Development and Validation of Machine Learning-Based Risk Prediction Model for Stroke-Associated Pneumonia in Elderly Hemorrhagic Stroke

Yi Cao¹Haipeng Deng¹Shaoyun Liu¹Xi Zeng¹Yangyang Gou²Weiting Zhang²Yixinyuan Li²Hua Yang^1*Min Peng^3*

• ¹Department of Neurosurgery, Affiliated Hospital of Guizhou Medical University, Guiyang, China
• ²School of Nursing, Affiliated Hospital of Guizhou Medical University, Guiyang, China
• ³Department of Nursing Quality Management, Affiliated Hospital of Guizhou Medical University, Guiyang, China

Objective: To develop and validate an ML - based model for predicting SAP risk in elderly hemorrhagic stroke patients.Methods: We retrospectively collected elderly hemorrhagic stroke patients from three tertiary hospitals in Guiyang (Jan 2019 - Dec 2022) as the modeling cohort, split into training and internal validation sets (7:3). External validation used data from Jan - Dec 2023. After regression analyses, four ML models (Logistic Regression, XGBoost, Naive Bayes, SVM) were built. ROC curves, AUC, Delong's or Bootstrap tests, and various performance metrics were used for evaluation. Calibration curves were used to assess model calibration.Results: 788 patients were enrolled (training: 462, internal validation: 196, external validation: 130). The SAP incidence was 46.7% (368/788). Risk factors included advanced age, smoking, low GCS and Braden scores, and nasogastric tube. The LR model showed the best and most stable performance. It achieved AUCs of 0.883 (training), 0.855 (internal validation), and 0.882 (external validation). H - L test P - values were 0.381, 0.142, and 0.066, indicating satisfactory calibration.Conclusions: A multi - center SAP risk prediction model for elderly hemorrhagic stroke patients was constructed and validated. The LR model had the best performance. Easily - obtainable factors like age, smoking etc. were identified. The model has good generalization ability, and a nomogram was drawn for clinical use to reduce SAP incidence and improve prognosis.