Clinical laboratories may soon have a simpler way to implement one of the most accurate methods for estimating low-density lipoprotein cholesterol (LDL-C). Researchers at Johns Hopkins have developed a machine learning implementation of the Martin-Hopkins equation that preserves the performance of the original calculation while simplifying deployment across laboratory information systems.
Published in JAMA Cardiology, the new equation uses the same inputs as a standard lipid panel but replaces the original lookup-table approach with a transparent, single-line equation. According to the investigators, the approach could help laboratories that previously lacked the information technology infrastructure needed to implement the original Martin-Hopkins equation.
Designed for routine laboratory implementation
Although direct LDL cholesterol measurement by ultracentrifugation remains the research gold standard, it is too time-consuming and expensive for routine clinical testing. Most laboratories instead estimate LDL-C using equations based on total cholesterol, HDL cholesterol, and triglycerides.
The Friedewald equation, introduced in the 1970s, remains widely used but can underestimate LDL-C in patients with low LDL cholesterol and elevated triglycerides, potentially affecting treatment decisions for high-risk patients.
The original Martin-Hopkins equation, introduced in 2013, addressed this limitation by individualizing the relationship between triglycerides and very-low-density lipoprotein cholesterol (VLDL-C). However, because it relies on a lookup table, implementation has been challenging for some laboratories.
"We heard from labs that the easiest thing is if they could have a single line of code that they could plug into the lab IT system similar to the Friedewald equation," said senior author Seth Martin, MD, MHS, director of the Advanced Lipid Disorders Program and Digital Health Lab at the Johns Hopkins Ciccarone Center for the Prevention of Cardiovascular Disease, in a Johns Hopkins Medicine video interview.
To simplify implementation, the investigators developed a transparent machine learning model using multivariate adaptive regression splines (MARS). Unlike many artificial intelligence models, the equation is fixed rather than adaptive and does not require retraining once implemented.
Researchers trained the model using more than 3.2 million lipid profiles from the Very Large Database of Lipids and evaluated it in an additional 1.6 million samples. The equation was further validated using a Mayo Clinic reference laboratory dataset and samples from the FOURIER clinical trial, with calculated LDL-C values compared with ultracentrifugation measurements.
The machine learning equation differed from the original Martin-Hopkins equation by just 0.5 mg/dL. Both equations correctly classified 90% of patients into the appropriate treatment category, outperforming the Sampson-NIH, modified Sampson-NIH, and Friedewald equations. The advantage was greatest among patients with LDL-C below 70 mg/dL and triglycerides between 200 mg/dL and 399 mg/dL, where underestimation can influence eligibility for lipid-lowering therapy.
How laboratories can maintain confidence after implementation
For laboratories adopting the new calculation, Martin emphasized that reliability depends on maintaining the quality of the lipid measurements used as inputs rather than modifying the equation itself.
"It is important to recognize that this is a stable, transparent equation rather than a dynamic model that changes over time," Martin said to Today's Clinical Lab. "Once implemented, the equation itself does not require retraining or recalibration."
Instead, laboratories should focus on ensuring continued accuracy and standardization of total cholesterol, HDL cholesterol, and triglyceride measurements through established quality assurance practices. Martin noted that participation in the CDC Lipid Standardization Program and adherence to laboratory quality processes can help maintain confidence in results over time.
Communicating changes in LDL-C results
The researchers also emphasized the importance of helping clinicians understand why the updated calculation was developed and how it supports current approaches to cardiovascular risk reduction.
Martin described the transition as a progression from a simpler, one-size-fits-all approximation developed decades ago toward a more personalized approach that improves accuracy using the same conventional lipid panel.
"The current change can be understood as a maturation of LDL-C estimation science from a simpler one-size-fits-all approximation developed in the 1970s to a modernized precision approach that maximizes the accuracy obtainable from the conventional lipid panel," Martin said.
He added that the equation was developed from a large and diverse dataset, validated across multiple populations, and designed to avoid repeated cycles of competing formulas derived from smaller datasets.
Applying machine learning beyond LDL cholesterol
The study also demonstrates how machine learning can improve laboratory medicine beyond creating complex predictive models.
"One important lesson is that machine learning can be used not only to create highly accurate models but also to make evidence-based calculations easier to implement at scale," Martin said.
"The goal is not necessarily complexity; sometimes machine learning can help simplify a process while preserving or improving performance."
The machine learning equation and the original Martin-Hopkins equation have no patent or intellectual property restrictions, allowing laboratories to implement the calculation without licensing barriers. The authors said broader adoption could help clinicians and patients access more accurate LDL-C estimates to guide evidence-based cardiovascular risk reduction.
“If we provide the most accurate LDL results, and we don’t falsely reassure someone, we provide them with the result closest to truth, and then they act on that by starting the evidence-based therapies recommended by guidelines.” Martin said in a video interview. “Ultimately, the clinical impact is reduced heart attacks, reduced strokes, and lives saved.”
This article was created with the assistance of Generative AI and has undergone editorial review before publishing.
