Enhancing Engagement with Digital Mental Health Care

加强与数字精神卫生保健的接触 摘要 数字精神卫生（DMH）是利用技术通过快速疾病改善人口福祉 检测、结果测量和护理1。尽管一些随机临床试验已经证明， 数字心理健康工具非常有效2-6，大多数消费者不能持续使用这些工具7-9。的 领域目前缺乏对DMH工具参与的理解，参与如何与幸福相关联， 以及什么样的实践能有效地维持参与。在美国心理健康协会 (MHA)，Talkspace（TS）和华盛顿大学（UW），我们提出了一个自然和实验， 理论驱动的研究计划10，11，目的是了解1）消费者如何参与自助 和临床医生辅助DMH的变化和存在什么样的参与模式，2）模式之间的关联， 参与度和重要的消费者成果，以及3）个性化策略的有效性， 与DMH治疗有关。这项研究将前瞻性地跟踪一个大的，自然的MHA样本， TS消费者，并将应用机器学习，以用户为中心的设计策略，以及微随机和 序贯多分配随机（SMART）试验，以解决这些目标。按照惯例， 平台，消费者将完成在线心理健康筛查和评估，我们将能够 根据疾病状态和症状严重程度对参与者进行分类。我们将使用的样本不会是 受诊断或合并症的限制。参与者将年满10岁，并进入MHA和TS 未来4年的平台。为了测试第一个目标，我们将确定至少10万个 过去访问过MHA和TS平台的消费者。将对参与者数据进行统计分析 揭示基于人口统计学、症状和平台的各组参与度和辍学率差异 活动对于aim 2，我们将使用监督机器学习技术来识别基于消费者的子类型。 人口统计学、DMH参与模式、脱离原因、现有MHA和TS的成功 参与策略，以及对DMH工具的满意度，这些都是未来参与模式的预测。 最后，基于目标2的结果，在目标3中，我们将采用以用户为中心的设计进行焦点小组讨论 战略，以确定和共同建立潜在的有效参与战略，为特定的客户亚型。我们 然后将进行一系列微观随机和SMART试验，以确定哪种理论驱动的参与 与用户共同设计的战略最适合根据目标2制定的子类型。我们将测试 这些战略的有效性，1）防止脱离那些更有可能有穷人， 脱离接触后的结果，2）改善从动机到意志的运动，3）提高最佳剂量 DMH的参与，从而改善心理健康的结果。这些数据将使用 采用纵向混合效应模型和效应编码来估计每种策略对客户的有效性 参与行为和心理健康结果。

Enhancing Engagement with Digital Mental Health Care Abstract Digital mental health (DMH) is the use of technology to improve population well-being through rapid disease detection, outcome measurement, and care 1. Although several randomized clinical trials have demonstrated that digital mental health tools are highly effective 2-6, most consumers do not sustain their use of these tools 7-9. The field currently lacks an understanding of DMH tool engagement, how engagement is associated with well-being, and what practices are effective at sustaining engagement. In this partnership between Mental Health America (MHA), Talkspace (TS) and the University of Washington (UW), we propose a naturalistic and experimental, theory-driven program 10,11 of research, with the aim of understanding 1) how consumer engagement in self-help and clinician assisted DMH varies and what engagement patterns exist, 2) the association between patterns of engagement and important consumer outcomes, and 3) the effectiveness of personalized strategies for optimal engagement with DMH treatment. This study will prospectively follow a large, naturalistic sample of MHA and TS consumers, and will apply machine learning, user-centered design strategies, and micro-randomized and sequential multiple assignment randomized (SMART) trials to address these aims. As is usual practice for both platforms, consumers will complete online mental health screening and assessment, and we will be able to classify participants by disease status and symptom severity. The sample we will be working with will not be limited by diagnosis or co-morbidities. Participants will be 10 years old and older and enter the MHA and TS platforms prospectively over 4 years. In order to test the first aim, we will identify a minimum of 100,000 consumers who have accessed MHA and TS platforms in the past. Participant data will be analyzed statistically to reveal differences in engagement and dropout across groups based on demographics, symptoms and platform activity. For aim 2, we will use supervised machine learning techniques to identify subtypes based on consumer demographics, engagement patterns with DMH, reasons for disengagement, success of existing MHA and TS engagement strategies, and satisfaction with the DMH tools, that are predictive of future engagement patterns. Finally, based on the outcomes from aim 2, in aim 3 we will conduct focus groups applying user-centered design strategies to identify and co-build potentially effective engagement strategies for particular client subtypes. We will then conduct a series of micro-randomized and SMART trials to determine which theory-driven engagement strategies, co-designed with users, have the greatest fit with subtypes developed under aim 2. We will test the effectiveness of these strategies to 1) prevent disengagement from those who are more likely to have poor outcomes after disengagement, 2) improve movement from motivation to volition and, 3) enhance optimal dose of DMH engagement and consequently improve mental health outcomes. These data will be analyzed using longitudinal mixed effects models with effect coding to estimate the effectiveness of each strategy on client engagement behavior and mental health outcomes.