Improving diabetes and depression self-management via adaptive mobile messaging

项目总结/摘要 糖尿病和抑郁症是主要的公共卫生问题，不成比例地影响种族/族裔少数群体 和美国的低收入人群。抑郁症和糖尿病的有效干预措施存在，但没有 尽管对这两种情况的治疗建议（特别是体力活动）相似，但经常合并使用。 特别是在资源受限的环境中，移动的健康（mHealth）技术具有成本效益， 提供自我管理支持的可行方法， 社会经济地位。现有的移动健康干预措施已取得初步成功，但仍存在困难。 持续参与。当与机器学习算法相结合时，健康信息可以适应 根据个人独特的个人资料来具体激励他们。在目标1中，我们将整合来自 针对糖尿病，抑郁症和身体活动的干预措施，应用用户设计方法。我们将利用 将现有的健康短信平台作为这一干预措施的基础。这将被称为糖尿病和精神 健康适应性通知跟踪和评估（DIAMANTE）研究。在目标2中，我们将测试移动健康 糖尿病和抑郁症的干预，将使用自适应机器学习生成消息 从患者步数数据（通过智能手机应用程序被动收集）和患者 输入血糖和情绪评分我们将比较这种适应性的，个性化的干预与静态的 消息传递干预，典型的许多现有的文本消息传递干预。在目标3中，我们将重新随机 对接受护士外展的无应答参与者使用顺序、多重分配、随机 试验（SMART）设计。我们将利用SMART设计来节省更昂贵的一对一护士 为那些不再参与该计划但最需要支持的患者提供外展服务。我们将测试 这项干预有350名来自英语和西班牙语安全网的患者。的主要结局 HbA 1c水平和PHQ-9评分。这项研究的结果将有助于我们了解 利用机器学习算法个性化内容以及提供临床医生支持的影响 为那些接受移动的保健干预的人提供服务。由于我们是在一个资源有限的 环境，这项研究的结果将适用于更广泛的人群。

Project Summary/Abstract Diabetes and depression are major public health problems that disproportionately affect racial/ethnic minorities and low-income individuals in the US. Efficacious interventions for depression and diabetes exist but are not often combined despite similar treatment recommendations (specifically physical activity) for both conditions. Especially in resource-constrained environments, mobile health (mHealth) technologies are cost effective and feasible methods for delivering self-management support given the more ubiquitous penetration across socioeconomic status. Existing mHealth interventions have shown preliminary success but have had difficulty sustaining engagement. When combined with machine learning algorithms, health messages can be adapted to specifically motivate individuals based on their unique profiles. In Aim 1, we will integrate content from interventions targeting diabetes, depression, and physical activity applying user design methods. We will utilize the existing HealthySMS platform as the basis for this intervention. This will be called the Diabetes and Mental Health Adaptive Notification Tracking and Evaluation (DIAMANTE) study. In Aim 2, we will test an mHealth intervention for diabetes and depression that will generate messages using an adaptive machine learning algorithm that learns from patient step count data (collected passively via a smartphone app) and patient entered blood glucose and mood ratings. We will compare this adaptive, personalized intervention with a static messaging intervention, typical of many existing text messaging interventions. In Aim 3, we will rerandomize non-responsive participants to receiving nurse outreach using a sequential, multiple assignment, randomized trial (SMART) design. We will leverage the SMART design to conserve more expensive one-on-one nurse outreach for the patients who are no longer engaged in the program and need the most support. We will test this intervention with 350 patients from a safety net setting in English and Spanish. The primary outcomes for this study are HbA1c levels and PHQ-9 scores. The results of this study will help us understand the impact of personalizing content utilizing machine learning algorithms as well as the impact of providing clinician support for those receiving mobile health interventions. Since we are testing this intervention in a resource-constrained environment, the results of this study will be relevant for a broader population.