Association of LDL Cholesterol, Non–HDL Cholesterol, and Apolipoprotein B Levels With Risk of Cardiovascular Events Among Patients Treated With Statins:  A Meta-analysis FREE

Quiz Ref IDStatin therapy is the cornerstone of pharmacological therapy for the primary and secondary prevention of cardiovascular disease. All currently available guidelines state that low-density lipoprotein cholesterol (LDL-C) levels should be used as the primary target to initiate and titrate lipid-lowering therapy.^1- 2 However, trials investigating the efficacy of statin therapy have shown that the cardiovascular benefits of statins may go beyond their influence on LDL-C levels. Thus, LDL-C may not be the best lipid parameter to predict cardiovascular risk or to quantify the atheroprotective effect of statin therapy.³Quiz Ref IDExperimental evidence supports a more important role for apolipoprotein B (apoB) and apoB-containing lipoproteins than for LDL-C content in mediating atherogenesis. Lipoproteins containing apoB must first enter the arterial wall and undergo oxidative modification before they can contribute to atherogenesis. This modification affects the structure of the apoB molecule or the phospholipid membrane of these lipoproteins, yielding ligands for the scavenger receptors of macrophages in the arterial wall.⁴ Subsequently, cholesterol accumulation and crystallization in macrophage cytoplasm leads to the formation of foam cells and progression to atherosclerotic plaque.⁵

Several alternative lipid and apolipoprotein parameters have been proposed as alternatives for LDL-C, most prominently apoB and non–high-density lipoprotein cholesterol (non–HDL-C), both markers of the concentration of atherogenic apoB-containing lipoproteins. Several population-based studies have shown that these 2 parameters are more strongly associated with cardiovascular risk than LDL-C.^6- 9 Similar observations have been reported in some statin trials, although less extensively.^10- 12 Therefore, the role of lipid or apolipoprotein parameters other than LDL-C as targets for statin therapy remains debated and clinically relevant.

It was the objective of this meta-analysis to assess whether among statin-treated patients, non–HDL-C and apoB were more strongly associated with the risk of future cardiovascular events than LDL-C. The secondary objective was to explore whether non–HDL-C and apoB explained a larger proportion of the atheroprotective effects of statin therapy than LDL-C. These objectives were addressed by performing a meta-analysis of individual patient data from large randomized controlled statin trials.

The literature was searched to identify all randomized controlled trials (RCTs) that assigned study participants in at least 1 of the study groups to statin therapy. Trials in which total cholesterol, LDL-C, HDL-C, triglycerides, and apolipoproteins were measured at baseline and during statin therapy in the entire study population were selected. Trials with a mean follow-up for cardiovascular events shorter than 2 years and those including fewer than 1000 participants were excluded.

The literature search was undertaken in PubMed using the following search terms: statin, hydroxymethylglutaryl coenzyme A reductase inhibitor, simvastatin, lovastatin, fluvastatin, pravastatin, atorvastatin, rosuvastatin, cholesterol, apolipoprotein, coronary heart disease, coronary artery disease and cardiovascular disease. The results were limited to randomized trials and English language. The first search was performed on January 4, 2009, and the updated search until December 31, 2011, was performed on January 20, 2012. Two authors (B.J.A. and S.M.B.) independently screened all abstracts for randomized controlled trials fulfilling the inclusion criteria. If an abstract described a subanalysis of a potentially relevant trial, the original publication was traced. Results were compared and inconsistencies were resolved by consensus. Investigators were contacted and asked to provide individual patient data. The requested patient characteristics included sex, age, smoking status, body mass index, diabetes mellitus, systolic and diastolic blood pressure, fasting glucose, total cholesterol, LDL-C, HDL-C, triglycerides, and apolipoprotein A-I and B at baseline and at 1-year follow-up, study medication, and patient history of stable coronary artery disease, myocardial infarction, percutaneous coronary intervention, or coronary artery bypass grafting. The following outcomes (and time to event) were also collected: fatal and nonfatal myocardial infarction, fatal other coronary artery disease, hospitalization for unstable angina, fatal and nonfatal stroke, peripheral artery disease, and congestive heart failure. Data were harmonized into a pooled database and this database was independently validated against the original files. Quality of the included trials was assessed by the Delphi score.¹³ This meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, and a checklist was provided at the time of manuscript submission.¹⁴

