Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Protocol - Intervention

Ezetimibe for primary hypercholesterolemia

Authors' declarations of interest

Version published: 18 October 2006 Version history

https://doi.org/10.1002/14651858.CD006298

This is not the most recent version

view the current version 30 June 2015
Collapse all Expand all

Abstract

This is a protocol for a Cochrane Review (Intervention). The objectives are as follows:

To assess the effects of ezetimibe for primary hypercholesterolemia.

Background

Description of the condition

Primary hypercholesterolemia comprises a group of disorders characterized by persistent high levels of cholesterol in plasma. With the exception of familial hypercholesterolemia (FH) the remaining forms have not been well characterized biochemically and are caused by a mix of both genetic and dietary factors (Grundy 1986). They become defined only by excluding secondary causes such as untreated‐hypothyroidism, uncontrolled‐diabetes mellitus, hepatic or kidney disease and use of steroids (NCEP 2001). In contrast, FH is caused by a mutation of the gene that codifies the low density lipoprotein (LDL) receptor or its ligand (apolipoprotein B‐100) afflicting one in 500 Caucasians with various degrees of severity (Austin 2004; Choumerianou 2005). Hypercholesterolemia is considered one of the main risk factors for many forms of cardiovascular disease (that is coronary heart disease (CHD), cerebrovascular disease and peripheral artery disease), which share atherosclerosis as the underlying pathology (De Backer 2003). In conjunction with alcohol abuse and arterial hypertension, hypercholesterolemia is the main contributor to disease worldwide both in developed countries as well as in developing ones (Ezzati 2002).

