Predictors of severe coronavirus disease 2019 pneumonia

BACKGROUND: Early warning of severe coronavirus disease 2019 (COVID-19) pneumonia on admission is critical for reducing mortality. PURPOSE: The purpose of this study was to identify the risk factors for predicting severe COVID-19 pneumonia on admission. MATERIALS AND METHODS: Computed tomography (CT) scans on admission and initial clinical data were collected from 213 patients with COVID-19 pneumonia. Semi-quantitative CT scoring was performed, multiplying the CT patterns by their extent. CT patterns were graded on a four-point scale: 0, normal attenuation; 1, ground-glass opacities (GGOs); 2, mixed patterns of GGO and consolidation; and 3, consolidation. The extent of patterns was visually estimated as the percentage (to the nearest 10%) of the affected pulmonary lobe. Inter-observer agreement was evaluated using the inter-class correlation coefficient. CT scores and clinical data were compared between severe and nonsevere patients using parametric and nonparametric statistics, as appropriate. The least absolute shrinkage and selection operator (LASSO) with 10-fold cross-validation and logistic regression was used to select the risk factors and construct a predictive model. RESULTS: Age, respiratory rate, hypertension, procalcitonin, D-dimer, lactate dehydrogenase, high-sensitivity C-reactive protein (hs-CRP), cystatin C, brain natriuretic peptide (pro-BNP), and CT score were higher in severe COVID-19 infection. LASSO analysis revealed that the CT score coupled with hs-CRP was optimal for predicting progression to severe pneumonia. The areas under the curves for validation and testing data were 0.85 and 0.82, respectively, with sensitivity of 89.5% and 75.0%, specificity of 75.4% and 98.1%, and accuracy of 77.2% and 95.3%. CONCLUSION: The CT score combined with hs-CRP on admission predicted severe COVID-19 pneumonia.