Abstract
Background and Aims
Despite growing evidence that apolipoprotein B (apoB) is the most accurate marker of atherosclerotic cardiovascular disease (ASCVD) risk, its adoption in clinical practice has been low. This investigation sought to determine whether low-density lipoprotein cholesterol (LDL-C), non–high-density lipoprotein cholesterol (HDL-C), and triglycerides are sufficient for routine cardiovascular care.
Methods
A sample of 293 876 UK Biobank adults (age: 40–73 years, 42% men), free of cardiovascular disease, with a median follow-up for new-onset ASCVD of 11 years was included. Distribution of apoB at pre-specified levels of LDL-C, non-HDL-C, and triglycerides was examined graphically, and 10-year ASCVD event rates were compared for high vs. low apoB. Residuals of apoB were constructed after regressing apoB on LDL-C, non-HDL-C, and log-transformed triglycerides and used as predictors in a proportional hazards regression model for new-onset ASCVD adjusted for standard risk factors, including HDL-C.
Results
ApoB was highly correlated with LDL-C and non-HDL-C (Pearson’s r = .96, P < .001 for both) but less so with log triglycerides (r = .42, P < .001). However, apoB ranges necessary to capture 95% of all observations at pre-specified levels of LDL-C, non-HDL-C, or triglycerides were wide, spanning 85.8–108.8 md/dL when LDL-C 130 mg/dL, 88.3–112.4 mg/dL when non-HDL-C 160 mg/dL, and 67.8–147.4 md/dL when triglycerides 115 mg/dL. At these levels (±10 mg/dL), 10-year ASCVD rates for apoB above mean + 1 SD vs. below mean − 1 SD were 7.3 vs. 4.0 for LDL-C, 6.4 vs. 4.6 for non-HDL-C, and 7.0 vs. 4.6 for triglycerides (all P < .001). With 19 982 new-onset ASCVD events on follow-up, in the adjusted model, residual apoB remained statistically significant after accounting for LDL-C and HDL-C (hazard ratio 1.06, 95% confidence interval 1.0–1.07), after accounting for non-HDL-C and HDL-C (hazard ratio 1.04, 95% confidence interval 1.03–1.06), and after accounting for triglycerides and HDL-C (hazard ratio 1.13, 95% confidence interval 1.12–1.15). None of the residuals of LDL-C, non-HDL-C, or of log triglycerides remained significant when apoB was included in the model.
Conclusions
High variability of apoB at individual levels of LDL-C, non-HDL-C, and triglycerides coupled with meaningful differences in 10-year ASCVD rates and significant residual information contained in apoB for prediction of new-onset ASCVD events demonstrate that LDL-C, non-HDL-C, and triglycerides are not adequate proxies for apoB in clinical care.
Keywords: apoB, Non-HDL-C, Triglycerides, Discordance analysis
Structured Graphical Abstract
Structured Graphical Abstract.
Illustration of residual discordance and variance analysis of apoB and LDL-C/non-HDL-C in UK Biobank. LDL-C, low-density lipoprotein cholesterol; non-HDL-C, non–high-density lipoprotein cholesterol; apoB, apolipoprotein B.
See the editorial comment for this article ‘ApoB triumphs once more over LDL-C and non-HDL-C in risk prediction: ready for guidelines?’, by M.B. Mortensen, https://doi.org/10.1093/eurheartj/ehae257.
Introduction
In 2019, the evidence from prospective observational studies, randomized clinical trials, and Mendelian randomization became sufficient for the European Atherosclerosis Society/European Society of Cardiology Guidelines to conclude that apolipoprotein B (apoB) was a more accurate marker of cardiovascular risk than either low-density lipoprotein cholesterol (LDL-C) or non–high-density lipoprotein cholesterol (non-HDL-C).1 They also concluded that apoB could be measured inexpensively and more accurately and precisely than either LDL-C or non-HDL-C. However, LDL-C was retained as the primary marker for clinical decision-making, as is the case in all the other major guidelines. Moreover, all reports since then, with one exception,2 including observational studies and randomized clinical trials,3–13 have confirmed that apoB is a more accurate marker of cardiovascular risk than LDL-C or non-HDL-C, reinforcing the validity of the evaluation by the 2019 EAS/ESC Guidelines as to the superiority of apoB.
Nevertheless, neither non-HDL-C nor apoB has displaced LDL-C in routine clinical care or in guideline priority. This reluctance to change might be explained by the belief that despite the documented statistical superiority of apoB, the extremely high correlation among LDL-C, non-HDL-C, and apoB would make it unlikely that practical decision-making would be meaningfully altered by the introduction into routine clinical care of either apoB or non-HDL-C.
Furthermore, standard statistical measures that quantify the added prognostic value of new biomarkers in risk prediction models were not designed to compare highly correlated and biologically tightly related variables such as LDL-C, non-HDL-C, or apoB.14 Finally, when recommended by guidelines and consensus groups, measurement of apoB has been prioritized or restricted to patients with hypertriglyceridaemia, the presumption being cholesterol-depleted apoB particles are only common in such individuals. If this restriction is valid, triglycerides should add significantly to the predictive value of apoB.
In this analysis, we focused on two key aspects of surrogacy: the extent to which one marker contains added information after the other is already accounted for and the degree of variability of one marker at the pre-specified levels of the other. The first set of analyses, termed ‘discordance analysis’, was designed specifically to determine which of two highly correlated variables is more closely related to the outcome of interest.15 Yet, discordance analysis, like other standard epidemiologic analyses, relies on estimates derived from groups whereas clinical decisions are made in individual cases to maximize individual benefit and to minimize individual risk. Thus, when comparing predictive power of two correlated markers, it is imperative to also estimate the variability in distribution of one marker as a function of the other. Hence, a second set of analyses was conducted to determine how large was the variance in apoB at different levels of LDL-C, non-HDL-C, and triglycerides, and whether this variance was associated with differences in cardiovascular risk.
