Predicting rTMS treatment response in depression: use of machine learning models to identify the roles of metabolic and clinical factors.

Background: Repetitive transcranial magnetic stimulation (rTMS) is an effective treatment for depression in patients with major depressive disorder (MDD) and bipolar disorder (BD), but accurate prediction of treatment response remains a challenge. This study aims to identify key metabolic and clinical factors that serve as predictors of rTMS efficacy in patients with MDD and BD.

Methods: Sixty-eight patients (28 with BD) underwent 12 sessions of rTMS targeting the left dorsolateral prefrontal cortex. After treatment, patients were divided into remission and non-remission groups. Recursive feature elimination (RFE) and four decision tree-based machine learning algorithms were applied to develop prediction models.

Results: The Extra Trees classifier was the best predictor of remission at Week 4, with an area under the curve (AUC) of 0.766, accuracy of 0.726, and F1-score of 0.768. Key predictors included body mass index (BMI), baseline Hamilton Depression Rating Scale (HDRS) score, ratio of low-density lipoprotein (LDL) to high-density lipoprotein (HDL), systolic blood pressure (SBP), and high-sensitivity C-reactive protein (hsCRP). At Week 12, the Gradient Boosting model performed best (AUC = 0.863, accuracy = 0.773, F1-score = 0.817). Key predictors included homeostasis model assessment of insulin resistance (HOMA-IR), brain-derived neurotrophic factor (BDNF), Homeostasis Model Assessment of Beta-cell Function (HOMA-beta), total cholesterol, and LDL/HDL ratio.

Conclusion: These results suggest that metabolic and clinical factors may serve as predictors of rTMS outcomes, and decision tree-based machine learning models may be utilized for individualized treatment prediction.