Construction of a composite risk index combining triglyceride glucose-body mass index and smoking behavior and its application in predicting prehypertension risk
LI Quan-yi
LI Meng-huan
JIANG Wei
PAN Ping
PAN Yang
SUN Qiu
Abstract：Objective To construct a comprehensive risk index(CRI)integrating insulin resistance marker(TyG-BMI)and smoking behavior,and develop a high-performance machine learning model for prehypertension risk prediction.Methods A retrospective cohort of 986 examinees from Nanjing Brain Hospital(2020-2025)was enrolled.The innovative CRI formula was established using an improved snow goose algorithm(ISGA)optimized via Cubic chaotic initialization and dimension-by-dimension Gaussian mutation to determine optimal decision variables.The core predictors were screened by Lasso regression,and six machine learning models were constructed.Class imbalance was addressed by SMOTE,and hyperparameters were tuned.The evaluation metrics included area under the receiver operating characteristic curve(ROC-AUC),area under the precision-recall curve(PR-AUC),calibration curves and decision curve analysis(DCA).SHAP interpreted model features.Results Lasso regression identified six key predictors for prehypertension risk:CRI,age,waist circumference,family history of hypertension,high density lipoprotein cholesterol(HDL-C)and glucose.Six machine learning models-Logistic Regression(LR),Support Vector Machine(SVM),Random Forest,Backpropagation Neural Network(BPNN),XGBoost and LightGBM were subsequently developed using these variables.The comparative analysis demonstrated LightGBM's superior discriminative performance,achieving a test set ROC-AUC of 0.9495 and PR-AUC of 0.9526,significantly outperforming the other models.DCA revealed substantial net benefit within the 1%to 90%threshold range,and the calibration curve exhibited a Brier score of 0.1084.SHAP analysis prioritized core predictors as follows:CRI,age,waist circumference,family history of hypertension,HDL-C and fasting blood glucose.Conclusion The chronic disease risk assessment model developed in this study,integrating swarm intelligence optimization and explainable AI(XAI),features a core innovative design based on integer-constrained weights.The model enables clinicians to perform reliable,quantifiable individual risk stratification.It thereby provides a practical tool foundation for optimizing early intervention strategies and healthcare resource allocation.
Keywords：Triglyceride glucose indexBody mass indexSmokeComprehensive risk indexPrehypertensionRisk predicting modelMachine learningLight GBM
Publication Date：2025-12-20
Online Publishing Date：2026-01-04(First online date of this platform, not the publication date of the document)
Pages：9( 1121-1129 )