Methods
Between 2006 and 2010, the UK Biobank recruited 502 413 women and men aged 37–73 years from primary care lists.16 Data collection methods, including lipid assays, are described elsewhere. From the full cohort, we excluded participants with prevalent cardiovascular disease (N = 32 140) or those receiving lipid-lowering therapy (N = 69 322) at baseline examination. We excluded 60 139 participants with missing records for baseline lipid measurements (apoB, triglycerides, non-HDL-C, or HDL-C) and 40 405 with missing records for sex, systolic blood pressure (SBP), glycated haemoglobin (HbA1c), hypertension or diabetes treatment, body mass index (BMI), and smoking status. Finally, participants with triglycerides ≥ 400 mg/dL, apoB outside the 20–400 mg/dL range, or LDL-C ≥ 250 mg/dL were also excluded (N = 6524) as were 7 patients less than 40. The final analytic sample included 293 883 adults . The primary outcome was atherosclerotic cardiovascular disease (ASCVD), defined in Supplementary data online, Table S1. Analyses of the role of metabolic risk factors in the UK Biobank have been approved by the McGill University Health Center Research Ethics Board.
Statistical analysis
Values for baseline characteristics of participants were expressed as median and interquartile range for continuous variables and numbers and percentages for categorical data. The strength of association between lipid levels was assessed using Pearson’s correlation coefficients. To better understand the variability of one lipid marker at a given value of another, we plotted histograms of apoB at LDL-C levels of 70, 100, 130, 160, and 190 mg/dL as well as non-HDL-C levels of 100, 130, 160, 190, and 220 mg/dL and at triglyceride levels of 50, 75, 115, 175, and 250 mg/dL. These were selected to represent similar percentiles of the respective distributions. For each LDL-C, non-HDL-C, and triglyceride value, we also calculated apoB levels at 2 SDs below and above the mean value, which capture ∼95% of the observed apoB measurements that correspond to that particular value of LDL-C, non-HDL-C, or triglyceride.
To illustrate the potential impact of differing apoB at selected levels of the other lipid markers, we calculated 10-year rates of ASCVD for individuals with apoB at least 1 SD above the mean vs. apoB at least 1 SD below the mean at the LDL-C, non-HDL-C, and triglyceride levels specified above but enlarged by ±10 mg/dL to achieve sufficient sample size in the subgroups (bands of ±5 and ±15 mg/dL were used as sensitivity analyses). We also estimated net correct reclassification for ASCVD events and non-events based on ‘high’ (1 SD above the mean) and ‘low’ (1 SD below the mean) apoB levels.
The effect of primary exposure variables (LDL-C, non-HDL-C, apoB, and log-transformed triglycerides) was quantified using Cox proportional hazards models with incidence of ASCVD as the primary outcome. The resulting hazard ratio (HR) estimates, and the corresponding 95% confidence intervals (CIs), were expressed per 1 SD. We adjusted for age, sex, SBP, antihypertensive treatment, HbA1c, diabetes medication, HDL-C, and smoking. Model fit was described using the likelihood ratio χ2 statistic.
To accurately quantify the effect of apoB vs. LDL-C, non-HDL-C, and triglycerides, we proceeded as follows. Because of the high correlation between apoB and LDL-C or non-HDL-C, we regressed apoB on LDL-C or non-HDL-C and used the resulting residual to represent the amount of apoB that is not explained by LDL-C or non-HDL-C in a model alongside LDL-C or non-HDL-C.8 Similarly, we regressed LDL-C or non-HDL-C on apoB, and the resulting residual was used alongside apoB to represent the portion of LDL-C or non-HDL-C not explained by apoB. We proceeded in a similar fashion with apoB and log triglycerides, including apoB and residual of log triglycerides in one model and residual apoB and log triglycerides in another. All models included HDL-C as a key exposure.
Hypotheses were tested at the two-sided alpha level of .05. Analyses were performed on SAS v.9.4 (SAS Institute, Cary, NC) and R software version 4.2.4 (www.r-project.org).
Results
Baseline characteristics
The median age of the 293 876 participants at baseline was 56 years; 42% were men (Table 1). Median BMI was 26 kg/m2, median SBP was 135 mmHg, and median HbA1c was 5.3% (5.1, 5.6). A total of 12.3% of participants reported use of antihypertensive medications and 58.5% reported smoking, now or in the past. Median and interquartile ranges were 1.05 g/L (.90, 1.20) for apoB, 3.66 mmol/L (3.15, 4.22) for LDL-C, 4.33 mmol/L (3.69, 5.03) for non-HDL-C, 1.44 mmol/L (1.21, 1.72) for HDL-C, and 1.42 mmol/L (1.01, 2.02) for triglycerides (Table 1). The standard deviation for apoB was 22.7 mg/dL.
Table 1.
