A Novel Risk System Integrating Clustering-derived Subtype and TyG Index for Predicting Postoperative Delirium in Colorectal Cancer Patients

This observational study aims to address the unmet clinical need for a simple, colorectal cancer-specific tool to predict postoperative delirium (POD), a prevalent complication linked to prolonged hospitalization and compromised recovery.

Eligible patients will undergo preoperative data collection, including demographic details, comorbidities, and laboratory tests (e.g., triglycerides, glucose for TyG index calculation, inflammatory and metabolic markers). Unsupervised K-means clustering will be applied to these multidimensional data to identify latent clinical subtypes, capturing complex interactions between metabolic status, inflammatory responses, and clinical characteristics that may influence POD risk.

Postoperatively, standardized POD assessment will be conducted twice daily for 7 days using the validated 3D-CAM tool, ensuring consistent identification of POD cases. No experimental interventions will be implemented-all participants receive routine preoperative evaluation, surgical care, and postoperative management per clinical guidelines.

Statistical analyses will first screen independent POD predictors via univariate and multivariate logistic regression. A scoring scale will then be developed by assigning weights to core predictors (clustering-derived subtypes, TyG index, and key clinical factors) based on their predictive strength (odds ratios). The scale's performance will be validated for discriminative ability (AUC) and calibration to ensure reliability in clinical practice.

The study's primary output is a user-friendly risk scoring tool that enables clinicians to rapidly assess POD risk preoperatively, facilitating targeted preventive strategies and improving patient outcomes without adding complexity to clinical workflows.