Data_Sheet_1_Machine learning approach to classifying declines of physical function and muscle strength associated with cognitive function in older women: gait characteristics based on three speeds.docx

BackgroundThe aging process is associated with a cognitive and physical declines that affects neuromotor control, memory, executive functions, and motor abilities. Previous studies have made efforts to find biomarkers, utilizing complex factors such as gait as indicators of cognitive and physical health in older adults. However, while gait involves various complex factors, such as attention and the integration of sensory input, cognitive-related motor planning and execution, and the musculoskeletal system, research on biomarkers that simultaneously considers multiple factors is scarce. This study aimed to extract gait features through stepwise regression, based on three speeds, and evaluate the accuracy of machine-learning (ML) models based on the selected features to solve classification problems caused by declines in cognitive function (Cog) and physical function (PF), and in Cog and muscle strength (MS).MethodsCognitive assessments, five times sit-to-stand, and handgrip strength were performed to evaluate the Cog, PF, and MS of 198 women aged 65 years or older. For gait assessment, all participants walked along a 19-meter straight path at three speeds [preferred walking speed (PWS), slower walking speed (SWS), and faster walking speed (FWS)]. The extracted gait features based on the three speeds were selected using stepwise regression.ResultsThe ML model accuracies were revealed as follows: 91.2% for the random forest model when using all gait features and 91.9% when using the three features (walking speed and coefficient of variation of the left double support phase at FWS and the right double support phase at SWS) selected for the Cog+PF+ and Cog–PF– classification. In addition, support vector machine showed a Cog+MS+ and Cog–MS– classification problem with 93.6% accuracy when using all gait features and two selected features (left step time at PWS and gait asymmetry at SWS).ConclusionOur study provides insights into the gait characteristics of older women with decreased Cog, PF, and MS, based on the three walking speeds and ML analysis using selected gait features, and may help improve objective classification and evaluation according to declines in Cog, PF, and MS among older women.

背景：衰老过程与认知和物理功能的衰退相关，这种衰退影响神经运动控制、记忆力、执行功能和运动能力。既往研究致力于寻找生物标志物，利用诸如步态等复杂因素作为评估老年人认知和身体健康的指标。然而，尽管步态涉及多种复杂因素，如注意力、感觉输入的整合、与认知相关的运动规划和执行，以及肌肉骨骼系统，但综合考虑多个因素进行生物标志物研究的工作尚显匮乏。本研究旨在通过逐步回归方法，基于三种速度提取步态特征，并评估基于所选特征的机器学习（ML）模型在解决由认知功能（Cog）和身体功能（PF）的下降以及Cog和肌肉力量（MS）下降所引起的分类问题时的准确性。方法：对198名65岁或以上的女性进行了认知评估、五次坐站测试和握力测试，以评估Cog、PF和MS。对于步态评估，所有参与者以三种速度（首选步行速度（PWS）、较慢步行速度（SWS）和较快步行速度（FWS））沿着19米长的直线行走。基于三种速度提取的步态特征通过逐步回归方法进行选择。结果：ML模型的准确率如下：在使用所有步态特征时，随机森林模型的准确率为91.2%，在使用三种特征（FWS时的步行速度和左双支撑相变系数变异性以及SWS时的右双支撑相变系数变异性）时，Cog+PF+和Cog–PF–分类的准确率为91.9%。此外，支持向量机在使用所有步态特征和两个所选特征（PWS时的左步时间以及SWS时的步态不对称性）时，Cog+MS+和Cog–MS–分类问题的准确率为93.6%。结论：本研究基于三种步行速度和所选步态特征的机器学习分析，揭示了认知、身体功能和肌肉力量下降的老年女性的步态特征，并可能有助于根据认知、身体功能和肌肉力量的下降情况，提高对老年女性的客观分类和评估。