Prevention & lipidsCohort study
Tailoring Risk Prediction Models to Local Populations
- Retrospective cohort study
- EHR-based
- prediction model development and validation
Authors' conclusion
The interpretable ML approach presented in this article enhanced the accuracy of the AHA-PREVENT model when applied to a local population while still preserving the risk associations found by the original model.
Study context
- Population
- New England EHR cohort of patients without prior atherosclerotic cardiovascular disease (ASCVD), event period 2007–2016test set 38,137
- Participants
- 95,326 participants
- Exposure
- MLM-PREVENT vs AHA-PREVENT equations
- Outcome
- ASCVD events (nonfatal MI, CAD, ischemic stroke or CV death); calibration, discrimination (Harrell C index) and reclassification
- Follow-up
- 10-year
Main finding
As reported in the abstract
Calibration (modified Hosmer-Lemeshow)
- overall; AHA-PREVENT χ23 > 16.3, P < .001
- χ23 2.2; P = 0.53
- men; AHA-PREVENT χ23 > 16.3, P < .001
- χ23 2.1; P = 0.55
- women; AHA-PREVENT χ23 > 16.3, P < .001
- χ23 6.5; P = 0.09
Observational data: associations, not effects.
Further findings
| Outcome | MLM-PREVENT | AHA-PREVENT |
|---|---|---|
| Reclassification at 7.5% risk threshold | 11.5% | Not reported in the abstract |
Calibration by race
- MLM-PREVENT maintained or improved calibration in Asian, Black and White individuals
Discrimination
- Both models performed equally well (approximate ΔC index)
- ±0.01
Original publication
Tailoring Risk Prediction Models to Local Populations. JAMA Cardiol. Published 1 November 2024. doi:10.1001/jamacardio.2024.2912
Summarised from the published abstract; every number is checked against it