Background
Medical research and clinical trials have shown that the low-density lipoprotein cholesterol (LDL-C) concentration is causally related to an increased risk of coronary artery disease [
1,
2]. In addition, a report by the National Cholesterol Education Program Adult Treatment Panel III notes that the level of LDL-C is the primary variable that is used to predict cardiovascular disease [
1]. One well-known formula for calculating this, the Friedewald formula (FF), is of doubtful accuracy and precision, and thus other approaches have been developed, such as DeLong’s formula (DF) [
3,
4], Teerakanchana’s multiple regression (MR) [
5], Balal’s formula (BF), which is derived from the FF [
6], Tsai’s formula (TF) [
7], calibrated from TF (CTF) [
8], and Tsai’s multiple regression (TMR) [
8]. All of these formulae measure the LDL-C concentrations based on total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), and triglyceride (TG) concentrations [
9‐
12]. Several studies compare the various methods used to assess the LDL-C concentration, and this is likely due to rising healthcare expenditures as well as an increasing demand for quality healthcare. It is thus highly desirable to identify an accurate, a cost-effective method to determine the LDL-C concentration.
Most clinical trials employ the FF [
3], which uses TC, HDL-C, and TG to measure the levels of LDL-C [
5]; thus, it can be applied to the clinical treatment and prevention of atherosclerotic disease [
6,
8]. However, the FF has produced inaccurate results in some cases, and it is not recommended for use in the presence of hypertriglyceridemia (>400 mg/dL) or type III hyperlipoproteinemia [
13]. This method also tends to underestimate LDL-C concentrations [
6,
14‐
18] when the triglyceride concentration is normal [
19,
20] or less than 400 mg/dL [
4,
6,
21,
22]. Balal et al. [
6] thus revised the FF for use with renal transplant recipients by considering those with TG concentrations lower than 400 mg/dL to calculate LDL-C levels. Teerakanchana et al. [
5] developed a multiple regression formula by using a multiple linear regression model to test different data sets. Tsai et al. [
8] further took into account residual cholesterol (RC), which consists of high-density lipoprotein cholesterol (HDL-C), and revised the FF by using TG = 1/8 instead of TG = 1/5, which represents very-low-density lipoprotein cholesterol (VLDL-C).
LDL-C can now be measured directly using advanced technologies, and while the time and cost of these technologies continue to decrease, their costs remain relatively high compared to using formulae to produce estimates. LDL-C concentration may thus be determined in hospitals, at least in part, through best practice measures, and TMR is a valuable method for providing benchmarking data [
8]. However, no studies to date have explored the use of formulae to estimate LDL-C concentration among subjects of different ages and genders. Measuring LDL-C without considering age and gender may produce misleading results, because one formula may perform well with one age group or gender, but perform poorly with others. This study thus compares all seven formulae shown in Table
1 in terms of mean squared error (MSE), as well as underestimation and overestimation of LDL-C concentrations, using subjects of various ages and both genders.
Table 1
Comparison of seven LDL-C formulae
| FF: |
LDL-C = TC- (HDL-C) - (TG/5) |
| BF: |
LDL-C =8.018 + 0.99(LDL-C predicted by FF) |
| DF: |
LDL-C = TC- (HDL-C)- 0.16TG |
Teerakanchanna et al. [ 5] | MR: |
LDL-C = 0.910TC - 0.634(HDL-C) - 0.111TG - 6.755 |
| TF: |
LDL-C = TC- (HDL-C) - (TG/8) |
| CTF: |
LDL-C =0.276 + 0.997(LDL-C predicted by TF) |
| TMR: |
LDL-C =0.988TC - 0.853(HDL-C) - 0.107TG - 8.703 |
Discussion
The results shown in Figures
1 and
2 indicate that the FF has relatively low accuracy. Although it exhibits relatively good performance among older women (aged 45 and above) with TG ≤ 400 mg/dL, its overall performance is worse than that of the other formulae. The formula with the best performance is TMR, followed by TF, CTF, and MR, with no significant differences among them, and the TF and CTF values in particular being virtually identical. Due to the properties of the multiple regression equation, the coefficients are more complex for MR and TMR. In terms of ease of use, TF is the preferred formula.
