In January 2003, the Justification for the Use of statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER) investigators began to enroll 17,802 men and women with no evidence of cardiovascular disease and normal to low levels of low-density lipoprotein (LDL) cholesterol into a controversial trial testing whether subjects with enhanced inflammatory responses might benefit from aggressive statin therapy. At study entry, the mean LDL cholesterol level for enrolled participants was just above 100 mg/dl, nearly half had Framingham scores of ≤10%, and none were considered candidates for statin therapy under any current set of prevention guidelines. However, all trial participants had levels of the inflammatory biomarker high-sensitivity C-reactive protein (hsCRP) ≥2 mg/L, putting them at substantial risk for future cardiovascular disease.
As the JUPITER investigators have since reported in a series of rigorously peer-reviewed reports published in the New England Journal of Medicine , The Lancet , the Annals of Internal Medicine , Circulation , the Journal of the American College of Cardiology , and The American Journal of Cardiology , random allocation to rosuvastatin 20 mg compared to placebo within the trial population resulted in a 54% reduction in myocardial infarction (p = 0.0002), a 51% reduction in ischemic stroke (p = 0.004), a 46% reduction in need for bypass surgery or angioplasty (p <0.0001), a 43% reduction in venous thromboembolism (p = 0.002), and a 20% reduction in all-cause mortality (p = 0.02). For the JUPITER primary end point of major arterial vascular events, the p value for the overall 44% relative risk reduction was <0.0000001 ( Table 1 ).
| End Point | Randomized to Rosuvastatin (n = 8,901) | Randomized to Placebo (n = 8,901) | HR (95% CI) | p Value | ||
|---|---|---|---|---|---|---|
| n | Rate ⁎ | n | Rate ⁎ | |||
| Primary trial end point | 142 | 0.77 | 251 | 1.36 | 0.56 (0.46–0.69) | <0.00001 |
| Nonfatal myocardial infarction | 22 | 0.12 | 62 | 0.33 | 0.35 (0.22–0.58) | <0.00001 |
| Any myocardial infarction | 31 | 0.17 | 68 | 0.37 | 0.46 (0.30–0.70) | 0.0002 |
| Nonfatal stroke | 30 | 0.16 | 58 | 0.31 | 0.52 (0.33–0.80) | 0.003 |
| Any stroke | 33 | 0.18 | 64 | 0.34 | 0.52 (0.34–0.79) | 0.002 |
| Arterial revascularization | 71 | 0.38 | 131 | 0.71 | 0.54 (0.41–0.72) | <0.0001 |
| Hospitalization for unstable angina | 16 | 0.09 | 27 | 0.14 | 0.59 (0.32–1.10) | 0.09 |
| Arterial revascularization or hospitalization for unstable angina | 76 | 0.41 | 143 | 0.77 | 0.53 (0.40–0.70) | <0.00001 |
| Myocardial infarction, stroke, or confirmed cardiovascular death | 83 | 0.45 | 157 | 0.85 | 0.53 (0.40–0.69) | <0.00001 |
| Confirmed cardiovascular death | 35 | 0.19 | 43 | 0.23 | 0.82 (0.52–1.27) | 0.37 |
| Sudden death | 16 | 0.08 | 25 | 0.13 | 0.64 (0.34–1.20) | 0.16 |
| Total mortality | 198 | 1.00 | 247 | 1.25 | 0.80 (0.67–0.97) | 0.02 |
In these publications, we have also shown that the relative benefits of rosuvastatin in those with elevated hsCRP are as large for women as for men and that rosuvastatin is highly effective in the elderly, in non-Caucasians, and in those with reduced renal function. Furthermore, we have shown that the number needed to treat (NNT) for those with low levels of LDL cholesterol but elevated hsCRP is if anything smaller than the NNT for the treatment of either dyslipidemia or high blood pressure in comparable primary prevention patients. Others have shown that the cost-effectiveness of the JUPITER trial is not only similar to that already accepted for the treatment of hyperlipidemia but for many patient subgroups is in fact cost saving to society. On the basis of these data, national guidelines in several countries have endorsed the use of statin therapy in primary prevention in those with elevated levels of hsCRP, and the United States Food and Drug Administration has formally expanded the labeling of statin therapy to include those with elevated hsCRP and ≥1 additional risk factor, even if LDL cholesterol levels are already low.
Given the broad acceptance of these data in academic publications, national guideline committees, and federal regulatory agencies, it came as a surprise to the JUPITER Steering Committee to read 3 reports published simultaneously in the Archives of Internal Medicine that substantially mischaracterized and misinterpreted the trial data. These reports—an “original investigation,” a “special article,” and an editorial —together took up 12 pages of print. One or more of these reports described the trial as “flawed” on the basis of the authors’ perceptions that there were errors in the trial design; that early stopping of the trial by its independent Data and Safety Monitoring Board was inappropriate; that there were inconsistencies in the trial data with regard to mortality, statin therapy, and hsCRP; that Bayesian analyses had been incorrectly considered; and that the study results were a result of “pervasive” industry influence.
