OUP user menu

Efficacy and safety of statin treatment for cardiovascular disease: a network meta-analysis of 170 255 patients from 76 randomized trials

E.J. Mills, P. WU, G. Chong, I. Ghement, S. Singh, E.A. Akl, O. Eyawo, G. Guyatt, O. Berwanger, M. Briel
DOI: http://dx.doi.org/10.1093/qjmed/hcq165 109-124 First published online: 7 October 2010


Background: Statins represent the largest selling class of cardiovascular drug in the world. Previous randomized trials (RCTs) have demonstrated important clinical benefits with statin therapy.

Aim: We combined evidence from all RCTs comparing a statin with placebo or usual care among patients with and without prior coronary heart disease (CHD) to determine clinical outcomes.

Design: We searched independently, in duplicate, 12 electronic databases (from inception to August 2010), including full text journal content databases, to identify all statin versus inert control RCTs. We included RCTs of any statin versus any non-drug control in any populations. We abstracted data in duplicate on reported major clinical events and adverse events. We performed a random-effects meta-analysis and meta-regression. We performed a mixed treatment comparison using Bayesian methods.

Results: We included a total of 76 RCTs involving 170 255 participants. There were a total of 14 878 deaths. Statin therapy reduced all-cause mortality, Relative Risk (RR) 0.90 [95% confidence interval (CI) 0.86–0.94, P ≤ 0.0001, I 2 = 17%]; cardiovascular disease (CVD) mortality (RR 0.80, 95% CI 0.74–0.87, P < 0.0001, I 2 = 27%); fatal myocardial infarction (MI) (RR 0.82, 95% CI 0.75–0.91, P < 0.0001, I 2 = 21%); non-fatal MI (RR 0.74, 95% CI 0.67–0.81, P ≤ 0.001, I 2 = 45%); revascularization (RR 0.76, 95% CI 0.70–0.81, P ≤ 0.0001); and a composite of fatal and non-fatal strokes (0.86, 95% CI 0.78–0.95, P = 0.004, I 2 = 41%). Adverse events were generally mild, but 17 RCTs reported on increased risk of development of incident diabetes [Odds Ratio (OR) 1.09; 95% CI 1.02–1.17, P = 0.001, I 2 = 11%]. Studies did not yield important differences across populations. We did not find any differing treatment effects between statins.

Discussion: Statin therapies offer clear benefits across broad populations. As generic formulations become more available efforts to expand access should be a priority.


For over 15 years, randomized trials of 3-hydroxymethyl-3-methylglutaryl coenzyme A reductase inhibitors (statins), have evaluated their impact on cardiovascular morbidity and overall mortality in patients with stable coronary artery disease.1 Since then, statins have been extensively studied in a large variety of patient populations including both primary and secondary prevention of cardiovascular disease (CVD).2,3 There is a widespread interest in the use of statins for broad populations given their effectiveness and relatively inexpensive costs now that three of them (lovastatin, simvastatin and pravastatin) are available in generic form. Statins are currently the largest selling prescription drug worldwide and may one day be widely available over-the-counter (OTC),4 with a 10 mg tablet of simvastatin already on sale OTC in the UK.

Clinicians have recognized that much of a statins therapeutic effect is derived from its low-density lipoprotein (LDL)-lowering effects.5 The greater the LDL reduction, the greater the clinical benefit in terms of risk reduction for CVD events.5 In addition, there is evidence that statins, beyond their LDL-lowering effects, reduce vascular inflammation, improve endothelial function and decrease thrombus formation.6–8 The role of these so-called pleiotropic effects of statins is less well established and it remains unclear if there are differences among available statins translating into different clinical benefit.

Large, up-to-date systematic reviews with meta-analyses are essential to provide clinicians, health economists and policy makers with the most reliable, critically appraised and precise estimates of treatment effects and to monitor for rare adverse events. Therefore, we updated previous meta-analyses of statin trials3,5,9–15 in an effort to assemble the totality of randomized trial (RCT) evidence to date in order to quantify the effects of statin therapy on a wide range of clinical outcomes and populations. Our primary outcome of interest is CVD mortality. We additionally examined whether specific statins exerted important therapeutic differences across the class of drugs adjusted for LDL-lowering effects.


Eligibility criteria

We included any RCT of atorvastatin, fluvastatin, lovastatin, pitavastatin, pravastatin, rosuvastatin and simvastatin for CVD event prevention among both primary and secondary prevention populations. We did not include cerivastatin as it has been withdrawn from the market due to serious adverse events. Studies had to compare a statin to placebo, standard therapy or no-treatment and report on any of the following clinically important cardiovascular outcomes: All-cause mortality; CVD mortality; fatal myocardial infarction (MI); Non-fatal MI; major CV events (stroke, revascularization). We excluded studies only reporting on surrogate outcomes [e.g. LDL and high-density lipoprotein (HDL) levels] and follow-up studies where randomization had been subverted.16 We additionally excluded head–head statin evaluations.

Search strategy

In consultation with a medical librarian, we established a search strategy (available from authors upon request). We searched independently, in duplicate, the following 12 databases (from inception to August 2010): MEDLINE, EMBASE, Cochrane CENTRAL, AMED, CINAHL, TOXNET, Development and Reproductive Toxicology, Hazardous Substances Databank, Psych-info and Web of Science, databases that included the full text of journals, ScienceDirect and Ingenta, including articles in full text from ∼1700 journals since 1993). In addition, we searched the bibliographies of published systematic reviews3,5,9–15 and health technology assessments.17–19 Finally, we searched our own comprehensive rolling database of statin trials, updated monthly. We also contacted the authors of all trials for study clarifications, where required, and the authors of the only individual patient data meta-analysis of statins, that included 14 trials.5,15 Searches were not limited by language, sex or age.

Study selection

Two investigators (E.M., P.W.) working independently, in duplicate, scanned all abstracts and obtained the full text reports of records, that indicated or suggested that the study was a RCT evaluating statin therapy on the outcomes of interest. After obtaining full reports of the candidate trials (either in full peer-reviewed publication or press article) the same reviewers independently assessed eligibility from full text papers.

Data collection

The same two reviewers conducted data extraction independently using a standardized pre-piloted form. The reviewers collected information about the statin and type of interventions tested, the population studied (age, sex, underlying conditions), the treatment effect on specified outcomes, absolute and proportion change in LDL, HDL and total cholesterol and the length of follow-up. Study evaluation included general methodological quality features, including sequence generation, blinding, use of intent-to-treat analysis, percentage follow-up and allocation concealment.20 We extracted data on the incidence of the following clinical outcomes: all-cause mortality, CVD mortality, MI mortality, stroke mortality, non-CVD mortality, major CVD, MI, strokes, revascularization, cancers, rhabdomylosis, diabetes, aspartate and alanine aminotransferase (AST/ALT), and creatinine kinase (CK) increases beyond the upper limit of normal. We entered the data into an electronic database such that duplicate entries existed for each study; when the two entries did not match, we resolved differences through discussion and consensus.

Data analysis

In order to assess inter-rater reliability on inclusion of articles, we calculated the Phi (ϕ) statistic, which provides a measure of inter-observer agreement independent of chance.21 We calculated the Relative Risk [RR] and appropriate 95% Confidence Intervals [CIs] of outcomes according to the number of events reported in the original studies or sub-studies intent-to-treat analyses. Where studies did not report intent-to-treat, we analyzed outcomes as all-patients randomized.22 In the case of an individual patient data meta-analysis of 14 trials, we included outcomes as reported by the meta-analysis, in correspondence with the study’s authors. In the event of zero outcome events in one arm of a trial, we applied the Haldane method and added 0.5 to each arm.23 We pooled studies as an analysis of all-statins combined using the DerSimonian–Laird random effects method,24 which recognizes and anchors studies as a sample of all potential studies, and incorporates an additional between-study component to the estimate of variability.25 We conducted a sensitivity analysis to determine if individual statins exerted differing effects using a mixed-treatment comparison and also on whether baseline population risks differed in treatment outcomes using a Breslow–Day test.26 For adverse events, we calculated event rates using Peto’s Odds Ratio (OR).27 Peto’s odds ratios appears to provide the least biased estimates and CI coverage with rare events.28 Forest plots are displayed for each all-statins analysis of our primary analyses and a combined forest plot for secondary outcomes, showing pooled estimates with 95% CIs, and the overall DerSimmonian–Laird pooled estimate. We calculated the I2 statistic for each all-statin analysis as a measure of the proportion of the overall variation that is attributable to between-study heterogeneity.29 We conducted a multivariable meta-regression analysis to examine the impact of the following co-variates, all chosen a priori: absolute LDL change; proportion of individuals in trials that were men; had a history of coronary heart disease (CHD), had baseline diabetes, or were hypertensive and current smokers.30

In order to evaluate the relative effectiveness of each study drug on CVD mortality, we used the Lu-Ades method for combining indirect evidence in mixed-treatment comparisons.31 We estimated the posterior densities for all unknown parameters using MCMC (Markov chain Monte Carlo) for each model. Each chain used 20 000 iterations with a burn-in of 20 000, thin of 5 and updates varying between 80 and 110. We used the same seed number (SEED = 314 159, equivalent to 10 pi) for all chains. The choice of burn-in was chosen according to Gelman–Rubin approach.32 We applied the covariate of LDL-C change and also statin dosing (high or moderate determined by the Canadian Compendium of Pharmaceuticals and Specialties),33 using an approach developed by Cooper et al.34 We assessed convergence based on trace plots and time series plots. The accuracy of the posterior estimates was done by calculating the Monte Carlo error for each parameter. As a rule of thumb, the Monte Carlo error for each parameter of interest is less than ∼5% of the sample standard deviation. All results for the mixed-treatment analysis are reported as posterior means with corresponding 95% credibility intervals (CrIs). CrIs are the Bayesian equivalent of classical CIs. We assessed the fit of our model using the Deviance Information Criterion (DIC), a measure of model fit that penalizes model complexity. This criterion advocates selecting the model with the lowest DIC value among a series of competing models for the same data, as this model is believed to provide the best fit to the data. DIC’s were not importantly different across models.

Finally, we conducted a trial sequential analysis to determine the strength of information for our meta-analysis on the primary outcome of CVD mortality to determine the conservative number of patients required to provide an authoritative answer of therapeutic efficacy.35 We applied a Lan–DeMets (LD) sequential monitoring boundary that assumes a 4% control event rate, 20% relative risk reduction, 90% power and a two-sided α = 0.05. We plotted the trial sequential analysis to display the heterogeneity-corrected optimal information size (HOIS). Analyses were conducted using StatsDirect (version 2.5.2, www.statsdirect.com), Stata (version 9, www.stata.com) and in WinBUGS version 1.4 (Medical Research Council Biostatistics Unit, Cambridge).

Role of the funding source

No funding sources had a role in study design, data collection, data analysis, data interpretation or writing of the report. The writing group had full access to all the data in the study and had final responsibility for the decision to submit for publication.


We included a total of 76 RCTs meeting our inclusion criteria (Figure 1). Data were available on 170 255 participants. Women represented ∼26% of trial participants. The average age of included participants was 59.6 [standard deviation (SD) 5.93], ranging from 38 to 75. Trials used four distinct controls as an inert control. These included placebo (52 RCTs),1,36–85 usual care (18 RCTs),8,86–102 no treatment (four RCTs)103–106 and conventional therapy (2 RCTS).89,107 Trials followed patients for an average of 2.7 years (SD 1.60), ranging from 0.5 years to 6.1 years.40 The mean pre-treatment LDL cholesterol was 4.61 mmol/l (179.79 mg/dl) and ranged from 2.43 mmol/l105 (94.77 mg/dl) to 5 mmol/l (195 mg/dl).51 Table 1 displays the study characteristics.

Figure 1.

Flow diagram of included studies.

