Machine Learning may Identify High-Risk Diabetic Cardiomyopathy Phenotype: Study

• Clustering analysis identified three phenogroups among diabetic patients. Phenogroup-3, consisting of 27% of the ARIC cohort, had a significantly higher 5-year HF incidence rate of 12.1% compared with Phenogroups 1 and 2, which had HF incidence rates of 3.1% and 4.6%, respectively. This high-risk phenotype was associated with higher NT-proBNP levels, increased left ventricular mass, larger left atrial size, and poorer diastolic function-all important markers of heightened HF risk.

• The DeepNN classifier had excellent predictive capability and identified 16% of high-risk DbCM cases in the CHS cohort and 29% in the UT Southwestern EHR cohort. Of great interest, the incidence of HF was significantly increased among the high-risk DbCM phenotype individuals identified by the classifier, with hazard ratios of 1.61 (95% CI 1.18–2.19) in the CHS cohort and 1.34 (95% CI 1.08–1.65) in the UT Southwestern EHR cohort.

This study has shown the potential of machine learning-based approaches to identify individuals with diabetes as having a high-risk DbCM phenotype, an opportunity for targeted HF prevention strategies. Researchers believe that predictive models in clinical practice will manage and mitigate the risk of HF in diabetic patients by earlier and more aggressive intervention strategies.

Reference: