Machine learning and diabetes are popular topics with Health Tech Insider readers. The growing global prevalence of Type 2 diabetes (T2D) drives tech development to diagnose and manage the disease. Machine learning models make headlines as researchers explore myriad applications for AI in medicine. The disease and the methodology often come together. In 2017, data science researchers at The Arnhold Institute for Global Health of the Icahn School of Medicine at Mount Sinai used machine learning to discover new insights about diabetes. The same year IBM scientists used deep learning to teach a computer system to diagnose diabetic retinopathy. In 2018, SweetchHealth Ltd. and WellSpan Health developed a system to predict diabetes using AI with personalized data from smartphones and wearables.
A new study published in Medical Care shows machine learning can predict which patients with T2D are at greatest risk of poor glycemic control. Harvard’s Sanjay Basu, MD, Ph.D., postulated that where people live affects their health. Basu used individual variables (covariants) such as age, gender, income, and employment status and social determinants of health (SDH): the conditions where people live, learn, work, and play. The project used a sample population of 1,015,808 people with T2D. The researchers analyzed the data with standard statistical regression models and several machine learning models. Overall the regression model achieved 68.4% predictive accuracy, but the best machine learning model was 90.6% accurate in determining which patients would have control issues.
Basu concluded that the study proved that predictive machine learning models identify potential risks for diabetes control. The models could help health care organizations target programs and community outreach to reach individuals with the greatest risk. This could lead to more efficient use of resources to reduce the rate of life-altering (and expensive) secondary complications from diabetes.
