Establishment and validation of a clinical prediction model for perioperative pneumonia in elderly patients with hip fractures combined with preoperative stroke

Background Hip fractures in the elderly are associated with alarmingly high disability and mortality rates, which severely impair patients’ quality of life. Patients with a history of stroke face a significantly increased risk of perioperative pneumonia and a threefold higher risk of death. This study aimed to establish a clinical prediction model for perioperative pneumonia in elderly patients with hip fractures and preoperative stroke. Methods A total of 698 patients (244 in the pneumonia group and 454 in the non-pneumonia group) were retrieved from medical records and randomly divided into a training set and a validation set at a 7:3 ratio. The Least Absolute Shrinkage and Selection Operator (LASSO) was used for variable selection, and a nomogram prediction model was constructed. The model’s discriminative ability, calibration, and clinical utility were evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Shapley Additive explanations (SHAP) was employed to identify core predictive variables. Additionally, the predictive performance of 10 machine learning models was systematically compared. Results Pulmonary hypertension, respiratory failure, chronic obstructive pulmonary disease (COPD), surgical type, age, albumin level, hemoglobin level, and brain natriuretic peptide (BNP) level were identified as independent risk factors for perioperative pneumonia. The nomogram model had an area under the ROC curve (AUC) of 0.9203 in the training set and 0.7356 in the validation set. Calibration curves demonstrated good consistency between the model’s predicted probabilities and actual pneumonia risk. Decision curve analysis showed that the nomogram had clinical utility within the moderate-risk threshold range. SHAP analysis further identified albumin, hemoglobin, age, and BNP as core predictive variables. Among the machine learning models, logistic regression and linear discriminant analysis (LDA) exhibited optimal performance (both with an AUC of 0.743), achieving accuracies of 0.712 and 0.708, respectively. All models had a recall exceeding 0.680, precision ranging from 0.650 to 0.660, and high F1 scores. Conclusion This study established a risk prediction model for perioperative pneumonia in elderly patients with hip fractures and preoperative stroke using objective clinical indicators. The model shows good predictive performance and clinical applicability, enabling individualized risk assessment and early intervention for this patient population, with the potential to improve outcomes in high-risk individuals.