This project aims to develop a risk prediction model and explore potential mechanisms underlying perioperative neurocognitive disorders (PND) in adults undergoing general anesthesia, using data from the UK Biobank. Kolmogorov-Arnold Networks (KANs), a recently introduced and interpretable deep learning architecture, will be employed to capture complex nonlinear interactions across clinical and biological domains. The study will incorporate a wide array of variables, including demographic and lifestyle factors, perioperative clinical exposures, anesthetic techniques, intraoperative medications, physiological monitoring data, surgical procedures, cognitive performance measures, comorbidities, medication history, imaging findings, and relevant biological markers. The primary objectives are: (1) to construct a robust and interpretable machine learning model for predicting the risk of PND; (2) to identify modifiable clinical and biological risk factors associated with postoperative cognitive impairment; and (3) to explore plausible mechanistic pathways linking perioperative events with long-term neurocognitive outcomes through multi-level data integration. The scientific rationale stems from the growing recognition of PND as a major perioperative complication that impacts recovery, independence, and long-term brain health. Early identification of at-risk individuals and deeper understanding of the contributing mechanisms may facilitate tailored anesthetic strategies, preventive interventions, and evidence-based perioperative decision-making. The findings of this study will be disseminated primarily through publication in peer-reviewed academic journals in the fields of perioperative medicine, anesthesiology, and computational medicine. Additional dissemination will include presentations at relevant international conferences and, where appropriate, data-sharing in line with UK Biobank policies.
