Grant: Project Grant

The administration of universal health care produces a wide variety of information that is routinely collected and covers large segments of the population: administrative health data. Canada has rich administrative health data from sources such as health insurance registries, inpatient hospital care, day surgery, emergency departments, and outpatient physician services. Administrative health data helps decision-makers to improve health policies, budget, and administer health services to ultimately improve population health. The assessment of data quality is fundamental to using administrative data, as high quality data are necessary for monitoring how frequently diseases occur and evaluating the quality of care in healthcare settings. The Discharge Abstract Database (DAD) is a hospital administrative database that is used locally and internationally for the research and evaluation of healthcare systems. Studies on patient safety have shown that countries with better data quality have higher rates of adverse events. This illustrates the persistent issue of underreporting or misreporting of conditions in DAD, which potentially underestimates disease frequency. To address this, the Canadian Institute of Health Information has developed a framework to investigate data quality; however, there remains an urgent need for intelligent and cost-effective tools that quickly assess and identify data quality issues in the large volume of complex health data being generated. This study will work these stakeholders to ensure this need is met.Using machine learning and data quality mining methods, this project will develop a tool that automatically assesses data quality and identifies issues. Internal and external validation will be conducted using reabstracted data from Canada and Australia and medical chart review to test the tool. The application of this tool will avoid the 'garbage in, garbage out' phenomenon often observed in data used for healthcare management and research.