Identifying subtypes of type 2 diabetes mellitus with machine learning: development, internal validation, prognostic validation and medication burden in linked electronic health records in 420 448 individuals
Authors' conclusion
In the largest study using ML to date in incident T2D, we identified four distinct subtypes, with potential future implications for etiology, therapeutics, and risk prediction.
Study context
- Population
- Individuals ≥18 years with incident T2D in population-based electronic health records (2006–2020; Clinical Practice Research Datalink); 3787 factors incl. demography, history, examination, biomarkers and medicationsincident T2D
- Participants
- 420,448 individuals
- Exposure
- Machine learning (ML)-derived T2D subtypes: early onset, late onset, cardiometabolic vs metabolic
- Outcome
- T2D subtypes identified by 9 unsupervised ML methods
Methodological detailsmachine learning clustering, linked electronic health records, internal and prognostic validation
- Machine learning clustering
- Linked electronic health records
- Internal and prognostic validation
Main finding
As reported in the abstract
Four T2D subtypes identified: metabolic, early onset, late onset and cardiometabolic
4
Internal validity: subtypes predicted with high accuracy
F1 score >0.98
Observational data: associations, not effects.
Further findings
| Outcome | HR (CI) | |
|---|---|---|
| 5-year all-cause mortality · 5 yearsLate onsetvs MetabolicHighest risk | 1.95 (1.85–2.05) | |
| 5-year all-cause mortality · 5 yearsEarly onsetvs MetabolicLowest risk | 1.18 (1.11–1.27) | |
| 5-year hospitalization · 5 yearsLate onsetvs MetabolicHighest risk | 1.66 (1.58–1.75) | |
| 5-year hospitalization · 5 yearsEarly onsetvs MetabolicLowest risk | 0.85 (0.80–0.90) | |
| 0.512Hazard ratio (log scale) |
| Outcome | Estimate |
|---|---|
| Medication prescribing within 5 years of T2D onsetLate onset vs Metabolic · most likely to be prescribed medications | 1.31 (CI 1.28–1.35) |
| Medication prescribing within 5 years of T2D onsetEarly onset vs Metabolic · least likely to be prescribed medications | 0.83 (CI 0.81–0.85) |
- Medication prescribing within 5 years of T2D onset: adjusted for age, sex, and pre-T2D medications
New chronic disease incidence · 5 years
- Highest in late-onset subtype and lowest in early-onset subtype
Original publication
Identifying subtypes of type 2 diabetes mellitus with machine learning: development, internal validation, prognostic validation and medication burden in linked electronic health records in 420 448 individuals. BMJ Open Diabetes Res Care. Published 4 June 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →