Lukkahatai Nada, Chen Wanqi, Kawi Jennifer, Wu Hulin, Campbell Claudia M, Thrul Johannes, Huang Xinran, Christo Paul, Johnson Constance M
School of Nursing, Johns Hopkins University, Baltimore, MD 21205, USA.
School of Public Health, University of Texas Health Science Center at Houston, Houston, TX 77030, USA.
Clin Tradit Med Pharmacol. 2025 Jun;6(2). doi: 10.1016/j.ctmp.2025.200215. Epub 2025 Apr 14.
Chronic low back pain (cLBP) is a major cause of disability, with varied patient responses to treatments. Auricular point acupressure (APA) has shown potential as a non-pharmacological intervention, but individual responses may differ significantly.
This study aimed to determine the predictability of baseline characteristics, including functional disability, symptom severity, and treatment expectancy, on clinically significant responses to APA in reducing pain and improving function.
A secondary analysis was performed using data from a randomized controlled trial with 263 cLBP patients. Participants were randomly assigned to targeted APA (T-APA), non-targeted APA (NT-APA), or to a control group. APA responders were defined as those with at least a 1.5-point reduction in pain intensity or a 2.5-point improvement in the Roland-Morris Disability Questionnaire (RMDQ). Predictors of response were assessed using logistic regression and machine learning models, including the Random Forest and Support Vector Machine (SVM).
Baseline pain, physical function, sleep disturbance, and treatment expectancy were key predictors. The Random Forest model had the highest accuracy for T-APA; however, logistic regression performed best in NT-APA. SVM was most accurate in the control group, with predictive accuracy varying by group (AUC 60.9%-80%). The Least Absolute Shrinkage and Selection Operator (LASSO) method was found to be overly aggressive, often eliminating important variables.
This study highlights the variability in APA treatment responses for cLBP. While predictive models provide useful insights, further research with larger datasets is needed to improve prediction accuracy and generalizability, enhancing personalized treatment approaches for cLBP.
慢性下腰痛（cLBP）是导致残疾的主要原因，患者对治疗的反应各不相同。耳穴按压（APA）已显示出作为一种非药物干预手段的潜力，但个体反应可能存在显著差异。
本研究旨在确定包括功能障碍、症状严重程度和治疗期望在内的基线特征对APA在减轻疼痛和改善功能方面的临床显著反应的预测能力。
使用一项针对263例cLBP患者的随机对照试验的数据进行二次分析。参与者被随机分配到靶向APA（T-APA）组、非靶向APA（NT-APA）组或对照组。APA反应者定义为疼痛强度至少降低1.5分或罗兰-莫里斯残疾问卷（RMDQ）得分提高2.5分的患者。使用逻辑回归和机器学习模型（包括随机森林和支持向量机（SVM））评估反应的预测因素。
基线疼痛、身体功能、睡眠障碍和治疗期望是关键预测因素。随机森林模型对T-APA的预测准确率最高；然而，逻辑回归在NT-APA组中表现最佳。SVM在对照组中最准确，预测准确率因组而异（曲线下面积为60.9%-80%）。发现最小绝对收缩和选择算子（LASSO）方法过于激进，经常消除重要变量。
本研究强调了cLBP患者对APA治疗反应的变异性。虽然预测模型提供了有用的见解，但需要使用更大的数据集进行进一步研究，以提高预测准确性和可推广性，从而加强cLBP的个性化治疗方法。