Description of the intervention

The connection between total cholesterol and CHD (that is myocardial infarction (MI), unstable and stable angina) has been acknowledged for many years from several observational studies (Castelli 1986; Chen 1991; Stamler 1986). In these studies it was not possible to identify a threshold below which a lower serum cholesterol concentration was not associated with a lower risk of CHD (Law 1994). In that regard, the relation between cholesterol and CHD has been described as log‐linear, that is "... one in which the percentual difference in risk associated with a given difference in usual cholesterol concentration is similar at all concentration of cholesterol" (Chen 1991). In comparing data within groups of individuals of various backgrounds it was possible to estimate that a difference in cholesterol concentration of 20 mg/dl is associated with a relative difference of 17% in CHD mortality (Verschuren 1995). In addition, several experimental studies have proved that cholesterol‐lowering therapy reduces the risk of CHD in primary prevention as well as in secondary prevention. As an example in "Heart Protection Study" (HPS 2002; HPS 2003) which included over 20,000 individuals at high coronary risk (that is prior coronary disease, other occlusive arterial disease or diabetes) the daily intake of 40 mg of simvastatin during five years reduced the risk ratio of MI (fatal and non fatal) and of any revascularisation procedure (both coronary and non coronary) by 27% and 24% respectively in comparison with the placebo intake. In the West of Scotland Coronary Prevention Study (Shepherd 1995) the daily intake of 40 mg of pravastatin during five years prevented seven deaths from cardiovascular causes, 20 MIs, 14 coronary angiograms and eight revascularisation procedures for each 1000 individuals without a prior history of MI. In this context of primary prevention the risk ratio reduction of MI or death from CHD as against placebo intake was 31%.
In contrast to what was observed in CHD research, observational studies have failed to find a direct connection between cholesterol concentration and cerebrovascular disease risk (specially ischemic stroke risk). No evidence could be found even when examining populations in high risk of stroke as in Eastern Asian countries (Chen 1991; ESCHDCRG 1998) or considering FH patients (Hutter 2004). However, several experimental studies have reported that cholesterol‐lowering therapy using statins could be associated to an important reduction of stroke risk. For example in the above mentioned HPS (HPS 2002) the reduction of the risk ratio for both fatal and no‐fatal stroke was 25% as compared to placebo. A similar result was found in the diabetic people subgroup (HPS 2003). In addition the 'Myocardial Ischemia Reduction with Aggressive Cholesterol Lowering' (MIRACL) study reported a small difference that was yet statistically significant in the same outcome for a follow up of only four months (Schwartz 2001).
The strategic implementation of a cost‐effective cholesterol‐lowering therapy for both community and public health services is not easy. The log‐linear association between total cholesterol and CHD risk does not allow us to determine a unique cholesterol target that could be considered universally desirable. Additionally, it means that proportional benefit of a determined reduction in cholesterol to be constant whatever the initial CHD risk and that in this case just the absolute benefits vary, being higher for higher basal risk (Grundy 2004; McKenney 2005; Ramsay 1996). This has made it necessary to approach hypercholesterolemia in a joint vision regarding both cholesterol levels and baseline risks for CHD. Therefore, one should consider the absolute benefit of the intervention as expected for each patient as well as the proportion of population that could require treatment when applying a community based strategy (Ramsay 1996). Both U.S. and European guidelines have initially approached the problem by determining risk categories for CHD and assigning different targets of LDL cholesterol for each one of them (De Backer 2003; Grundy 2004; NCEP 1988; NCEP 2001). However, taking into account the most recent evidence, the determination of risk for cardiovascular disease has replaced at present the CHD risk determination (De Backer 2003; Grundy 2004; HPS 2002); this concept has partially blurred the concept of primary and secondary prevention for CHD.
Besides LDL cholesterol‐lowering other factors can be useful in the treatment of people with hypercholesterolemia. In this regard, the level of high density lipoprotein (HDL) cholesterol is inversely associated with CHD risk (Castelli 1986), therefore its increase can provide an additional benefit. The ratio total‐cholesterol/HDL has been found to be a more specific and precise predictor for CHD than the isolated measure of LDL cholesterol, both with similar sensitivity (Wang 2001). The 'Veterans Affair High‐Density Lipoprotein Intervention Trial' (VA‐HIT) reported a reduction of th risk ratio for MI or death from CHD of 11% for each increase of 5 mg/dl in the HDL cholesterol level by using gemfibrozil as opposed to using a placebo, keeping constant the LDL cholesterol levels in both trial arms (Robins 2001).
At present statins are the standard therapy for people with hypercholesterolemia due to their greater efficacy and safety in reducing cholesterol concentration. In fact, the studies using these drugs support a great part of our present knowledge about cholesterol‐lowering therapy. However cholesterol‐independent pleiotropic effects of statins have been reported and much has been discussed about their impact on clinical practice. Among these effects, statins capacity to reduce levels of super‐oxide radicals, markers of oxidative stress and susceptibility of LDL cholesterol to oxidation have been mentioned, effects that can only be partially explained by a reduction in LDL cholesterol (Choumerianou 2005). In animal models, the use of statins can reduce macrophage infiltration, the presence of monocyte chemoattractant protein‐1 (MCP‐1) and the proportion of nuclear factor kappa‐B activated (which is responsible for starting transcription of MCP‐1) in the atherosclerotic plaque (Bustos 1998). The presence of inflammatory cells in this plaque is considered evidence of instability as these cells can enzymatically degrade the fibrous cap unleashing its rupture and a process of fibrin and platelets aggregation that leads to the sudden occlusion (total or partial) of the affected blood‐vessel (Yeghiazarians 2000). However, due to the different magnitude of cholesterol lowering and varying tolerance between statins and other therapies it is very difficult to carry out a head‐to‐head trial to assess the impact of these pleiotropic effects on clinical outcomes, independently of LDL cholesterol reduction. In a meta‐analysis at the end of the 90s (Gould 1998), authors concluded that statins reduce CHD and total mortality more than other cholesterol‐lowering therapies because they reduce more efficaciously the LDL cholesterol, without increasing no‐CHD mortality. Recently, in a post‐hoc analysis of the 'Antihypertensive and Lipid‐Lowering Treatment to Prevent Heart Attack Trial' (ALLHAT), the effects found in clinical outcomes of CHD could be explained exclusively by the modest reduction of cholesterol achieved (ALLHAT 2002). In the case of cerebrovascular disease, the lack of a direct link between cholesterol levels and stroke risk in observational studies, despite a clear beneficial effect by the statins, has led to the opinion that the anti‐oxidant or anti‐inflammatory effects and the improvement of endothelial homeostasis might be more relevant in this context (Vaughan 2003). Many of these effects involve the covalent activation by isoprenylation of small GTP‐binding proteins and heterotrimeric G‐proteins in a mevalonate‐dependent mechanism that is thought to be inhibited by statins.