View this table:
Table 1

Study characteristics

StudyYearPatient status/ conditionTreatment comparisons (mg/day)Follow-up, yearsRandomized IndividualsAge, mean, yearsMen (%)Prior CHD (%)Diabetes (%)Hyper tension (%)Current smokers (%)baseline, mean mg/dl (change)
CCAIT361994CHDL20–80 vs. placebo23315381100143727172 (–47)41 (1.0)
CLAPT861999CHDL20–80 vs. usual care222654100100144211182 (–36)42 (1.5)
FATS statin871990CHDL40 vs. usual care2.5984710010003624185 (–78)36 (3.5)
MARS371993CHDL73 vs. placebo2.227058911000080156 (–65)43 (2.5)
ACAPS611994Atherosclerotic–carotid stenosisL10–40 vs. placebo391961.752022912155 (–42)52 (2.6)
AFCAPS621998Primary preventionL20–40 vs. placebo5.266055885062212150 (–40)37 (1.8)
EXCEL811991Primary preventionL20–80 vs. placebo0.9824556593382401818045
ATHEROMA882005CHDP10–20 vs. usual care33735983100154220142 (–27)50 (0.5)
CARE381996CHDP40 vs. placebo541595986100144221139 (–41)39 (2.5)
GISSI-P statin892000CHDP20 vs. usual care242716086100143712151 (–24)46 (1.0)
HARP391994CHDP40 vs. placebo2.579588910010490137 (–58)42 (5.4)
L-CAD902000CHDP20–40 vs. usual care2126578010003267174 (–50)32 (–2.0)
LIPID401998CHDP40 vs. placebo6.19014628310094210150 (–38)36 (1.8)
Makuuchi H912005CHDP10–20 vs. usual care4.53355984100335242141 (–20)41 (3.2)
PCS412003CHDP10 vs. placebo5.41206092100185968128 (–15)43 (0.4)
PLAC I,II421995CHDP40 vs. placebo3559588010014116163 (–47)41 (1.9)
PREDICT431997CHDP40 vs. placebo0.5695588410073134156 (–41)47 (2.0)
PTT922002CHDP40 vs. usual care0.5775283100142312133 (–52)39 (5.0)
REGRESS441995CHDP40 vs. placebo28845610010002828166 (–49)36 (3.5)
OACIS-LIPID1032008CHDP10 vs. no statin0.753536377100314858150 (–19)48 (–13)
PROSPER592002Elderly patientsP40 vs. placebo3.25 80075.44844116227146 (–46)50 (5.8)
ALLHAT-LLT632002Primary preventionP40 vs. placebo4.810 35566.451143510023145 (–22)48 (2.6)
CAIUS661996Atherosclerotic–carotid stenosisP40 vs. placebo33055553024180 (–43)53 (1.9)
FAST982002Atherosclerotic–carotid stenosisP1 vs. usual care216466.131.314.222.841.559.3166 (–39)57 (1.8)
KAPS691995Atherosclerotic–carotid stenosisP40 vs. placebo344757.4100823326185 (–62)44 (3.9)
MEGA992006Primary preventionP10–20 vs. usual care5.3783258.3320214221156 (–23)58 (2.3)
WOSCOPS701995Primary preventionP40 vs. placebo4.9659555.2100011644192 (–46)44 (3.1)
PHYLLIS712004Atherosclerotic–carotid stenosisP40 vs. placebo2.650858.4400010016181 (–36)53 (0.93)
PMSG741993Primary preventionP20 vs. placebo0.5106255777504829180 (–47)44 (2.3)
KLIS222000Primary preventionP-NA vs. conventional treatment5.1434958100023.5943.7839.76254
Kobashigawa1001995Transplant patientsP20–40 vs. usual care197525348000158 (–42)43 (9)
ALLIANCE932004CHDA40–80 vs. usual care4.32442618210022019146 (–16)41 (–0.8)
Colivicchi82002CHDA80 vs. usual care181695810057890130 (–20)40 (1.0)
ESTABLISH942004CHDA20 vs. usual care0.5706286100335461124 (–50)45 (–2.0)
GREACE952002CHDA10–80 vs. usual care31600597810020430179 (–74)39 (2.2)
Wojnicz R962006CHDA40 vs. usual care0.574388110000NR158 (–35)40 (–12)
Yamada T452007CHDA10 vs. placebo3386479%10022%20%37%117 (–32)NR
ATAHEB1042008CHDA20 vs. no statin11067045100995NR130 (–30)41 (–14)
Vrtovec B1052008CHDA10 vs. no statin11106361100NRNRNR95 (–7)35 (–4)
Sdringola S462008CHDA80 vs. placebo0.51456590100NRNR14.00130 (–3.0)44 (–1.3)
Mohler722003Elderly patientsA10 vs. placebo123468771840125 (–50)46 (1.0)
Mohler722003Elderly patientsA80 vs. placebo123468771840125 (–50)46 (1.0)
ASCOT-LLA642004Primary preventionA10 vs. placebo3.310 30563.28102510033132 (–44)51 (1.3)
ASPEN652006DiabeticsA10 vs. placebo4.2524116166211005512113 (–33)47 (1.1)
CARDS672004DiabeticsA10 vs. placebo4283861.66801008422116 (–43)54 (1.9)
4D752005DiabeticsA20 vs. placebo4125565.754291009125 (–29)36 (13)
Stegmayr1012005Renal disease patientsA10 vs. usual care2.8143696728336565135 (–18)44 (0)
SPARCL772006Previous stroke (without peripheral artery disease)A80 vs. placebo4.9473163600176219133 (–55)50 (1.1)
Sola782006Congestive heart failure not secondary to CHDA20 vs. placebo110854630000121 (–25)43 (1)
FLARE471999CHDF80 vs. placebo0.8834618210043329152 (–47)41 (0.77)
FLORIDA482002CHDF80 vs. placebo15406183100112452137 (–41)46 (2.1)
LCAS491997CHDF40 vs. placebo2.6429598110048220145 (–29)44 (1.6)
LIPS502002CHDF80 vs. placebo3.916776084100123927131 (–40)38 (0)
LiSA511999CHDF40–80 vs. placebo1365606210062910195 (–38)55 (1.6)
O'Rourke522004Transplant patientsF40 vs. placebo1795285100NR800177 (–45)NR
HYRIM682004Atherosclerotic - carotid stenosisF40 vs. placebo456857.31000100183.87 (0.76)1.27 (0.30)
ALERT732003Transplant patientsF40–80 vs. placebo5.121025066101975184.1 (1)1.3 (0.5)
BCAPS792001Atherosclerotic - carotid stenosisF40 vs. placebo37936246431231161 (–37)53 (0)
4S11994CHDS20 vs. placebo5.44444598110042626188 (–68)46 (3.9)
Christenson532001CHDS20–40 vs. placebo2776380100114946166 (–38)52 (3.5)
CIS541997CHDS20–40 vs. placebo2.3254491001000084165 (–57)44 (2.8)
MAAS551994CHDS20 vs. placebo438155881000024170 (–54)43 (4.2)
SCAT562000CHDS10–40 vs. placebo44606189100113615129 (–45)38 (1.5)
Petronio A1062005CHDS20 vs. no statin171617510007046114 (–20)NR
HPS602002Primary preventionS40 vs. placebo520 5367565294114131 (–39)41 (1.2)
Beishuizen802004DiabeticsS40 vs. placebo2250594701005124134 (–42)47 (0.39)
Krum H572007CHDR10–40 vs. placebo0.58660801007NRNR124.8 (–39)
CORONA582007CHDR10 vs. placebo35011737610029639138 (–36)48 (1.4)
GISSI-HF852008CHFR10 vs. placebo3.945746782100265514123 (–40)122 (–2)
JUPITER762008Primary preventionR20 vs. placebo417 8026661.800015.810849
AURORA822009Renal disease patientsR10 vs. placebo3.2277364624026NR15100 (35)45 (15)
MUSASHI-AMI972006CHDAS vs. usual care1.14866479100306054133 (–28)46 (0.5)
SALTIRE832005Primary preventionA80 vs. placebo215568702546628130 (30)
Sahni1071991CHDL 20–40 vs. Conventional therapy0.51576070100244330140 (40)36 (10)
Wenke1021997Transplant patientsS 5–20 vs. Usual care47248892710105
Lewis842007Liver diseaseP80 vs. placebo0.732650520140 (30)48 (10)
MIRACL1082001CHDA80 vs. placebo0.333086656510023552812446

