Introduction

Patients with cardiovascular disease (CVD), the most common form being coronary heart disease (CHD), are at high risk of events such as myocardial infarction, angina, and stroke. Risk status may be defined according to the American College of Cardiology (ACC), the American Heart Association (AHA), and the European Society of Cardiology (ESC) which consider patients with CHD or CHD risk equivalents as high risk [1, 2]. To reduce lipid levels, these high-risk patients are commonly managed with statins as a first-line treatment [1,2,3,4,5]. Many of these patients, however, do not reach low-density lipoprotein cholesterol (LDL-C) goals with statin monotherapy alone and may require titration to higher statin doses, switching to a more potent statin, or use of combination lipid-lowering therapy, such as ezetimibe and PCSK9 inhibitors [1, 3,4,5,6,7,8,9,10,11].

To date, no single head-to-head randomized controlled trial (RCT) has been conducted to evaluate additional treatment options in high-risk patients on existing statin therapy. When RCT evidence is available comparing subsets of treatments within the larger patient population framework, it is possible to propose a network of connected evidence. The results of these trials can then be synthesized by means of a network meta-analysis (NMA). Conceptually, the NMA pools the results of trials on a single intervention and uses multiple pairwise comparisons to estimate the relative treatment effects of all interventions included in the network of evidence. Thus, relative efficacy can be estimated between interventions for which no head-to-head RCT evidence currently exists.

Previous systematic reviews and meta-analyses have also analyzed the efficacy of ezetimibe as a supplementary treatment to statin monotherapy. A 2007 meta-analysis of five studies compared ezetimibe as an addition to a statin versus placebo in addition to a statin; results suggested that the addition of ezetimibe to ongoing statin therapy lowered LDL-C significantly more in patients who were not at LDL-C goal on previous treatment compared with those adhering only to statin monotherapy [12]. A 2011 systematic review and meta-analysis of 13 studies analyzed the LDL-C reduction of ezetimibe in combination with a statin versus doubling of statin dose; these results also suggested that ezetimibe as an add-on treatment was significantly more effective in lipid lowering than doubling of statin [13]. While a 2012 analysis looked at ezetimibe plus statin versus different statin monotherapies, including switching to another statin, it was a pooled analysis designed to assess the factors that might affect a patient’s response to lipid-altering therapy, rather than a systematic review and meta-analysis [14]. While this analysis suggested that patient characteristics had a limited influence on the lowering of LDL-C, it also did not examine clinical outcomes.

The LDL-C-lowering efficacy of ezetimibe in combination with statin therapy has not yet been reviewed simultaneously alongside two different statin monotherapies (doubling statin dose or switching to higher-potency statin). The purpose of this study, therefore, was to evaluate, through an NMA, the efficacy of adding ezetimibe to existing simvastatin, atorvastatin, or rosuvastatin therapy compared to doubling the statin dose or switching to a higher-potency dose of another statin in patients with hypocholesterolemia and at high risk of CVD. The evidence was based on RCTs identified by means of a systematic literature review.

Methods

Literature search and eligibility criteria

MEDLINE, EMBASE, and the Cochrane Register of Controlled Trials (CENTRAL) were systematically searched in May 2014 to identify RCTs evaluating changes in LDL-C and total cholesterol in patients with hypercholesterolemia and high or very high CVD risk, previously treated with atorvastatin, rosuvastatin, or simvastatin monotherapy. Studies evaluating combination ezetimibe and statin therapy, with either simvastatin, atorvastatin, or rosuvastatin, were considered eligible for inclusion. Placebo was included as a valid comparator. Only studies published in English after 1990 were considered. We excluded crossover or titration studies if outcomes were only reported at study end, but included these studies if outcomes were reported at the time of titration or crossover. Search strategies are presented in Appendix 1.

Data extraction and outcomes

Two researchers, independently and in duplicate, screened all abstracts to ascertain whether they met predefined inclusion criteria. Abstracts that either met inclusion criteria or for which decisions were unclear were evaluated at the full text level.

