Development of a secure, cloud-based platform to improve record linkage & cross-agency collaboration for the public sector: using deep learning & scalable data integrations to combat the opioid crisis

项目概要/摘要： 该SBIR第一阶段提案旨在为一种新的、多租户的、基于云的安全 专门为地方政府机构提供共享数据集并将其链接的工具的平台 准确，高质量和低成本。OpenLattice平台将专注于减少药物过量， 使药物治疗不那么支离破碎。跨医疗提供者和法律链接的个人层面数据集 执法可以支持对阿片类药物之前的处方途径和治疗轨迹的分析 过量、进入治疗、中断和恢复。然而，在个人层面上链接数据， 事实证明，与使用聚合级数据相比， 重复数据删除会对许多机构数据库进行重复数据删除。每天有91名美国人死亡， 阿片类药物过量和护理系统遍布多个机构，需要更大和高质量的 数据共享不可否认。我们的测试合作伙伴，用于评估拟议创新的有效性， 缅因州的波特兰成瘾协作组织（GPAC），医院、警察局、监狱、戒毒所的合作伙伴 治疗中心和中途之家已经在合作减少药物过量。这项建议旨在 展示概念验证：（i）在多个政府部门之间扩展高质量的数据集成， 域通过标准化的实体数据模型，和（ii）改善使用神经网络的记录链接。 首先，OpenLattice正在开发一个开源本体和集成脚本，以标准化集成 OpenLattice的数据库。随着个人定制需求的下降， 客户并将新数据集成到平台中，成本将大大削减， 为全国较小的县和城市提供数据解决方案的障碍，这些县和城市在历史上一直面临自定义 集成、系统更新、数据存储费用和附加费用高昂。OpenLattice平台还 能够使用现有的ETL工具，并与警察调度系统、紧急医疗系统、 致力于数据共享的合作伙伴之间的电话、医疗记录和在线处方系统 和合作。其次，OpenLattice正在开发一种新的专有记录链接算法， 采用了一种有前途但尚未在商业上测试过的技术：多层感知器神经网络， 也就是通常所说的深度学习在试点研究中，链接算法已经得到了验证 成功的竞争-有时甚至是超越-当前最先进的连接技术。在第一阶段， OpenLattice将继续改进本体、集成工具和深度学习神经网络， 在具有不同数据类型和格式的公开数据集上进行测试，并手动确认结果。 一旦成功，这些创新将解决改善治疗临床实践的关键障碍。 通过为治疗中的人提供更全面的连续护理，减少阿片类药物成瘾。

Project Summary/Abstract: This SBIR Phase I proposal aims to fund research and development for a new, multitenant secure cloud-based platform specifically tailored to provide local governmental agencies with tools to share datasets and link them accurately, at high quality and low cost. The OpenLattice platform will focus on reducing drug overdoses and making drug treatment less fractured. Individual-level datasets linked across medical providers and law enforcement can support analyses of prescribing pathways and treatment trajectories that precede opioid overdose, entry into treatment, disruption, and recovery. However, linking data at the individual level has proven to be a difficult and resource-intensive endeavor compared to use of aggregate-level data, with issues with deduplication plaguing many institutional databases. With 91 American deaths recorded daily from opioid overdoses and systems of care spread across multiple institutions, the need for greater and high-quality data sharing is undeniable. Our test partner for assessing the efficacy of proposed innovations is the Greater Portland Addiction Collaborative (GPAC) in Maine, a partnership of hospitals, a police department, jail, detox treatment centers and halfway houses already working together to reduce drug overdoses. This proposal aims to demonstrate proof of concept for (i) scaling high-quality data integrations across multiple governmental domains via a standardized entity data model, and (ii) improving record linkage using neural networks. Firstly, OpenLattice is developing an open source ontology and integration scripts to standardize integration of datasets into OpenLattice's database. As the individual customization requirements decline for onboarding customers and integrating new data into the platform, costs will be greatly slashed, removing a significant barrier to data solutions for smaller counties and cities across the country, who have historically faced custom integrations, system updates, data storage fees and add-ons at high cost. The OpenLattice platform also enables use of existing ETL tools and seamless integration with police dispatch systems, emergency medical calls, healthcare records, and online prescription systems across partners who have committed to data sharing and collaboration. Secondly, OpenLattice is developing a new, proprietary algorithm for record linkage that employs a promising but as-yet commercially untested technique: a multilayer perceptron neural network, more commonly known as deep learning. In pilot research, the linking algorithm has already demonstrated success rivaling—and sometimes exceeding—current state of the art linking technologies. In Phase I, OpenLattice will continue to improve ontologies, integration tools, and the deep learning neural network, and test on publicly available datasets with dissimilar data types and formats, with manual confirmation of results. When successful, these innovations will address critical barriers to improving clinical practice in treating opioid addiction by enabling a more comprehensive continuum of care for those in treatment.