Machine Learning Predicts Significant Improvement in Motor Aphasia with Tongue Acupuncture

Qixiu Wang¹Guoce Zhan²Hongfei Zhou^2*Jun Liu^2*

• ¹Liaoning University of Traditional Chinese Medicine, Shenyang, China
• ²Affiliated Hospital of Liaoning University of Traditional Chinese Medicine, Shenyang, Liaoning Province, China

Objective: Motor aphasia is a common language disorder that significantly disrupts patients' communication abilities and quality of life. Recent studies have shown that acupuncture treatment is effective for motor aphasia, but in clinical practice, the selection of acupuncture points for motor aphasia is diverse and lacks a unified standard. Therefore, by analyzing a range of clinical parameters encompassing multiple acupuncture points, we identified independent predictive factors for recovery from motor aphasia following acupuncture treatment. Materials and Methods: This retrospective case-control study included 144 patients with motor aphasia at Liaoning University of Traditional Chinese Medicine Affiliated Hospital (2019–2023). Propensity score matching (PSM) balanced baseline characteristics (age, gender, disease factors, comorbidities) using 1:1 nearest neighbor matching (caliper=0.2 SD). LASSO, Random Survival Forest, and Gradient Boosting Machine algorithms selected 44 variables, and a multivariate Cox regression model assessed treatment outcomes. Results: After PSM, baseline characteristics were balanced between the treatment group (tongue acupuncture, n=40) and the control group (n=40) (SMD<0.1). Cross-analysis using LASSO, RSF, and GBM confirmed that age, time to rehabilitation start (TSR), and tongue acupuncture treatment are key predictive factors. Multivariate Cox regression analysis revealed that age ≥60 years (HR=0.10, 95% CI: 0.02-0.50, p=0.005) and TSR ≥12 days (HR=0.41, 95% CI: 0.20-0.82, p=0.031) are risk factors for recovery, while tongue acupuncture treatment (HR=2.92, 95% CI: 1.29-6.62, p=0.010) significantly improved treatment outcomes. Model performance was robust, with AUC values of 0.91±0.07, 0.89±0.08, and 0.89±0.07 for LASSO, RSF, and GBM, respectively, and Cox model AUC of 0.88. Patients were categorized into low-risk (age < 60 years, TSR < 12 days, receiving tongue acupuncture) and high-risk groups, with significant differences observed (HR=0.31, 95% CI: 0.16-0.61, p<0.001). Conclusion: Tongue acupuncture enhances motor aphasia recovery, while older age and delayed rehabilitation hinder it. PSM and machine learning ensured robust predictions, supporting early tongue acupuncture. Future multicenter studies will further validate these findings.