Prevention & lipidsDiagnostic or prognostic study
Development and Validation of a Simplified Martin-Hopkins Low-density Lipoprotein Cholesterol (LDL-C) Equation Using Machine Learning
- Diagnostic accuracy study
- cross-sectional
- machine learning
- randomly assigned training and test sets
- external validation
Authors' conclusion
Study results suggest that the new machine learning-based LDL-C equation provided results comparable with those of the original Martin-Hopkins method while simplifying implementation to a single equation.
Study context
- Population
- Adult and pediatric patients with complete lipid panel data, Very Large Database of Lipids (Vertical Auto Profile ultracentrifugation, 2015–2019)
- Participants
- 4,939,528 participants
- Comparison
- LDL-C-MH-MARS vs LDL-C-MH
- Outcome
- Accuracy of estimated vs ultracentrifugation-measured LDL-C (bias, RMSE, concordance by guideline-based categories)
Performance
As reported in the abstract
Median difference
−0.5 mg/dL
By group
LDL-C-MH-MARS
−0.1 (−2.1 to 1.8) mg/dL
Median bias (IQR) · root mean square error (RMSE) 4.7 mg/dL · Correctly classified 89.7%
LDL-C-MH
4.9 mg/dL
RMSE · Correctly classified 89.6%
LDL-C-S
5.8 mg/dL
RMSE · Correctly classified 86.3%
LDL-C-MS
6.0 mg/dL
RMSE · Correctly classified 84.7%
LDL-C-F
7.2 mg/dL
RMSE · Correctly classified 83.1%
Further findings
| Outcome | Values |
|---|---|
| Underestimation when classifying LDL-C <70 mg/dL, TG 200 to 399 mg/dL | LDL-C-MH-MARS 16% · LDL-C-MH 17% · LDL-C-MS 28% · LDL-C-S 39% · LDL-C-F 60% |
External validation (Mayo Clinic; FOURIER)
- Similar patterns; LDL-C-MH-MARS and LDL-C-MH highest accuracy
Original publication
Summarised from the published abstract; every number is checked against it