Methodological quality of included studies

We found that the reporting quality of studies varied. Twenty-six studies reported how randomization sequence was generated in their primary publication.1,8,37,44,45,50,52,58–60,63,64,67,71,75,76,80,82,83,85,94,96,98,99,104,105 Eighteen studies reported on how allocation to groups was concealed.38,59,60,63,64,66,67,70,73,75,76,83,91,93,97,99,105,106 Most of the studies (64) reported on loss-to-follow up1,36,38,40–46,48–50,52,53,55–65,67,69–82,8,83–88,90–97,99–104,106–108 and only four studies37,49,78,80 reported that the primary results were based on a per-protocol analysis rather than intent to treat. Sixty-one studies reported on at least one specific group being blinded in the trial, typically patients and caregivers.1,8,36–52,54–56,58–62,64–91,95,96,103,106–108

All cause and cause-specific mortality

There were a total of 14 878 deaths including a total of 7864 from confirmed vascular causes (Figure 2). In all trials combined, there were a total of 7004 (8.1%) deaths among the 86 328 patients receiving a statin and 7713 (9.5%) deaths among 80 365 patients receiving a control intervention. Combined, this represents a 10% reduction in all-cause mortality (RR 0.90, 95% CI 0.86–0.94, P ≤ 0.0001, I2 = 17%). Each 10% change in absolute LDL levels was associated with a 1.1% (95% CI 0.3–1.19, P = 0.003) risk reduction.

Figure 2.

Forest plot of mortality across statins.

The large risk reduction in all-cause mortality was chiefly attributed to the 20% risk reduction in CVD deaths [3605 (4.1%) of statin-treated patients vs. 4248 (5.1%) control-treated patients: RR 0.80, 95% CI 0.74–0.87, P < 0.0001, I2 = 27%]. Applying a univariate regression, each 10% change in LDL levels was associated with a 5.6% (95% CI 2–8%, P ≤ 0.001) risk reduction of CV death. This effect diminished in the multivariable analysis. Table 2 displays the impact of a priori chosen covariates on CVD mortality.

View this table:
Table 2

Meta-regression, impact of co-variates on estimates of CV death

Independent variable n = 51Parameter (95% CI)Relative increase in RR (95%CI) (every 10 U increase in predictor)P-valueR2
Intercept−0.68 (–1.14 to –0.22)0.0050.45
Delta-LDL0.001 (–0.003 to 0.005)0.51
Men (%)0.001 (–0.004 to 0.007)0.60
Prior CHD (%)0.002 (–0.00003 to 0.004)2% (0–4%)0.05
Diabetes (%)0.003 (–0.001 to 0.007)0.15
Hypertension (%)0.005 (0.002 to 0.008)5% (2–8%)0.0003
Current smokers (%)0.0008 (–0.005 to 0.007)0.78
High dose−0.096 (–0.194 to 0.001)0.054

We also found a consistent reduction in fatal MI with an 18% risk reduction (RR 0.82, 95% CI 0.75–0.91, P < 0.0001, I2 = 21%). We found a statistically non-significant reduction in deaths from stroke (RR 0.92, 95% CI 0.80–1.07, P = 0.55) and in non-CVD causes (RR 0.95, 95% CI 0.90–1.00, P = 0.07).

Risk factors across underlying conditions

We assessed whether our pooling of data from all CVD trials across disease conditions was reasonable and divided the RCTs into their specific primary disease populations assessing CVD death. We included 42 CHD RCTs;1,8,36–51,53–58,86–97,103–108 7 atherosclerosis;61,66,68,69,71,79,98 11 primary prevention;60,62–64,70,74,76,81,83,89,99 4 diabetic patients;65,67,75,80 2 elderly patients; 59,72 2 renal disease; 82,101 4 transplant patient;52,73,100,102 1 previous stroke;77 2 RCTs of congestive heart failure;78,85 and 1 RCT with hypercholesterolemic patients with chronic liver disease.84 Studies did not yield an importantly different direction of effect dependent on populations (heterogeneity P = 0.07) (Table 3).

