Cardiometabolic & kidneyDiagnostic or prognostic study
Development and Validation of a Machine Learning-Based Approach to Identify High-Risk Diabetic Cardiomyopathy Phenotype
- Machine learning development and validation study (cohort)
- unsupervised hierarchical clustering
- deep neural network classifier
- external validation
Authors' conclusion
Machine learning-based techniques may identify 16% to 29% of individuals with diabetes as having a high-risk DbCM phenotype
Study context
- Population
- Individuals with DM free of CVD and other potential aetiologies of cardiomyopathy (Atherosclerosis Risk in Communities (ARIC) training cohort; CHS and University of Texas (UT) Southwestern electronic health record (EHR) validation cohorts)ARIC training cohort
- Participants
- 1,199 individuals
- Comparison
- High-risk diabetic cardiomyopathy (DbCM) phenotype vs other phenogroups
- Phenogroup-3n = 324
- Outcome
- HF incidence by phenogroup
- Follow-up
- 5-year
Performance
As reported in the abstract
5-year HF incidence
Further findings
| Outcome | Values | HR (95% CI) | |
|---|---|---|---|
| HF incidence, high-risk DbCM vs without, external validation — High-risk DbCM vs Without high-risk DbCM · CHS | 1.61 (1.18–2.19) | ||
| HF incidence, high-risk DbCM vs without, external validation — High-risk DbCM vs Without high-risk DbCM · UT Southwestern EHR cohort | 1.34 (1.08–1.65) | ||
| 0.512Hazard ratio (log scale) |
| Outcome | Values | Estimate |
|---|---|---|
| Proportion classified as DbCM by DeepNN in validation cohorts | CHS 16% · UT Southwestern EHR 29% |
- Proportion classified as DbCM by deep neural network (DeepNN) in validation cohorts: CHS n = 802; EHR cohort n = 5071
Original publication
Development and Validation of a Machine Learning-Based Approach to Identify High-Risk Diabetic Cardiomyopathy Phenotype. Eur J Heart Fail. Published 6 September 2024. doi:10.1002/ejhf.3443
Summarised from the published abstract; every number is checked against it