Baseline characteristics
| N | 293 876 |
| Age (years) | 56.00 [49.00, 62.00] |
| Male | 123 313 (42.0) |
| Incidence ASCVD | 19 982 (6.8) |
| Follow-up (years) | 10.95 [10.11, 11.66] |
| Body mass index (kg/m2) | 26.26 [23.77, 29.27] |
| Systolic blood pressure (mmHg) | 135.00 [123.50, 148.00] |
| Blood pressure medication | 36 264 (12.3) |
| Ever smoked | 171 831 (58.5) |
| HbA1C (%) | 5.33 [5.12, 5.55] |
| Diabetes medication | 2389 (.8) |
| Total cholesterol (mmol/L) | 5.82 [5.16, 6.54] |
| HDL-C (mmol/L) | 1.44 [1.21, 1.72] |
| LDL-C (mmol/L) | 3.66 [3.15, 4.22] |
| TG (mmol/L) | 1.42 [1.01, 2.02] |
| apoB (g/L) | 1.05 [.90, 1.20] |
| Non-HDL-C (mmol/L) | 4.33 [3.69, 5.03] |
Baseline characteristics of the cohort are represented as median [interquantile range] or N (%).
apoB, apolipoprotein B; ASCVD, atherosclerotic cardiovascular disease; HbA1C, haemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; Non-HDL-C, non–high-density lipoprotein cholesterol; TG, triglyceride.
Correlations and variability
ApoB was highly correlated with both LDL-C and non-HDL-C (see Supplementary data online, Table S2), Pearson’s r = .96 (P < .001) for both. However, as illustrated in Figure 1, this high correlation did not imply narrow ranges of apoB at pre-selected levels of LDL-C or non-HDL-C. Indeed, when LDL-C is 70 mg/dL, to capture ∼95% of all apoB values, requires a range that extends from 40.8 to 67.0 mg/dL. The corresponding ranges were 68.4–87.0 mg/dL when LDL-C 100 mg/dL, 85.8–108.8 md/dL when LDL-C 130 mg/dL, 106.3–131.7 mg/dL when LDL-C 160 mg/dL, and 124.6–156.4 mg/dL when LDL-C 190 mg/dL. Similarly, when non-HDL-C is 100 mg/dL, to capture ∼95% of all apoB values, requires a range that extends from 52.3 to 78.0 mg/dL. The corresponding ranges were 73.1–94.9 mg/dL when non-HDL-C 130 mg/dL, 88.3–112.4 md/dL when non-HDL-C 160 mg/dL, 104.8–130.8 mg/dL when non-HDL-C 190 mg/dL, and 119.4–150.8 mg/dL when non-HDL-C 220 mg/dL. apoB was also positively, but less strongly, correlated with log triglycerides (r = .42, P < .001). The apoB ranges necessary to capture ∼95% of all observations were also wide: 54.7–119.6 mg/dL for triglycerides 50 mg/dL, 65–132.0 mg/dL for triglycerides 75 mg/dL, 67.8–147.4 mg/dL for triglycerides 115 mg/dL, 70.1–158.1 mg/dL for triglycerides 175 mg/dL, and 79.3–167.6 mg/dL for triglycerides 250 mg/dL (see Supplementary data online, Figure S1).
Figure 1.
Distribution of apoB at different levels of low-density lipoprotein cholesterol and non–high-density lipoprotein cholesterol. The figure depicts density plots of apoB levels at pre-selected levels of low-density lipoprotein cholesterol (A) and non–high-density lipoprotein cholesterol (B). For each value of low-density lipoprotein cholesterol (A) or non–high-density lipoprotein cholesterol (B), the figure displays a histogram of apoB values for all individuals with that value of low-density lipoprotein cholesterol or non–high-density lipoprotein cholesterol ±1. We note the location of the histograms shifting from left to right as the values of low-density lipoprotein cholesterol (A) or non–high-density lipoprotein cholesterol (B) increase. At the same time, the range covered by the histograms remains wide
apoB was weakly negatively correlated with HDL-C (r = −.11, P < .001), and similar correlation was observed for non-HDL-C and HDL-C. When non-HDL-C was regressed on apoB, the resulting residual was uncorrelated with apoB, and when apoB was regressed on non-HDL-C, the resulting residual was uncorrelated with non-HDL-C. Regressing apoB on log triglycerides and vice versa also led to uncorrelated residuals (see Supplementary data online, Table S2).
Ten-year. event rates at high vs. low apoB levels
Figure 2 presents 10-year Kaplan–Meier rates of ASCVD for individuals with apoB at least 1 SD above the mean vs. at least 1 SD below the mean at the pre-specified LDL-C, non-HDL-C, and triglyceride levels specified above ±10 mg/dL. We observe higher ASCVD rates when apoB is high: for example, when LDL-C = 130 ± 10 mg/dL, 10-year ASCVD rates were 7.3 vs. 4.0 (P < .001) for high vs. low apoB (Figure 2A). The corresponding rates were 6.4 vs. 4.6 (P < .001) when non-HDL-C = 160 ± 10 mg/dL (Figure 2B) and 7.0 vs. 4.6 (P < .001) when triglycerides = 115 ± 10 mg/dL (see Supplementary data online, Figure S2). Results remained consistent when we varied the bands around the pre-selected values of LDL-C, non-HDL-C, and triglycerides (see Supplementary data online, Tables S3–S5). An alternative view of this analysis is presented in Supplementary data online, Table S6, which gives the numbers of individuals with and without ASCVD events reclassified based on high vs. low apoB levels. Net correct reclassification of events based on apoB ranged from 3% to 10% (see Supplementary data online, Table S6).
Figure 2.
