Prediction Model for MINS After Major Hepatobiliary Surgery

Overview

This multi-center, prospective observational study aims to develop and validate an interpretable prediction model for Myocardial Injury After Noncardiac Surgery (MINS) in patients undergoing major hepatobiliary surgery. The study adopts a nested modeling strategy, starting with baseline risk factors (e.g., RCRI) and stepwise incorporating hepatic inflow occlusion strategies (specifically comparing SPVO vs. Pringle maneuver) and routine intraoperative biomarkers. The model's performance will be evaluated using AUC, Net Reclassification Improvement (NRI), and Decision Curve Analysis (DCA), followed by interpretability analysis using SHAP values and external validation in an independent cohort.

Study Design and Methodology: The study consists of four consecutive phases aimed at constructing a robust and interpretable prediction model for MINS. 1. Multi-center Cohort Standardization: Based on clinical data from multiple participating centers, the investigators will establish a standardized structural dataset. Strict inclusion and exclusion criteria will be applied. The process involves rigorous data cleaning and normalization to harmonize demographics, surgical operation details, and perioperative hemodynamic parameters across different centers, laying the foundation for model construction. 2. Nested Modeling and Performance Evaluation: A nested modeling strategy will be employed to assess the incremental predictive value of specific surgical and biological variables: Model A (Baseline): Constructed using standard baseline variables such as the Revised Cardiac Risk Index (RCRI). Model B (+Surgical Technique): Incorporates hepatic inflow occlusion strategies, specifically comparing SPVO (Selective Pringle Vascular Occlusion) vs. Pringle maneuver, along with occlusion duration and frequency. Model C (Full Model): Further incorporates MINS-related biomarkers. Model performance will be comprehensively evaluated using: Discrimination: Area Under the Receiver Operating Characteristic Curve (AUC). Calibration: Calibration plots. Clinical Utility: Net Reclassification Improvement (NRI) and Decision Curve Analysis (DCA) to assess the improvement in risk stratification and clinical net benefit after adding new variables. 3. Model Interpretability Analysis: To enhance the transparency of the model ("White-box" approach), SHAP (SHapley Additive exPlanations) values or similar methods will be utilized. This will quantify and visualize the specific contribution (weight) of key variables, such as SPVO usage, to the individual risk prediction, aligning the statistical results with clinical medical reasoning. 4. External Validation: The final model will undergo validation using an independent external clinical cohort. This step aims to test the stability and generalizability of the model across different center data, defining its applicable scope in real-world clinical scenarios.