Venous thromboembolic disease (i.e., pulmonary embolism and deep vein thrombosis) is a common cause of premature death and morbidity [1]–[3], yet our knowledge about how to safely prevent it is limited. During recent years, statins have emerged as one of the most effective treatments to reduce the burden of arterial cardiovascular disease worldwide [4]. Because of their remarkably good safety profile and declining costs, there has been some interest in their potential use for prevention of other conditions, such as venous thromboembolic events [5]–[7]. Venous and arterial thrombosis often co-occur [8],[9] and seem to share some common risk factors [10]. These epidemiological findings together with experimental evidence revealing novel mechanisms for the beneficial effect of statins unrelated to their low density lipoprotein (LDL) cholesterol lowering effect [11]–[13] have raised hopes that statins may also protect against venous thromboembolic events.

Until recently, clinical evidence for the effect of statins on venous thromboembolism was largely confined to non-randomised studies (with somewhat contradictory conclusions) [14],[15]. In 2009, however, secondary analyses of the JUPITER trial, in which 17,802 apparently healthy men and women were randomly allocated to receive either rosuvastatin 20 mg daily or matching placebo, provided direct randomised evidence that statin therapy might reduce the risk of venous thromboembolic events [5]. In this trial, allocation to rosuvastatin was associated with a reduction in the risk of venous thromboembolic events of 43% (95% CI 14%–63%) during a median 1.9 y follow-up [5], but this was based on relatively few patients with a venous thromboembolic event (34 versus 60), and so may partly (or even wholly) reflect the play of chance [16]. This uncertainty was reflected by calls for confirmatory evidence from other studies [17],[18]. An opportunity to obtain such evidence is provided by the routinely collected adverse event reports in existing statin trials. We therefore performed a meta-analysis of all larger scale trials of a statin versus control, and of a more intensive versus a less intensive statin regimen [19], which collected, but did not necessarily publish, information on the incidence of venous thromboembolic events during follow-up.

Out of 4,736 abstracts reviewed, 231 papers describing 112 trials were retrieved for further examination, of which 92 had both a follow-up duration of 6 mo or longer and had included 100 patients or more (Figure 1). Of these 92 trials, 47 trials comprising 54,643 participants and about 189,800 person-years' follow-up were excluded either because no venous thromboembolic events were recorded (i.e., zero events in both groups after interrogation of trial database) or because such information was not accessible to the trial investigators at the time. A further 16 trials comprising 7,846 participants and 18,200 person-years follow-up were excluded because there was no response to the data request. Our final database therefore included 29 trials comprising 146,353 participants and about 613,800 person-years follow-up. Of these, published information about venous thromboembolic events was available (at the time of our database search) from just two trials [5],[26], but unpublished information was provided by authors for a further 27 trials. 22 trials compared the effect of statin with control (105,759 randomised participants and 422,000 person-years follow-up) [26]–[46], and seven trials compared a more intensive with a standard dose statin (40,594 randomised participants and 191,000 person-years follow-up) [47]–[53], with no overlap between the two trial groups [39]. The characteristics of the 29 included trials are shown in Table 1. Risk of bias for individual trials is summarised in Table 2. Risk of bias for venous thromboembolic events was deemed low for all included trials.

The primary analyses were restricted to the 22 trials that compared a statin with a control regimen. In these trials, an episode of venous thromboembolic event occurred in 986 patients. Overall, there was no clear evidence that statin therapy reduced the risk of venous thromboembolic events (465 [0.9%] statin versus 521 [1.0%] control, OR = 0.89 [95% CI 0.78–1.01]; p = 0.08) (Figure 2). There was no evidence of heterogeneity in estimated effect size between the trials (heterogeneity χ²_21 =23; p = 0.34) but a moderate degree of statistical inconsistency between the trials could not be ruled out (I^2 =0%, 95% CI 0%–43%). Since it was the result from the JUPITER trial that motivated us to perform this meta-analysis, it could be argued that that result should be considered only as “hypothesis generating,” and that including it in the main analyses may have led to a summary point estimate, CI, and p-value that are appreciably biased [54],[55]. Excluding this trial, however, had little effect on the overall result (431 [0.9%] versus 461 [1.0%], OR 0.93; 95% CI 0.82–1.07; p = 0.32) (Figure 2). Virtually identical results were seen when the individual trial results were weighted by the 1-y LDL cholesterol difference (in all 22 trials: OR 0.90 per mmol/l LDL cholesterol reduction, 95% CI 0.81–1.00; p = 0.05).

