Qualitative and quantitative chest CT parameters as predictors of specific mortality in COVID-19 patients

Abstract

Purpose: To test the association between death and both qualitative and quantitative CT parameters obtained visually and by software in coronavirus disease (COVID-19) early outbreak. Methods: The study analyzed retrospectively patients underwent chest CT at hospital admission for COVID-19 pneumonia suspicion, between February 21 and March 6, 2020. CT was performed in case of hypoxemia or moderate-to-severe dyspnea. CT scans were analyzed for quantitative and qualitative features obtained visually and by software. Cox proportional hazards regression analysis examined the association between variables and overall survival (OS). Three models were built for stratification of mortality risk: clinical, clinical/visual CT evaluation, and clinical/software-based CT assessment. AUC for each model was used to assess performance in predicting death. Results: The study included 248 patients (70% males, median age 68 years). Death occurred in 78/248 (32%) patients. Visual pneumonia extent > 40% (HR 2.15, 95% CI 1.2–3.85, P = 0.01), %high attenuation area – 700 HU > 35% (HR 2.17, 95% CI 1.2–3.94, P = 0.01), exudative consolidations (HR 2.85–2.93, 95% CI 1.61–5.05/1.66–5.16, P < 0.001), visual CAC score > 1 (HR 2.76–3.32, 95% CI 1.4–5.45/1.71–6.46, P < 0.01/P < 0.001), and CT classified as COVID-19 and other disease (HR 1.92–2.03, 95% CI 1.01–3.67/1.06–3.9, P = 0.04/P = 0.03) were significantly associated with shorter OS. Models including CT parameters (AUC 0.911–0.913, 95% CI 0.873–0.95/0.875–0.952) were better predictors of death as compared to clinical model (AUC 0.869, 95% CI 0.816–0.922; P = 0.04 for both models). Conclusions: In COVID-19 patients, qualitative and quantitative chest CT parameters obtained visually or by software are predictors of mortality. Predictive models including CT metrics were better predictors of death in comparison to clinical model.