How epidemiology and AI can help with planning and selecting decision support tools
Researchers from Western University, Canada have demonstrated the value of using artificial intelligence (AI) methods for descriptive epidemiology when developing data-driven decision support tools in health care.
Data-driven decision support tools are technology that provide information to help people make decisions. These tools use data from various sources to identify gaps, patterns or trends, or to make prediction of future events. Although becoming increasingly common within health care, their selection, development, and evaluation remains challenging to do well.
In Canada, the Alliance for Healthier Communities that provides team-based primary health care through Community Health Centres across Ontario to clients who otherwise experience barriers to regular care, are exploring how to effectively use their care-derived data to inform and improve care delivery.
To help support this process, Jaky Kueper, PhD and her colleagues Jennifer Rayner, PhD, Merrick Zwarenstein, MBBCH, PhD, and Dan Lizotte, PhD performed a large-scale study to describe sociodemographic, clinical, and healthcare use characteristics of the adult primary care population of the Alliance using electronic health record (EHR) data from 2009-2019. Their findings have not only provided a foundation for the development of decision support tools for the Alliance, but also highlight the value of using descriptive epidemiology early on in the development process of decision support tool initiatives more generally. Traditional descriptive epidemiology (although not easy to do well) often uses simple statistics to capture things like case counts or trends in diseases over time. For complex populations like those in primary health care, these methods may not capture the full picture.
The research team have demonstrated how AI, more specifically unsupervised machine learning, can be used to capture more complex patterns (e.g. condition co-occurrences and care provider teams) and provide complementary information to traditional descriptive epidemiology approaches when planning and selecting decision support tool initiatives. Their findings are informing conversations within the Alliance for Healthier Communities and also contributing towards gaps in general understanding about the characteristics and functions in primary health care.
Jaky explained that “Epidemiology and machine learning are complementary fields, and the more we learn how to integrate their various tools and approaches to conceptualizing and solving problems, the better able we will be to derive value from health care data.”
Jaky Kueper, PhD, Department of Epidemiology and Biostatistics, Schulich School of Medicine & Dentistry, Canada and Department of Computer Science, Faculty of Science, Western University, London, Ontario, Canada
Kueper, J. K., Rayner, J., Zwarenstein, M. and Lizotte, D. (2022) “Describing a complex primary health care population to support future decision support initiatives”, International Journal of Population Data Science, 7(1). doi: 10.23889/ijpds.v7i1.1756.