Bearish AMRN Report: by Medical Research Collaborative, LLC (sample pages)
Following the publication of the REDUCE-IT trial results in November of 2018[1] and the issues that came to light therein, Amarin Corp. (NASDAQ: AMRN) posted a series of apologetics on their website.[2] We found many of these to be lacking in scientific rigor.
For example, the results of a post-hoc subgroup analysis were offered to explain away the observed increase in LDL-C in the placebo group (which was first suggested by the REDUCE-IT trial investigators as possibly due to dosing with mineral oil) and the potential for such to have exacerbated the control event rate.[3]
According to the sponsor (Amarin Corp.), the analysis suggests a lower risk for icosapent ethyl (IPE), aka "Vascepa," group compared with placebo group, regardless of whether or not there was an increase in LDL-C level among patients in the placebo group. The sponsor argues this is proof that statin absorption was either not affected by dosing with mineral oil, or even if it was, had no material impact on the risk of atherosclerotic cardiovascular disease (ASCVD) events, since even those placebo group subjects with no change or a decrease in LDL-C (hence, we assume their statin meds were working just fine) still performed worse than IPE group.
However, the critical drawback to this subgroup analysis is that it was not pre-specified, and therefore fails to take into account the baseline risk of the placebo subgroup that showed “no change/decrease in LDL-C” relative to the baseline risk of the placebo subgroup that showed an “increase in LDL-C” relative to the IPE group. As such, it remains uninterpretable, given that the baseline characteristics (major prognostic factors) between the subgroups explored are not equivalent. An adjusted analysis was not provided either, which might have induced at least some confidence in the finding.[4]
The sponsor was also not forthcoming on the number of subjects in each placebo subgroup (number at risk info is conspicuously absent from both charts). Upon closer inspection, it appears that the “no change/decreased LDL-C” subgroup was much smaller than the “increased LDL-C” subgroup, inflating inter- and intra-group variability (a negative consequence of low statistical power[5]); zooming-in on the original plots that they provide[6] roughly reveals the upticks on the curve:
By our count, it appears to be around 75 key secondary endpoint events for the “no change/decreased LDL-C” placebo subgroup, which leaves about 530 events for the “increased LDL-C” placebo subgroup (in the NEJM paper they enumerate 605 total key secondary endpoint events in placebo group). This is also probably why they avoid any mention of p-values associated with the stated HRs.
Comparing time-to-event rates between these placebo subgroups and the IPE group and drawing conclusions based on such is therefore flawed, due to a lack of equivalent baseline risk between groups and the potential for exaggeration of observed effects from the smaller sample. The fact that the second chart above shows a marked increase in the incidence of “hard” major coronary events (MCE) in the “no change/decreased LDL-C” subgroup beyond 3-years compared to the “increased LDL-C” subgroup (a counter-intuitive observation) infers that the former is much more likely to be a higher risk group than either of the other two (i.e. higher percentage of secondary prevention patients, higher overall BMI, greater prevalence of male subjects and smokers, higher overall levels of atherogenic and inflammatory markers, etc.). The hierarchy of baseline risk by group would then be (from highest to lowest): “no change/decreased LDL-C” group > IPE group > “increased LDL-C” group. It would be expected, then, for the "no change/decreased LDL-C" group to perform worse than the IPE group, as they are a higher baseline risk group (and apparently much smaller in number). We cannot reasonably conclude anything more than this without grossly overinterpreting these data.[7]
Thus, the analysis fails to prove: (a) that statin malabsorption did not occur in placebo group subjects, or (b) if statin malabsorption did occur, it did not meaningfully impact the placebo group’s performance. Post-hoc subgroup analyses such as these—especially when unadjusted—are innately unreliable, and appropriate for hypothesis generating only.[8]
On the second point, regarding the JELIS trial,[9] there are major limitations to the study that make generalizability difficult. It was open-label, and open-label trials are notorious for reporting exaggerated treatment effects;[10],[11] and the only significant individual finding from the trial was a reduction in unstable angina, with well over half of the primary endpoint events comprised of these (193 vs 147 events, p=0.014). Therefore, there is an increased likelihood the results were due to performance bias and/or detection bias, resulting from changes in patients’ behavior, physicians’ treatment, or event ascertainment.[12],[13],[14] It was also comprised of a 100% Japanese population of patients with poorly controlled LDL-C (182 mg/dL at baseline); nearly 70% were women; all were given very low dose statin regimens, even for an all-Asian population (pravastatin 10 mg/day or simvastatin 5 mg/day), and around 27% of the subjects discontinued statin use during the study (whether or not more control group subjects discontinued statin use earlier than ethyl ester EPA (EPA-E) group subjects was not disclosed).
Importantly, in JELIS, the MCE composite endpoint itself was different from that in REDUCE-IT. It included revascularization and hospitalization for unstable angina, but did not include strokes. In order to make an apt comparison with the REDUCE-IT trial—ignoring for a moment the numerous differences in populations and background therapies—the same composite endpoint should be used. When this is done, with strokes included, the RRR in JELIS decreases sharply from 19% to just 11.5%, insignificant from 914 events ([324 vs 262 non-stroke MCE; HR 0.81] + [162 vs 166 stroke events; HR 1.02] = blended HR 0.885). Comparing the reported 19% RRR in JELIS with the 25% RRR in REDUCE-IT is therefore misleading.
