Johns Hopkins Publishes Machine‑Learning Version of Martin–Hopkins LDL‑C Equation
An open‑access formula that reproduces the original equation within about 0.5 mg/dL is designed for single‑line laboratory implementation to help clinicians meet 2026 LDL‑C treatment thresholds.
Overview
- The study, published in JAMA Cardiology, presents a machine‑learning simplification of the Martin–Hopkins method that uses multivariate adaptive regression splines to estimate low‑density lipoprotein cholesterol (LDL‑C).
- Developed and tested on roughly 4.9 million U.S. lipid samples from the Very Large Database of Lipids, the model matched the original Martin–Hopkins outputs with a median difference of about 0.5 mg/dL.
- Both the machine‑learning and original Martin–Hopkins equations correctly classified 90% of samples into guideline treatment categories and performed best for high‑risk profiles with triglycerides 200–399 mg/dL and LDL‑C under 70 mg/dL.
- The authors released open‑access code that can be implemented as a single line in laboratory information systems to simplify use and support the 2026 guideline preference for Martin‑Hopkins calculations.
- Study disclosures note funding from the David & June Trone Foundation and use of FOURIER trial samples funded by Amgen, and authors recommend broader external testing outside U.S. datasets before widespread adoption.