View this table:
Table 3

CVD deaths across populations in included studies

PopulationRR (95% CI)
CHD0.82 (0.76–0.88)
Atherosclerotic0.51 (0.22–1.18)
Primary prevention0.81 (0.75–0.87)
Diabetes0.85 (0.70–1.03)
Elderly0.79 (0.60–1.02)
Renal disease1.01 (0.89–1.16)
Transplant0.68 (0.45–1.03)
Previous stroke1.02 (0.66–1.68)
Congestive heart failure1.01 (0.91–1.13)
  • Heterogeneity P-value = 0.07.

Major cardiovascular events

There were 6318 non-fatal MIs reported in 58 RCTs enrolling 146 041 patients.1,8,37–45,47–52,54–56,58–63,66,67,69–77,79,81,82,84,86–94,95,97–99,101,103,107,108 Overall, there was a highly significant 26% reduction in non-fatal MI [2810 (3.6%) statin vs. 3508 (4.9%) control: RR 0.74, 95% CI 0.67–0.81, P ≤ 0.001, I2 = 45%]. Data were also available on revascularization from 44 RCTs enrolling 118 296 individuals.1,38–41,43,44,47–49,53–56,58–60,62,66,67,69,70,72–77,82,87–91,93–95,98,99,101,103,106–108 We found a highly significant effect of statins on revascularization status [3723 (6.2%) statin vs. 4816 (8.1%) control: RR 0.76, 95% CI 0.70–0.81, P ≤ 0.0001, I2 = 44%].

In addition to assessing fatal strokes, we evaluated fatal and nonfatal strokes excluding transient ischemic events and included data from 53 RCTs enrolling 154 818 individuals.8,36–38,40–44,48,50,52,54–56,58–61,63–65,67,69–77,79,81,82,86,88–95,97–99,101,103,107,108 We found a strongly significant effect favoring statins [2201 (2.7%) statins vs. 2516 (3.4%) controls: RR 0.86, 95% CI 0.78–0.95, P = 0.004, I2 = 41%]. Due to concern that statins raise hemorrhagic stroke risk, we evaluated the number of hemorrhagic strokes reported in 14 RCTs enrolling 61 045 individuals.1,38,40,52,58,60,74,75,82,86,89,91,99,107 We found a low incidence of hemorrhagic strokes [267 (0.86%) statins vs. 310 (1.03%) controls: RR 0.86, 95% CI 0.73–1.01, P = 0.07, I2 = 0%], and our analysis indicated that statins did not increase the risk.

Impact of individual statins on CVD mortality

We assessed the impact of individual statins on CVD death. Figure 3 displays the geometric distribution of the RCTs (Figure 3). We excluded one trial of mixed statins in this analysis.97 Our analysis included 15 RCTs assessing atorvastatin (n = 29 931);8,45,46,64,65,67,72,83,93–96,104,105,108 nine assessing fluvastatin (n = 7383);47–52,68,73,79 eight RCTs evaluating lovastatin (n = 16 827);36,37,61,62,81,86,87,107 25 RCTs evaluating pravastatin (n = 51 011);38–44,59,63,66,69–71,74,84,88–92,98–100,103 five RCTs evaluating rosuvastatin (n = 30 245);57,58,76,82,85 and nine RCTs assessing simvastatin (n = 26 545).1,53–56,60,80,102,106

Figure 3.

Geometric distribution of network of evidence. Geometric distribution of included RCTs in mixed-treatment analysis. Each node in the network represents a drug treatment and each arm is weighted by the number of trials of that intervention versus the common control comparator.

Mixed-treatment comparison

We applied a mixed-treatment comparison adjusting for LDL-C changes. Table 4 displays the mixed-treatment comparisons between statins. We did not find a significant difference between any statins. A Bayesian probability estimate suggests that certain statins exert a minimally important difference over other statins (Table 5).

View this table:
Table 4

Mixed-treatment comparison for CV deaths

Treatment comparisonOR (95% Credible Interval)
Pravastatin vs. Control0.78 (0.65–0.93)
Atorvastatin vs. Control0.80 (0.65–0.96)
Fluvastatin vs. Control0.61 (0.41–0.88)
Simvastatin vs. Control0.74 (0.56–0.98)
Lovastatin vs. Control0.73 (0.43–1.22)
Rosuvastatin vs. Control0.88 (0.73–1.06)
Atorvastatin vs. Pravastatin1.02 (0.79–1.33)
Fluvastatin vs. Pravastatin0.78 (0.51–1.19)
Simvastatin vs. Pravastatin0.95 (0.68–1.33)
Lovastatin vs. Pravastatin0.94 (0.55–1.60)
Rosuvastatin vs. Pravastatin1.12 (0.87–1.46)
Fluvastatin vs. Atorvastatin0.76 (0.50–1.18)
Simvastatin vs. Atorvastatin0.93 (0.66–1.31)
Lovastatin vs. Atorvastatin0.91 (0.53–1.58)
Rosuvastatin vs. Atorvastatin1.10 (0.84–1.44)
Simvastatin vs. Fluvastatin1.22 (0.76–1.97)
Lovastatin vs. Fluvastatin1.20 (0.63–2.27)
Rosuvastatin vs. Fluvastatin1.44 (0.94–2.20)
Lovastatin vs. Simvastatin0.98 (0.55–1.76)
Rosuvastatin vs. Simvastatin1.18 (0.85–1.66)
Rosuvastatin vs. Lovastatin1.20 (0.69–2.09)
View this table:
Table 5