Adverse effects of the intervention

In minor early studies (with short term follow up) the use of ezetimibe did not alter the withdrawal rates due to serious adverse events. However, isolated cases of angioedema, rash and increase in the enzyme markers of liver damage were reported (Husereau 2003). In human pharmacokinetic studies relevant pharmacologic interactions with other drugs (including both lipid‐lowering and glucose‐lowering medications) have not been found (Kosoglou 2005). Due to its wide hepatic metabolism, patients with hepatic failure could reach greater concentrations in blood in direct relation to the severity of the hepatopathy (Kosoglou 2005). The safety profile of ezetimibe in long‐term use has not been sufficiently assessed.

How the intervention might work

Ezetimibe (1‐(4‐fluorophenyl)‐3(R)‐[3‐(4‐fluorophenyl)‐3(S)‐hydroxypropyl]‐4(S)‐(4‐hydroxyphenyl)‐2‐azetidinone) is a new drug, the first in its class, that inhibits the intestinal uptake of dietary and biliary cholesterol (Kosoglou 2005; Patel 2004). After the discovery of this drug it became possible to determine the presence of a specific transport protein of cholesterol in the wall of the small intestine, which is also responsible for the absorption of dietary plant sterols (Patel 2004). Ezetimibe is extensive metabolized within the intestinal mucous membrane and the liver, originating its active form (Kosoglou 2005). Ezetimibe‐glucuronide accounts for most of the total ezetimibe concentrations measured in plasma (80% to 90%) and both unconjugated and conjugated ezetimibe are highly bound to human plasma proteins. Ezetimibe is eliminated mainly by hepatobiliary excretion and its extensive enterohepatic recycling allows it to be administered once a day (Patel 2004). Only around 11% of the administered dose are eliminated through urine (Kosoglou 2005).
Cholesterol absorbed through the intestine has a dual origin, dietary and biliary. Ezetimibe seems to inhibit the specific intestinal transport of cholesterol, preventing the intake of cholesterol by more than 50% (Patel 2004). In this manner, ezetimibe can contribute to reduce the body pool of cholesterol through a mechanism that completes the other cholesterol‐lowering therapies like statins.
Unlike statins, ezetimibe seems to lack pleiotropic effects. Recently, a comparative study between simvastatin and ezetimibe in normocholesterolemic patients with chronic heart failure has been published. In this study the assignment to simvastatin reduced vascular oxidant stress and improved endothelium‐dependent vasodilation in comparison with the ezetimibe intake by a cholesterol‐independent mechanism since cholesterol reduction was similar in both therapies (Landmesser 2005).

Why it is important to do this review

Hypercholesterolemia is one of the more widely spread risk factors for cardiovascular disease in developed countries as well as in developing ones. There is general agreement that a cholesterol‐lowering therapy must be started early in these patients in order to reduce the associated mortality and morbidity. However, many of these therapies show a low profile of safety and tolerance. At present, statins are considered the drug of first choice due to their tolerance and to the fact that in standard doses it is possible to achieve a reduction of LDL cholesterol of about 30% to 40%, which is impossible to get with any other cholesterol‐lowering therapy alone (McKenney 2005). In spite of the efficacy of these drugs and the possibility of using combined therapy, many patients fail in achieving the goals defined in the guidelines for clinical practice. In the frame of primary care in the United States less than 50% of all persons with primary hypercholesterolemia and 18% of patients with CHD achieved their targeted LDL cholesterol according to the National Cholesterol Education Program (Pearson 2000). Even considering those that receive a combined therapy (only 10.2% of the total), success is only reached in about 40% of cases. Both a poor patient compliance with dietary and medication advice as well as the reluctance of physicians to prescribe high doses of statins or drug combination come together in this sub‐optimal lipid management. In this context, the role of ezetimibe in clinical practice as a new cholesterol‐lowering therapy must be carefully assessed.
The 'Canadian Coordinating Office for Health Technology Assessment' (CCOTHA) published in 2003 a review about ezetimibe. This review failed in finding controlled trials of sufficient duration or statistical power to detect differences in patient‐oriented outcomes such as death and hospitalisation (Husereau 2003). In a recent meta‐analysis supported by a grant from Merck/Schering‐Plough Pharmaceuticals (ezetimibe/simvastatin manufacturer) the cholesterol‐lowering efficacy of ezetimibe plus simvastatin versus rosuvastatin was indirectly compared (Catapano 2005). The combined treatment produced a higher reduction in the LDL cholesterol, triglycerides, total cholesterol and apolipoprotein‐B concentrations with a similar increase in the HDL cholesterol levels. In spite of the test of heterogeneity of effects across studies being statistically significant in most circumstances, the assessment of the methodological quality of the included studies, subgroup and sensitivity analyses and assessments of reporting biases were not published. Assessments of clinical outcomes like mortality, cardiovascular events or adverse events were not included in the publication.
We focus the present systematic review on patient‐oriented outcomes instead of lipid serum measurements in order to assess the role of ezetimibe in clinical practice. Also, an additional effort to assess ezetimibe safety in both short and long‐term use by the meta‐analytic approach will be done.

