Velmovitsky Pedro Elkind, Alencar Paulo, Leatherdale Scott T, Cowan Donald, Morita Plinio Pelegrini
School of Public Health Sciences, University of Waterloo, Waterloo, ON, Canada.
David R. Cheriton School of Computer Science, University of Waterloo, Waterloo, ON, Canada.
Digit Health. 2024 Jun 8;10:20552076241249931. doi: 10.1177/20552076241249931. eCollection 2024 Jan-Dec.
Public health surveillance involves the collection, analysis and dissemination of data to improve population health. The main sources of data for public health decision-making are surveys, typically comprised of self-report which may be subject to biases, costs and delays. To complement subjective data, objective measures from sensors could potentially be used. Specifically, advancements in personal mobile and wearable technologies enable the collection of real-time and continuous health data.
In this context, the goal of this work is to apply a mobile health platform (MHP) that extracts health data from the Apple Health repository to collect data in daily-life scenarios and use it for the prediction of stress, a major public health issue.
A pilot study was conducted with 45 participants over 2 weeks, using the MHP to collect stress-related data from Apple Health and perceived stress self-reports. Apple, Withings and Empatica devices were distributed to participants and collected a wide range of data, including heart rate, sleep, blood pressure, temperature, and weight. These were used to train random forests and support vector machines. The SMOTE technique was used to handle imbalanced datasets.
Accuracy and f1-macro scores were in line with state-of-the-art models for stress prediction above 60% for the majority of analyses and samples analysed. Apple Watch sleep features were particularly good predictors, with most models with these data achieving results around 70%.
A system such as the MHP could be used for public health data collection, complementing traditional self-reporting methods when possible. The data collected with the system was promising for monitoring and predicting stress in a population.
公共卫生监测涉及数据的收集、分析和传播，以改善人群健康。公共卫生决策的数据主要来源是调查，通常由自我报告组成，这可能会受到偏差、成本和延迟的影响。为了补充主观数据，可以潜在地使用来自传感器的客观测量方法。具体而言，个人移动和可穿戴技术的进步使得能够收集实时和连续的健康数据。
在此背景下，本研究的目标是应用一个移动健康平台（MHP），该平台从苹果健康存储库中提取健康数据，以在日常生活场景中收集数据，并将其用于预测压力，这是一个主要的公共卫生问题。
对45名参与者进行了为期2周的试点研究，使用MHP从苹果健康和感知压力自我报告中收集与压力相关的数据。向参与者分发了苹果、Withings和Empatica设备，收集了包括心率、睡眠、血压、体温和体重在内的广泛数据。这些数据用于训练随机森林和支持向量机。采用SMOTE技术处理不平衡数据集。
对于大多数分析和样本，准确率和f1宏分数与用于压力预测的最先进模型一致，超过60%。苹果手表的睡眠功能是特别好的预测指标，大多数包含这些数据的模型都取得了约70%的结果。
诸如MHP这样的系统可用于公共卫生数据收集，在可能的情况下补充传统的自我报告方法。该系统收集的数据有望用于监测和预测人群中的压力。
