Predicting traditional Chinese medicine constitutions in adults aged ≥ 65 years: A machine learning approach

This study aimed to predict biased traditional Chinese medicine (TCM) constitutions among individuals aged ≥ 65 years using machine learning models and to identify the key predictors of biased TCM constitutions.

This cross-sectional study enrolled 4403 older adults in Shanghai, China. Demographic, lifestyle and clinical data were collected. Six machine learning models were trained and compared: random forest (RF), gradient boosting machine (GBM), support vector machine (SVM), extreme gradient boosting (XGBoost), adaptive boosting classifier (AdaBoost) and logistic regression (LR).

Among these 4403 participants, 29.2% presented with biased TCM constitutions. RF demonstrated the highest predictive performance with an area under the curve (AUC) of 0.847 (indicating excellent discrimination), followed by GBM (AUC = 0.842), XGBoost (AUC = 0.840), AdaBoost (AUC = 0.830), SVM (AUC = 0.764) and LR (AUC = 0.759). Key predictors included age, heart rate, and specific blood parameters such as monocytes, alanine aminotransferase, platelet distribution width, total bilirubin, and creatinine.

The high prevalence of biased TCM constitutions among elderly adults underscores the need for targeted health management strategies. Machine learning models, particularly RF, can accurately predict biased TCM constitutions, enabling early identification of at-risk individuals. The identified predictors provide valuable insights for developing personalised preventive strategies and inform future research on TCM-based elderly healthcare. Please cite this article as: Sun C, Yu Z, Ge ZY, Wang WJ, Wang BY, Song HL, Xie GQ, Zhao HL, Zhang Y, Xu XL. Predicting traditional Chinese medicine constitutions in adults aged ≥ 65 years: A machine learning approach. J Integr Med. 2026; 24(1):98-104.