Lipid and apolipoprotein levels at baseline and at 1-year follow-up were obtained from the participating trials. For on-statin measurements, the 1-year time point was chosen because this was the first uniform time point when apolipoproteins were measured in all participating trials. Quiz Ref IDNon–HDL-C was calculated as total minus HDL-C. Cholesterol levels reported in mmol/L were converted to mg/dL by multiplying by 38.7, and triglycerides levels reported in mmol/L were converted to mg/dL by multiplying by 88.5. The primary disease outcome of this meta-analysis was time to first major cardiovascular event, which was defined as fatal or nonfatal myocardial infarction, fatal other coronary artery disease, hospitalization for unstable angina, and fatal or nonfatal stroke. Additional analyses were performed for the prediction of time to first major coronary events (fatal or nonfatal myocardial infarction, fatal other coronary artery disease, and hospitalization for unstable angina) and time to major cerebrovascular event (fatal or nonfatal stroke).

Baseline characteristics, levels of lipids and apolipoproteins at baseline and on-trial, as well as absolute changes and percentage changes between on-trial and baseline levels were calculated for each trial separately. The risk of major cardiovascular events, major coronary events, and major cerebrovascular events associated with on-statin lipid and apolipoprotein levels was quantified using Cox proportional hazards models. Study participants allocated to placebo were excluded. Analyses were performed for LDL-C, non–HDL-C, and apoB separately. Hazard ratios (HRs) and corresponding 95% CIs were calculated per 1-SD increase of the respective on-statin lipid or apolipoprotein; this was the principal outcome on which conclusions were based.

Similar analyses were performed for patients in the top quartile of the on-statin lipid or apolipoprotein distribution, using those in the bottom quartile as the reference standard. A trend test was used to test for linear trend across quartiles. Analyses were adjusted for sex, age, smoking (current vs not), diabetes mellitus, systolic blood pressure, and trial. Analyses were not additionally adjusted for prevalent coronary artery disease since all trials enrolled either 0% or 100% patients with prevalent disease, so adjustment for trial implies adjustment for prevalent coronary artery disease.

Analyses were performed for the outcomes major cardiovascular events, major coronary events, and major cerebrovascular events. Because the Cox proportional hazards models were based on identical populations and included LDL-C, non–HDL-C, or apoB in addition to identical covariates, they were highly correlated and not independent. Therefore the approach to test for a significant difference between 2 models should be to assess whether the 95% CI of the difference between 2 models excludes zero. Bootstrapping was therefore performed on the difference between log-transformed HR of the adjusted Cox regression models (both by 1-SD increase and for top vs bottom quartile) for non–HDL-C vs LDL-C, non–HDL-C vs apoB, and apoB vs LDL-C. This approach yields a mean difference and its 95% CI, and thus an estimate of the statistical significance of this difference. HRs were also calculated for 4 categories of statin-treated patients based on whether or not they reached the LDL-C target 100 mg/dL and the non–HDL-C target 130 mg/dL. Forest plots were constructed and interaction terms were calculated to assess whether adjusted HRs per 1-SD increase of each lipid or apolipoprotein parameter differed significantly between subgroups based on baseline characteristics, lipid levels, or trial.

The proportion of the treatment effect of the statin intervention (either statin vs placebo or high-dose vs usual-dose statin) that was explained by LDL-C, non–HDL-C, or apoB was calculated as previously described.¹² Bootstrapping was performed on the difference between the proportions of treatment effect explained by non–HDL-C vs LDL-C, non–HDL-C vs apoB, and apoB vs LDL-C.

