PLS regression models for quantification of TC, TG, HDL-C and apoB
The lipid methyl and methylene region of proton NMR spectra of human serum encodes detailed information about the concentrations and lipid compositions of the multiplicity of lipoproteins of different size and density that transport lipids in blood [6]. We used the NMR data from this spectral region, obtained from thousands of patient serum samples that had also undergone standard chemical analysis for lipids and apolipoproteins, to create separate PLS regression models for TC, TG, HDL-C, and apoB to enable their simultaneous quantification during clinical NMR LipoProfile testing.
The characteristics of the sample sets used to build the PLS models are given in Table 1. Not only were very large numbers of samples included in each training set (> 3,500 for TC, TG, and HDL-C; 969 for apoB), the ranges of lipid and apoB values were also very large so as to encompass the wide diversity of normolipidemic and dyslipidemic samples encountered in clinical practice. Samples with low HDL-C pose an added challenge because the low levels can arise either from low HDL particle numbers and/or from HDL particles with abnormally low amounts of cholesterol per particle as typically found in sera from hypertriglyceridemic individuals. To optimize quantification of HDL-C, we created two PLS models, one for use with normal TG samples (TG < 250 mg/dL) and the other for samples with high TG (≥250 mg/dL). The ELP assay software uses the PLS-determined TG level of the sample to automatically select which of the two PLS models to use for calculating HDL-C. As shown in Table 1 and Fig. 1 for the training sample datasets, the 5 created PLS models produce NMR-derived values of TC, TG, HDL-C, and apoB that strongly correlate with chemically-measured values (r ≥0.98).
Table 1
Sample sets used to build PLS models and the model performance in these sample sets
| Sample Set Characteristics | Model Performance |
PLS Model | N | Mean (mg/dL) | SD (mg/dL) | Range (mg/dL) | Latent Variables | R | RMSECV (mg/dL) | CVCV % |
TC | 3746 | 181 | 42 | 64–476 | 25 | 0.987 | 6.67 | 3.68 |
TG | 3734 | 127 | 68 | 24–886 | 27 | 0.996 | 5.74 | 4.51 |
HDL-C (1) | 3453 | 54 | 16 | 14–167 | 31 | 0.988 | 2.50 | 4.60 |
HDL-C (2) | 1354 | 38 | 13 | 3–104 | 29 | 0.976 | 2.28 | 6.01 |
ApoB | 969 | 94 | 27 | 35–305 | 23 | 0.978 | 5.53 | 5.91 |
TC, total cholesterol; TG, triglycerides; HDL-C, HDL cholesterol; ApoB, apolipoprotein B; HDL-C (1), HDL-C model used for samples with TG < 250 mg/dL; HDL-C (2), HDL-C model used for samples with TG ≥250 mg/dL; R, correlation coefficient; RMSECV, root mean square error of cross validation; CVCV, coefficient of variation of cross validation = RMSECV/Mean. |
Independent validation of ELP assay performance
Evaluation of assay sensitivity, linearity and precision
The limits of blank (LOB) were calculated to be 18.9, 11.9, 11.3, and 15.7 mg/dL for TC, TG, HDL-C, and apoB, respectively. The corresponding analytical sensitivity or limits of detection (LOD) were 22.2, 13.8, 12.9, and 17.9 mg/dL, respectively. Testing of several pools with varying analyte concentrations gave functional sensitivity or limits of quantitation (LOQ) of 23.5, 15.2, 12.9, and 17.9 mg/dL, respectively.
To evaluate linearity over the biological ranges of the ELP analytes, several serum pools with widely varying analyte concentrations were prepared and tested. Plots of expected versus PLS-measured NMR values are shown in Fig. 2, demonstrating excellent linearity over a wide range of concentrations for each ELP analyte.
Serum pools with varying levels (low, medium and high) of each of the analytes were used to evaluate intra-assay (within-run) and inter-assay (within-lab) precision. Results are summarized in Table 2. The CV% for intra-assay precision for all of the analytes ranged from 1.0 to 3.8% for the low, 1.0 to 1.7% for the medium, and 0.9 to 1.3% for the high pools. The CV% for inter-assay precision ranged from 1.4 to 3.6% for the low, 1.2 to 2.3% for the medium, and 1.0 to 1.9% for the high pools.