Ten-year risk of atherosclerotic cardiovascular disease for high vs. low apoB at different levels of low-density lipoprotein cholesterol and non–high-density lipoprotein cholesterol. The figure depicts 10-year Kaplan–Meier rates of atherosclerotic cardiovascular disease for high (at least 1 SD above the mean; right bar) vs. low (at least 1 SD below the mean; left bar) apoB levels at pre-selected levels of low-density lipoprotein cholesterol ±10 mg/dL (A) and non–high-density lipoprotein cholesterol ±10 mg/dL (B)
Associations on follow-up
Over a median of 11 years, there were 19 982 ASCVD events. In a model adjusted for age, male sex, BMI, smoking, SBP, HbA1c, blood pressure, and glucose-lowering medications, addition of each, LDL-C, non-HDL-C, and apoB, led to significant improvement in model fit (see Supplementary data online, Table S7). The magnitude of this improvement was the largest for apoB (χ2 = 374 vs. 327 for LDL-C and 337 for non-HDL-C, P < .01 vs. apoB; Supplementary data online, Table S7).
apoB and HDL-C but not residual of LDL-C were significantly associated with new-onset ASCVD with HRs per 1 SD change of 1.15 (95% CI: 1.13, 1.17) for apoB, .85 (95% CI: .83, .87) for HDL-C, and .99 (95% CI: .97, 1.00) for residual LDL-C (Table 2). In a model adjusted for the same predictors, but now including LDL-C and residual apoB, the effect of HDL-C remained the same and both LDL-C and residual apoB were significantly associated with the outcome: HR 1.14 (95% CI: 1.12, 1.16) for LDL-C and HR 1.06 (95% CI: 1.04, 1.07) for residual apoB. Similar results were observed for non-HDL-C (Table 3).
Table 2.
Hazard ratios and 95% confidence intervals for incident atherosclerotic cardiovascular disease in adjusted Cox proportional hazards models with apoB and low-density lipoprotein cholesterol
| Variables | HR (95% CI) | P-value | HR (95% CI) | P-value |
|---|---|---|---|---|
| apoB | 1.15 (1.13, 1.17) | <.001 | ||
| LDL-C residual | .99 (.97, 1.00) | .054 | ||
| LDL-C | 1.14 (1.12, 1.16) | <.001 | ||
| apoB residual | 1.06 (1.04, 1.07) | <.001 | ||
| HDL-C | .85 (.83, .87) | <.001 | .85 (.83, .87) | <.001 |
| Body mass index | 1.08 (1.07, 1.10) | <.001 | 1.08 (1.07, 1.10) | <.001 |
| Age | 1.07 (1.06, 1.07) | <.001 | 1.07 (1.06, 1.07) | <.001 |
| Male sex | 1.71 (1.66, 1.77) | <.001 | 1.71 (1.66, 1.76) | <.001 |
| Ever smoked | 1.24 (1.20, 1.27) | <.001 | 1.24 (1.20, 1.27) | <.001 |
| Systolic blood pressure | 1.18 (1.17, 1.20) | <.001 | 1.18 (1.17, 1.20) | <.001 |
| Blood pressure medications | 1.51 (1.46, 1.57) | <.001 | 1.51 (1.46, 1.57) | <.001 |
| HbA1C | 1.08 (1.07, 1.09) | <.001 | 1.08 (1.07, 1.09) | <.001 |
| Diabetes medication | 1.58 (1.42, 1.75) | <.001 | 1.58 (1.42, 1.75) | <.001 |
Hazard ratios for apolipoprotein B, model residuals, LDL-C, HDL, BMI, SBP, and HbA1C are expressed per 1 SD.
apoB, apolipoprotein B; ASCVD, atherosclerotic cardiovascular disease; HbA1C, haemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol.
Table 3.
Hazard ratios and 95% confidence intervals for incident atherosclerotic cardiovascular disease in adjusted Cox proportional hazards models with apoB and non–high-density lipoprotein cholesterol
| Variables | HR (95% CI) | P-value | HR (95% CI) | P-value |
|---|---|---|---|---|
| apoB | 1.15 (1.13, 1.17) | <.001 | ||
| Non-HDL-C residual | 1.00 (.99, 1.01) | .939 | ||
| Non-HDL-C | 1.14 (1.13, 1.16) | <.001 | ||
| apoB residual | 1.04 (1.03, 1.06) | <.001 | ||
| HDL-C | .84 (.83, .86) | <.001 | .84 (.83, .86) | <.001 |
| Body mass index | 1.08 (1.07, 1.10) | <.001 | 1.08 (1.07, 1.10) | <.001 |
| Age | 1.07 (1.06, 1.07) | <.001 | 1.07 (1.06, 1.07) | <.001 |
| Male sex | 1.71 (1.66, 1.77) | <.001 | 1.71 (1.66, 1.77) | <.001 |
| Ever smoked | 1.24 (1.20, 1.28) | <.001 | 1.24 (1.20, 1.28) | <.001 |
| Systolic blood pressure | 1.18 (1.17, 1.20) | <.001 | 1.18 (1.17, 1.20) | <.001 |
| Blood pressure medications | 1.52 (1.46, 1.57) | <.001 | 1.52 (1.46, 1.57) | <.001 |
| HbA1C | 1.08 (1.07, 1.09) | <.001 | 1.08 (1.07, 1.09) | <.001 |
| Diabetes medication | 1.59 (1.43, 1.76) | <.001 | 1.59 (1.43, 1.76) | <.001 |
Hazard ratios for apolipoprotein B, model residuals, non-HDL, HDL, BMI, SBP, and HbA1C are expressed per 1 SD.
apoB, apolipoprotein B; ASCVD, atherosclerotic cardiovascular disease; HbA1C, haemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; non-HDL-C, non–high-density lipoprotein cholesterol.