Objectives

To assess the effects of ezetimibe for primary hypercholesterolemia.

Methods

Criteria for considering studies for this review

Types of studies

We will include randomised controlled trials (published and non‐published) regardless of the duration of the studies. Trials using inadequate randomisation methods as alternation, date of birth, date of admission or hospital numbers will be excluded (Jadad 1996).

Types of participants

People with primary hypercholesterolemia, as defined in inclusion criteria or deducible from the inclusion or exclusion criteria. If primary hypercholesterolemia is not explicitly mentioned, trial participants meeting any of the following points will be included:

  • people with genetics disorders defined as familial hypercholesterolemia (Austin 2004; Choumerianou 2005; Hutter 2004);

  • people with hypercholesterolemia (in agreement with any current or past definition), if secondary causes of hypercholesterolemia (in accordance with the 'Adult Treatment Panel III of the National Cholesterol Education Program' (NCEP 2001) were excluded or sufficiently treated.

Current and past guidelines recommend different levels of either total or LDL cholesterol to consider drug therapy in people with hypercholesterolemia (De Backer 2003; Grundy 2004; NCEP 1988; NCEP 2001). Inclusion of studies using different definitions of hypercholesterolemia can be the cause of significant variability. These differences will be considered and explored in a sensitivity analysis.

Types of interventions

Comparisons

  • ezetimibe versus placebo or usual care.

  • ezetimibe plus statins versus statins.

  • ezetimibe versus statins.

Exclusion criteria

Trials that permit using other lipid‐lowering agents (except ezetimibe and statins) will be excluded.

Types of outcome measures

Primary outcomes

  • death from all causes.

  • death from cardiovascular causes, that is death from myocardial infarction, stroke (any stroke, haemorrhagic and non‐haemorrhagic), complications of peripheral arterial disease, or sudden death.

Secondary outcomes

  • non‐fatal myocardial infarction.

  • admission for unstable angina.

  • non‐fatal stroke (haemorrhagic and non‐haemorrhagic).

  • amputation due to peripheral arterial disease.

  • vascular procedures (coronary and no‐coronary revascularisation).

  • any adverse event (for example liver function impairment, angioedema, rash).

  • quality of life, only measured using validated scales.

  • costs

  • surrogate outcomes measures (total serum cholesterol, LDL cholesterol, HDL cholesterol and triglycerides).

Covariates and effect modifiers measures

We will assess treatment compliance measure as an effect modifier.

Timing of outcome assessment

Ideally, long‐term follow up can is more desirable to assess outcomes like mortality or rare adverse events. In treatment of people with hypercholesterolemia the full reduction in risk of coronary heart disease (CHD) could be achieved within five years (Law 1994). As an example, in the Heart Protection Study the assessment at the first year of follow up found a non‐significant difference for major cardiovascular events (HPS 2002). In contrast, the MIRACL study found a significant difference for re‐admission due to unstable angina (secondary outcome) within only four months of follow up. On the other hand, changes in serum lipid measures can be assessed in the short term. In this case, a minimum time of six weeks is proposed by the Adult Treatment Panel III (NCEP 2001) and is strongly correlated with long time effects over lipids control (Andrews 2001). Additionally, this time period could be sufficient to assess the rate of some adverse events like allergic reactions.