Table 2
Within-run and within-lab precision of ELP assay results
| | TC (mg/dL) | TG (mg/dL) | HDL-C (mg/dL) | ApoB (mg/dL) | LDL-C (mg/dL) |
Within-runa | | | | | | |
Low | Mean | 159.3 | 128.4 | 36.6 | 76.7 | 57.0 |
| SD | 2.6 | 1.0 | 0.9 | 0.9 | 2.2 |
| CV% | 1.6 | 1.0 | 2.4 | 1.2 | 3.8 |
Medium | Mean | 196.3 | 157.7 | 49.9 | 105.5 | 102.1 |
| SD | 2.6 | 1.6 | 0.8 | 1.1 | 1.7 |
| CV% | 1.3 | 1.0 | 1.6 | 1.1 | 1.7 |
High | Mean | 275.8 | 317.4 | 91.2 | 133.8 | 181.2 |
| SD | 2.6 | 3.1 | 1.2 | 1.4 | 2.3 |
| CV% | 0.9 | 1.0 | 1.3 | 1.1 | 1.3 |
Within-labb | | | | | | |
Low | Mean | 166.6 | 130.6 | 36.7 | 78.9 | 58.4 |
| SD | 2.7 | 1.8 | 1.0 | 1.9 | 2.1 |
| CV% | 1.6 | 1.4 | 2.8 | 2.4 | 3.6 |
Medium | Mean | 197.1 | 160.9 | 49.3 | 109.4 | 104.4 |
| SD | 2.8 | 1.9 | 1.2 | 2.4 | 2.1 |
| CV% | 1.4 | 1.2 | 2.3 | 2.2 | 2.0 |
High | Mean | 279.6 | 320.3 | 91.7 | 137.9 | 184.9 |
| SD | 3.2 | 3.1 | 1.3 | 2.6 | 2.6 |
| CV% | 1.1 | 1.0 | 1.4 | 1.9 | 1.4 |
a Serum pools tested in 1 run of 20 replicates. b Serum pools tested in 2 runs of duplicates per day for 20 days (n = 80 per analyte). TC, total cholesterol; TG, triglycerides; HDL-C, HDL cholesterol; ApoB, apolipoprotein B; CV%, coefficient of variation expressed as percent |
Reproducibility of results generated in three clinical laboratories
The reproducibility of ELP assay results obtained in three clinical laboratory sites were evaluated using serum pools with analyte levels at or around their established medical decision limits. Results shown in Supplemental Table 1 (see Additional file 1) indicate very good agreement of ELP analyte concentrations obtained on different Vantera analyzers at the 3 sites, with CV% values generally below 5%.
Method comparison
Independent sets of clinical serum specimens were analyzed to compare results generated by the NMR ELP assay and standard chemistry assays. The method comparison study included 281 samples for TC, 270 for TG, 514 for HDL-C and 266 for apoB, each tested in singlicate. Concentrations obtained by NMR and chemistry testing were highly correlated (r ≥0.98) with slopes ranging from 0.970 to 0.982 and intercepts from − 3.9 to 7.3 mg/dL (Fig. 3).
Taking into account the LOQ, linearity and method comparison results, the reportable ranges for the analytes measured by the ELP assay are 66 ─ 868 mg/dL for TC, 35 ─ 950 mg/dL for TG, 14 ─ 152 mg/dL for HDL-C, and 35 ─ 366 mg/dL for apoB.
Evaluation of potentially interfering substances
A total of 27 endogenous (e.g. bilirubin, hemoglobin) and exogenous (over-the-counter and prescription drugs) substances were tested in vitro for potential interference with ELP assay HDL-C, TC and TG results, while 38 substances were tested for potential interference with apoB results. All substances were tested at concentrations prescribed by CLSI guidelines. The data in Supplemental Table 2 (see Additional file 1), showing the highest substance concentrations tested that did not elicit interference with TC, TG, HDL-C, apoB and LDL-C results, indicate that none of the substances interfered with the NMR ELP assay at naturally-occurring levels (endogenous) or at therapeutic concentrations (exogenous).