In the seven trials that compared a more intensive versus a standard statin regimen, there was no evidence that higher dose statin therapy reduced the risk of venous thromboembolic events compared with standard dose statin therapy (198 [1.0%] versus 202 [1.0%], respectively, OR 0.98; 95% CI 0.80–1.20; p = 0.87) and there was no evidence that the effect varied within these trials (heterogeneity χ₆^2 =4.5; p = 0.61) (Figure 3). However a moderate to large degree of statistical inconsistency between the trials could not be ruled out (I^2 =0%, 95% uncertainty level 0%–61%). The effect estimates weighted for 1-y LDL cholesterol differences were similar (weighted OR 0.99, 95% CI 0.66–1.51, per 1 mmol/l LDL cholesterol reduction).

To assess a possible differential effect of statins (or higher dose statins) on pulmonary embolism and deep vein thrombosis, we estimated the effects on each outcome separately. There was no good evidence that the effect of statin therapy differed by the type of outcome (χ²_1 =0.4, p = 0.54 for heterogeneity for the trials of statin versus control, and χ²_1 =3.4, p = 0.06 for heterogeneity for the trials of more intensive versus standard dose statin) (Figure 4). Nor was there evidence that either statin therapy or higher dose statin therapy significantly reduced the risk of either type of outcome in isolation (deep vein thrombosis: 266 versus 311, OR 0.85 [99% CI 0.69–1.06] for trials of statin versus control and 88 versus 106, OR 0.83 [99% CI 0.57–1.21] for trials of more versus less statin; pulmonary embolism: 205 versus 222, OR 0.92 [99% CI 0.72–1.19] for trials of statin versus control and 127 versus 107, OR 1.19 [99% CI 0.84–1.68] for trials of more versus less statin) (Figure 4).

In trials that exclusively studied patients with no previous history of cardiovascular disease [5],[27],[32],[37],[43]–[45], statin therapy was associated with a significant 38% reduction in the risk of venous thromboembolism (OR 0.62, 99% CI 0.41–0.94), which appeared to differ from the non-significant 4% reduction (OR 0.96, 99% CI 0.80–1.15) seen in other trials (p-value for heterogeneity between two groups of trials = 0.01) (Figure 5). However, there were no significant differences between the two groups of trials when the hypothesis-generating JUPITER trial was excluded from these analyses (p-value for heterogeneity between two groups of trials after exclusion of JUPITER = 0.20).

There was no good evidence that the effect of statin therapy on venous thromboembolism varied depending on type of statin studied (χ²_5 =10.7, p = 0.06) (Figure S1), particularly when the results from the JUPITER trial were excluded (χ²_5 =5.8, p = 0.32) (Figure S2).

In this study, we gathered information from over 100,000 participants in 22 randomised trials of statin therapy versus control and 40,000 participants in seven randomised trials of intensive versus standard dose statin therapy, which together involve about 14 times as many venous thromboembolic events as previously reported in the JUPITER study [5]. Overall, the results from this meta-analysis do not support the suggestion that statins [5] (or higher doses of statins [19],[23]) reduce the risk of venous thromboembolic events substantially, although a more moderate reduction in risk up to about one-fifth cannot be ruled out.

Our meta-analysis has several strengths compared to previous reports. Compared with previous publications that are either based on non-randomised comparisons or have been based on one single randomised trial with limited events, our findings are based on substantially more first events from randomised trials and, importantly, include previously unpublished as well as published data. Inclusion of unpublished data helps avoid the well-documented problems caused by the preferential publication of positive findings (i.e., publication bias) [16],[56]. However, there is a perception among some that such unpublished data may be of inferior quality compared with data already in the public domain, despite the lack of evidence supporting such a view [21],[57]. In our study, reports from all trials had previously been subject to external peer review and the risk of bias in these trials for the outcomes of interest was judged to be low in our assessments (Table 2). Most of the trials had a double-blind design (i.e., trial participants and investigators were unaware of the treatment allocation) and, perhaps more importantly, venous thromboembolic events had been collected routinely as part of the safety and efficacy monitoring done in each trial (at which point there were no specific hypotheses related to venous thromboembolic events) [26]. The use of routine unadjudicated events from these properly designed and conducted randomised trials is also unlikely to have resulted in any substantial biases because any under- or over-reporting of events (or, for that matter, lack of independent confirmation) would be expected to affect both treatment arms equally [20],[58],[59]. While systematic over-reporting of events unaffected by statin therapy would tend to bias treatment effects towards the null, the magnitude of the bias would again likely be small because OR estimates are surprisingly robust to such errors. For example, even if as much as one-quarter of the reported venous thromboembolic events in our study were “false” events that were unaffected by statin treatment (and hence equally distributed by treatment group), the estimate of the OR among the trials of statin versus control would have been expected to change from 0.89 (shown in Figure 2) to 0.86 as result of removal of the false events. Confirmation of diagnosis of symptomatic venous thromboembolism in routine clinical settings already depends on objective measures such as imaging and biochemistry. It therefore seems implausible that over-reporting of unrelated outcomes in these trials of the magnitude needed to result in substantial bias in OR estimates would have occurred. This is further supported by a recent study in which treatment effects based on adjudicated versus unadjudicated vascular events in ten trials involving over 9,000 events were compared, and found to be virtually identical (OR for reported versus adjudicated outcomes 1.00, 95% CI 0.97–1.02) [59]. Indeed, if anything, one might expect the reported outcomes in our study to be more consistent with “real-world” settings in effectiveness studies and, therefore, more relevant to policy and practice.