We consider four timing of outcome assessments:

  • short term: less than three months;

  • medium term: between three months and one year;

  • long term: between one and five years;

  • very long term: more than five years.

All outcomes will be assessed in the medium, long and very long term. Also, serum lipid measures and adverse reactions will be assessed in the short term.

Search methods for identification of studies

Electronic searches

We will use electronic search to identify relevant randomised trials (as specified above), reviews/meta‐analyses (for identification of additional trials), and economic analyses (to be able to comment on costs). No language restrictions will be applied. The next databases will be searched:

  • The Cochrane Library (latest issue);

  • MEDLINE (PubMed interface ‐ until recent);

  • EMBASE (until recent);

  • LILACS (until recent).

We will also search databases of ongoing trials:

  • Current Controlled Trials (www.controlled‐trials.com)

  • The National Research Register (http://www.update‐software.com/National/)

  • Clinical Research Studies (http://clinicalstudies.info.NIH.gov)

The search strategy was developed by the Metabolic and Endocrine Disorders Group as a modification of the highly sensitive search strategy for retrieving reports of controlled trials (Robinson 2002). The described search strategy (see for a detailed search strategy Appendix 1) will be used for MEDLINE. For use with EMBASE and The Cochrane Library this strategy will be slightly adapted.

Additional key words of relevance may be identified during any of the electronic or other searches. If this is the case, electronic search strategies will be modified to incorporate these terms. Studies published in any language will be included.

Searching other resources

Authors of relevant identified studies, authors of reviews, and ezetimibe manufacturers (Merck/Schering‐Plough Pharmaceuticals) will be contacted in order to obtain additional references, unpublished trials, or ongoing trials.

We will try to identify additional studies by searching the reference lists of relevant trials and reviews or meta‐analysis identified.

Data collection and analysis

Selection of studies

Two authors (LC and AC) will independently check the titles, abstract sections, keywords, description or comments of all registers retrieved through the search strategy (see above). The full text of all studies considered potentially relevant by at least one author will be obtained and considered by means of an inclusion or exclusion form. We will probe this form in a pilot test and edit it if necessary. Interrater agreement for studies inclusion will be measured using the Cohen's kappa statistic (Cohen 1960). In the case of disagreement or doubt we will contact study authors to obtain further information that permits resolving it. The final inclusion or exclusion decision will be made by consensus. An adapted QUOROM (quality of reporting of meta‐analyses) flow‐chart of study selection will be attached (Moher 1999).

Data extraction and management

Two authors (LC and AC) will collect all data from each study using data extraction forms, which will be probed in a pilot test and edited if necessary (for details see 'Characteristics of included studies' and Appendix 3Appendix 4; Appendix 5; Appendix 6; Appendix 7).

All data will be collected with original units and transformed for comparison if necessary. In case of differences in data extraction, we will refer back to the original paper. We will collect the following data:

  • general information: author, publication (reference or draft), other duplicate or preliminary publications, year of publication, language, country where the study was conducted, stated aim of study, sources of support or sponsors, contact address.

  • trial characteristics: inclusion and exclusion criteria, definition of hypercholesterolemia (level of either total or LDL cholesterol used as cut off), run‐in period (dietary advise, metabolic control in people with diabetes), sampling method, power calculation, randomisation and allocation concealment (method used), follow‐up, blindness of patients and treatment administrators (method and adequacy). Moreover we will register if the analysis was in accordance with intention‐to‐treat principle, rate of crossover, missing data, drop‐outs and withdraws, analyses by subgroups (pre‐definite and post‐hoc), patients compliance (and method of assessment), final estimate of methodological quality.

  • interventions: type, dose, regimen and duration of therapy and follow up in both intervention and control group; other co‐interventions permitted.

  • patients: baseline characteristics in intervention and control group: total number of participants, age, gender ratio, baseline cardiovascular risk, people with diabetes, hypertension, CHD, stroke (haemorrhagic and no‐haemorrhagic) and peripheral vascular disease, proportion of smokers, basal levels (mean and standard device (SD)) of total cholesterol, LDL cholesterol, HDL cholesterol and triglycerides.