When considering apoB and residual log triglycerides as well as log triglycerides and residual apoB, HDL-C was significant in both models (HR .85, 95% CI: .83, .86 ). The HRs were 1.15 (95% CI: 1.14, 1.17) for apoB and 1.01 (.99, 1.03) for residual log triglycerides but 1.07 (1.05, 1.09) for log triglycerides and 1.13 (1.12, 1.15) for residual apoB (see Supplementary data online, Table S8).
Discussion
Notwithstanding that the correlation coefficient in UK Biobank between apoB and either LDL-C or non-HDL-C is .96, discordance analysis based on residuals demonstrates that apoB is a more accurate marker of cardiovascular risk than either LDL-C or non-HDL-C in this data base (Structured Graphical Abstract). This finding accords with multiple previous reports. This means that when values of LDL-C or non-HDL-C differ with apoB, apoB more closely corresponds to cardiovascular risk than either LDL-C or non-HDL-C. But are these differences clinically meaningful? Welsh and Sattar,17 for example, have argued that the correlation coefficient between non-HDL-C and apoB is so high that even if apoB is overall more accurate, the cost and educational effort to introduce apoB into clinical care are not justified. Given the correlation coefficient between LDL-C and non-HDL-C is even higher (.98), this restriction would seem to apply to non-HDL-C as well.
The challenge is legitimate if we accept that markers, which are highly correlated statistically, are equivalent clinically. However, the results of this study demonstrate that despite extremely high correlations between apoB and LDL-C or non-HDL-C, at the individual patient level, there remains significant variability in values for apoB at any given level of LDL-C or non-HDL-C. The issue becomes how often such differences might be clinically meaningful.
Discordance analysis was designed to compare the predictive accuracy of two highly correlated variables.15 Groups are created in which the levels of the two correlated variables differ. Cardiovascular risk in these groups is then estimated to determine which variable aligns more closely with observed risk. The groups can be constructed in multiple ways: division based on the medians of the markers,4–7,10,18,19 based on population percentiles of the markers,11 or based on residual analysis.3,8 Residual analysis, which is the approach used in this analysis and which calculates and compares the differences from the regression relating the two markers, is the least arbitrary and theoretically captures the maximum of the differing information between markers and is therefore the strongest methodologically. Importantly, with a single exception,2 the results of all the prospective observational discordance analyses, irrespective of the method employed, are consistent: namely, that the number of apoB particles is a more accurate index of cardiovascular risk than the mass of cholesterol within them. The present analysis adds further strength to this overall body of evidence and establishes that in this specific database, when apoB and either LDL-C or non-HDL-C differ in their estimates of cardiovascular risk—that is, when they are discordant—apoB is the variable that should be relied on clinically.
Nevertheless, as Welsh and Sattar17 pointed out, this is not sufficient to determine whether either LDL-C or non-HDL-C is an adequate clinical surrogate for apoB. Are the differences sufficiently great, sufficiently often, to be clinically meaningful as well as statistically significant? To answer this question, we first determined the extent to which values of one marker vary as a function of the other marker. Our findings challenge the common understanding of the correlation coefficient. In general, high correlation is interpreted as meaning that if we obtain a high value using one marker, we will also obtain a similarly high value for the other, highly correlated marker. While generally correct, in this instance, Figure 1 demonstrates substantial variance in individual values of apoB for a given value of either LDL-C or non-HDL-C. This establishes that high correlation does not necessarily imply equivalence of absolute levels of different markers in individual cases. Moreover, this example provides a precedent that quantitation of discordance should be a necessary step in any comparison of highly correlated markers.
Next, we determined the potential clinical implications of this individual variance between markers. To do so, we have calculated the range that includes 95% of the values of apoB for each of multiple values of LDL-C and non-HDL-C. Histograms at each point demonstrated these values were apparently normally distributed. At a level of LDL-C of 70 mg/dL, the absolute values of apoB could vary from 40.8 to 67.0 mg/dL, an absolute difference of 27.2 mg/dL, a value >1 SD of apoB for the American population (22.7 mg/dL). Similarly, on average, the range in values of apoB for non-HDL-C was ∼25 mg/dL, which is also just over 1 SD of apoB for this population. Similar findings were documented at all the levels of LDL-C and non-HDL-C examined. Because the variance was normally distributed, this means not only that the variance between levels of apoB at given values of LDL-C or non-HDL-C was large but that the subgroups of those with either values either 1 SD higher or lower than the mean would also be large.
Our results extend the analysis of Cole et al.,20 who demonstrated substantial differences between measured levels of LDL-C and calculated levels of non-HDL-C and values of LDL-C and non-HDL-C determined from population percentile equivalents of apoB. At all ranges examined, there was substantial variance between the values of LDL-C or non-HDL-C and apoB that would not be present if the markers were equivalent.
The present analysis further extends these observations by demonstrating how, at multiple ‘fixed’ (±10 mg/dL) levels of LDL-C and non-HDL-C, higher levels of apoB were associated with higher cardiovascular risk than lower levels of apoB. The relative differences were similar across the entire ranges of LDL-C and non-HDL-C and were statistically significant except at the lowest levels. Given the lower event rates and group sizes, this is not surprising. Moreover, our sensitivity analyses at either smaller or greater bands around the ranges of LDL-C or non-HDL-C confirmed these trends. These novel subgroup analyses are consistent with and supported by the core overall analysis that demonstrates that apoB is a more accurate marker of cardiovascular risk than either LDL-C or non-HDL-C.
