Validity of the Apple Watch® for monitoring push counts in people using manual wheelchairs

Objective: A recent Apple Watch® activity-monitoring innovation permits manual wheelchair users to monitor daily push counts. This study evaluated the validity of the Apple Watch® push count estimate. Design: Criterion validity. Setting: Southern Finland and Southeast Queensland, Australia. Participants: Twenty-six manual wheelchair users from Finland and Australia were filmed completing a standardized battery of activities while wearing the Apple Watch® (dominant wrist). Outcome Measures: Wheelchair pushes as determined by the Apple Watch® were compared to directly observed pushes. Results: Agreement between Apple Watch® push counts and directly observed pushes was evaluated using Intraclass correlation coefficients (ICC), Pearson correlations and Bland-Altman analyses. Apple Watch® pushes and directly observed push counts were strongly correlated (ICC = 0.77, P r = 0.84, P ® underestimated push counts (M = −103; 95% ULoA = 217; LLoA = −423 pushes). Mean absolute percentage error was 13.5% which is comparable to studies evaluating agreement between pedometer-based step counts and directly observed steps. Conclusion: Apple Watch® push-count estimates are acceptable for personal, self-monitoring purposes and for research entailing group-level analyses, but less acceptable where accurate push-count measures for an individual is required.

研究目标：近期推出的苹果手表（Apple Watch®）活动监测创新功能，可支持手动轮椅使用者追踪每日轮椅推杆次数。本研究旨在评估苹果手表推杆次数估算结果的有效性。
研究设计：效标效度（Criterion validity）研究。
研究地点：芬兰南部与澳大利亚昆士兰州东南部。
研究对象：招募26名来自芬兰与澳大利亚的手动轮椅使用者，受试者佩戴苹果手表（佩戴于优势手手腕）完成标准化活动组合测试，测试过程全程录像记录。
结局指标：将苹果手表记录的轮椅推杆次数与人工直接观测得到的推杆次数进行对比分析。
结果：采用组内相关系数（Intraclass correlation coefficients, ICC）、皮尔逊相关系数（Pearson correlations）与布兰德-奥特曼分析（Bland-Altman analyses）评估苹果手表推杆次数与人工直接观测推杆次数之间的一致性。结果显示，二者具有较强相关性（ICC=0.77, P<0.001；皮尔逊相关系数r=0.84, P<0.001）。布兰德-奥特曼分析结果表明，苹果手表存在推杆次数低估现象（均值偏差M=-103；95%一致性界限上限（ULoA）=217次，下限（LLoA）=-423次）。平均绝对百分比误差为13.5%，该结果与评估计步器步数计数与人工观测步数间一致性的相关研究结果相当。
结论：苹果手表的推杆次数估算结果可用于个人自我监测以及群体水平的研究分析，但在需要获取个体精准推杆次数的场景中，其准确性尚不满足应用要求。