TOPLINE

A machine learning-based equation estimated low-density lipoprotein cholesterol (LDL-C) with accuracy similar to that of the Martin-Hopkins method, outperformed other equations, and reliably classified patients into treatment categories with minimal bias.

METHODOLOGY

• Researchers evaluated whether a simplified machine learning-based alternative to the Martin-Hopkins LDL-C equation could be developed for clinical care. They used a cross-sectional database involving lipid measurements collected from clinical facilities across the US from October 2015 to June 2019.
• More than 4.9 million patients (mean age, 56 years; 53% women) with complete lipid panel data and triglyceride concentrations < 400 mg/dL were randomly assigned to the training (n = 3,292,889) and test (n = 1,646,639) datasets.
• Researchers used multivariate adaptive regression splines (MARS) to estimate very LDL-C (VLDL-C) from total cholesterol, triglycerides, and high-density lipoprotein cholesterol (HDL-C). They then calculated LDL-C as total cholesterol minus HDL-C minus estimated VLDL-C (MARS equation).
• The MARS equation was compared with the Friedewald, Sampson-National Institutes of Health (NIH), Modified Sampson, and Martin-Hopkins equations. Accuracy was assessed using bias, root mean square error, and concordance with clinical guideline-based LDL-C categories.
• The results were externally validated using ultracentrifugation-measured LDL-C values from the Mayo Clinic (n = 20,740) and the FOURIER trial (n = 12,742).

TAKEAWAY

• The MARS equation demonstrated a minimal median bias of -0.1 mg/dL and differed from the original Martin-Hopkins equation by a median of -0.5 mg/dL.
• The root mean square error was smallest for the MARS equation (4.7 mg/dL), followed by the Martin-Hopkins (4.9 mg/dL), Sampson-NIH (5.8 mg/dL), Modified Sampson (6.0 mg/dL), and Friedewald (7.2 mg/dL) equations.
• The proportion of patients correctly classified according to clinical categories was nearly identical for the MARS (89.7%) and Martin-Hopkins (89.6%) equations but lower for the other equations.
• Validation using data from the Mayo Clinic and the FOURIER trial produced similar findings, with the MARS and Martin-Hopkins equations showing the highest accuracy.

IN PRACTICE

“Given its high accuracy and straightforward implementation as a single line of code in laboratory information systems, it is an alternative option to consider implementing in practice,” the researchers of the study wrote.

SOURCE

The study was led by Jihwan Park, PhD, Johns Hopkins University Bloomberg School of Public Health, Baltimore. It was published online on July 15 in JAMA Cardiology.

LIMITATIONS

The study lacked data on race, ethnicity, comorbidities, obesity, and lipid-lowering treatment. The ultracentrifugation-based assay may have underrecovered VLDL cholesterol at high triglyceride concentrations.

DISCLOSURES

The lipids database received funding from the David and June Trone Family Foundation. Amgen funded the FOURIER trial. Several authors reported receiving grants, personal fees, or research support from multiple pharmaceutical companies including Amgen, Merck, and AstraZeneca. One author reported serving as the deputy editor of JAMA Cardiology. Additional disclosures are reported in the original article.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.