CVD mortality and the probability that each treatment is associated with lowest mortality

TreatmentAbsolute treatment effect (%)Probability that treatment is best

Adverse events

Data were available from 34 RCTs on first incident cancers recorded after randomization [median follow-up 3.9 years (interquartile range 2.6–4.9)].1,37,38,40,44,47,49,50,55,58–60,62,63,66,67,69–73,75–77,80,85,88–90,93,99–102 The incidence of cancers was not different between statin groups and control groups [3860 (4.5%) vs. 3703 (4.7%): OR 0.99, 95% CI 0.94–1.04, P = 0.76, I2 = 0%]. Rhabdomyolysis information was available from 35 RCTs enrolling a total of 135 243 individuals.1,8,37,38,40,44,47,50,55,57–60,62,64,65,67–70,73–77,80,81,83,84,89,93,95,99,101,104 We did not find a significant difference between groups [176 (0.25%) statins vs. 168 (0.25%) controls: OR 1.04, 95% CI 0.82–1.30, P = 0.73, I2 = 0%].

We evaluated incident diabetes available from 17 RCTs enrolling 111 003 individuals.1,40,44,58–60,64,70,74,76,88,99 62,63,85,89 When we evaluated new incident diabetes [2215 (3.8%) statins vs. 2048 (3.5%) controls: OR 1.09; 95% CI 1.02–1.16, P = 0.008, I2 = 26%], we found a significantly increased rate of diabetes. Finally, we examined the impact of statins on elevated AST from 23 RCTs and found a significant association (OR 1.12, 95% CI 1.03–1.22, P = 0.005); the impact of statins on ALT increases from 18 RCTs (OR 1.30, 95% CI 1.13–1.50, P ≤ 0.001, I2 = 0%); and the impact of statins on CK increases beyond normal from 19 RCTs (OR 1.07, 95% CI 0.78–1.46, P = 0.66, I2 = 29%). Figure 4 graphically displays the adverse event effect sizes.

Figure 4.

Adverse events associated with statin use in included trials.

Trial sequential monitoring

We applied the TSM evaluation to determine the strength of inference about statins in preventing our primary outcome, CV deaths (Figure 5). We display that based on events accumulating up to 2001, there is conclusive evidence of CV death protection across broad populations.

Figure 5.

Trial sequential analysis plot, CVD mortality.


Our meta-analysis demonstrated consistent benefits from LDL-lowering effects attributed to statin therapy. Our analysis demonstrates that statin therapy reduces major CVD events and all-cause mortality. Risks associated with statins appear limited to changes in biochemical profiles rather than clinical events, although there is now reason to explore the extent to which statins may contribute to increased incidence of diabetes. Reasons for possible increased risk of developing diabetes are poorly understood and genome-wide scans have not identified an association between genes involved in moderating LDL cholesterol and statin pharmacodynamics.109

There are several strengths to consider in our analysis. First, our study is the largest evaluation of statins to date. The findings of our analysis are remarkably similar to the Cholesterol Treatment Trialists Collaboration (CTTC), an individual patient data meta-analysis that has now published findings on LDL-lowering effects and outcomes among diabetic patients.5 While, we included 62 more RCTs than the CTTC analysis, our analysis is based on secondary data and we did not have access to individual level data. We applied rigorous searching, based on our ongoing statins database, and extracted data in such a manner as to reduce the risk of error. We considered the strength of evidence using trial sequential monitoring and found that clear evidence existed in 2001 of statins in CVD risk populations. We demonstrated a harmful effect associated with diabetes incidence, first highlighted in the Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER) trial.76 A recent meta-analysis of 13 RCTs, published while this manuscript was under review, found an estimate of RR 1.09 (95% CI 1.02–1.17).110 We compared our previous search with their included studies and added an additional seven studies to our review, an addition of three from theirs. We modified our manuscript as a result. Finally, we applied a mixed-treatment comparison to evaluate the relative effectiveness of specific statins. There are also limitations to consider in our analysis, mainly related to the limitations of the included studies. As with any meta-analysis that uses published manuscripts as the data source, it is possible that the original papers were poorly reported. Examples of this include where the papers may report on all-cause mortality, but not specific elements of the cause of death. Similarly, manuscripts may include a composite endpoint, for example, stroke death and non-fatal stroke, and it is occasionally impossible to extract data on the individual components of the composite.111 We included only trials evaluating statins with inert controls rather than head-to-head trials. Previous analysis of head-to-head trials have demonstrated that these trials evaluate dosing rather than the effectiveness of individual statins.112–115 Our analysis of cancer trials involved a median follow-up of 3.9 years (interquartile range 2.6–4.9). It is possible that longer periods of follow-up would find differing effects as cancers may take years to develop. Finally, while we found concerning evidence of increased diabetes incidence, this appears to be poorly monitored in clinical trials.

