess the goodness of fit, ROC curve analysis was also conducted. Results: All models’ highest odd ratio (OR) for mortality outcome was related to CT severity score. The 15-score model achieved the greatest OR, and the 24-score model the smallest OR (1.158 vs 1.086). All three CT severity scoring models can significantly distinguish between COVID-19 survivors and expired patients (P ≤.001). However, the area under a ROC curve (AUC) was the highest for a 20-score severity model (20-score, AUC = 0.672; 24-score, AUC = 0.663; 15-score, AUC = 0.646). Conclusion: A chest CT scoring system where the sum of scores from five pulmonary lobes ranges between 0 and 20 is more effective in predicting mortality outcomes in COVID-19 patients.