Combined predictive value of TyG index and PLR for mental health in Chinese adults: A machine learning approach

To investigate the independent and combined associations of the triglyceride-glucose (TyG) index and platelet-to-lymphocyte ratio (PLR) with mental health in Chinese adults, and to evaluate their predictive value using both traditional regression and machine learning approaches.

Data from 725 adults at Qilu Hospital of Shandong University were analyzed. Mental health was evaluated using Symptom Checklist-90 (SCL-90). Fasting blood samples were used to calculate the TyG index and PLR. Linear and logistic regressions evaluated associations with overall and domain-specific mental health. The predictive performance of the TyG index and PLR was further evaluated using logistic regression and six machine learning classifiers, based on an 80/20 train-test split with cross-validation. Multiple performance metrics-including AUC, sensitivity, specificity, and MCC-were reported, and LASSO regression was applied to identify key predictors.

The TyG index was positively associated with the SCL-90 total score, somatization, interpersonal sensitivity, depression, anxiety, and hostility (P < 0.05); PLR showed similar associations and was also associated with phobic anxiety and psychoticism (P < 0.05). Individuals in the high-TyG/high-PLR group had significantly higher scores across all SCL-90 dimensions (P < 0.05), except for obsessive-compulsive symptoms and paranoid ideation. Among the six machine learning models, LASSO regression demonstrated the best overall predictive performance for mental health problems (AUC = 0.744), showing balanced sensitivity, specificity, F1-score, and MCC. PLR and the TyG index were identified as the strongest positive predictors. Compared with traditional logistic regression, machine learning models showed superior discriminative ability and enabled the assessment of variable importance.

The TyG index and PLR are independently and jointly associated with mental health indicators in Chinese adults. Their combination may enhance the ability to identify individuals at elevated risk for mental health problems and could serve as a useful biomarker pair in predictive applications.