  • outcomes and results: results in both intervention and control group will be registered: total deaths and deaths from cardiovascular causes (MI, stroke (any stroke, haemorrhagic and no‐haemorrhagic stroke), complication of peripheral vascular disease or sudden death), no‐fatal events from cardiovascular causes (MI, stroke, amputation, coronary and no‐coronary vascular procedure, admission for unstable angina), total adverse events, withdrawals for adverse events, quality of life (and method of measure), compliance, costs and either final serum lipids measures (total cholesterol, LDL cholesterol (and method of measure), HDL cholesterol and triglycerides) or change in serum lipids. Other outcomes assessed in the study will be registered. Also the number of participants included in outcome assessment and timing of outcome assessment will be registered.

Assessment of risk of bias in included studies

We will use a simple approach to assess the methodological quality (validity) of each study, according to Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2005). In particular, the following criteria will be studied:

Minimisation of selection bias

Randomisation method and allocation concealment:

  • Adequate: that is methods that allow each study participant to have the same chance of receiving each intervention and provide some assurance that allocations were not know until the point of allocation (for example centralised or pharmacy‐controlled randomisation, on‐site computer systems combined with allocation kept in a locked, unreadable computer file that can be accessed only after the characteristics of an enrolled participant have been entered, sequentially numbered, sealed, opaque envelopes, etc). Partially adequate: that is methods that allow each study participant to have the same chance of receiving each intervention but dot not provide assurance that allocations were not know prior the point of allocation (for example flip coin, open random number list, serially numbered envelopes not sealed and opaque, etc).

  • Inadequate: quasi‐randomisation methods like alternation, date of birth, date of admission, hospital numbers or similar considered as inappropriate randomisation methods (Jadad 1996).

Minimisation of performance bias

Blinding of patients and people administering the treatment:

  • Adequate: study described as double‐blind when the method of blinding used was appropriate. We will consider that the method of blinding was appropriate if was stated that neither the person administering the treatment nor the study participant could identify the intervention being assessed, or if in the absence of such a statement the use of active placebos, identical placebos, or dummies is mentioned (Jadad 1996).

  • Partially adequate: study described as single‐blind or double‐blind when the method of blinding used was partially reported or inappropriate.

  • Inadequate: open‐label study or when method of blinding was not reported.

Minimisation of attrition bias

  • Adequate: studies with intention‐to‐treat analysis where exclusions were less than 10% and differences in exclusion between groups were less than 5% (with adequate reporting of withdrawals and dropouts).

  • Partially adequate: studies without intention‐to‐treat analysis but exclusions were less than 10% and differences in exclusion between groups were less than 5% (with adequate reporting).

  • Inadequate: no intention‐to‐treat analysis performed, no reporting of exclusions (withdrawals and dropouts), exclusions of 10% or more, or wide differences in exclusion between groups (more than 5%).

Minimisation of detection bias

Blinding of outcome assessors is to date not a common practice in the conduct of randomised trials and this blinding is more likely to be particularly important in research with subjective outcome measures (Higgins 2005). Moreover some quality scales like Jadad's scale do not include this criterion (Jadad 1996); therefore we will not use this item for the overall quality assessment.

Based on these criteria, studies will be broadly subdivided into the next three categories (Higgins 2005):
A ‐ all quality criteria met (low risk of bias), that is all criteria are "adequate";
B ‐ one or more of the quality criteria only partly met (moderate risk of bias), that is at least one criterion is "partially adequate", but none is "inadequate";
C ‐ one or more criteria not met (high risk of bias), that is at least one criterion is "inadequate".

Each study report(s) will be assessed independently by two reviewers (LC, EC). Interrater agreement for assessment of methodological quality will be measured using the Cohen's kappa statistic (Cohen 1960). In cases of disagreement a judgement will be made based on consensus. We will further on explore the influence of individual quality criteria in a sensitivity analysis (see under 'sensitivity analyses').

Measures of treatment effect

Data will be summarised statistically if they are available, sufficiently similar, and of sufficient quality. Results will be divided in all possible comparisons first (that is ezetimibe versus placebo, ezetimibe plus statins versus placebo and ezetimibe versus statins), then sub‐divided into all possible outcomes (that is all‐cause and specific mortality, number of cardiovascular events, number of adverse events, drop‐outs related adverse events, quality of life, costs and serum lipid measures) and finally into all possible follow‐up periods (that is short term (only serum lipids measures and adverse events), medium, long and very long term).

