Enhanced data extraction and modelling from electronic medical records and phenotyping for clinical care, and research: Case studies in management of medication stewardship.
Flagship Program: Intelligent Decision Support to Improve Value and Efficiency
Project Description
Current methods of healthcare delivery are no longer sustainable. Sizeable investments have been made in implementing electronic medical records (EMRs). Unfortunately, many of these EMRs are simply being used as “electronic paper”: an electronic store of what before was on paper charts. However, the data contained in EMRs could be exploited in an aggregated way to enable data-driven decision making and facilitate research and to improve the quality and efficiency of patient care in real time. This is currently not done largely because the current methods of data extraction are technically challenging and labour intensive and the training and governance for real time analytics in hospitals is immature. The aim of this project is to demonstrate and refine data-extraction methods and to deliver real-time analytics to Australia’s largest health care service.
Electronic medical records capture the full gamut of health-related information in real-time, positioning them as a rich repository of generalizable patient and health service level data. Despite these notable advantages; several limitations need to be addressed including (1) complex query and extraction procedures inherent to storing ‘Big Data’ using the current SQL data model (2) un-indexed, and unsearchable unstructured data (3) data fragmentation (storage of the same field in multiple back-end locations) and (4) lack of semantic search functionality (i.e. search beyond simple keyword matching). This project will develop a new, efficient capability for data extraction from electronic medical records, starting with Queensland’s integrated Electronic Medical Record (iEMR), along with demonstration of its use within clinical analytics methods that address clinical workflow activities and evaluate the impact. The algorithms will collate and index information from both the existing structured query language (SQL) fields and currently unindexed, but invaluable unstructured information to create a robust data model (phenobank) that can be accurately searched using natural language for both clinical and research purposes. The developed algorithms and methods will be validated across two case studies and the impact on healthcare outcomes evaluated
Project Objectives
- Validate extraction algorithms for unstructured and semi-structured data. Map this data to established ontologies where applicable and associate them with appropriate meta-data. This will include findings of diagnostic and interventional procedures, functional assessments.
- Develop a data dictionary for elements that are not already covered by existing dictionaries across data relevant to the two case-studies. Such elements will be standardised, and documented in a manner consistent with the Queensland Clinical Data Dictionary methodology
- Develop portable, vendor-agnostic methods to validate and make clear the accuracy of both the search functionality and the data model for identifying patients and/or clinical events/outcomes. Apply and evaluate the methods on a selected iEMR site (Surgical Treatment and Rehabilitation Service (STARS) in Metro North Hospital and Healthcare Service.
- Develop a suite of clinical analytics methods, aligned to two selected clinical workflow activities, that demonstrate the utility and benefit of the devised real-time data extraction methods.
- Develop suitable methods to govern the interpretation of the insights provided by the clinical analytics methods, to ensure their safe uptake within clinical practice and evaluate the outcome using quantitative and qualitative analysis