Despite the administrative and regulatory emphasis on increasing healthcare quality and reducing hospital readmissions, the disparate and disconnected electronic health record systems in US remain largely incompatible. In addition, data is collected at varying geographic locations with different scales. They hinder the development of learning healthcare systems in US for healthcare quality improvement and cost reduction. It was estimated that by providing interoperability standard that allows different stakeholders to exchange health related data and information, the cost can be tremendously reduced. Thus, Fast Healthcare Interoperability Resources (FHIR), which integrates heterogeneous data sources from local, regional, and national partners to enable the development of a new generation of intelligent health decision support systems, has recently been introduced as a new interoperability standard for health systems and information exchange. This project aims to develop an intelligent computational-health platform that harmonizes heterogeneous data sources such as various electronic medical records (EMRs), ontologies, and population-health databases, and deploys fast data analytics models to improve clinical point-of-care and global public health policy decision support.
Specifically, the development of this intelligent computational-health platform includes creating new resources under FHIR and new systems under Substitutable Medical Applications and Reusable Technologies based on FHIR (SMART-on-FHIR) for healthcare, leveraging common data models to harmonize different health data sources into FHIR-enabled databases, and comparing with baseline electronic data capture databases. The resulting platform with harmonized heterogeneous datasets and advanced data analytics can enable the following functions for healthcare: (i) on decision support at the point-of-care, heterogeneous EMRs covering large patient population are integrated to identify both time series data and discrete clinical features to predict acute medical event for future patients; (ii) on clinical decision making that requires extensive medical knowledge for interpretation, probabilistic methods such as Markov Logic Network are used to create graphs and rules. Such rules are then combined with EMR data mining to improve clinical relevance of final decisions; and (iii) on public health policy decision support, incompatible electronic death records in different States in US are integrated with EMRs to discover the causes of high prevalent deaths for public health surveillance. Thus, this EAGER project not only integrates different EMRs, knowledge database, and population databases for analytics development, but also assesses the capabilities and constraints of the FHIR standard.
