Validating Sensor-based Approaches for Monitoring Eating Behavior and Energy Intake by Accounting for Real-World Factors that Impact Accuracy and Acceptability

项目摘要/摘要 能量摄入(EI)在慢性病的病因和预防中起着关键作用。 超重/肥胖和2型糖尿病等疾病。自我监控是自我管理的基石 监管途径，以减少伤害，但目前流行的方法是负担和不准确的，这限制了我们的 了解饮食模式并对其进行干预以改善健康的能力。显然有必要 创新的解决方案，可以不引人注意地监测和可靠地估计日常生活背景下的EI。10+ 多年来，我们团队一直在研究腕表设备(如智能手表)用于被动监控的效用 通过测量食物存在的主手手腕运动的加速度和旋转来测量进食行为 送到嘴里。通过几项研究，我们改进了使用手腕运动模式的方法。 识别个人在进餐/零食时的进食手势(“咬”食物，“喝”饮料)。我们已经展示了 我们可以使用进食手势计数来估计进食水平的Ei，通过使用高级建模来估计 每口千卡路里(KPB)和每杯饮料千卡(KPD)(例如，EI=#口×KPB+#饮品x KPD)。我们是 即将使这种方法广泛应用于临床，但我们在传感器方面的最新进展- 基于EI的估计需要验证，然后该方法才能在现实世界中真正可行。在这个项目中，我们 将最终解决3个障碍：1)我们的方法必须在各种环境和高度 有代表性的样本；2)我们使用摄食手势来估计Ei的模型必须考虑到变化 环境，如不同类型的食物或食物来源，可能影响EI；以及3)我们必须最大限度地 测量方法的可接受性。拟议的研究将验证我们基于传感器的Ei估计。 方法在不同的样本中，跨越三个环境(自助餐厅、家庭和自由生活)，包括 将用户对食品和饮料(例如，高能量密度食品、零卡路里饮料)的投入降至最低 上下文(例如，食物来源、时间)，并使用两种不同的传感器(商用智能手表和智能 铃声)。我们将进行两次受控数据收集，其中一顿饭被录像，而参与者 在自助餐厅佩戴智能手表和智能戒指：在自助餐厅佩戴N=300，在参与者家庭佩戴N=240。所有参与者 (n=540)将在4天的日常生活中佩戴这两种设备并完成远程食物摄影 (自由生活)。我们将根据使用视频(自助餐厅)捕获的地面事实来评估基于传感器的EI估计 和家庭)和远程食物摄影方法(自由生活)。我们将利用我们的发现来创建一个实用的 指导研究人员/临床医生实施基于传感器的EI自我监控方案的平台 最大限度地提高精确度和可接受性(选择手腕与戒指传感器、用户输入类型和自我识别的长度 监控)。我们的平台最终将通过改变我们开发和开发的方式来支持精准营养方面的工作 评估与健康相关的干预措施，最终提高针对EI的干预措施的质量。

Project Summary/Abstract Energy intake (EI) plays a critical role in the etiology and prevention of prevalent and debilitating chronic diseases such as overweight/obesity and type 2 diabetes. Self-monitoring is the cornerstone of the self- regulation approach for reducing EI, but prevailing methods are burdensome and inaccurate which limits our ability to understand eating patterns and intervene on them to improve health. There is a clear need for innovative solutions that can unobtrusively monitor and reliably estimate EI in the context of daily life. For 10+ years, our group has researched the utility of a wrist-watch device (e.g., smartwatch) to passively monitor eating behavior by measuring the acceleration and rotation of dominant-hand wrist motion of food being brought to the mouth. Through several studies we have refined our approach for using patterns of wrist motion to identify individual intake gestures ("bite" of food, "drink" of beverage) during meals/snacks. We have shown that we can use intake gesture count to estimate meal-level EI by using advanced modeling to estimate kilocalories per bite (KPB) and kilocalories per drink (KPD) (e.g., EI = #bites x KPB + #drinks x KPD). We are on the cusp of making this approach widely available for clinical application, but our latest advances in sensor- based EI estimation require validation before the method is truly viable in real-world settings. In this project we will definitively address 3 final barriers: 1) Our approach must be validated across settings and among a highly representative sample; 2) Our models that use intake gestures to estimate EI must account for varying contexts, such as different types of foods or food sources, that could influence EI; and 3) We must maximize acceptability of the measurement methods. The proposed study will validate our sensor-based EI estimation methods among a diverse sample, across three settings (cafeteria, home-based, and free-living), incorporating minimal user input on foods and beverages (e.g., high energy density foods, zero calorie beverages) and contexts (e.g., food source, time of day), and using two different sensors (commercial smartwatch and smart ring). We will conduct two controlled data collections in which a single meal is video recorded while participants wear the smartwatch and smart ring: N=300 in a cafeteria and N=240 in participant homes. All participants (N=540) will then wear both devices and complete remote food photography during 4 days of everyday life (free living). We will evaluate sensor-based estimates of EI against ground truth captured using video (cafeteria and home) and remote food photography method (free-living). We will use our findings to create a practical platform to guide researchers/clinicians implementing a sensor-based EI self-monitoring protocol that maximizes accuracy and acceptability (selecting wrist vs. ring sensor, type of user input, and length of self- monitoring). Our platform will ultimately support work in precision nutrition by transforming how we develop and evaluate health-related interventions, and ultimately improve the quality of interventions targeting EI.