Dichotomous data

To analyse dichotomous variables (that is mortality (global and specific), persons who experienced at least one non‐fatal cardiovascular event, persons who experienced at least one adverse event and withdrawal related adverse events) will be analysed by use of the relative risks (RR) and 95% confidence interval 95% (CI). Regarding cardiovascular mortality, changes in absolute risk are dependent on baseline risk (see background), therefore we will estimate reduction in absolute risk (for example number‐need‐to‐treat to benefit or NNT) in the analysis by subgroups of risk (see below). In case of adverse events we will calculate the number‐need‐to‐harm (NNH).

Continuous data

Continuous data such as serum lipid levels (that is total cholesterol, LDL cholesterol, HDL cholesterol and triglycerides) will be evaluated by means of the (weighted) mean difference and 95% CIs as measures of treatment effect. We will transform lipid measures in mmol/L to mg/dl using a correction factor different for cholesterol and triglycerides (NCEP 2001). If measures of quality of life were performed using different scales we will use the standardised mean difference and 95% CIs.

Conversion of final values to change score from trials is a difficult, error‐prone task, and may best be avoided. Mixing change score and final values in an analysis does not change the direction or size of the point estimate in these analyses, suggesting that both could give unbiased estimates of treatment effect in a meta‐analysis (Deeks 2003; Green 2001). Therefore, continuous variables will be entered into Review Manager, as they are presented in the included studies (that is as either change or end point scores).

Unit of analysis issues

Studies with multiple treatment groups

In case of studies with multiple treatment groups we will define a single "intervention group" and "control group" for all possible comparisons (that is ezetimibe versus placebo, ezetimibe plus statins versus placebo and ezetimibe versus statins). In all cases, the intervention group will be that one which uses the greater doses of ezetimibe. For ezetimibe plus statins versus placebo and ezetimibe versus statins comparisons, the control group will be the one which uses the greater doses of statins.

Repeated observations on participants

In case of studies which report more than one result for the same outcome at different timing we will select a single result (longest time) for all definite follow up (that is short, medium, long and very long term).

Dealing with missing data

In case of missing data in full reports, duplicate publications and preliminary report, the study will be awaiting assessment until further information may be obtained of authors. Exclusions for unreported data could introduce bias.

Dealing with duplicate publications

We will make an additional effort to identify duplicate publications as well as preliminary reports. We will consider all papers with the same number of participants, identical follow up, identical interventions (type, doses, etc), the same countries or centres and similar authors like duplicate publications. In the case of duplicate publications and companion papers of a primary study, we will try to maximise yield of information by simultaneous evaluation of all available data. In cases of doubt, the original publication (usually the oldest version) will obtain priority. If we find disagreements in published results we will contact study authors. If papers differ in the methods of analyses we will primarily consider those papers where an analysis by the intention‐to‐treatment principle was performed.

Assessment of heterogeneity

Heterogeneity will be tested for using the Z score and the chi‐squared statistic with significance being set at P < 0.10. Possible sources of heterogeneity will be assessed by subgroup and sensitivity analyses as described below. Quantification of the effect of heterogeneity will be assessed by means of I2 (Higgins 2002). I2 demonstrates the percentage of total variation across studies due to heterogeneity and will be used to judge the consistency of evidence. I2 values of 50% and more indicate a substantial level of heterogeneity (Higgins 2003)

Assessment of reporting biases

To investigate the impact of publication bias, we will perform funnel plots for all relevant outcomes. Test for funnel plots asymmetry (Begg's and Egger's test) also will be performed using StataTM (Sterne 2001).