Multiple recommendations prioritize or restrict the measurement of apoB to subjects with elevated triglycerides based on the assumption this is when cholesterol-depleted apoB particles would be present and that apoB would be more informative than LDL-C.1,21 Therefore, triglycerides should add clinically relevant information to apoB. This analysis offers no support for this recommendation. After factoring the information contained by triglycerides, residual information contained in the apoB was still statistically significant. This was not true for residual information contained in triglycerides after factoring in apoB. This finding should not be surprising given that cholesterol-depleted apoB particles are common at all levels of plasma triglyceride.22
It is also noteworthy that HDL-C remains significantly predictive of risk in all scenarios examined. Based on the failure of the cholesterol ester transfer inhibitors to demonstrate any clinical benefit related to increase in HDL-C23 and the Mendelian randomization studies, which did not support a causal relation between HDL-C and cardiovascular risk,24,25 interest in the potential role of HDL-C as a determinant of the outcome of atherosclerosis has plummeted. Our results, however, are consistent with almost all previous analyses in demonstrating a strong inverse association between HDL-C and cardiovascular risk, an association which must still be explained.
Our analysis relies on a large population-based cohort (UKB). However, it is limited to one European country with racial and ethnic subgroups that represent that country. Furthermore, it is important to note that the focus of this paper is to determine the most accurate lipid marker that should be measured in clinical care and not how to construct the best risk prediction algorithms. Nevertheless, with a trial comparing care guided by different lipid parameters neither likely nor feasible, this observational cohort study provides strong evidence for the superiority of apoB over LDL-C, non-HDL-C, or triglycerides for guiding clinical care to prevent new-onset ASCVD.
Additional cost has been a major argument against broad clinical utilization of apoB. In the USA, measuring apoB as well as a lipid panel at each visit would raise the cost of care by ∼1%.26 However, given that apoB is a more accurate measure of the adequacy of lipid-lowering therapy than LDL-C or non-HDL-C,1,3–13,27 there would be no reason to measure a standard lipid panel at follow-up visits. Accordingly, making apoB the prime measure of lipid therapy would simplify care and minimize additional cost. Notably, this analysis does not take account the improvement in outcome that might reasonably be anticipated from the implementation of a more accurate measure of the adequacy of therapy. Also, newer therapies, such as the PCSK9 inhibitors, are costly, and therefore, the test that most accurately identifies residual cardiovascular risk—apoB—is the test that should be used to select patients for such therapies.
In summary, the present prospective discordance analysis adds to the evidence that apoB is a more accurate marker of cardiovascular risk than LDL-C, non-HDL-C, or triglycerides. Moreover, the present results demonstrate the wide range of values of LDL-C, non-HDL-C, and triglyceride for a given value of apoB and the significant differences in risk among such subjects that becomes evident if apoB is measured. Given the established benefit of lipid-lowering therapies, the imprecision in clinical decision-making that occurs at an individual level if apoB is not measured is sufficient to be clinically unacceptable. These results also explain why treatment of subjects for primary or secondary prevention based on apoB should be more effective than treatment based on LDL-C or non-HDL-C.28 Taken together with the evidence that apoB can be measured more accurately, more precisely, and more selectively than LDL-C or non-HDL-C,1,29–31 this study demonstrates that neither LDL-C nor non-HDL-C is an adequate surrogate for apoB in the assessment and clinical care of individual patients.
Supplementary data
Supplementary data are available at European Heart Journal online.
Supplementary Material
Contributor Information
Allan D Sniderman, Mike and Valeria Rosenbloom Centre for Cardiovascular Prevention, Department of Medicine, McGill University Health Centre-Royal Victoria Hospital, 1001 Boulevard Décarie, Montreal, Québec H4A 3J1, Canada.
Line Dufresne, Mike and Valeria Rosenbloom Centre for Cardiovascular Prevention, Department of Medicine, McGill University Health Centre-Royal Victoria Hospital, 1001 Boulevard Décarie, Montreal, Québec H4A 3J1, Canada.
Karol M Pencina, Mike and Valeria Rosenbloom Centre for Cardiovascular Prevention, Department of Medicine, McGill University Health Centre-Royal Victoria Hospital, 1001 Boulevard Décarie, Montreal, Québec H4A 3J1, Canada; Section on Men’s Health, Aging and Metabolism, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA.
Selin Bilgic, Mike and Valeria Rosenbloom Centre for Cardiovascular Prevention, Department of Medicine, McGill University Health Centre-Royal Victoria Hospital, 1001 Boulevard Décarie, Montreal, Québec H4A 3J1, Canada.
George Thanassoulis, Mike and Valeria Rosenbloom Centre for Cardiovascular Prevention, Department of Medicine, McGill University Health Centre-Royal Victoria Hospital, 1001 Boulevard Décarie, Montreal, Québec H4A 3J1, Canada.
Michael J Pencina, Mike and Valeria Rosenbloom Centre for Cardiovascular Prevention, Department of Medicine, McGill University Health Centre-Royal Victoria Hospital, 1001 Boulevard Décarie, Montreal, Québec H4A 3J1, Canada; Department of Biostatistics and Bioinformatics, Duke University School of Medicine, DCRI, Durham, NC, USA.
Declarations
Disclosure of Interest
A.D.S. reports unrestricted grants from the Doggone Foundation and the Mike and Valeria Rosenbloom Foundation. L.D. has no conflict of interest. K.M.P. reports funding from the non-profit Doggone Foundation and past consulting from Cleerly Inc. S.B. has received a research scholarship for her MSc from the Mike and Valeria Rosenbloom Foundation. G.T. has participated in advisory boards or speaker bureaus for Amgen, Canadian Cardiovascular Society Dyslipidemia Guideline Group, Eli Lilly, HLS Therapeutics, New Amsterdam, Novartis, and Regeneron/Sanofi and has received grant funding from the Canadian Institutes of Health Research, Fonds de Recherche Quebec—Santé, and Heart & Research Foundation Canada. M.J.P. reports funding from the non-profit Doggone Foundation and a grant from NIH/NINDS, consulting fees from Eli Lilly, past consulting from Cleerly Inc., and past advisory board for Janssen.
