Abstract Background Patients at high risk for Atherosclerotic Cardiovascular Disease (ASCVD), who do not achieve adequate lowering of LDL-cholesterol (LDL-C) with maximally tolerated statins or who are statin-intolerant, are often treated with proprotein convertase subtilisin/kexin (type 9) serine protease inhibitor (PCSK9-inh) therapy. The LDL-C reference method, Beta-Quantitation (BQ), is not routinely performed in these patients. Because low values are measured less accurately with current methods, a variety of equations have been developed to calculate LDL-C from the standard lipid panel test results (total cholesterol (TC), triglycerides (TG), and high-density lipoprotein cholesterol (HDL-C)). We further modified the Sampson LDL-C (S-LDL-C) equation by a novel approach by combining logistic regression (LR) and linear regression to minimize errors when compared to BQ for low LDL-C. Methods A dataset of BQ results from Mayo Medical Labs was divided randomly into a training set (TG ≤ 1500 mg/dL, TC ≤ 1050 mg/dL, n=19,639) and a validation set (TG ≤ 800 mg/dL, BQ-LDL-C ≤ 130 mg/dL, n=12,910). The training dataset was analyzed by LR, using BQ-LDL-C to classify patients as being above or below the 70 mg/dL cut-point. Dependent variables in the LR model were the S-LDL-C, HDL-C, TC, TG and an interaction term (TG x NonHDL-C). The sum of the exponential terms in the trained LR equation for the dependent variables were then fitted by least squares linear regression to match BQ-LDL-C to develop the modified S-LDL-C equation (mS-LDL-C). We used the validation dataset to compare Friedewald (F-LDL-C), extended Martin (eM-LDL-C), Sampson (S-LDL-C) and modified Sampson (mS-LDL-C) equations by Cohen’s kappa coefficient for agreement with BQ-LDL-C and normalized Matthew’s correlation coefficient (MCC), which considers any imbalances in the number of false high versus false low results. Results The mS-LDL-C equation (mS-LDL-C = (TC/0.9423 - HDL-C/0.9582 - TG/9.397 - ((TGxNonHDL-C)/1305) + TG2/8569) - 5.688), primarily differs from the S-LDL-C equation in the value of the coefficients that include TG as a variable. For S-LDL-C ≤100 mg/dL, mS-LDL-C had the best concordance with BQ-LDL-C in terms of the total percent of patients misclassified as either high or low (mS-LDL-C: 21.4%, S-LDL-C: 24.4%, eM-LDL-C: 25.4% F-LDL-C: 30.8%). mS-LDL-C also had the best kappa score when compared to BQ (mS-LDL-C: 0.833, S-LDL-C: 0.810, eM-LDL-C: 0.789, F-LDL-C: 0.743). MCC was also the highest for mS-LDL-C (mS-LDL-C: 0.916, S-LDL-C: 0.906, eM-LDL-C: 0.894, F-LDL-C: 0.879). Finally, mS-LDL-C also showed the best concordance with BQ-LDL-C when an alternative lower LDL-C therapeutic decision cut-point of 55 mg/dL is used (kappa of 0.766 versus 0.724) but showed an increasingly poorer performance than S-LDL-C for higher LDL-C whenever it exceeded 150 mg/dL. Conclusions A strategy to maximize the overall accuracy of estimated LDL-C could provide an automatic reflex in a laboratory information system to the mS-LDL-C equation whenever S-LDL-C is <100 mg/dL. By doing so, the use of the more accurate mS-LDL-C equation for low LDL-C should improve PCSK9-inh therapeutic decision making over the other LDL-C equations.
Read full abstract