Research Questions:
Which integrated set of clinically accessible preoperative, intraoperative, and postoperative variables most accurately predicts PND risk in elderly surgical patients?
Does a machine learning model derived from UK Biobank data maintain predictive performance when externally validated in a prospective, independent cohort?
Can a simplified risk stratification tool improve early identification of high-risk patients in routine clinical practice?
Objectives:
1.Primary: To develop and internally validate a PND risk prediction model for elderly patients using UK Biobank data, integrating demographic, clinical, surgical, and cognitive reserve variables through machine learning (e.g., XGBoost, logistic regression) and rigorous TRIPOD-compliant methodology.
2.Secondary: To externally validate the model in a prospectively recruited cohort of Chinese elderly surgical patients , evaluating its discriminative power (AUC), calibration, and clinical utility across diverse healthcare settings.
Scientific Rationale:
Perioperative neurocognitive disorders (PND) are a common and devastating complication in older surgical patients, with an incidence up to 65%, leading to long-term cognitive decline and increased dementia risk. Existing models are limited by retrospective designs, narrow variable selection, and lack of external validation. While other UK Biobank projects focus on novel algorithms or multi-omics mechanisms, this study emphasizes immediate clinical applicability by leveraging routinely available perioperative data. By combining UK Biobank’s large-scale data with prospective validation, this research addresses critical gaps in model generalizability and translational potential, directly supporting preoperative risk stratification and personalized interventions for vulnerable aging populations.