Data Availability
The data analysed are from the UK Biobank, which is a publicly accessible databank.
Funding
This research was supported by an unconditional grant from the Doggone Foundation.
Ethical Approval
Ethical approval was obtained from the McGill University Health Centre Research Ethics Board.
Pre-registered Clinical Trial Number
None supplied.
References
- 1. Mach F, Baigent C, Catapano AL, Koskinas KC, Casula M, Badimon L, et al. 2019 ESC/EAS guidelines for the management of dyslipidaemias: lipid modification to reduce cardiovascular risk. Eur Heart J 2020;41:111–88. 10.1093/eurheartj/ehz455 [DOI] [PubMed] [Google Scholar]
- 2. Helgadottir A, Thorleifsson G, Snaebjarnarson A, Stefansdottir L, Sveinbjornsson G, Tragante V, et al. Cholesterol not particle concentration mediates the atherogenic risk conferred by apolipoprotein B particles—a Mendelian randomization analysis. Eur J Prev Cardiol 2022;29:2374–85. 10.1093/eurjpc/zwac219 [DOI] [PubMed] [Google Scholar]
- 3. Lawler PR, Akinkuolie AO, Ridker PM, Sniderman AD, Buring JE, Glynn RJ, et al. Discordance between circulating atherogenic cholesterol mass and lipoprotein particle concentration in relation to future coronary events in women. Clin Chem 2017;63:870–9. 10.1373/clinchem.2016.264515 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Wilkins JT, Li RC, Sniderman A, Chan C, Lloyd-Jones DM. Discordance between apolipoprotein B and LDL-cholesterol in young adults predicts coronary artery calcification: the CARDIA study. J Am Coll Cardiol 2016;67:193–201. 10.1016/j.jacc.2015.10.055 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Cao J, Nomura SO, Steffen BT, Guan W, Remaley AT, Karger AB, et al. Apolipoprotein B discordance with low-density lipoprotein cholesterol and non-high-density lipoprotein cholesterol in relation to coronary artery calcification in the Multi-Ethnic Study of Atherosclerosis (MESA). J Clin Lipidol 2020;14:109–121.e5. 10.1016/j.jacl.2019.11.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Johannesen CDL, Mortensen MB, Langsted A, Nordestgaard BG. Apolipoprotein B and non-HDL cholesterol better reflect residual risk than LDL cholesterol in statin-treated patients. J Am Coll Cardiol 2021;77:1439–50. 10.1016/j.jacc.2021.01.027 [DOI] [PubMed] [Google Scholar]
- 7. Welsh C, Celis-Morales CA, Brown R, Mackay DF, Lewsey J, Mark PB, et al. Comparison of conventional lipoprotein tests and apolipoproteins in the prediction of cardiovascular disease. Circulation 2019;140:542–52. 10.1161/CIRCULATIONAHA.119.041149 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Pencina MJ, D'Agostino RB, Zdrojewski T, Williams K, Thanassoulis G, Furberg CD, et al. Apolipoprotein B improves risk assessment of future coronary heart disease in the Framingham Heart Study beyond LDL-C and non-HDL-C. Eur J Prev Cardiol 2015;22:1321–7. 10.1177/2047487315569411 [DOI] [PubMed] [Google Scholar]
- 9. Su X, Cai X, Pan Y, Sun J, Jing J, Wang M, et al. Discordance of apolipoprotein B with low-density lipoprotein cholesterol or non-high density lipoprotein cholesterol and coronary atherosclerosis. Eur J Prev Cardiol 2022;29:2349–58. 10.1093/eurjpc/zwac223 [DOI] [PubMed] [Google Scholar]
- 10. Kim C-W, Hong S, Chang Y, Lee JA, Shin H, Ryu S. Discordance between apolipoprotein B and low-density lipoprotein cholesterol and progression of coronary artery calcification in middle age. Circ J 2021;85:900–7. 10.1253/circj.CJ-20-0692 [DOI] [PubMed] [Google Scholar]
- 11. Murphy A, Faria-Neto JR, Al-Rasadi K, Blom D, Catapano A, Cuevas A, et al. World Heart Federation Cholesterol Roadmap. Glob Heart 2017;12:179–197.e5. 10.1016/j.gheart.2017.03.002 [DOI] [PubMed] [Google Scholar]
- 12. Hagström E, Steg PG, Szarek M, Bhatt DL, Bittner VA, Danchin N, et al. Apolipoprotein B, residual cardiovascular risk after acute coronary syndrome, and effects of alirocumab. Circulation 2022;146:657–72. 10.1161/circulationaha.121.057807 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Marston NA, Giugliano RP, Melloni GEM, Park JG, Morrill V, Blazing MA, et al. Association of apolipoprotein B-containing lipoproteins and risk of myocardial infarction in individuals with and without atherosclerosis: distinguishing between particle concentration, type, and content. JAMA Cardiol 2022;7:250–6. 10.1001/jamacardio.2021.5083 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Pencina MJ, Navar AM, Wojdyla D, Sanchez RJ, Khan I, Elassal J, et al. Quantifying importance of major risk factors for coronary heart disease. Circulation 2019;139:1603–11. 10.1161/CIRCULATIONAHA.117.031855 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Sniderman AD, St-Pierre AC, Cantin B, Dagenais GR, Despres J-P, Lamarche B. Concordance/discordance between plasma apolipoprotein B levels and the cholesterol indexes of atherosclerotic risk. Am J Cardiol 2003;91:1173–7. 10.1016/s0002-9149(03)00262-5 [DOI] [PubMed] [Google Scholar]
