Is Machine Learning-derived Low-Density Lipoprotein Cholesterol estimation more reliable than standard closed form equations? Insights from a laboratory database by comparison with a direct homogeneous assay

Abstract

BackgroundThere is no consensus on the best method to estimate Low Density Lipoprotein-Cholesterol (LDL-C) in routine laboratories. MethodsWe conducted a retrospective study to compare the performances of a Machine Learning (ML) algorithm using the K-Nearest Neighbors (LDL-KNN) method with that of the Friedewald formula (LDL-F), the Martin-Hopkins equation (LDL-NF), the de Cordova equation (LDL-CO) and the Sampson equation (LDL-SA) against direct homogeneous LDL-C assay (LDL-D) in patients who presented to the Laboratories of Hôtel Dieu de France university hospital in Beirut, Lebanon, from September 2017 to July 2020. Agreements between methods were analyzed using Intraclass Correlation Coefficients (ICC) and the Bland-Altman method of agreement. Results31,922 observations from 19,279 subjects were included, with a mean age of 52 ± 18 years and 10,075 (52.3%) females. All methods except LDL-F and LDL-CO exhibited an overall ICC beyond the 0.9 cut-off. LDL-SA, LDL-NF and LDL-KNN were less susceptible to triglyceridemia than LDL-F and LDL-CO, with LDL-KNN resulting in the lesser fraction of points beyond the Bland-Altman limits of agreement. ConclusionAn ML algorithm using LDL-KNN is promising for the estimation of LDL-C as it agrees better with LDL-D than closed form equations, especially in mild and severe hypertriglyceridemia.