From the final set of included studies, details on study design, patient baseline characteristics (e.g., age, sex, weight, comorbidities), interventions, and percent change from baseline (CFB) in LDL-C and total cholesterol were extracted. Relative to patients’ previous treatment, either pre-enrollment or as part of a trial run-in period, trial arms were classified as maintaining, increasing, switching, or adding new treatments. Potency relationships between statins were classified according to the clinical guidelines published by the National Institute for Clinical Excellence (NICE), presented in Appendix 2 [15]. The primary outcome of interest was the percent CFB in LDL-C, with CFB in total cholesterol explored as a secondary outcome. Safety outcomes were not considered in the current analysis.

Statistical analyses

A Bayesian NMA was used to synthesize the results of the included studies, with the purpose of estimating the relative treatment effects for each intervention represented in the evidence base with respect to CFB in LDL-C and total cholesterol. This statistical approach combines both direct and indirect evidence to estimate the relative treatment effects between each intervention in the network, while weighting trials according to sample size [16]. Inconsistency between direct and indirect evidence in the network was evaluated using edge splitting, with three-sided loops evaluated according to the Bucher test [17]. To estimate these effects for the continuous outcomes of interest, a model with a normal likelihood and identity link was used.

To not influence the observed results by the prior distribution, non-informative priors were used (\({d}_{AK}~\text{normal} \left(\text{0,10000}\right)\)for (pooled) treatment effects with NMA models, and \({d}_{XY}~\text{normal} \left(\text{0,10000}\right)\) for (pooled) treatment effects with independent-means models; \(\sigma ~\text{uniform} \left(0,u\right)\) for between-study heterogeneity with u set at five times the range of observed treatment effects across studies included in the NMA, and \({\mu }_{jb}~\text{normal} \left(\text{0,10000}\right)\) for nuisance parameters of the models).

Both fixed and random effects models were fitted to the data using Bayesian Markov chain Monte Carlo methods and were evaluated according to the deviance information criteria (DIC), a measure of model fit [18]. A more complex model will generally be a better fit to the data and will result in a smaller residual deviance. Thus, the model with the lowest DIC is preferred.

All analyses were performed using R version 3.0.3 (http://www.r-project.org/) and OpenBUGS version 3.2.3 (OpenBUGS Project Management Group).

For each outcome, we presented mean differences and 95% credible intervals (CrI) from the posterior distribution of relative treatment effects for all interventions relative to each other in a cross table. The use of non-informative priors allowed the 95% CrI to be interpreted similarly to the 95% confidence interval of the frequentist framework. Modeled outcomes, which apply the relative treatment effect for each intervention to an anchor comparator, were also presented, where each bar represents the estimated CFB and whiskers represent the corresponding 95% CrI. As each CrI reflects both uncertainty in the estimation of the effect of the anchor comparator as well as the relative treatment effect, these figures should not be used for comparative purposes. For all analyses, the anchor comparator was doubling the existing statin therapy.

Results

Study identification and selection

The systematic search of the clinical literature databases returned 2,960 citations. After duplicate abstract screening, 691 citations were selected for full text review. The final review resulted in 37 publications [19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48] being selected, describing 35 RCTs. The flow of information through the screening process is shown in Fig. 1. An overview of the baseline patient characteristics of the included studies is presented in Appendix 3.

Fig. 1
figure 1

Flow of information

Full size image

Evidence base and results of NMA, by population

Simvastatin-experienced patients

Thirteen RCTs [19,20,21,22,23,24,25,26,27,28,29,30,31] were identified evaluating outcomes in patients previously treated with simvastatin therapy. Most of the included trials specified treatment durations of 6 or 8 weeks, though two trials [20, 23] followed patients for 16 weeks. Trial inclusion criteria differed with respect to minimum required LDL-C levels as well as CHD risk requirements, with many trials specifying documented CHD and others specifying a presence of risk factors for developing CHD, such as diabetes or hypertension. Patients with liver disease, renal disease, or recent cardiovascular events were excluded in most trials. Patient baseline characteristics were consistent across trials with respect to mean age and BMI, though some variation was observed in the proportions of males, smokers, and patients with hypertension. Baseline LDL-C varied from 91 to 169 mg/dL and total cholesterol varied from 154 to 253 mg/dL.