Statistical heterogeneity across studies was quantified using the Cochran Q statistic and I² statistic. The I² statistic is derived from the Q statistic ([Q − df /Q ]*100) and provides a measure of the proportion of the overall variation attributable to between-study heterogeneity.¹⁵ The potential for publication bias was addressed by drawing funnel plots and visual assessment. Proportionality of hazards over time was graphically checked by plotting the cumulative hazards over time for all quartiles against each other. This was done for LDL-C, non–HDL-C, and apoB separately. A P value of less than .05 was considered statistically significant. Bootstrapping was performed in R. All other statistical analyses were performed using SPSS (version 17.0).

The literature search yielded a total of 1643 potentially relevant abstracts (eFigure). The large majority of articles were excluded because they were not primary reports on an RCT fulfilling the inclusion criteria.

We identified 28 potentially relevant RCTs. A total of 17 were excluded because apolipoprotein levels were not measured. One RCT was excluded because apolipoprotein levels were not measured during statin therapy, and 2 additional RCTs were excluded because apolipoproteins were measured in only a subset. A total of 8 trials fulfilled all inclusion criteria: the Scandinavian Simvastatin Survival Study (4S),¹⁶ the Air Force/Texas Coronary Atherosclerosis Prevention Study (AFCAPS-TexCAPS),¹⁷ the Long-term Intervention with Pravastatin in Ischaemic Disease (LIPID) trial,¹⁸ the Collaborative Atorvastatin Diabetes Study (CARDS),¹⁹ the Treating to New Targets (TNT) trial,²⁰ the Incremental Decrease in Endpoints through Aggressive Lipid Lowering (IDEAL) trial,²¹ the Stroke Prevention by Aggressive Reduction in Cholesterol Levels (SPARCL) trial,²² and the Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER).²³ Individual patient data were obtained from all 8 trials. Study characteristics of these 8 trials are presented in eTable 1.

Trials were of high quality with a median Delphi score of 9 (range 6-9). Heterogeneity between trials with regard to the association with risk of major cardiovascular events was low for LDL-C (Q  = 6.94, P = .43, I² = 0), non–HDL-C (Q  = 6.05, P = .53, I² = 0) and apoB (Q  = 9.55, P = .22, I² = 26%). Visual assessment of funnel plots suggested bias, although it is unlikely that this was publication bias since the primary objective of our meta-analysis was different from the primary objective of the included trials (statin efficacy). Bias was based on the fact that the associations between all 3 lipid parameters and risk of cardiovascular events were stronger in 2 trials contributing low numbers of events (AFCAPS-TexCAPS, JUPITER), which may have been related to their inclusion criteria. The proportionality assumptions were satisfied.

Data were available from 62 154 study participants of the 8 included trials. Baseline characteristics are presented in eTable 2. Levels of lipids and apolipoproteins at baseline and at 1-year on trial, as well as the absolute changes and percentage changes between baseline and on-trial levels for each trial separately are presented in eTable 3. A total of 38 153 study participants were randomized to a statin group and had a complete set of lipid and apolipoprotein levels during statin treatment available. Among these individuals, a total of 158 (0.4%) developed a fatal myocardial infarction and 1678 (4.4%) developed a non-fatal myocardial infarction during follow-up. Fatal other coronary artery disease occurred in 615 study participants (1.6%) and fatal or nonfatal stroke occurred in 1029 study participants (2.7%). A total of 2806 participants (7.4%) were hospitalized for unstable angina. A total of 6286 major cardiovascular events occurred in 5387 study participants (event rate 14.1%). Of these, 4577 experienced 1 event, 728 experienced 2 events, 75 experienced 3 events, and 7 experienced 4 events.

