Big data vs accurate data in health research: Large-scale physical activity monitoring, smartphones, wearable devices and risk of unconscious bias
Big data vs accurate data in health research: Large-scale physical activity monitoring, smartphones, wearable devices and risk of unconscious bias
复制标题
DOI：
10.1016/j.mehy.2018.07.015
复制
发表时间：
2018-10-01
影响因子：
4.7
通讯作者：
Lord, S. R.
中科院分区：
文献类型：
--
作者：
Brodie, M. A.;Pliner, E. M.;Lord, S. R.
科学知识进步的基础是公正、准确且经过验证的测量技术。最近的联合国和具有里程碑意义的《自然》出版物强调了移动技术的全球应用以及大数据在鼓励人们锻炼身体和影响健康政策方面的巨大潜力。然而，人们对智能手机健康应用程序的不一致感到担忧。大数据有很多好处，但嘈杂的数据可能会导致错误的结论。针对低质量数据的可用性不断增加；我们呼吁对健康研究中用大数据替代准确数据的有效性进行严格辩论。我们评估了来自 111 个国家的 717,527 人之前使用的智能手机应用程序的计步准确性。我们的新数据（来自 48 名参与者；年龄 21-59 岁；体重指数 17.7-33.5 kg/m(2)）显示 Apple 手机存在显着的低估（15-66%）。与可穿戴设备在步行等典型跑步机上的普遍积极表现相反，我们观察到 Android 和 Apple 手机的误差范围非常大（步数的 0-200%）。无意识的偏见（开发人员对通常行为的看法）可能嵌入到许多未经验证的智能手机应用程序中。消费级可穿戴设备似乎不适合检测步态缓慢、短或非刻板模式的人的脚步。具体而言，存在系统性地低估肥胖者、女性或来自不同种族群体的人的步数的风险，从而在报告缺乏身体活动与肥胖之间的关联时导致偏差。需要更多的研究来开发适合全球异质人口中所有人的智能手机应用程序。
Fundamental to the advancement of scientific knowledge is unbiased, accurate and validated measurement techniques. Recent United Nations and landmark Nature publications highlight the global uptake of mobile technology and the staggering potential for big data to encourage people to be physically active and to influence health policy.However, concerns exist about inconsistencies in smartphone health apps. Big data has many benefits, but noisy data may lead to wrong conclusions. In reaction to the increasing availability of low quality data; we call for a rigorous debate into the validity of substituting big data for accurate data in health research.We evaluated the step counting accuracy of a smartphone app previously used by 717,527 people from 111 countries. Our new data (from 48 participants; aged 21-59 years; body mass index 17.7-33.5 kg/m(2)) revealed significant (15-66%) undercounting by Apple phones. In contrast to the generally positive performances of wearable devices for stereotypical treadmill like walking, we observed extraordinarily large (0-200% of steps taken) error ranges for both Android and Apple phones.Unconscious bias (developers' perceptions of usual behaviour) may be embedded into many unvalidated smartphone apps. Consumer-grade wearable devices appear unsuitable to detect steps in people with slow, short or non-stereotypical gait patterns. Specifically, there is a risk of systematically undercounting the steps by obese people, females or people from different ethnic groups resulting in biases when reporting associations between physical inactivity and obesity. More research is required to develop smartphone apps suitable for all people of the heterogeneous global population.