A clinical-radiological model for predicting in-hospital mortality in acute exacerbation of COPD

Background Quantitative computed tomography (QCT) has emerged as a promising tool for chronic obstructive pulmonary disease (COPD) phenotyping. However, its prognostic value in acute exacerbations of COPD (AECOPD) remains insufficiently explored, and existing risk stratification models have limited integration of imaging-derived metrics. This study aimed to develop and validate a prognostic model that integrates multidimensional clinical variables with QCT-derived metrics, and to compare its predictive performance with traditional clinical scores for in-hospital mortality in AECOPD.Methods In this retrospective cohort study, consecutive patients aged ≥40 years hospitalized with severe AECOPD were enrolled from September 2010 to October 2024, with collection of demographic characteristics, clinical features, laboratory indices, and chest CT imaging data. The cohort was randomly split into training and testing sets (7:3). Candidate predictors were selected using least absolute shrinkage and selection operator (LASSO) regression followed by multivariable logistic regression. Four classification models were developed and evaluated using 10-fold cross-validation. Discriminative performance was assessed by area under the receiver operating characteristic curve (AUC) and compared with conventional clinical scores.Results A total of 465 patients with AECOPD were included, of whom 47 (10.1%) died during hospitalization. Five independent predictors were identified: confusion, lymphocyte count <0.8 × 109/L, chronic heart failure, white blood cell count ≥10 × 109/L, and reduced pulmonary vascular fractal dimension. The logistic regression model demonstrated the best performance, achieving an AUC of 0.852 (95% CI, 0.738–0.965) in the testing set, significantly outperforming other models and established clinical scores.Conclusions A predictive model integrating QCT-derived metrics with clinical features improved AECOPD in-hospital mortality prediction over traditional scores, supporting its utility for individualized risk stratification and management.