According to Tsai’s analyses, the FF tends to underestimate LDL-C concentration by 10.1 mg/dL on average [
7], while Balal et al. [
6] report that the FF underestimates it by 8 mg/dL, and other studies have shown similar results [
14‐
18]. Tsai’s results also showed that the difference in the maximum and minimum for the FF is larger than that of the other formulae, and concluded that it is unsuitable for research on epidemiological or causal relationships [
7].
For all cases examined with/without TG ≥ 400 mg/dL in this study, BF, the formula proposed by Balal et al. [
6], provided better results than the FF, although it was still not as good as the other formulae. Tsai et al. [
7] report that BF has exactly the same R
2 as the FF, suggesting that BF only calibrated the underestimation of the FF. These results demonstrate that while the calibrated formula, acquired from the regression of the estimated value and the measured value, could produce an average estimated error that approaches zero and hence reduce the estimated bias, this still would not make the estimation more precise [
7]. In addition, an LDL-C formula is primarily used to precisely estimate the LDL-C concentration for individuals, and while reducing the group estimated bias is important, this only reduces part of the individual estimated bias by expanding another part of it, and the standard deviation of estimated error is not improved. As shown in this study, BF is not able to replace the FF or improve its shortcomings.
As noted above, the best performance for the FF was in subjects with TG ≤ 400 mg/dL, although even among these it was outperformed by the other formulae, which provided stable results when age and gender were taken into account.
Based on a multiple linear regression analysis of 1,016 cases, Teerakanchana et al. [
5] obtained the formula LDL-C = 0.910TC - 0.634(HDL-C) - 0.111TG - 6.755. Tsai et al. [
8] also analyzed training data with multiple linear regression, and found that LDL-C = 0.9882TC - 0.8526(HDL-C) - 0.1065TG - 8.7029, with an R
2 value similar to that of MR (R
2 = 0.9649) and TF (R
2 = 0.9608). In the present study, the R
2 values for MR and TMR were determined to be 0.9648 and 0.9597, respectively; thus, there was no substantial difference between them in this respect. Since multiple linear regression analysis, TMR, is far more complex than TF, it is suggested that TF be used in most cases.
Because LDL-C tests tend to be time-consuming and inconvenient, the FF of LDL-C = TC - (HDL-C) - (VLDL-C) is often clinically applied to produce estimates of this value [
3]. This formula assumes that the VLDL-C of healthy adults, except those with type III hyperlipidemia, is TG/5 [
3,
23,
24] without chylomicrons. However, when using FF, VLDL-C would be overestimated, causing the underestimation of LDL-C, when TG chylomicrons and related remnants appear in plasma [
25]. FF also assumes that TC only contains LDL-C, HDL-C, and VLDL-C, although it likely contains other constituents as well. For example, it has been shown that TC also contains intermediate-density lipoprotein cholesterol (IDL-C), chylomicrons, VLDL-C remnants, lipoprotein(a) [Lp(a)], Lp-X, and some fats that cannot be quantified with current methods [
26]. In this case, when the contents other than HDL-C and LDL-C in TC are defined as RC, then RC = TC - (HDL-C) - (LDL-C) would be more accurate than using VLDL-C to estimate the RC. When TG has a specific relationship with RC, it would be more reasonable to estimate RC using TG [
8]. When the regression analysis takes into account that TC contains LDL-C, HDL-C, VLDL-C, IDL-C, chylomicrons, Lp(a), Lp-X, and other non-quantifiable fats, Tsai et al. suggest revising the FF using TG = 1/8 instead of TG = 1/5 [
8].
Research limitations
In this study, participants with diabetes, secondary dyslipidemias (e.g., dyslipidemia due to renal, liver, or thyroid disease), and those who were taking statins or other lipid-modifying agents at the time of the enrollment were not excluded. In addition, the extrapolation of findings to other populations could introduce errors. The experimental benchmarking is therefore deemed specific for the Taiwanese cohort in this study.
Some subjects with heritable hyperlipidemia have extremely high TG. However, the current study had few cases with TG > 1500 mg/dL; these were not included in the analyses. In addition, some related studies were carried out after the subjects had fasted for 12 hours [
27,
28], while in this study the subjects fasted for 8 hours, and this may have produced some discrepancies with previous results, which is an issue that requires further examination.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.
The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
CHT – conception, design, acquisition of data, drafting of manuscript, final approval. HHW – conception, analysis and interpretation of data, revising manuscript, final approval. SJW-conception, design, interpretation of data, drafting of manuscript, revising manuscript, final approval.