All 13 RCTs (N = 4535) reported CFB outcomes for LDL-C (Fig. 2a). The results of the NMA are presented in Table 1 and modeled outcomes are presented in Fig. 3a. All interventions were more effective in lowering LDL-C relative to maintaining the baseline simvastatin dose. The addition of ezetimibe to simvastatin was significantly more efficacious than doubling the simvastatin dose (MD − 13.62%, CrI: − 19.99, − 6.91) or switching to a higher-potency dose of rosuvastatin (MD − 12.03%, CrI: − 19.37, − 4.73). There was no statistical difference between the addition of ezetimibe and a quadruple dose of the base simvastatin or switching to a higher-potency dose of atorvastatin.

Fig. 2
figure 2

Network diagrams of evidence for change from baseline in LDL-C. a Simvastatin experienced; b Atorvastatin experienced; c Rosuvastatin experienced. HP higher potency; LP lower potency; EP equal potency; SIM simvastatin; RO rosuvastatin; AT atorvastatin; EZ ezetimibe. Doses in brackets indicate the baseline or run-in statin dose

Full size image
Table 1 Mean differences in mean percent change from baseline in LDL-C among simvastatin-experienced patients from random-effects network meta-analysis
Full size table
Fig. 3
figure 3

Modeled outcomes for change from baseline in LDL-C. HP higher potency, LP lower potency, EP equal potency, SIM simvastatin, RO rosuvastatin, AT atorvastatin, EZ ezetimibe

Full size image

Changes in total cholesterol were reported by 11 RCTs (N = 3458), as presented in Appendix 4. All interventions, with the exception of doubling simvastatin, significantly lowered total cholesterol relative to maintaining the baseline dose. The addition of ezetimibe lowered total cholesterol relative to doubling the base simvastatin dose (MD − 8.43%, CrI − 16.01, − 0.73) and switching to a higher-potency dose of rosuvastatin (MD − 8.79%, CrI − 17.08, − 0.84). No statistically important difference was found between the addition of ezetimibe and a quadruple dose of simvastatin or switching to a higher-potency dose of atorvastatin.

Atorvastatin-experienced patients

We identified 16 RCTs reporting outcomes for patients previously treated with atorvastatin [20, 25, 27, 29, 32,33,34,35,36,37,38,39,40,41,42,43,44,45]. One trial, however, did not include any study arms that could be incorporated into the network of evidence [32]. There was some variation in the baseline dose of atorvastatin, though the majority of trials included patients previously treated with 10 mg of atorvastatin. Most studies specified treatment durations between 6 and 12 weeks, though a single study only followed patients for 4 weeks [43]. There was inconsistency with respect to exclusion criteria for patients with liver disease, kidney disease, or diabetes. Baseline characteristics were comparable across study arms with respect to mean age and BMI, though there was variation with respect to the proportion of males, smokers, and mean LDL-C (82–187 mg/dL) and mean total cholesterol (160–264 mg/dL) at enrollment.

Sixteen RCTs (N = 7167) reported mean CFB in LDL-C in atorvastatin-experienced patients (Fig. 2b). The results of the NMA are presented in Table 2, with modeled outcomes available in Fig. 3b. All interventions significantly lowered LDL-C relative to maintaining the baseline atorvastatin dose. The addition of ezetimibe to atorvastatin significantly lowered LDL-C relative to doubling the base atorvastatin dose (MD − 14.71%, CrI − 16.46, − 12.95) or switching to an equal- or higher-potency dose of rosuvastatin (MD − 15.78%, CrI − 19.21, − 12.45). No statistically meaningful differences were found between the fixed-dose and loose combination of simvastatin and ezetimibe. Adding ezetimibe to atorvastatin produced similar reductions in LDL-C as switching to lower-potency simvastatin and adding ezetimibe.

Table 2 Mean differences in mean percent change from baseline in LDL-C among atorvastatin-experienced patients from random-effects network meta-analysis
Full size table

Changes in total cholesterol were reported by 14 RCTs (N = 5775), as presented in Appendix 4. All interventions, with the exception of switching to equal potency rosuvastatin, lowered total cholesterol relative to maintaining the baseline atorvastatin dose. The addition of ezetimibe was statistically more efficacious in lowering total cholesterol than either doubling the atorvastatin dose (MD − 9.41%, CrI: − 10.89, − 7.92) or switching to a higher-potency dose of rosuvastatin (MD − 11.61%, CrI: − 15.19, − 7.88).

