Researchers at Johns Hopkins University have developed a machine learning-based version of the widely used Martin-Hopkins equation that simplifies low-density lipoprotein cholesterol (LDL-C) calculations without sacrificing accuracy.
The new approach is JAMA Cardiology, Easier laboratory estimation of LDL-C may improve treatment decisions for patients at risk for cardiovascular disease.
“We optimized the calculation of LDL cholesterol to make this formula more accessible and easier to perform for all laboratories,” said Seth Martin, MD, senior study author and director of the Progressive Dyslipidemia Program and Digital Health Lab at the Johns Hopkins Siccarone Center for Cardiovascular Disease Prevention. “Our goal is to help clinicians and patients prevent heart attacks and strokes and make better decisions about starting lifesaving treatments.”
LDL-C is the main cause of atherosclerotic cardiovascular disease (ASCVD) and is a major therapeutic target. Current guidelines recommend using LDL-C cutoffs such as 70 mg/dL or 55 mg/dL (multiply by 0.0259 to convert to mmol/L) in ASCVD patients to guide clinical lipid management.
The gold standard for measuring LDL-C concentration is preparative ultracentrifugation, but this method is expensive and time-consuming. Therefore, LDL-C concentrations are usually estimated in routine clinical practice.
One of the most accurate methods to estimate LDL-C concentration is the Martin-Hopkins method, which is recommended for clinical use in the United States, Europe, and South America. However, implementation can be difficult as it requires the user to search a large table for adjustable coefficients based on the patient’s triglyceride and non-high-density lipoprotein cholesterol levels.
“A low-cholesterol and high-triglyceride lipid profile is the ultimate stress test for LDL cholesterol calculations,” says Martin. He explains that based on various formulas, a difference of 5, 10, or 20 mg/dL can change eligibility for treatment with treatments such as PCSK9 inhibitors, which have been shown to significantly lower LDL cholesterol levels. “It is cutting-edge examples of this kind that will benefit most from more accurate results,” he added.
To overcome this barrier and ease implementation, Martin and team used a transparent machine learning approach (multivariate adaptive regression splines) to create a simplified formula-based LDL-C equation.
They trained and tested their tool on data from 4,939,528 adults and children (mean age 56 years, 53% female) with complete lipid panel test results. These samples are representative of the US population, have a median LDL cholesterol level of 114 mg/dL, and were obtained from a very large lipid database.
The researchers report: JAMA Cardiology The machine learning version of the Martin-Hopkins equation estimated LDL-C concentrations similar to the original equation, with a minimum difference of 0.5 mg/dL.
Both Martin-Hopkins equations classified 90% of the samples into the correct treatment category. Of the commonly used tools for LDL-C estimation, the Sampson-NIH equation correctly classified 86%, the modified Sampson-NIH equation 85%, and the Friedewald equation 83% into the correct category.
Importantly, the researchers found that the Martin-Hopkins equation was the most accurate for classifying high-risk patients with low LDL cholesterol levels, Martin said.
When evaluating people with triglyceride levels between 200 mg/dL and 399 mg/dL and LDL cholesterol levels below 70 mg/dL, the Martin Hopkins machine learning equation correctly classified 84% of high-risk samples, the original Martin Hopkins equation classified 83%, the modified Sampson NIH equation classified 72%, and the Sampson NIH The equations have been classified. 61%, the Friedewald equation classified 40%.
Martin and his coauthors conclude: [the Martin-Hopkins machine learning equation] This is an alternative option to consider for actual implementation. ”
