This is a multicenter ambispective observational cohort study led by Beijing Anzhen Hospital, Capital Medical University, with China-Japan Friendship Hospital serving as the external validation center.
Major adverse cardiovascular events (MACE) occurring after surgery are common complications among older adults undergoing hip fracture surgery and are associated with poor long-term outcomes, including increased one-year mortality. Previous studies have not systematically evaluated perioperative risk factors or developed prediction models specifically for this population. In addition, conventional statistical approaches may have limited ability to account for complex interactions and nonlinear associations among multiple perioperative variables. Therefore, this study aims to identify risk factors for postoperative MACE and to develop and externally validate a machine learning-based prediction model.
Patients aged 60 years or older who undergo surgical treatment for hip fracture will be included. The derivation cohort, established at Beijing Anzhen Hospital, will be used for model development and internal validation, whereas the independent external validation cohort, established at China-Japan Friendship Hospital, will be used to evaluate the generalizability and predictive performance of the developed model. The study will adopt a bidirectional cohort design: one portion of the study population will be identified retrospectively from electronic medical records, and the remaining participants will be consecutively enrolled prospectively following ethics committee approval. Postoperative outcomes will be ascertained from the end of surgery until the earliest occurrence of hospital discharge, death, or postoperative day 30.
Standardized, de-identified data will be collected, including demographic characteristics, comorbidities, laboratory findings, surgical and anesthetic information, perioperative medications, and postoperative MACE outcomes. No study-specific intervention will be assigned, and all patients will receive routine clinical care. Data will be managed in accordance with institutional privacy policies and applicable data protection requirements.
Candidate predictors will be evaluated using univariable analyses, least absolute shrinkage and selection operator regression, and random forest-based feature selection. Prediction models, including logistic regression, XGBoost, and LightGBM, will be developed in the derivation cohort. Internal validation will be performed using resampling methods, and external validation will be conducted using data from the independent validation center. Model performance will be assessed in terms of discrimination, calibration, and clinical utility using the area under the receiver operating characteristic curve, calibration plots, and other appropriate performance measures.