WEDNESDAY, Oct. 30, 2024 (HealthDay News) -- Machine learning can he;p identify individuals with diabetes at high risk for diabetic cardiomyopathy (DbCM), according to a study published online Sept. 6 in the European Journal of Heart Failure.
Matthew W. Segar, M.D., from the Texas Heart Institute in Houston, and colleagues developed and validated a machine learning-based clustering approach to identify high-risk DbCM based on echocardiographic and cardiac biomarker parameters. The training model included 1,199 individuals with diabetes participating in the Atherosclerosis Risk in Communities cohort who were free of cardiovascular disease. The model was validated using data from 802 participants in the Cardiovascular Health Study and a separate cohort of 5,071 electronic medical records.
The researchers found that phenogroup-3, the high-risk DbCM phenotype (324 patients), had significantly higher five-year heart failure incidence than other phenogroups (12.1 versus 4.6 percent for phenogroup 2 and 3.1 percent for phenogroup 1). For the high-risk DbCM phenotype, key echocardiographic predictors included higher N-terminal pro-B-type natriuretic peptide levels, increased left ventricular mass and left atrial size, and worse diastolic function. In the validation cohorts, the deep neural network classifier identified 16 and 29 percent of participants with DbCM, respectively. Participants with the high-risk DbCM phenotype in the external validation cohorts had a significantly higher incidence of heart failure.
"A machine learning-based clustering approach to identify DbCM may promote a risk-based approach for the use of effective but potentially expensive heart failure preventive therapies among the highest-risk subgroup of individuals with diabetes who are more likely to benefit," the authors write.
Several authors disclosed ties to the pharmaceutical and biotechnology industries.
Abstract/Full Text
