Prognostic value of deep learning-based coronary artery calcium score and quantitative pneumonia burden in patients hospitalized with COVID-19

Background Severe COVID-19 outcomes are influenced by both pulmonary involvement and underlying cardiovascular disease. Deep learning (DL) techniques can rapidly quantify pneumonia burden and coronary artery calcium score (CACS) from routine chest computed tomography (CT), offering potential for early risk stratification. We aimed to evaluate the prognostic value of DL-based CACS and quantitative pneumonia burden for predicting in-hospital mortality in patients hospitalized with COVID-19. Methods This single-center retrospective study evaluated 1,050 PCR-confirmed SARS-CoV-2 patients hospitalized between 1 September and 31 December 2020 who underwent chest CT. DL algorithms quantified pneumonia burden and CACS. CTs from 300 patients were used to train and tune the CACS model; 388 patients formed the test cohort. Patients were stratified by CACS into six CACS categories. The primary outcome was in-hospital mortality. Multivariate logistic regression and ROC analysis assessed predictive performance. Results In-hospital mortality occurred in 74 patients. Mortality increased with higher CACS: 8.2% in CACS = 0 vs. 27.3% in CACS> 1,000. A CT-based model including pneumonia burden and CACS demonstrated strong predictive power (AUC: 0.77; 95%CI: 0.71-0.83), which improved with the addition of clinical variables (AUC: 0.85; 95%CI: 0.81-0.90; p < 0.001). However, CACS did not independently predict mortality beyond age. In multivariate analysis, pneumonia burden (OR: 1.05; 95%CI: 1.04-1.07; p < 0.001), age, and immunodeficiency remained significant predictors. Conclusions DL-based CT quantification of pneumonia burden and CACS provides strong prognostic value for in-hospital mortality. Pneumonia burden remains an independent predictor, while CACS does not offer additional value over age.