Supplementary Material for: A 20-Second Video-Based Assessment of Cognitive Frailty: Results from a Cohort Study within the Precision Aging Network

Background: Cognitive frailty, the concurrent presence of mild cognitive impairment (MCI) and physical frailty, poses a significant risk for adverse outcomes in older adults. Traditional assessments that rely on extensive walking tests or specialized equipment, are impractical for routine or remote evaluations. This study evaluated a 20-second video-based Upper Frailty Meter (vFM) test, incorporating dual-task conditions, as a feasible tool for identifying cognitive frailty. Methods: Data from 413 participants aged 50–79 years in the Healthy Minds for Life cohort were analyzed across four sites: the University of Arizona, Johns Hopkins University, Emory University, and the University of Miami. Cognitive function was measured using the Montreal Cognitive Assessment (MoCA), whereas frailty indices were derived from the vFM test. Participants performed repetitive elbow flexion-extension under single-task (physical task only) and dual-task (physical task with concurrent cognitive exercise) conditions. Frailty phenotypes, including slowness, weakness, and exhaustion, were quantified using AI-based video kinematic analysis. Logistic regression and receiver operating characteristic (ROC) analyses evaluated the model's predictive accuracy for cognitive frailty. Results: Participants classified as cognitive frailty group (n=53, 12.8%) demonstrated significantly higher frailty index scores compared to robust individuals (p<0.001). Among all vFM derived parameters, the dual-task slowness phenotype demonstrated the strongest correlation with MoCA scores (r = -0.282, p < 0.001) and emerged as the most predictive single marker for distinguishing the cognitive frailty group, demonstrating high clinical applicability (Area Under the Curve [AUC] = 0.87). Combining single-task and dual-task metrics further enhanced predictive accuracy (AUC = 0.91), achieving sensitivity and specificity rates exceeding 85%. This combined approach significantly differentiated cognitive frailty from robust status, outperforming models based on age alone or single-task metrics. Conclusions: The 20-second vFM test offers a practical, non-invasive, easy-to-implement, and accessible solution for objectively evaluating cognitive frailty, demonstrating high predictive accuracy in distinguishing at-risk individuals. Its integration into telehealth platforms could enhance early detection and enable timely interventions, promoting healthier aging trajectories. Further longitudinal studies are recommended to validate its utility in tracking cognitive and physical decline over time.背景：认知衰弱（cognitive frailty）指轻度认知障碍（mild cognitive impairment, MCI）与躯体衰弱共存的状态，会显著提升老年人出现不良预后的风险。传统评估手段依赖长距离行走测试或专业设备，难以适配常规或远程评估场景。本研究对融合双任务范式的20秒视频版上肢衰弱评估仪（video-based Upper Frailty Meter, vFM）测试进行了评估，旨在将其作为识别认知衰弱的可行工具。 方法：本研究对「健康终生心智」（Healthy Minds for Life）队列中413名年龄介于50至79岁的受试者数据展开分析，受试机构涵盖亚利桑那大学、约翰·霍普金斯大学、埃默里大学与迈阿密大学四所院校。认知功能采用蒙特利尔认知评估量表（Montreal Cognitive Assessment, MoCA）进行测评，躯体衰弱指数则通过vFM测试计算得出。受试者需分别完成单任务（仅躯体运动任务）与双任务（躯体运动任务同步伴随认知训练）条件下的重复性肘关节屈伸动作。基于人工智能的视频运动学分析可量化动作迟缓、肌力减退与疲劳感等衰弱表型。本研究采用逻辑回归与受试者工作特征（Receiver Operating Characteristic, ROC）曲线分析，评估模型对认知衰弱的预测准确率。 结果：被归类为认知衰弱组的受试者（n=53，占比12.8%），其衰弱指数得分显著高于躯体健康个体（p<0.001）。在所有vFM衍生参数中，双任务动作迟缓表型与MoCA得分的相关性最强（r=-0.282，p<0.001），同时也是区分认知衰弱组的最优单一预测指标，展现出较高的临床应用价值（曲线下面积（Area Under the Curve, AUC）=0.87）。融合单任务与双任务指标可进一步提升预测准确率（AUC=0.91），灵敏度与特异度均超过85%。该联合模型可显著区分认知衰弱与躯体健康状态，其表现优于仅基于年龄或单任务指标的模型。 结论：这项20秒vFM测试为认知衰弱的客观评估提供了一种实用、无创、易操作且可及性强的解决方案，在区分高危人群方面展现出较高的预测准确率。将其整合至远程医疗平台，可助力早期筛查与及时干预，推动老年人健康老龄化进程。未来需开展纵向研究以验证其在追踪认知与躯体功能随时间衰退方面的应用价值。