To address these issues, we (PMR, the JUPITER principal investigator, and RJG, the academic study statistician) wrote an extended response intended for publication in the Archives of Internal Medicine . However, an editorial decision was made at that journal to limit our response to a brief letter to the editor, a decision we believed provided insufficient opportunity to address the complexity of the issues raised. We thus present our full response here, with hopes that clinicians will judge for themselves the validity of the trial and in turn make independent decisions regarding the utility of statin therapy for their primary prevention patients who do not have overt hyperlipidemia yet who are at risk for heart attack and stroke at least in part because of elevated levels of hsCRP.
The “Flawed” JUPITER Trial Design
In the most critical of the 3 reports, de Lorgeril et al put aside 20 years of randomized placebo-controlled statin trial evidence and stated that “the results of recent cholesterol-lowering drug trials on decreasing morbidity and mortality among persons with or without coronary heart disease have been consistently negative.” It is only by disregarding overwhelming previous evidence of benefit of statins that de Lorgeril et al can make the argument that JUPITER, also a positive trial, is a “unique” outlier that requires “critical reappraisal.” De Lorgeril et al then went on to claim that the published data from JUPITER “do not support the use of statin treatment for primary prevention of cardiovascular disease” and show “no differences between the 2 groups on most objective criteria,” despite the highly significant 44% reduction in the trial primary end point (p <0.0000001).
We find it of interest that de Lorgeril et al stated without reservation that “the trial is flawed,” when JUPITER addressed the core clinical question posed using the most robust and widely accepted form of clinical scientific inquiry available, the randomized, placebo-controlled, double-blind trial. In total, the JUPITER trial enrolled 17,802 participants, half of whom were randomly allocated to rosuvastatin 20 mg/day and half to matching placebo. Reflecting the investigators’ interest in evaluating patient groups underrepresented in previous statin trials, JUPITER enrolled 6,801 women, 5,695 participants aged ≥70 years, and >5,000 non-Caucasians. Contrary to the characterization of de Lorgeril et al (and as noted at the Food and Drug Administration hearings), the trial used high levels of scientific rigor, achieved exemplary participant follow-up, used rigorous end point adjudication, prespecified a highly conservative monitoring plan in the charter of its Data and Safety Monitoring Board, and followed a prespecified analysis plan conducted exclusively on an intention-to-treat basis.
The JUPITER trial did not explore the potential role of statin therapy in those with LDL cholesterol <130 mg/dl and hsCRP <2 mg/L, and thus data from the trial cannot be generalized to this group. The only available data in such subjects derive from post hoc analyses of the Air Force/Texas Coronary Atherosclerosis Prevention Study (AFCAPS/TexCAPS), in which such subjects were found to have very low absolute event rates and no clinical benefit from random allocation to lovastatin, despite substantial reductions in LDL cholesterol. Subsequent post hoc analyses of the Controlled Rosuvastatin Multinational Trial in Heart Failure (CORONA) of rosuvastatin in those with heart failure have also observed benefit in those with hsCRP ≥2 mg/L but no benefit in those with lower levels of hsCRP. Observational evidence from many sources, most recently from the Atherosclerosis Risk in Communities (ARIC) study, also confirm that absolute risks are small in patients with low LDL cholesterol and low hsCRP but that absolute risks are high (and missed by traditional Framingham scoring) in those with low LDL cholesterol but elevated hsCRP. The decision by the JUPITER investigators to rely on hsCRP as a marker for high risk has been validated by overview data from 54 prospective cohort studies in which the magnitude of relative hazard associated with a 1-SD increase in hsCRP was found to be greater than that of a comparable SD increase in either total or high-density lipoprotein cholesterol.
Early Stopping of the JUPITER Trial and Bayesian Data Interpretation
In their recent critiques, de Lorgeril et al, Kaul et al, and Green all suggested that early stopping of JUPITER by its Data and Safety Monitoring Board substantially overestimated trial benefits. Furthermore, in their commentary, de Lorgeril et al suggested that the JUPITER trial’s stopping rules were not prespecified and that the end point used to define those stopping rules was not defined.
Quite to the contrary, a comprehensive charter was agreed on between the JUPITER investigators and the members of the independent Data and Safety Monitoring Board before randomization of any JUPITER participants. That charter describes not only a requirement that any consideration of early stopping be based on statistically extreme findings in accordance with a conservative and principled O’Brien-Fleming boundary based on a Lan-DeMets spending function but also that early stopping “would require proof beyond reasonable doubt that prolonged use of rosuvastatin was clearly indicated or clearly contra-indicated for all or some specific types of patients.” That charter further clearly indicates that decisions for early stopping were to be based on analyses of the primary trial end point.
