Two-cohort machine learning approach for predicting the risk of secondary hyperlipidemia in patients with depression

Secondary hyperlipidemia is a common and serious complication in patients with clinically diagnosed depressive disorders, yet early screening tools are lacking. This study aims to develop and validate a machine learning-based model to predict the risk of secondary hyperlipidemia in this population.

We conduct a retrospective study of 627 patients (mean age: 44.5 ± 13.5 years; 51.9% female). LASSO regression was utilized for feature selection, followed by the development of seven predictive models is used. Model performance is evaluated using the Area Under the Curve (AUC), calibration curves, and Decision Curve Analysis (DCA). The SHapley Additive exPlanations (SHAP) method provides individual-level interpretation.

Six core predictors are identified: Body Mass Index (BMI), weekly physical activity, long-term medication, emotion regulation disorder, C-reactive protein (CRP), and Fasting Plasma Glucose (FPG). Among the evaluated models, the Decision Tree model demonstrates the most clinically appropriate and generalizable performance, with an AUC of 0.87 (95% CI: 0.82-0.92) in external validation.

The integration of machine learning with routine clinical data provides a highly accurate and interpretable tool for the early identification of secondary hyperlipidemia in patients with depression. This approach may facilitate personalized preventive interventions and improve long-term metabolic health outcomes in clinical practice.