Architectures and Patterns for Moving Towards the Use of High-Frequency, Low-Fidelity Data in Healthcare

Abstract

The U.S. healthcare data has undergone significant transformations as computerized technologies have evolved in recent decades. Data trends in healthcare have transitioned from less frequent, higher-fidelity provider-documented electronic health records (EHR) to more frequent, lower-fidelity patient and device-generated data. In particular, prevalent Internet of Things (IoT) devices and apps collect enormous amounts of information associated with individuals’ health statuses, physical activities, and environmental triggers to chronic conditions. As a result, health-related data generated from IoT devices today is now exceeding EHR data in terms of volume and frequency. It is important, however, to integrate these data into healthcare decisions since they reflect various aspects of citizens’ lifestyles and well-being in a comprehensive and continuous manner. Given these trends, a key problem facing heathcare researchers and practitioners is how to successfully migrate towards the use of the high-frequency, low-fidelity (HFQ) data in the healthcare domain.

Addressing this problem requires research that focuses on the following issues: (1) how to scalably extract insights from large volumes of health data, (2) how to integrate and share HFQ data along with learned insights from those data, and (3) after an integrated health system is created, what methods and techniques are needed to evolve it to adopt more efficient technology or perform upgrades and updates as needed. This dissertation presents software architectures and patterns targeting these issues. First, we propose a machine learning based filtering architecture for drawing insights from HFQ data at scale. Second, we describe a data sharing framework based on distributed ledger technologies (DLT) to address technical requirements defined by the Office of the National Coordinator for Health IT (ONC). Lastly, we document a design pattern sequence for effectively designing and maintaining a DLT-enabled healthcare data sharing system in a secure and evolvable manner.