All studied lipid and apolipoprotein parameters were statistically significantly associated with the risk of major cardiovascular events. The HRs per 1-SD increase in LDL-C (32 mg/dL), non–HDL-C (36 mg/dL), and apoB (27 mg/dL) were 1.13 (95% CI, 1.10-1.17), 1.16 (95% CI, 1.12-1.19) and 1.14 (95% CI, 1.11-1.18), respectively (Table 1). Patients in the top quartile of the LDL-C distribution had an HR of 1.26 (95% CI, 1.14-1.39) compared with those in the bottom quartile. Associations were stronger for non–HDL-C (HR, 1.42; 95% CI, 1.29-1.56) and for apoB (HR, 1.33; 95% CI, 1.22-1.45) (P for linear trend across quartiles <.001 for each). Bootstrap analyses showed that the difference between LDL-C and non–HDL-C in predicting the risk of major cardiovascular events per 1-SD increase was statistically significant (P = .002). The difference between non–HDL-C and apoB was also statistically significant (P = .02), whereas the difference between LDL-C and apoB was not statistically significant (P = .21). Analyses for the major coronary events outcome showed similar associations as for major cardiovascular events (eTable 4). Associations with risk of major cerebrovascular events were not as strong.

Subgroup analyses showed that the associations between lipid or apolipoprotein parameters and risk of major cardiovascular events did not have statistically significant difference between primary and secondary prevention trials (P for interaction >.78 for each; Figure 1). The strength of the associations with LDL-C, non–HDL-C, or apoB also did not differ between clinically relevant subgroups, including those with vs without diabetes mellitus, hypertriglyceridemia, and low HDL-C (P for interaction >.10 for each). There was also no overall statistically significant difference between the individual trials (P for interaction >.26 for each; Figure 2).

In Figure 3, HRs for the risk of major cardiovascular events are presented for 4 patient categories divided by the LDL-C target of 100 mg/dL and the non–HDL-C target of 130 mg/dL. Statin-treated patients reaching the non–HDL-C target but not the LDL-C target had an HR of 1.01 (95% CI, 0.92-1.12; P = .85) compared with those reaching both targets. Patients reaching the LDL-C target but not the non–HDL-C target had an HR of 1.32 (95% CI, 1.17-1.50; P < .001).

The proportions of the treatment effect of the respective trial interventions (either statin vs placebo or high-dose vs moderate-dose statin therapy) that are explained by changes in LDL-C, non–HDL-C, and apoB are shown in Table 2. Statin-induced changes in LDL-C levels explained 50% of the effect of statin treatment, whereas non–HDL-C explained 64% and apoB explained 54%. The proportion of treatment effect explained by non–HDL-C was larger than by LDL-C (P < .001) and by apoB (P = .007). The proportion of treatment effect explained did not differ between LDL-C and apoB (P = .44).

Quiz Ref IDWe performed a meta-analysis to determine the association between on-statin LDL-C, non–HDL-C, and apoB and risk of future cardiovascular events. The main finding of this study is that among statin-treated patients, non–HDL-C had a stronger association with risk of major cardiovascular events than LDL-C and apoB. Changes in non–HDL-C also explained a larger proportion of the atheroprotective effect of statin intervention than did LDL-C and apoB.

The predictive value of different lipid and apolipoprotein parameters has been a topic of debate for decades. As early as 1980, Sniderman et al²⁴ suggested that apoB levels were more closely associated with the presence of coronary atherosclerosis compared with either LDL-C levels or triglyceride levels. The strong predictive value of apoB was confirmed in the apolipoprotein-related mortality risk (AMORIS) study,²⁵ which included more than 175 000 individuals. Although LDL-C was not measured by established methods in AMORIS, it was shown to correlate well with LDL-C as quantified by conventional methods in a subgroup. The INTERHEART study²⁶ has shown that the association between deciles of apoB and risk of first myocardial infarction was stronger than for non–HDL-C, but these 2 parameters were not compared with LDL-C. Another important limitation of INTERHEART resides in its cross-sectional design. A recent large-scale meta-analysis on individual patient data from 68 prospective studies showed that the HRs for risk of coronary artery disease were similar for non–HDL-C and apoB.⁹ These parameters were not compared with LDL-C as calculated by the Friedewald formula, as is commonly used in prospective studies and in clinical practice. More recently, another meta-analysis, which used study level data and a different set of inclusion criteria, addressed the same question. This meta-analysis concluded that apoB was superior to non–HDL-C and that both were superior to LDL-C as a predictor of cardiovascular risk.²⁷