Comparison of results from samples obtained using different blood collection tubes
ELP assay results were compared for specimens obtained using the following blood collection tubes: Greiner serum tube (serving as the comparator since it is the preferred collection tube for NMR LipoProfile testing), BD Vacutainer serum tubes (red-top), K2EDTA plasma tubes, and Na-heparin plasma tubes. Supplemental Table 3 in Additional file 1 shows the results of linear regression analyses comparing ELP values from each collection tube to those obtained using the Greiner tube. Slopes were generally ≥ 0.95 except for Na-heparin plasma (slope = 0.94), with excellent correlation coefficients (R2 = 1.00) for all tube types. No significant bias (> 10%) was observed for the 95% confidence intervals around the correlation slopes or intercepts.
Analyte stability in specimens stored at different temperatures
The stability of each analyte reported by the ELP assay at different storage temperatures and freeze-thaw cycles was evaluated in serum and plasma samples. Results were considered acceptable if the means were within 10% of the day 0 mean (Supplementary Table 4 in Additional file 1). All 5 analytes were stable in samples collected in Greiner tubes when stored at room temperature for up to 7 days, refrigerated for up to 14 days, frozen at -25 to -10 °C for up to 14 days and frozen at -70 °C for up to 6 years. All analytes were stable for 5 freeze-thaw cycles, except apoB collected in Greiner tubes which was stable for just one freeze-thaw cycle.
Accuracy assessed by analysis of NIST SRM 1951c standard
Table 3
Comparison of ELP and certified lipid values for the NIST SRM 1951c reference material
| Total Cholesterol (mg/dL) | Triglycerides (mg/dL) | HDL Cholesterol (mg/dL) | LDL Cholesterol (mg/dL) |
Level | ELP valuea | Ref. valueb | Biasc % | ELP value | Ref. value | Bias % | ELP value | Ref. value | Bias % | ELP value | Ref. value | Bias % |
1 | 157.3 (2.9) | 152.4 (1.8) | 3.2 | 153.0 (1.7) | 152.0 (3.2) | 0.7 | 43.3 (1.2) | 41.0 (0.9) | 5.6 | 87.0 (1.7) | 86.4 (1.4) | 0.7 |
2 | 243.0 (2.6) | 241.4 (2.8) | 0.7 | 139.7 (1.5) | 145.4 (3.2) | -3.9 | 66.3 (1.2) | 64.9 (1.7) | 2.2 | 152.0 (1.0) | 143.8 (2.1) | 5.7 |
aELP values are means (standard deviation) of triplicate measurement. bReference values are means (95% confidence intervals). cBias is percent difference between ELP and reference value. |
The accuracy of ELP assay lipid values was assessed by comparing results to the certified/reference values assigned to the NIST 1951c Standard Reference Material (SRM). The purpose of the NIST reference material is to evaluate the accuracy of clinical procedures for determination of TC, TG, HDL-C and LDL-C, as well as validating working or secondary reference materials. As shown in Table 3, ELP assay values were within 4% of the certified/reference values, except for low HDL-C (5.6% bias) and high LDL-C (5.7% bias).
Accuracy and precision over time assessed by the CDC Lipids Standardization Program
Participation in the CDC Lipids Standardization Program (LSP) provides external monitoring over time of analytical accuracy and precision of lipid and apolipoprotein testing as performed in clinical laboratory settings. Three blinded LSP serum standards traceable to the CDC Reference Laboratory were obtained quarterly and tested 4 times in duplicate at 3-week intervals. Table 4 summarizes the results of ELP testing for each quarter of 2019. Overall mean bias and CV% over time for TC, TG, HDL-C, and apoB were very low (all < 3%, except 3.6% bias for apoB). Precision and accuracy evaluations for TC, TG, and HDL-C have continuously passed the criteria set by the LSP since ELP assay participation began in 2018 (LSP does not set performance criteria for apoB).
