Machine learning model predicts high-risk diabetic cardiomyopathy, revealing potential for targeted prevention
Researchers at UT Southwestern Medical Center have developed a machine learning model that can identify patients with diabetic cardiomyopathy, a heart condition characterized by abnormal changes in the heart’s structure and function that predisposes them to increased risk of heart failure. The findings, published in the European Journal of Heart Failure, offer a data-driven method to detect a high-risk diabetic cardiomyopathy phenotype, enabling early interventions that could help prevent heart failure in this vulnerable population.
“This research is noteworthy because it uses machine learning to provide a comprehensive characterization of diabetic cardiomyopathy – a condition that has lacked a consensus definition – and identifies a high-risk phenotype that could guide more targeted heart failure prevention strategies in patients with diabetes,” said senior author Ambarish Pandey, M.D., Associate Professor of Internal Medicine in the Division of Cardiology at UT Southwestern.
Study senior author Ambarish Pandey, M.D., is Associate Professor of Internal Medicine in the Division of Cardiology at UT Southwestern.
Phenotypes are observable physical properties of individuals that give them specific biological traits, according to Dr. Pandey. He and his research colleagues used data from the Atherosclerosis Risk in Communities cohort, which included over 1,000 participants with diabetes but no history of cardiovascular disease. By analyzing a set of 25 echocardiographic parameters and cardiac biomarkers, the team identified three patient subgroups.
The study identified one of these subgroups, making up 27% of the cohort, as the high-risk phenotype. Patients in this group exhibited significantly elevated levels of NT-proBNP, a biomarker linked to heart stress, along with abnormal heart remodeling, such as increased left ventricular mass and impaired diastolic function. Most notably, the five-year incidence of heart failure in this group was 12.1%, significantly higher than in the other subgroups.
Based on these findings, the researchers developed a deep neural network classifier to identify diabetic cardiomyopathy. When validated on additional cohorts, including the Cardiovascular Health Study and UT Southwestern’s electronic health record database, the model identified between 16% and 29% of diabetic patients as having the high-risk phenotype. These patients consistently exhibited a higher incidence of heart failure.
“Clinically, this model could help target intensive preventive therapies, such as SGLT2 inhibitors, to patients most likely to benefit,” Dr. Pandey said, referring to a class of medications used to treat Type 2 diabetes. “It may also help enrich clinical trials of heart failure prevention strategies in diabetes patients.”
The study expands on earlier research into diabetic cardiomyopathy, which has been difficult to define due to its asymptomatic early stages and the wide range of effects it can have on the heart. Machine learning has introduced a way to pinpoint high-risk patients, potentially offering a more refined approach than traditional diagnostic methods.
“This builds on our previous work that evaluated the prevalence and prognostic implications of diabetic cardiomyopathy in community-dwelling adults,” Dr. Pandey said. “It extends those efforts by using machine learning to identify a more specific high-risk cardiomyopathy phenotype.”
By providing a new way to identify patients at risk for heart failure, the model could enable earlier and more aggressive interventions, improving patient outcomes and shaping future research in cardiovascular care.
“This research aligns with UTSW’s missions by leveraging strengths in data science and cardiovascular research to develop tools that could improve patient care and inform future clinical trials,” Dr. Pandey said.
Other UTSW researchers who contributed to this study were DuWayne Willett, M.D., Professor of Internal Medicine, and Muhammad Shariq Usman, M.D., resident in Internal Medicine.
The study was supported by funding from Applied Therapeutics.