[Prediction model and feature analysis of pneumonia hospitalization duration based on CT radiomics and clinical...

Objective: To analyze clinical indicators and imaging data of hospitalized pneumonia patients and develop prediction models for length of hospital stay based on CT radiomics features and clinical indicators. Methods: Patients admitted to the First Clinical School of Hainan Medical University for pneumonia treatment between November 2020 and May 2024 were enrolled. Clinical data and CT imaging were collected, and radiomics features were extracted. Patients were divided into three groups based on the length of stay (<8 d, 8-28 d,>28 d, and poor prognosis). Three prediction models were constructed using logistic regression (LR): clinical features model, imaging features model and a combined clinical and imaging features model. The predictive performance of three models was evaluated using the area under the receiver operating characteristic (ROC) curve and DeLong test, followed by feature analysis. The Kappa test was used to compare the consistency of the overall classification. Results: A total of 343 subjects were included, with an average age of 61.46±20.98 years. The area under the curve (AUC) values of the combined model in different hospitalization duration classifications (<8 d, 8-28 d,>28 d, and poor prognosis) were 0.73, 0.65 and 0.78 in the training set, and 0.71, 0.65 and 0.76 in the validation set, respectively. The AUC values of the clinical feature model in different hospitalization duration classifications were 0.63, 0.64 and 0.73 in the training set, and 0.60, 0.52 and 0.63 in the validation set, respectively. The AUC values of the imaging feature model in different hospitalization duration classifications (<8 d, 8-28 d,>28 d, and poor prognosis) were 0.67, 0.66 and 0.73 in the training set, and 0.68, 0.54 and 0.66 in the validation set, respectively. The DeLong test results showed that the combined model outperformed other models in hospitalization duration classification (validation set AUC, all P<0.05, Bonferroni correction). Kappa test results showed that the combined model achieved the highest consistency between predicted classifications and actual hospitalization classifications (K=0.615). Shape features, texture features and clinical features all contributed proportionally to the combined model. Conclusion: The combined model integrating radiomics features with clinical indicators can significantly enhance the predictive efficacy for the length of hospitalization in pneumonia patients.