Data synthesis

Data will be summarised statistically if they are available, sufficiently similar and of sufficient quality. We will perform meta‐analyses of data using both fixed and random‐effects methods only if tests for heterogeneity are not significant. These methods tend to obtain similar results in cases of low heterogeneity (Deeks 2003). In case of heterogeneity a random effects method may be more "conservative" since it has a larger confidence interval (Hearst 2001). However, if heterogeneity is due either to publication bias (that is, it failed to find studies with 'negative' results) or differences in the quality of included studies (that is if many studies with low quality exist that overvalue the intervention) a random effects method may increase the biased effect (Deeks 2003). In all cases, we will report the more conservative method.
We will use the DerSimonian and Laid method (DerSimonian 1986) like random‐effects method for meta‐analysis of all outcomes. For fixed effect methods, we will use the Mantel‐Haenszel method (Mantel‐Haenzel 1959) for dichotomous data (summary statistic: risk ratio) and the inverse variance method for continuous data (summary statistic: mean difference or standardised mean difference). We will calculate summary estimates of the treatment effect employing the inverse variance method if the outcome measures would be normally distributed, but if not we will use alternatives such as transformation and non‐parametric tests. We will perform meta‐analyses using Review Manager 4.2.8.

Subgroup analysis and investigation of heterogeneity

Subgroup analyses will only be performed if one of the primary outcome parameters demonstrates statistically significant differences between treatment groups. We will investigate the following characteristics (either participants characteristics as well as studies design) that may have influence on results:

  • participants: vascular baseline risk, age;

  • study design: follow‐up (and treatment duration), inclusion criteria and definition of hypercholesterolemia, doses of statins and ezetimibe.

We consider a‐priori these characteristics as possible sources of heterogeneity of main outcomes and we will assess even if tests for heterogeneity are non‐significant. In order to assess sources of heterogeneity we will perform subgroup and meta‐regression analyses.

Subgroup analysis

(1) Cardiovascular baseline risk: baseline risk of participants is directly linked with both inclusion criteria and definition of hypercholesterolemia used in each study. At present, in both U.S. and European guidelines, people with type 2 diabetes mellitus are included in high baseline risk categories just like people with cardiovascular diseases (De Backer 2003; NCEP 2001). This fact may blur partially the concept of primary and secondary prevention within each study. Also these guidelines are not easily comparable among themselves or with others. We will use different approaches to assess the effect of baseline risk on both mortality and non‐fatal cardiovascular events (in all cases we will calculate the NNT within each risk group):

  • we will consider in comparative shape all results reported on both primary and secondary prevention of cardiovascular disease;

  • people with or without coronary heart disease;

  • people with recent cardiovascular events (within three months prior to their inclusion in the trial);

  • people with diabetes mellitus;

  • people with familial hypercholesterolemia.

(2) Doses of statins: We will divide all studies in groups according to statins doses used. Appendix 2 defines equivalent statins doses, which was modified from Grundy et al (Grundy 2004). Dose1 corresponds to a standard dose to obtain a 30% to 40% reduction in LDL cholesterol concentration. Each subsequent dose (dose2 and dose3) was calculated doubling the prior dose. For every doubling of the dose above standard dose, an approximate 6% decrease in LDL cholesterol level can be obtained.

(3) Doses of ezetimibe: we will divide all studies in groups according to ezetimibe doses used.

Meta‐regression

The robustness of the results will also be tested by performing a meta‐regression depending on quantity and quality of studies. The following studies' characteristics will be used in order to assess their impact on treatment effect: rate of high baseline risk participants, mean of age, follow up, statins doses. Meta‐regression will be performed with StataTM (Sterne 2001).

Sensitivity analysis

We will perform sensitivity analyses in order to explore the influence of the following factors on effect size:

  • repeating the analysis excluding unpublished studies;.

  • repeating the analysis including only studies with methodological quality "A", as specified above;

  • repeating the analysis excluding studies using the following filters: statistical model, source of funding (industry versus other), country (developed versus developing countries);

  • repeating the analysis excluding studies with initial disagreement about its inclusion;

  • we will also perform an examination of the influence of individual studies with the metainf command in StataTM (Sterne 2001).

The robustness of the results will also be tested by repeating the analysis using different statistical models (like fixed and random‐effects methods), as specified above.

Economic issues

If we find analyses of costs in included studies, these will be summarized in the result section and the table of comparisons just like the other results. They will be summarized in accordance to the type of analysis performed (that is cost‐effectiveness, cost‐efficacy, etc). Additional studies retrieved through the search strategy will be considered in the discussion.