Lee Jongseok, Lee Hyelim, Cha Hwajung, Seok Jun, Jeong In Cheol
School of Artificial Intelligence Convergence, Hallym University, Chuncheon, Republic of Korea.
College of Medicine, Hallym University, Chuncheon, Republic of Korea.
PLoS One. 2025 Jul 3;20(7):e0327169. doi: 10.1371/journal.pone.0327169. eCollection 2025.
Low-density lipoprotein cholesterol (LDL-C), a major modifiable risk factor for cardiovascular diseases, is typically calculated using the Friedewald formula when triglyceride (TG) levels are below 400 mg/dL. Recent studies have demonstrated the superior accuracy of the Martin-Hopkins method across diverse populations. While this method estimates very-low-density lipoprotein cholesterol (VLDL-C) using strata-specific median TG/VLDL-C ratios, its reliance on median statistics raises questions about whether these ratios are truly optimal.
This study evaluated the performance of the Martin-Hopkins method compared to the Friedewald formula, focusing on its potential for improvement by applying optimal TG/VLDL-C ratios. Using data from 18,322 individuals in the Korea National Health and Nutrition Examination Survey (KNHANES), we derived strata-specific optimal TG/VLDL-C ratios designed to maximize concordance with directly measured LDL-C values, based on LDL-C categories defined by clinical guidelines. We compared the performance of four LDL-C estimation models: the Friedewald formula (LDL-CF), the original Martin-Hopkins method (LDL-CM-N), and two alternative models that applied TG/VLDL-C ratios derived from our data-one using median values (LDL-CKM-N) and the other using optimal values tailored to each stratum (LDL-CKO-N).
The Martin-Hopkins method showed significantly higher concordance than the Friedewald formula for TG levels < 400 mg/dL (79.6% for LDL-CF vs. 83.2% for LDL-CM-180, p < 0.001). Concordance improved by less than 2% for TG levels < 150 mg/dL (83.3% vs. 84.9%), but by approximately 10% for TG levels of 150-399 mg/dL (68.8% vs. 78.0%). The largest discrepancy was observed in classifying LDL-C levels < 70 mg/dL among individuals with TG levels of 150-399 mg/dL (47.5% for LDL-CF vs. 90.3% for LDL-CM-180). However, the overall concordance differed only modestly between the 10-cell and 180-cell Martin-Hopkins equations (82.8% for LDL-CM-10 vs. 83.2% for LDL-CM-180, a difference of 0.4%), indicating only a marginal benefit despite the substantial increase in the number of strata. Using optimal TG/VLDL-C ratios increased overall concordance compared to median ratios within the same stratification, with LDL-CKO-N estimates outperforming their LDL-CKM-N counterparts. However, this improvement was not statistically significant in LDL-C estimates derived from TG-only stratification.
Applying optimal TG/VLDL-C ratios within the Martin-Hopkins method improves accuracy compared to median ratios, particularly when stratifications incorporate both TG and non-high-density lipoprotein cholesterol (non-HDL-C) levels. This enhancement can be achieved without increasing the number of strata, offering a practical pathway to refine LDL-C estimation while avoiding excessive stratification. Our findings suggest that while median statistics may be sufficient for TG-only stratifications, they do not fully capture optimal TG/VLDL-C ratios for combined TG and non-HDL-C stratifications.
低密度脂蛋白胆固醇（LDL-C）是心血管疾病的主要可改变风险因素，当甘油三酯（TG）水平低于400mg/dL时，通常使用Friedewald公式计算。最近的研究表明，Martin-Hopkins方法在不同人群中具有更高的准确性。虽然该方法使用特定分层的TG/VLDL-C中位数比率来估计极低密度脂蛋白胆固醇（VLDL-C），但其对中位数统计的依赖引发了这些比率是否真正最优的问题。
本研究评估了Martin-Hopkins方法与Friedewald公式相比的性能，重点关注通过应用最优TG/VLDL-C比率进行改进的潜力。利用韩国国家健康与营养检查调查（KNHANES）中18322名个体的数据，我们根据临床指南定义的LDL-C类别，得出了旨在使与直接测量的LDL-C值的一致性最大化的特定分层最优TG/VLDL-C比率。我们比较了四种LDL-C估计模型的性能：Friedewald公式（LDL-CF）、原始的Martin-Hopkins方法（LDL-CM-N）以及两种应用我们数据得出的TG/VLDL-C比率的替代模型——一种使用中位数（LDL-CKM-N），另一种使用针对每个分层量身定制的最优值（LDL-CKO-N）。
对于TG水平<400mg/dL，Martin-Hopkins方法显示出比Friedewald公式显著更高的一致性（LDL-CF为79.6%，LDL-CM-180为83.2%，p<0.001）。对于TG水平<150mg/dL，一致性提高不到2%（83.3%对84.9%），但对于TG水平为150 - 399mg/dL，一致性提高了约10%（68.8%对78.0%）。在TG水平为150 - 399mg/dL的个体中，对LDL-C水平<70mg/dL进行分类时观察到最大差异（LDL-CF为47.5%，LDL-CM-180为90.3%）。然而，10单元格和180单元格的Martin-Hopkins方程之间的总体一致性差异不大（LDL-CM-10为82.8%，LDL-CM-180为83.2%，差异为0.4%），表明尽管分层数量大幅增加，但仅带来了边际益处。与相同分层内的中位数比率相比，使用最优TG/VLDL-C比率可提高总体一致性，LDL-CKO-N估计优于其LDL-CKM-N对应项。然而，在仅基于TG分层得出的LDL-C估计中，这种改进在统计学上并不显著。
与中位数比率相比，在Martin-Hopkins方法中应用最优TG/VLDL-C比率可提高准确性，特别是当分层同时纳入TG和非高密度脂蛋白胆固醇（non-HDL-C）水平时。这种改进可以在不增加分层数量的情况下实现，为优化LDL-C估计提供了一条实用途径，同时避免过度分层。我们的研究结果表明，虽然中位数统计对于仅基于TG的分层可能足够，但它们不能完全捕捉到TG和non-HDL-C联合分层的最优TG/VLDL-C比率。