- 16. Elliott P, Peakman TC; UK Biobank . The UK Biobank sample handling and storage protocol for the collection, processing and archiving of human blood and urine. Int J Epidemiol 2008;37:234–44. 10.1093/ije/dym276 [DOI] [PubMed] [Google Scholar]
- 17. Welsh P, Sattar N. To ApoB or not to ApoB: new arguments, but basis for widespread implementation remains elusive. Clin Chem 2023;69:3–5. 10.1093/clinchem/hvac183 [DOI] [PubMed] [Google Scholar]
- 18. Cromwell WC, Otvos JD, Keyes MJ, Pencina MJ, Sullivan L, Vasan RS, et al. LDL particle number and risk of future cardiovascular disease in the Framingham Offspring Study—implications for LDL management. J Clin Lipidol 2007;1:583–92. 10.1016/j.jacl.2007.10.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Otvos JD, Mora S, Shalaurova I, Greenland P, Mackey RH, Goff DC Jr. Clinical implications of discordance between low-density lipoprotein cholesterol and particle number. J Clin Lipidol 2011;5:105–13. 10.1016/j.jacl.2011.02.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Cole J, Otvos JD, Remaley AT. A translational tool to facilitate use of apolipoprotein B for clinical decision-making. Clin Chem 2023;69:41–7. 10.1093/clinchem/hvac161 [DOI] [PubMed] [Google Scholar]
- 21. Grundy SM, Stone NJ, Bailey AL, Beam C, Birtcher KK, Blumenthal RS, et al. 2018 AHA/ACC/AACVPR/AAPA/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA guideline on the management of blood cholesterol: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Circulation 2019;139:e1082–143. 10.1161/CIR.0000000000000625 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. De Marco D, Pencina K, Pencina M, Dufresne L, Thanassoulis G, Sniderman AD. Is hypertriglyceridemia a reliable indicator of cholesterol-depleted apoB particles? J Clin Lipidol 2023;17:452–7. 10.1016/j.jacl.2023.05.093 [DOI] [PubMed] [Google Scholar]
- 23. Nelson AJ, Sniderman AD, Ditmarsch M, Dicklin MR, Nicholls SJ, Davidson MH, et al. Cholesteryl ester transfer protein inhibition reduces major adverse cardiovascular events by lowering apolipoprotein B levels. Int J Mol Sci 2022;23:9417. 10.3390/ijms23169417 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Do R, Willer CJ, Schmidt EM, Sengupta S, Gao C, Peloso GM, et al. Common variants associated with plasma triglycerides and risk for coronary artery disease. Nat Genet 2013;45:1345–52. 10.1038/ng.2795 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Holmes MV, Asselbergs FW, Palmer TM, Drenos F, Lanktree MB, Nelson CP, et al. Mendelian randomization of blood lipids for coronary heart disease. Eur Heart J 2015;36:539–50. 10.1093/eurheartj/eht571 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Kohli-Lynch CN, Thanassoulis G, Moran AE, Sniderman AD. The clinical utility of apoB versus LDL-C/non-HDL-C. Clin Chim Acta 2020;508:103–8. 10.1016/j.cca.2020.05.001 [DOI] [PubMed] [Google Scholar]
- 27. Thanassoulis G, Williams K, Ye K, Brook R, Couture P, Lawler PR, et al. Relations of change in plasma levels of LDL-C, non-HDL-C and apoB with risk reduction from statin therapy: a meta-analysis of randomized trials. J Am Heart Assoc 2014;3:e000759. 10.1161/JAHA.113.000759 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Sniderman AD, Williams K, Contois JH, Monroe HM, McQueen MJ, de Graaf J, et al. A meta-analysis of low-density lipoprotein cholesterol, non-high-density lipoprotein cholesterol, and apolipoprotein B as markers of cardiovascular risk. Circ Cardiovasc Qual Outcomes 2011;4:337–45. 10.1161/CIRCOUTCOMES.110.959247 [DOI] [PubMed] [Google Scholar]
- 29. Langlois MR, Chapman MJ, Cobbaert C, Mora S, Remaley AT, Ros E, et al. Quantifying atherogenic lipoproteins: current and future challenges in the era of personalized medicine and very low concentrations of LDL cholesterol. A consensus statement from EAS and EFLM. Clin Chem 2018;64:1006–33. 10.1373/clinchem.2018.287037 [DOI] [PubMed] [Google Scholar]
- 30. Contois JH, McConnell JP, Sethi AA, Csako G, Devaraj S, Hoefner DM, et al. Apolipoprotein B and cardiovascular disease risk: position statement from the AACC Lipoproteins and Vascular Diseases Division Working Group on Best Practices. Clin Chem 2009;55:407–19. 10.1373/clinchem.2008.118356 [DOI] [PubMed] [Google Scholar]
- 31. Langlois MR, Nordestgaard BG, Langsted A, Chapman MJ, Aakre KM, Baum H, et al. Quantifying atherogenic lipoproteins for lipid-lowering strategies: consensus-based recommendations from EAS and EFLM. Clin Chem Lab Med 2020;58:496–517. 10.1515/cclm-2019-1253 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data analysed are from the UK Biobank, which is a publicly accessible databank.