Table 4
Accuracy and precision of ELP assay assessed during Q1-Q4 2019 by the CDC Lipids Standardization Program
| Total Cholesterol (mg/dL) | Triglycerides (mg/dL) | HDL Cholesterol (mg/dL) | Apolipoprotein B (mg/dL) |
2019 Quarter/ Serum Pool | ELP Valuea | CDC Targetb | Biasc % | CVd % | ELP Value | CDC Target | Bias % | CV % | ELP Value | CDC Target | Bias % | CV % | ELP Value | CDC Target | Bias % | CV % |
Q1/162 | 166.9 | 169.4 | -1.5 | 1.6 | 88.3 | 88.5 | -0.2 | 2.9 | 58.3 | 55.3 | 5.4 | 1.5 | 83.9 | 83.7 | 0.2 | 2.0 |
Q1/163 | 117.5 | 116.1 | 1.2 | 2.0 | 74.5 | 73.0 | 2.1 | 1.5 | 42.1 | 40.8 | 3.2 | 3.2 | 58.4 | 53.8 | 8.6 | 3.3 |
Q1/171 | 176.4 | 176.2 | 0.1 | 3.2 | 104.3 | 103.2 | 1.1 | 3.2 | 56.9 | 55.7 | 2.2 | 2.0 | 84.1 | 86.3 | -2.5 | 2.5 |
Q2/171 | 173.4 | 176.2 | -1.6 | 1.5 | 102.9 | 103.2 | -0.3 | 1.9 | 56.6 | 55.7 | 1.6 | 2.9 | 83.4 | 86.3 | -3.4 | 2.0 |
Q2/172 | 157.8 | 160.7 | -1.8 | 3.0 | 79.8 | 78.8 | 1.3 | 3.4 | 51.6 | 51.8 | -0.4 | 4.0 | 82.9 | 80.3 | 3.2 | 2.8 |
Q2/485 | 131.0 | 131.4 | -0.3 | 3.6 | 107.9 | 106.8 | 1.0 | 1.8 | 37.3 | 34.8 | 7.2 | 2.0 | 72.4 | 70.3 | 3.0 | 2.8 |
Q3/171 | 178.8 | 176.2 | 1.5 | 1.8 | 105.5 | 103.2 | 2.2 | 2.9 | 57.9 | 55.7 | 3.9 | 1.4 | 87.0 | 86.3 | 0.8 | 2.6 |
Q3/172 | 162.6 | 160.7 | 1.2 | 1.9 | 82.4 | 78.8 | 4.6 | 2.2 | 53.0 | 51.8 | 2.3 | 1.7 | 84.8 | 80.8 | 5.0 | 2.2 |
Q3/173 | 184.0 | 187.9 | -2.1 | 2.3 | 254.0 | 251.9 | 0.8 | 2.0 | 36.5 | 37.3 | -2.1 | 3.3 | 106.9 | 104.8 | 2.0 | 1.5 |
Q4/174 | 152.8 | 152.6 | 0.1 | 1.8 | 85.4 | 88.0 | -3.0 | 1.4 | 52.6 | 52.4 | 0.4 | 3.0 | 75.9 | 70.8 | 7.2 | 1.8 |
Q4/175 | 187.9 | 189.4 | -0.8 | 1.4 | 164.5 | 165.0 | -0.3 | 0.9 | 54.4 | 54.4 | 0.0 | 1.4 | 97.1 | 93.7 | 3.6 | 1.2 |
Q4/A100 | 182.4 | 183.5 | -0.6 | 1.8 | 103.3 | 105.1 | -1.7 | 1.3 | 57.3 | 55.5 | 3.2 | 1.8 | 90.1 | 87.0 | 3.6 | 3.3 |
Absolute Mean | | | 1.1 | 2.2 | | | 1.6 | 2.1 | | | 2.7 | 2.4 | | | 3.6 | 2.3 |
aELP values are the mean of 8 determinations from ELP testing conducted in duplicate at 3 week intervals. bTarget values of the CDC reference sera pools. cBias is the percent difference between ELP and CDC values. dPrecision of ELP measurement as given by the coefficient of variation (CV) of the 8 ELP determinations. |