In contrast to the extensive literature on the predictive value of lipids and apolipoproteins in population studies, the body of evidence on statin-treated patients is more limited. Investigators of the AFCAPS-TexCAPS trial¹⁰ were the first to report that among patients treated with statin therapy, apoB may be a more accurate predictor of risk of major cardiovascular events than LDL-C. Our results suggest that the association in AFCAPS-TexCAPS was somewhat stronger than in the other participating trials (although not significantly), which may have been caused by the fact that the inclusion criteria of AFCAPS-TexCAPS differed substantially from those of other trials. A subanalysis of the LIPID trial¹² confirmed that apoB was more strongly associated with cardiovascular risk than LDL-C, but non–HDL-C was not evaluated in that analysis. A combined analysis of the TNT and IDEAL trials¹¹ also confirmed a stronger predictive value of both apoB and non–HDL-C compared with LDL-C. A subanalysis of CARDS also reported that LDL-C was inferior to apoB and non–HDL-C but concluded that apoB was the most consistent goal for statin therapy.²⁸ In JUPITER,²⁹ on-treatment concentrations of non–HDL-C and apoB were comparable with LDL-C in the prediction of residual risk. In summary, our observations in statin-treated patients extend prior results from large population-based studies showing that non–HDL-C is more strongly associated with risk of future major cardiovascular events than LDL-C, but we did not find evidence that apoB performed better than non–HDL-C or LDL-C. We found no evidence that the predictive power of these 3 parameters differed between subgroups based on other characteristics such as diabetes mellitus or hypertriglyceridemia.

In participating trials, HDL-C was measured in the supernatant after heparin-manganese precipitation of apoB-containing lipoproteins according to the protocol of the Centers for Disease Control and Prevention.³⁰ This is important to recognize in extrapolating the results of our meta-analysis to clinical practice because many clinical laboratories do not currently use this method to measure HDL-C.³¹ Rather, they use homogeneous methods that can be fully automated. Although the majority of these assays perform reasonably well and generally meet the National Cholesterol Education Program criteria for precision, accuracy, and total error, some assays show discrepant results compared with reference methods. This means that the use of HDL-C as measured in clinical practice could result in the calculation of non–HDL-C, which may not provide an index of cardiovascular risk as accurate as that in clinical trials. Nevertheless, given the fact that HDL-C is used to calculate LDL-C as well, these limitations similarly apply to LDL-C. Quiz Ref IDConversely, the standardized methods for the measurement of apoB used in statin trials are similar to those that can be used in clinical laboratories. Implementation of apoB assays in routine clinical practice is suboptimal but this is not due to the quality of the assays but due to the limited attention given in clinical guidelines. Thus, the final rating on the comparative accuracy of the methods used to measure apoB and non–HDL-C in clinical practice may therefore depend on other information. Such a conclusion should consider not just biological and statistical evidence, but also financial, logistic, and feasibility arguments.

Several aspects of the design of this meta-analysis warrant comment. An important strength is the availability of individual patient data, which enabled individual-level patient analyses—a more appropriate and accurate option than study-level analyses. A limitation is the fact that the participating trials had different inclusion criteria. The different distributions across trials of baseline characteristics, and lipid levels in particular, may have affected the results of our meta-analysis. In addition, outcome definitions may have differed slightly between trials. Subgroup analyses, however, showed that the associations did not differ significantly between trials. It should also be noted that the results are based on patients included in trials and that these results cannot necessarily be extrapolated to patients in routine clinical practice. Another limitation is the use of on-statin lipid and apolipoprotein levels measured at 1-year follow-up. This time point was chosen because it was the first uniform time point when lipids and apolipoproteins were measured in all participating trials. Therefore, fatal cardiovascular events occurring in the first year of therapy are not accounted for in this analysis.

In conclusion, among statin-treated patients, levels of LDL-C, non–HDL-C, and apoB were each strongly associated with the risk of major cardiovascular events, but non–HDL-C was more strongly associated than LDL-C and apoB. Given the fact that many other arguments for the clinical applicability of non–HDL-C and LDL-C are identical, non–HDL-C may be a more appropriate target for statin therapy than LDL-C.