卒中后机器人康复与常规康复的运动恢复预测因素是否不同
本研究基于既往随机对照试验的回顾性数据,针对99例常规康复和91例机器人康复的卒中患者,分别训练机器学习模型预测FMA和ARAT两项运动结局,并比较基线预测因素在两种治疗模式中的贡献差异。结果显示两组预测误差相近,基线FMA在两组中均能强预测ARAT结局,忽视在机器人组对FMA预测影响更大,年龄仅在常规组为关键预测因素。
要点
- 基线FMA在两种治疗模式中均强预测ARAT结局。
- 忽视对FMA预测的影响在机器人组更突出。
- 年龄仅在常规康复组为两项结局的关键预测因素。
核心结论
- 不同治疗模式下基线预测因素的贡献存在差异,可为基于个体特征优化康复策略的临床决策支持系统提供依据。
研究设计与样本
- 基于既往评估机器人上肢康复的随机对照试验数据进行回顾性分析,数据清洗后纳入常规组99例、机器人组91例卒中患者。
- 训练并验证四种机器学习算法,采用嵌套交叉验证框架,分别针对机器人和常规治疗亚组开发模型。
- 以基线预测因素估计治疗后FMA和ARAT两项运动结局,并用SHAP分析评估各预测因素的贡献。
主要发现
- ARAT评分预测误差(中位数[绝对误差])在机器人组为5.0[6.0],常规组为3.0[6.0]。
- FMA预测误差在机器人组为5.0[5.0],常规组为5.0[7.0]。
- 基线FMA在两组中均为ARAT结局的强预测因素。
- 对于FMA预测,忽视在机器人组影响更大;年龄仅在常规组为两项结局的关键预测因素。
临床怎么用
- 可利用基线FMA预判两种模式下ARAT结局,为康复方案选择提供参考。
- 机器人康复决策时应关注忽视对FMA恢复预测的影响,常规康复则应重视年龄因素。
局限性
- 本研究为回顾性分析,数据来源于既往随机对照试验,样本量有限(常规组99例、机器人组91例)。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: While research on robotic rehabilitation has largely focused on evaluating device effectiveness, there remains a clear need to investigate patient characteristics that predict response to robot-assisted therapy. Recent advancements in machine learning (ML) techniques support such investigations by enabling the development of data-driven tools to assist in selecting appropriate rehabilitation treatments. This study aimed to develop two ML models, predicting two motor outcomes each, following either robot-assisted or conventional upper limb rehabilitation in post-stroke patients. It further sought to compare how baseline predictors contributed differently across the two treatment modalities. METHODS: We conducted a retrospective analysis of data from a previous randomized controlled trial evaluating robotic upper limb rehabilitation in post-stroke patients. Four ML algorithms were trained and validated using a nested cross-validation framework, with models developed separately for robotic and conventional treatment subgroups. Baseline predictors were used to estimate two post-treatment motor outcomes: Fugl-Meyer Assessment (FMA) and Action Research Arm Test (ARAT). SHAP analyses were performed to assess the contribution of each predictor to the models. RESULTS: After data cleaning, 99 patients in the conventional group and 91 in the robotic group were included. Prediction errors for ARAT score after training (expressed in median and absolute error) were 5.0 [6.0] in the robotic group and 3.0 [6.0] in the conventional group. For FMA prediction, results were 5.0 [5.0] and 5.0 [7.0], respectively. Baseline FMA was a strong predictor of ARAT outcomes in both groups. For FMA prediction, the presence of neglect emerged as more influential in the robotic group. Age was a key predictor of both outcomes, but only in the conventional group. CONCLUSIONS: The differing contributions of baseline predictors across treatment types provide clinically meaningful insights and support the development of clinical decision support systems aimed at optimizing rehabilitation strategies based on individual patient characteristics.
引用
- Campagnini S, Liuzzi P, Germanotta M, et al. Do predictors of motor recovery differ between robotic and conventional post-stroke rehabilitation? J Neuroeng Rehabil. 2026. PMID: 42121151. doi:10.1186/s12984-026-02014-6
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。