For the record, the prespecified statistical stopping boundary for the JUPITER trial was crossed at the first formal efficacy analysis in September 2007, when the trial had accrued 219 confirmed primary end points (nominal hazard ratio [HR] associated with rosuvastatin 0.59, 95% confidence interval [CI] 0.46 to 0.78). However, taking a conservative stance, the monitoring board determined that the evidentiary criteria specified in its charter of the additional requirement for “proof beyond reasonable doubt” were not yet met and voted to continue the trial. At its next meeting in March 2008, the trial had accrued 328 confirmed primary end points (nominal HR 0.57, 95% CI 0.46 to 0.72). At that point, the board voted unanimously to recommend early stopping on the basis of the overall strength of the evidence, the evidence provided in the incremental data accrued since its previous review, the consistency of findings across all subgroups, the evidence for secondary end points, as well as considerations regarding the considerable previous evidence demonstrating the efficacy and safety of statin therapy in the many previous trials of secondary and primary prevention.
Kaul et al and de Lorgeril et al are correct in that there is controversy within the clinical trials community regarding the best methods for data and safety monitoring boards to consider trial outcomes in an unbiased and independent manner. Unprincipled stopping of a trial early on the basis of an apparently large treatment benefit can, in theory, be associated with an overestimate of effect. However, Kaul et al, de Lorgeril et al, and Green failed to address the appropriateness of continuing a trial after the main study question has been answered definitively; trialists and study monitors have an ethical responsibility to stop experiments and inform their participants as well as society when uncertainty and equipoise no longer hold with respect to the question under evaluation. Moreover, all 3 commentators failed to cite evidence from simulation studies, which consistently indicate that stopping early on the basis of conservative guidelines as was done in the JUPITER trial does in fact yield valid treatment effects.
Irrespective of this, a fundamental issue not considered by Kaul et al, de Lorgeril et al, or Green is what the magnitude of effect early stopping of the JUPITER trial might have had and whether that effect is large enough to alter clinical interpretation. Evaluation of studies of early stopping suggests that the potential overestimates of benefit in trials stopped early vary on the basis of the total numbers of end points accrued, whether a strict set of guidelines is adhered to, and whether stopping rules are based only on statistical considerations or, as was the case in JUPITER, go beyond that to include additional levels of evidence, including consideration of previous trial data. With specific regard to the JUPITER trial, an independent statistical analysis performed by the Food and Drug Administration in its formal review of the trial data concluded that any potential overestimate of relative risk reduction due to early stopping in JUPITER was likely to be ≤1%. Hence, even if early stopping did bias the JUPITER results, the overall estimate of a highly significant 44% reduction in relative risk for the trial primary end point would not be reduced in any clinically meaningful manner.
Finally, in their discussions of the effects of early stopping, both Kaul et al and de Lorgeril et al incorrectly suggested that the large magnitude of benefit observed in JUPITER was “unexpected” on the basis of previous data and that the Data and Safety Monitoring Board did not place the observed effects in context of other statin trials. Kaul et al went further and performed a post hoc Bayesian analysis in which they compared the results of JUPITER to a second rosuvastatin trial conducted in patients with congestive heart failure. This is an unorthodox approach, because (1) those with heart failure are unlikely to share the same pathophysiology as those in a primary prevention settings, and (2) a more principled Bayesian approach would take into account the substantial body of previous statin data that in fact directly informs patient care. Most relevant, as described in several previous meta-analyses of statin studies that predate JUPITER, each percentage reduction in LDL cholesterol is approximately associated with a 1% reduction in relative risk. On that basis, rosuvastatin 20 mg, which led to a 50% reduction in LDL cholesterol in the trial, would be anticipated on an a priori basis to reduce by approximately half the rate of cardiovascular events, a prediction close to the actual effect observed. Furthermore, the 44% relative risk reduction in JUPITER was also virtually identical to the relative risk reduction observed in our previous retrospective analysis of participants with low LDL cholesterol and high hsCRP in the AFCAPS/TexCAPS trial. A Bayesian perspective along these lines, as was adopted by the JUPITER independent Data and Safety Monitoring Board, would have found no issues with the final JUPITER trial report.
Thus, contrary to the views expressed, there is no credible evidence that early stopping of the JUPITER trial had anything more than a marginal effect on the true estimates of efficacy. If anything, the magnitude of benefit in the trial was increasing over time, not decreasing ( Figure 1 ). Figure 1 also demonstrates nominal statistical significance of rosuvastatin compared to placebo within 6 months of trial initiation. These data are valuable for patient care, because they indicate (1) that the certain benefits of therapy accrue almost immediately after starting treatment and (2) that there is no evidence of a reduced effect over time.
