Project Abstract: To reduce costs and enhance health outcomes, it is of critical importance that patient data are systematically gathered, cleaned and analyzed, thereby allowing us to build more accurate, timely and reliable models for diagnosing, managing and treating diseases. One example application domain is asthma control and prediction. In USA alone, about 40 million people suffered from lifetime asthma (13% of the USA population) and 26 million people (8%) suffered from current asthma. Developing better predictive models for asthma attacks, can result in enhancing preventive strategies, improving patient outcomes, and significantly reducing healthcare costs due to reduced emergency care need. One of the key factors obstructing such models for health care management is the integration of patient data that are scattered across multiple organizations. This fundamental challenge is particularly acute for chronic diseases such as asthma where patients often receive care at multiple institutions within a region. Furthermore, single site studies may provide inaccurate picture due to data inaccuracies. For example, due to certain selection biases, number of patients from certain race group maybe underrepresented in one location. In addition, severity information of diseases may not be complete if all the emergency care visits are not recorded. Without proper record linkage and data duplication, many of the disease specific conditions may be over-represented. For instance, it is reported that after cross-institution deduplication, number of records related to diabetes reduced 24.0%, asthma reduced 28.0%, and myocardial infarction reduced 10.9%. Therefore, it is of paramount importance to merge records in a manner that mitigates duplication, as well as fragmentation, of an individual’s information. Although there have been efforts to implement health information exchanges to facilitate data integration and exchange, linking patient records across multiple health care organizations create significant security and privacy challenges. At the same time, as the usage of healthcare analytics and the data sharing increases, patient trust in the overall data analytics pipeline must be ensured by asking patients to make a “consent decision”. This consent decision concerns the sharing and accessing of the patient’s health data for treatment, payment, and health care operations purposes. As a result, our healthcare analytics research nowadays is at utmost need of a product that can manage patient consent while allowing secure and privacy-preserving linkage of health care data across multiple institutions. To address these challenges, we will develop a privacy-preserving solution that can 1) efficiently capture consent, use the captured consent information to gather patient data distributed across resources within a certain health organization efficiently and 2) link the data hosted by different users across disparate health organizations while protecting patient privacy and providing accountability. Although there exist some solutions for managing healthcare consent and privacy-preserving record linkage, they are not integrated. In addition, existing techniques either do not easily scale for large amounts of data and/or leak sensitive information during record linkage process. Finally, we are not aware of any existing tool that combines private-record linkage with private blockchains for providing accountability.

Narrative: Linking patient records distributed across multiple healthcare organizations while maintaining patient privacy, informed consent and accountability is critical for building better machine learning and data analytics models and improving healthcare outcomes. This project will offer a first solution for privacy-preserving consent aware record linkage that leverages recent developments in hardware, blockchains and cloud computing. Moreover, the new tool will enable the key capabilities for auditing the entire workflow for accountability.