Conducting meta-analyses in cardiac trials presents an important methodologic challenge. Many cardiovascular trials use composite endpoints of their primary endpoints, whereby they combine various endpoints, but with little frequency of the same endpoints among trials. For example, a trial may report a primary composite outcome of all-cause mortality, MI and rehospitalization. Such an endpoint is useful for identifying a primary outcome unlikely to occur in a clinical trial, thus conserving power, but is unhelpful if the authors fail to report the individual outcomes across the composite symptoms. We have previously reviewed the role of composite outcomes in cardiovascular trials and found that composite outcomes can be misleading, as they place similar weight upon minor outcomes (such as rehospitalization) with major outcomes (such as all-cause mortality).111,116 We do not believe that composite outcomes of incoherent outcomes should be pooled in a meta-analysis if the individual components of the composite are not provided. In this study, we chose not to utilize the common composite endpoint of coronary heart disease death plus non-fatal MI.

Our mixed-treatment comparison failed to demonstrate significant differences between statins. Previous efforts to assess differences between statins have been based on smaller numbers of trials.2,17–19,117 We believe that, based on using all available evidence, generic versions of statins exert similar therapeutic effects as brand-label statins, a finding consistent across populations.2 Using a Bayesian probability framework, it is possible that lovastatin exerts a larger therapeutic effect. However, for several reasons, this inference may be weakened. Lovastatin trials had frequently smaller sample sizes and were conducted predominantly in the early 1990s, before other statin trials, when less complex patients could be entered into the trials, and when other concomitant therapies (i.e. blood pressure lowering, diuretics, anticoagulants) may have been in less frequent use compared to more recent standards of care.

We applied the trial sequential analysis strategy to determine the strength of inference of the cumulative data on the primary endpoint. Conceptually, the trial sequential analysis is analogous to determining whether a single large trial is sufficiently powered and has a sufficient number of events to warrant stopping a trial due to conservative expectations that a treatment effect is overwhelmingly beyond chance. Due to heterogeneity across included trial populations, treatments and methods, meta-analysis sample size considerations additionally need adjustment for variation across trials.35 Such adjustments are analogous to adjustments for variation across centers in a multi-center trial. In our analysis, we found compelling evidence of effectiveness at approximately 2001, when 31 trials had been conducted. We found that a large number of further inert-controlled trials have been conducted since that period. As with the mixed-treatment comparison, there are several considerations and possible explanations for this phenomenon. First, as drugs are developed and approved for use within comparatively uncomplex patient groups, such as secondary prevention CVD patients, trialists and drug developers seek further opportunities to reduce the risk of events within similar, but more complex patient groups, such as diabetic, renal and transplant patients. Drug companies may seek additional recommendations for their drug and seek increasingly complex patient groups, sometimes with disappointing results.58 Other drug companies may seek to gain access to the drug market and display evidence of similar efficacy within similar populations.

The important individual patient data meta-analysis of 14 large trials, conducted by the CTTC collaboration, found a 12% proportional reduction in all-cause mortality per mmol/l reduction in LDL-C (RR 0.88, 95% CI 0.84–0.91; P ≤ 0.0001).5 Our study examined LDL-C reductions on CVD mortality and found that every 10% absolute reduction in LDL-C was associated with a 2% RR in CVD mortality, consistent with the CTTC analysis. However, in our multivariable regression analysis, this effect was diminished. Meta-analyses by publication are frequently limited in assessing continuous outcome changes as we did not have the individual level data. It is possible that if we had the individual level data we could demonstrate larger treatment effects associated with LDL-C and possibly HDL-C changes.6 Our study demonstrates a clear and consistent benefit of statins, regardless of product. Many of the trials we included were conducted in resource-limited settings. In many of these settings, statins have been an exclusive therapy for wealthier individuals. As generic formulations are now available, and demonstrate consistent effects, there should be a greater effort to expand access to therapy among populations that may have previously been unable to access them.

In conclusion, statins play an important role in reducing clinically relevant cardiovascular outcomes, most likely due to reducing LDL-C levels. Current guidelines aim to establish target LDL-C reductions to improve a patients long-term reduction in clinical events. Given the clear benefits of statins, adherence to statin therapy should now be a major concern for physicians. Efforts to ensure adherence may be learned from other fields of chronic diseases health care, including reminder systems for patients and possibly even resource intensive strategies such as pill-counts and pharmacy refill assessments. There are few interventions in health care that offer such favorable outcomes and so improving access to treatment and adherence to therapy should be a prime concern for physicians and public health. As statin therapy moves into generic formulations, costs are reduced and this may open an opportunity to share these clinically important treatments with those who were previously excluded due to cost.


We thank Dr Nicola Cooper for her assistance in providing us with the WinBUGS code.

Conflict of interest: None declared.


View Abstract