Another concern about inclusion of unpublished data in meta-analysis is incompleteness in trials gathered, which may itself become a source of bias [57]. In our study, we sought to obtain data from all larger trials, including repeated attempts at contacting study authors. However, we might still have missed relevant event information from at least 16 further trials where there was no response from investigators (in another 47 trials, either no events occurred or events were not available to the investigators themselves so would therefore not constitute a source of bias). However, these 16 trials would have contributed only about an additional 18,000 person-years of exposure (compared to about 600,000 person-years of exposure in the 29 included trials). In addition, if an important reduction in venous thromboembolism had been observed in any single trial for which data were not made available, it seems likely that the result may already have been published (as most of these trials were completed several years ago), and hence identified by our literature search. Thus, the relatively small amount of missing information is unlikely to have resulted in any material change to our conclusions. This is illustrated by a recent similar study that assessed the effect of statin on atrial fibrillation [20]. Additional data on this outcome from a larger statin trial became available only after the meta-analysis that combined both published and unpublished data were published and this new information was entirely consistent with the pooled evidence [60].

Nonetheless, it should be recognised that the total number of first events in our meta-analysis (about 1,000 first reports of venous thromboembolism) is still relatively modest, as reflected by the confidence intervals that are consistent with anything from no effect to a real reduction in risk of about one-fifth. This makes the results of any subgroup analyses particularly unreliable, and so they need to be interpreted with due caution [54]. Consequently, this study cannot reliably investigate the potential differential effect of statins (or higher dose statins) in certain subgroups of patients based on important baseline characteristics such as prior history of venous thromboembolic events and use of anti-platelet or anti-coagulant therapy, or by the underlying cause of such events (e.g., provoked by cancer or other events versus unprovoked events).

Similarly, the suggested heterogeneity in effects on venous thromboembolism by prior vascular disease (Figure 5) is far from definitive. While it may be considered biologically plausible for there to be a proportionally smaller effect among people with pre-existing vascular disease or other chronic conditions (if the risk of venous thromboembolic events were less amenable to statin therapy as the proportion of events that are provoked by causes such cancer or immobility increases), the difference seen between trials that included healthier populations and those that did not was not statistically significant when the hypothesis-generating JUPITER trial was excluded (Figure 5). The same observation was made with regards to type of statins used (Figures S1 and S2). Thus, to demonstrate such differences (if one were to exist) would require further evidence from randomised trials.

Might the results be biased in favour of statin therapy owing to the interdependency of venous and arterial thrombosis? Statins reduce the risk of arterial vascular events (including hospitalisations) substantially [4], and more intensive regimens produce further definite reductions in risk [61]. One might therefore expect the number of venous thromboembolic events (some of which are related to arterial cardiovascular events [8],[9]) to have been greater among patients allocated placebo/less intensive statin regimens. While this may be plausible, it seems likely that the incidence of venous thromboembolic events following a cardiovascular event would be low (for example only six out of the 94 events in JUPITER occurred following a cardiovascular event) and so unlikely to have introduced any substantial bias in favour of statins.

In conclusion, this study provides a more detailed assessment of the potential effects of statins (or higher dose statins) on venous thromboembolic events than has previously been possible. We were unable to confirm the large proportional reduction in risk suggested by some previous studies. However, a more modest but perhaps clinically worthwhile reduction in venous thromboembolic events in some or all types of patient cannot be ruled out.