目的:探讨机器学习模型与逐步线性回归(Stepwise linear regression,SLR)模型在亚急性期脑卒中患者康复后功能结局预测中的价值.方法:选取中国人民解放军联勤保障部队第九四五医院2013年1月～2023年12月收治的亚急性期脑卒中患者1046例为研究对象,取患者一般资料以及入院时功能独立性量表(Functional Independence Measure,FIM)评分构建SLR、回归树(Regression trees,RT)、集成学习(Ensemble learning,EL)、人工神经网络(Artificial neural network,ANN)、支持向量回归(Support vector regression,SVR)以及高斯过程回归(Gaussian process regression,GPR)预测模型,并采用10折交叉验证,比较各模型实际与预测出院FIM评分以及FIM增益的决定系数(R2)、均方根误差(Root Mean Squared Error,RMSE).结果:机器学习模型(R2:RT=0.75,EL=0.78,ANN=0.81,SVR=0.80,GPR=0.81)在预测FIM运动评分方面优于SLR(0.70).机器学习模型对FIM增益总分的预测准确性(R2:RT=0.48,EL=0.51,ANN=0.50,SVR=0.51,GPR=0.54)也优于SLR(0.22).结论:机器学习模型在预测FIM预后方面优于SLR;仅包含患者一般信息和入院FIM评分的机器学习模型的预测准确性优于既往研究,同时GPR对FIM预后的预测准确性最高.
The Value of Machine Learning Models in Predicting Functional Outcomes after Rehabilitation in Patients with Subacute Stroke
Objective:This study aims to evaluate the efficacy of machine learning models compared to stepwise linear regression(SLR)in predicting functional outcomes following rehabilitation in patients with subacute stroke.Methods:A total of 1,046 subacute stroke patients admitted to the 945th Hospital of the Joint Logistics Support Force from January 2013 to December 2023 were in-cluded in this study.Patient demographics and Functional Independence Measure(FIM)scores at admission were used to construct various predictive models including SLR,Regression Trees(RT),Ensemble Learning(EL),Artificial Neural Networks(ANN),Sup-port Vector Regression(SVR),and Gaussian Process Regression(GPR).These models were evaluated using 10-fold cross-validation to compare the actual and predicted discharge FIM scores and the coefficients of determination(R2)and Root Mean Squared Error(RMSE)of FIM gains.Results:Machine learning models demonstrated superior performance in predicting FIM motor scores(R2:RT=0.75,EL=0.78,ANN=0.81,SVR=0.80,GPR=0.81)compared to SLR(0.70).These models also showed higher accuracy in predict-ing total FIM gain scores(R2:RT=0.48,EL=0.51,ANN=0.50,SVR=0.51,GPR=0.54)than SLR(0.22).Conclusion:Machine learn-ing models outperform SLR in predicting FIM outcomes.The accuracy of predictions using only patient demographics and admis-sion FIM scores in machine learning models was superior to previous studies,with GPR showing the highest predictive accuracy for FIM outcomes.
strokesubacute phasemachine learningpredictionfunctional independence measurestepwise linear regressionarti-ficial neural network
国家科技期刊平台
NSTL
万方数据
