PSG-Based Machine Learning for Predicting Acupuncture Efficacy in Insomnia

Insomnia affects nearly one-third of adults worldwide, significantly impairing qualify of life and increasing the risk of psychiatric and cardiovascular comorbidities. Acupuncture is widely applied as a safe and effective treatment, but individual variability in therapeutic response remains a major barrier to precision medicine. This study investigates the feasibility of predicting treatment response to acupuncture using pre-treatment polysomnography (PSG) parameters combined with machine learning models. A cohort of patients with chronic insomnia undergo comprehensive PSG assessment prior to acupuncture intervention. Key sleep features are extracted, and multiple classifiers are trained and validated using stratified crossvalidation and leave-one-out strategies. Model performance is evaluated by accuracy, precision, recall and f1 score. The proposed framework achieves an accuracy of 90.0 % with leave-one-out validation, outperforming stratified cross-validation (80.0 %). Feature analysis reveals that wake after sleep onset, sleep maintenance efficiency, and sleep efficiency are critical predictors of treatment response. PSG-based machine learning provides a feasible approach for individualized prediction of acupuncture efficacy in insomnia, supporting the development of precision treatment strategies.