Cardiometabolic Index (CMI), Lipoprotein Combine Index (LCI), Conicity Index (CI), Weight-adjusted Waist Index (WWI), Waist-to-Hip-to-Height Ratio (WHHR), Body Surface Area (BSA) and the 10-year risk of hypertension: A machine learning approach in the Yazd Healthy Heart Project

Objective To evaluate and compare the power of novel indices in forecasting the 10-year risk of hypertension, and also to identify the most reliable predictor of hypertension using machine learning and conventional statistical techniques. Methodology: Data were obtained from 2,000 adults aged 20 to 74 years who were enrolled in the Yazd Healthy Heart Project and followed for 10 years. Participants underwent comprehensive assessments of anthropometric, biochemical, and lifestyle variables. The discriminative ability of each index was evaluated using Receiver Operating Characteristic (ROC) analysis, Cox regression models, and three machine learning algorithms: Random Forest, XGBoost, and LightGBM. Outcomes: The LCI and CMI demonstrated the strongest independent associations with hypertension (LCI: HR = 3.64, AUC = 0.72; CMI: HR = 2.93, AUC = 0.72). Among the machine learning models, XGBoost yielded the best predictive performance (AUC = 0.76, sensitivity = 76.0%, F1-score = 54.8%) and exhibited the smallest discrepancy between training and test results. Stratified analyses revealed that LCI was most predictive in middle-aged individuals, whereas CMI demonstrated greater predictive value in older adults. In contrast, the CI, WWI, and BSA lost statistical significance after multivariable adjustment. Conclusion The LCI and CMI outperformed traditional anthropometric measures in predicting hypertension across both conventional statistical analyses and machine learning models. Incorporating these indices into clinical screening protocols could enhance early detection and support more targeted prevention strategies.