机器学习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研究

The Value of Machine Learning Models in Predicting Functional Outcomes afterRehabilitation in Patients with Subacute Stroke

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(R^(2))、均方根误差(Root Mean Squared Error,RMSE)。结果:机器学习模型(R^(2):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(R^(2):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。

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(R^(2))and Root Mean Squared Error(RMSE)of FIM gains.Results:Machine learning models demonstrated superior performance in predicting FIM motor scores(R^(2):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(R^(2):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.