Optimizing population health outcomes in diabetic retinopathy through personalized and scalable screening strategies

项目摘要/摘要 根据国家眼科研究所(NEI)的说法，早期发现和及时治疗可以降低严重视力丧失的风险 糖尿病视网膜病变(DR)的发病率增加了95%。然而，糖尿病仍然是导致美国成年人失明的主要原因。到2030年， 预计将有5490万美国人患有糖尿病。威胁视力的糖尿病视网膜病变(VTDR)的发病率 在美国和全世界，糖尿病患者的比例分别为4.4%和10.2%，代表了2800多万人 冒着失明的危险。因为DR在早期阶段不会引起疼痛、视力丧失或其他症状，而且只有50%的人 糖尿病患者每年都会接受眼科检查，在视力丧失不可逆转之前，许多人都不会意识到自己的疾病。 视网膜护理公司(RCI)认为，这种公共健康失败的根本原因更多的是行为和教育，而不是 临床；完全坚持筛查和治疗将防止博士几乎所有的视力损失，然而，100% 坚持是不可行的、不实际的、也不划算的。消除灾难恢复的失明需要进行战略转变， 承认预防DR造成的视力损失并不需要100%的年筛查率；它只需要所有 患有VTDR的患者由眼科护理提供者进行评估，并遵守后续建议。 RCI经济高效地消除DR失明的方法是优先考虑最有可能需要 立即关注并投入必要的资源，以确保眼科护理提供者对其进行评估。建议数 该项目旨在通过VTDR风险预测、有针对性的患者参与、教育和 行为干预；以及优化整个系统，以实现人口利益的最大化。 在目标1中，RCI将利用我们现有的数据存储库和机器学习方法来预测VTDR风险 电子健康记录(EHR)和医疗保险索赔数据，初始目标是正确放置90%以上的 当按风险排序时，VTDR患者处于最高风险人群的一半。通过这样做，RCI可以专注于患者 为最有可能需要立即评估和治疗的患者提供参与资源，而不是转移注意力 向不太可能需要立即关注的患者提供资源。 在目标2中，RCI将使用综合行为模型框架下的混合方法来确定障碍和激励因素 筛选和评估其相对重要性，制定和实施一项调查工具，以吸引参与意愿 在基于激励因素和障碍消除的筛选中，并使用自然语言处理来检测障碍和 来自患者沟通的糖尿病眼部筛查的促进者。 在目标3中，RCI将创建一个基于代理的模拟工具，以指导护理协调建议 在时间和成本的限制下，最大限度地提高人口健康成果。该工具将确定最佳干预措施 策略，并适应成本、疾病负担、患者偏好、人群特征和其他方面的变化 参数。将进行优化的决策包括根据风险、干预时机、筛选等因素进行资源分配 形态、沟通策略、障碍消除策略，以及系统的其他可修改方面。

Project Summary / Abstract According to the National Eye Institute (NEI), early detection and timely treatment can reduce the risk of severe vision loss from diabetic retinopathy (DR) by 95%. Yet, DR remains the leading cause of blindness among American adults. By 2030, 54.9 million Americans are expected to have diabetes. The prevalence of vision-threatening diabetic retinopathy (VTDR) among people with diabetes is 4.4% and 10.2% in the US and worldwide, respectively, representing over 28 million people at risk of blindness. Because DR causes no pain, vision loss, or other symptoms at its early stages and only 50% of people with diabetes receive annual eye exams, many will be unaware of their disease until vision loss is irreversible. Retinal Care Inc. (RCI) believes that the root cause of this public health failure is more behavioral and educational than clinical; complete adherence to screening and treatment would prevent nearly all vision loss from DR. However, 100% adherence is not feasible, practical, or cost-effective. Eliminating blindness from DR requires a strategy shift that acknowledges that preventing vision loss from DR does not require a 100% annual screening rate; it requires only that all patients with VTDR are evaluated by an eye care provider and adhere to follow up recommendations. RCI's approach to cost-effectively eliminating blindness from DR is to prioritize patients who are most likely to require immediate attention and devote the resources necessary to ensure they are evaluated by an eye care provider. The proposed project is designed to accomplish this goal through VTDR risk prediction; targeted patient engagement, education, and behavioral interventions; and optimization of the full system to achieve maximum population benefit. In Aim 1, RCI will leverage our existing Data Repository and machine learning methods to predict VTDR risk using electronic health record (EHR) and healthcare insurance claims data, with the initial goal of correctly placing over 90% of patients with VTDR in the highest-risk half of the population when ordered by risk. In doing so, RCI can focus patient engagement resources on patients who are most likely to need immediate evaluation and treatment, rather than diverting resources to patients who are less likely to require immediate attention. In Aim 2, RCI will use mixed methods framed by the Integrated Behavior Model to identify barriers and motivators for screening and assess their relative importance, develop and implement a survey instrument to elicit willingness to participate in screening based on motivating factors and barrier removal, and use natural language processing to detect barriers and facilitators for diabetic eye screening from patient communications. In Aim 3, RCI will create an agent-based simulation tool to guide care coordination recommendations in a way that maximizes population health outcomes subject to constraints on time and cost. The tool will identify optimal intervention strategies and be adaptable to changes in cost, disease burden, patient preference, population characteristics, and other parameters. Decisions that will be optimized include resource allocation as a function of risk, intervention timing, screening modality, communication strategies, barrier removal strategies, and other modifiable aspects of the system.