Rosuvastatin-experienced patients

Five trials investigated patients currently treated with rosuvastatin [34, 42, 46,47,48]. Four trials reported patient outcomes after 12 weeks, with a single trial [46] reporting outcomes after 6 weeks. Four trials specified minimum LDL-C levels of 100 mg/dL, with one study each specifying minimums of 70 mg/dL [48] and 80 mg/dL [47]. These trials were well matched in terms of mean age, BMI, and proportion of males, though there was more variability in the proportion of patients with diabetes. Baseline lipid levels were relatively consistent across trials, though baseline mean LDL-C was below 100 mg/dL in one trial [48].

All five RCTs (N = 1074) were included in the network of evidence for LDL-C (Fig. 2c). The results of the fixed-effects NMA are presented in Table 3, with modeled outcomes in Fig. 3c. A fixed-effects model was applied as there were too few studies to inform the estimate of between-study heterogeneity. No study included a treatment arm where patients maintained their baseline rosuvastatin dose. All interventions significantly lowered LDL-C relative to doubling the base rosuvastatin dose. The addition of ezetimibe to rosuvastatin significantly lowered LDL-C relative to doubling the base rosuvastatin dose (MD − 14.96%, CrI: − 17.79, − 12.11). The addition of ezetimibe was not statistically different from switching to a quadruple dose of base rosuvastatin.

Table 3 Mean differences in mean percent change from baseline in LDL-C among rosuvastatin-experienced patients from fixed-effects network meta-analysis
Full size table

Four RCTs (N = 999) described changes in the mean percent CFB in total cholesterol, as presented in Appendix 4. The addition of ezetimibe to either the base dose of rosuvastatin or to a switch to an equipotent dose of atorvastatin with ezetimibe was found to significantly lower total cholesterol relative to doubling the base rosuvastatin dose. No statistically important difference was observed between the addition of ezetimibe and switching to a quadruple dose of rosuvastatin.

Discussion

The purpose of this study was to evaluate the relative efficacy of adding ezetimibe to current statin treatment, in terms of lowering LDL-C and total cholesterol, compared to doubling the statin dose or switching to a higher-potency dose of another statin in patients previously treated with simvastatin, atorvastatin, or rosuvastatin. This objective was addressed through an NMA, which simultaneously estimated the relative treatment effect for all interventions in a network of evidence. Treatment regimens were classified according to their relationship (switch, augmentation, equipotency) to prior statin using the NICE statin conversion table; it should be noted that the FDA uses a different conversion table. In all patient populations, the addition of ezetimibe resulted in a statistically larger reduction in LDL-C than doubling the prior statin dose. Among patients on simvastatin or atorvastatin, adding ezetimibe was more efficacious than switching to higher-potency rosuvastatin. In these populations, the addition of ezetimibe was similarly efficacious as quadrupling the statin dose; however, this regimen is not seen as an appropriate option for many patients. Similar results were found with respect to lowering of total cholesterol levels. It should be noted that in the simvastatin and atorvastatin populations, the percent reduction in LDL-C for doubling the prior statin was slightly larger than that predicted by the “rule of 6” [49]; for patients on simvastatin and atorvastatin, the reduction was 9.5% and 9.8%, respectively, although the CrI included 6% in the simvastatin population, so this result is still statistically consistent with the “rule of 6”. The Cholesterol Treatment Trialists (CTT) group has described the relationship between lowering LDL-C and reducing major vascular events for statin therapy. In 2010 an updated meta-analysis, which included individual patient data from 170,000 patients in 26 RCTs, described a relationship between lowered LDL-C and a protective effect against vascular events [50]. Twenty-one trials compared statins to controls, with an additional 5 trials comparing higher-intensity to lower-intensity statin treatments. In the overall meta-analysis, a reduction in relative risk of 22% (95% CI 20, 24) was estimated for every 1.0 mmol/L reduction in LDL-C. The results also suggested that these reductions in vascular risk can be achieved safely even in patients who already have low LDL-C concentrations. In a pooled analysis of over 11,000 patients in 17 RCTs, similar results were found; adding ezetimibe resulted in a statistically meaningful reduction in LDL-C, with more than twice the percent change in LDL as doubling the dose of ongoing statin [51].

