New Machine-Learning Equation Improves LDL Cholesterol Assessment

Accurate assessment of low-density lipoprotein (LDL) cholesterol is central to cardiovascular risk management, yet calculation methods can underestimate values in some patients. Laboratories widely use equations to estimate LDL cholesterol from routine lipid panels, guiding treatment decisions to lower risk. Underestimation may lead to missed opportunities for therapy, particularly in people with low LDL cholesterol and high triglycerides. New findings demonstrate that a simplified machine-learning equation can match the accuracy of an established method across millions of samples.

Johns Hopkins University School of Medicine researchers evaluated a streamlined, machine-learning version of the Martin-Hopkins equation, which laboratories in the U.S. and other countries use to estimate LDL cholesterol. The updated calculation and its code were published in JAMA Cardiology. The approach was designed to be broadly accessible so that clinical laboratories can deploy it without additional complexity.

The method replaces only the triglyceride component used in the original calculation, allowing laboratories to substitute it for the corresponding term in the long-standing Friedewald equation. Investigators assessed performance by comparing equation-derived LDL cholesterol with results obtained through ultracentrifugation, a gold-standard research measurement method. The authors emphasize that the code is open and transparent to support use across different laboratory information systems

The study drew on the Very Large Database of Lipids and analyzed 4.9 million U.S. adult and pediatric samples with a median LDL cholesterol of 114 mg/dL. More than 3.2 million samples were used to train the machine-learning model and 1.6 million were used for testing.

The researchers also evaluated performance in two additional validation sets: a reference laboratory dataset and a clinical trial dataset from patients who had used PCSK9 inhibitors. Accuracy was compared with the original Martin-Hopkins equation, as well as the Sampson–NIH, modified Sampson-NIH, and Friedewald equations.

Across the full cohort, the machine-learning version closely matched the original calculation, with a minimal average difference of 0.5 mg/dL. Both Martin-Hopkins equations correctly classified 90% of samples into the appropriate treatment category, compared with 86% for Sampson–NIH, 85% for modified Sampson–NIH, and 83% for Friedewald.

In a key subgroup with triglycerides of 200–399 mg/dL and LDL cholesterol below 70 mg/dL, the machine-learning equation correctly classified 84% of samples, compared with 83% for the original, 72% for modified Sampson–NIH, 61% for Sampson-NIH, and 40% for Friedewald.

“We’ve optimized the calculation of LDL cholesterol and made this equation accessible and easier for all labs to implement,” said Seth Martin, M.D., M.H.S., the senior study author and director of the Advanced Lipid Disorders Program and Digital Health Lab at the Johns Hopkins Ciccarone Center for the Prevention of Cardiovascular Disease. "Our goal is to enable clinicians and patients to make better decisions about starting treatments that prevent heart attacks and strokes and save lives."

“This updated equation is not only highly accurate, but it’s transparent and can be easily adopted by laboratories. We wanted to avoid creating a ‘black box’ equation that is opaque or invisible to most users,” said Mark Marzinke, Ph.D., a study author who is the medical director overseeing this testing in the Johns Hopkins Hospital Core Laboratories and a professor of pathology and medicine at Johns Hopkins.

Related Links
Johns Hopkins University School of Medicine