The recently published IMPROVE-IT trial [52] compared the combination of simvastatin and ezetimibe to simvastatin monotherapy in a population of 18,144 patients who had been recently hospitalized for acute coronary syndrome. Patients on combination therapy achieved a 0.43 mmol/L greater reduction in LDL-C than those on simvastatin monotherapy, signifying a 24% additional reduction in LDL-C in the combination group. The trial found an absolute risk difference of 2.0% (in favor of simvastatin plus ezetimibe) of a composite cardiovascular outcome (cardiovascular death, nonfatal myocardial infarction, unstable angina, or coronary revascularization); HR 0.936 (95% CI 0.89––0.99, p = 0.016). Utilizing similar methods as utilized in the CTT meta-analysis (with imputation for missing LDL-C values) to estimate the clinical benefit to a per mmol/L basis of LDL-C reduction with ezetimibe in IMPROVE-IT resulted in a hazard ratio (HR) of 0.80 [95% CI (0.68; 0.94)], which is consistent with the HR 0.78 (95% CI [0.76; 0.80], p < 0.0001) observed with statins in the meta-analysis performed by the CTT in 2010. These results further strengthen the evidence of the relationship between absolute reduction in LDL-C levels through up-regulating LDL receptors and lowering of cardiovascular risk beyond therapy with statins alone. The recent ACC consensus statement [53] further supports consideration of the addition of ezetimibe to statin therapy, given its benefits in reduction of cardiovascular outcomes and demonstrated safety profile.

The evidence supporting the efficacy of ezetimibe in combination with a statin to further lower LDL-C is important given the number of patients that do not reach the recommended LDL-C values. The Dyslipidemia International Study (DYSIS) II was an observational, cross-sectional study conducted in 21 countries in Asia/Pacific, Europe, and Middle East/Africa in 2012–2014 that evaluated lipid-lowering treatment and LDL-C goal attainment in two distinct cohorts: patients who survive an ACS event and in patients with a documented history of stable CHD. In the global ACS cohort, 24.8% of patients achieved an LDL-C < 70 mg/dl at admission for their ACS event and 34.4% attained the same goal at 4 month follow-up; at follow-up 87% received statin monotherapy with a mean atorvastatin equivalent dose of 32 mg/day. In the global CHD cohort, 30.6% of patients achieved an LDL-C < 70 mg/dl; 82% received statin monotherapy with a mean atorvastatin equivalent dose of 25 mg/day. These data indicate that either the statin can be intensified (same dose or switch) or, if a larger decrease in LDL-C is required, ezetimibe can be used in combination with a statin to ensure that more high-risk patients reach the recommended LDL-C values to prevent CV events.

The results of the current study were based on clinical evidence identified through an exhaustive, systematic review of the literature. While the timing of the search precluded the addition of IMPROVE-IT, the results of the current analysis are in line with the IMPROVE-IT study and serve to reinforce the findings of that study. The estimation of relative treatment effects through an NMA is an established methodology and is accepted by numerous health technology assessment agencies worldwide. Extensive validation of NMA models was also performed to investigate the possible sources of heterogeneity and inconsistency in the network of evidence. Despite this, there are some limitations to this analysis. Safety outcomes were not considered, as the purpose of the current study was restricted to determining the relative efficacy. Relatively few studies were identified in rosuvastatin-experienced patients, so these results must be interpreted with some caution. Some inconsistencies were observed between patient baseline characteristics across study arms in the atorvastatin- and simvastatin-experienced trials, such as in the proportion of males, smokers, and patients with hypertension. Some differences were also observed in baseline lipid levels of some study arms. Trials in rosuvastatin-experienced patients were more consistent, though there was some variation in the proportion of patients with diabetes. It is unclear whether these differences impacted outcome estimates. Future studies may investigate the impact of these differences through a meta-regression NMA approach.

Conclusion

Through a systematic search of the literature, we identified trials in high-risk patients previously treated with simvastatin, atorvastatin, and rosuvastatin, and evaluated CFB in LDL-C and total cholesterol for alternative treatment options. Regardless of the base statin, the addition of ezetimibe resulted in a statistically larger reduction in LDL-C compared to doubling the prior statin dose or switching to higher-potency rosuvastatin. Given the proven LDL-C lowering and CV benefit provided by ezetimibe, it remains an important option to enable more patients to reach the recommended LDL-C values and prevent CV events.