Not only this, but there was no perceived effect on coronary death (HR 0.94, p=0.81) in JELIS, and all-cause mortality trended worse for treatment group (265 vs 286 events, HR 1.09, p=0.33).
It can only be said that in this open-label, 100% Japanese, 70% female-gender study, 1.8 g/d EPA-E (aka “IPE”) significantly reduced unstable angina in subjects not optimally treated, with the potential for bias to have overinflated the result. JELIS is not a trial that reliably shows IPE can significantly reduce the risk of ASCVD events, particularly not in a patient population similar to that of REDUCE-IT.
Furthermore, a positive result from one study is often not reproduced in later trials. This is one reason that FDA normally requires at least two adequate and well-controlled studies as proof of efficacy and tolerability.[15]
For example, the GISSI-Prevenzione trial, which tested 1 g/d omega-3 ethyl esters (OM3-E) vs control in over 11,000 Italian post-MI subjects, demonstrated a significant 15% RRR in the primary MCE composite, which included all-cause mortality, MI and stroke. The study showed a 20% reduction in death from any cause and a 30% reduction in CV mortality (each also significant) compared with control. There was also a 44% reduction in sudden death (p<0.01). This led to Omacor 1 g/d (aka “Lovaza”) being indicated for secondary prevention (following MI) in major EU member states. The AHA also began recommending 1 g/d omega-3s for secondary prevention,[16] though there was no formal FDA approval for that indication. However, subsequent placebo-controlled trials did not reproduce the positive findings,[17] and as contrary evidence continued to mount,[18] the EMEA eventually decided to retract their approval.[19]
What happened here? Was the result due to the open-label design? It does not seem possible that that could have affected an endpoint like mortality. Was it due to differences in background therapy between GISSI-P and more modern trials? That could be a plausible explanation, as most of the subjects, especially early on in the study (when much of the benefit in mortality was realized), were not on statin therapy. It is possible that the reduction in sudden death in particular was the result of antiarrhythmic and antifibrillatory properties that have been ascribed to omega-3s.[20] These same properties are also ascribed to statins, particularly at higher doses.[21] There were likely untreated targets present due to lack of statin therapy in GISSI-P subjects. Later trials were conducted in an era where all secondary prevention subjects are given moderate to high-intensity statins following an MI (as well as numerous other treatments and cardiac medications[22]). JELIS enrolled all of its subjects between 1996 and 1999, similar to the GISSI-P trial (1993 – 1997).[23] Would the open-label JELIS trial results involving 1.8 g/d of EPA-E repeat in a placebo-controlled test with optimally treated modern-day subjects of Japanese descent randomized over 20 years after the fact? Perhaps not.
Generalizability of study results is crucial in informing drug regulation.[24] 1 g/d OM3-E offers no additional benefit to secondary or primary prevention patients already on extensive background therapy, even though it proved effective in the GISSI-P trial. The JELIS trial subjects and REDUCE-IT trial subjects are too different and were too differently treated to confer anything from the results of one trial to that of the other.
Another apologetic presented by proponents of icosapent ethyl (IPE), aka “Vascepa,” is in comparing studies such as ODYSSEY and FOURIER that demonstrated a regression to the mean in LDL-C with the REDUCE-IT and ANCHOR trials, arguing that the changes in markers seen in the mineral oil (MO) placebo groups of the latter two trials are also regressions to the mean. However, the regressions seen in other trials often involve one or two parameters only, and impact both treatment and control groups. By contrast, in the REDUCE-IT and ANCHOR trials, we find:
• Highly significant elevations in every atherogenic lipid/ lipoprotein and inflammatory marker tested (11 total in ANCHOR) in MO placebo groups of both trials; • the absence of any of these changes in those randomized to IPE groups in either trial; and • the confirmation of these effects in two robustly sized studies—even to similar degrees.
As elaborated on previously, a lead-in period helps prevent regressions to the mean from impacting data, which the ANCHOR trial had—and yet, all eleven atherogenic/inflammatory markers showed sharp increases in its placebo group. What has so far been reported from the REDUCE-IT trial is a repeat (and thus, confirmation) of the same phenomenon. The baseline characteristics, including background therapies, of subjects in both trials are extremely similar.
In our view, the only potentially relevant comparison in seeking to prove that the highly significant increases in atherogenic and inflammatory markers in the MO placebo groups in ANCHOR and REDUCE-IT could have been a mere regression to the mean would have to be another trial with an extensive lead-in period that also showed abrupt changes in markers in subjects of approximately equal baseline characteristics. The EVOLVE trial was the closest that we could find.
In EVOLVE, which had a similar lead-in stabilization period as ANCHOR, there was noted a 10% increase in median LDL-C in the olive oil placebo group.[25] However, other atherogenic markers were reduced or did not significantly change in this group. Also, the least squares mean value recorded in the study (which reflects changes across the entire arm) showed an insignificant 3% increase in LDL-C in the placebo group.[26] It seems then that the elevations in LDL-C primarily occurred in those placebo group subjects with baseline values near the median, and little change in the rest.
There could also have been some impact on LDL-C from the modest TG and VLDL-C lowering effect of olive oil (-10% and -11% median change, respectively) by mechanisms similar to those by which DHA can cause an increase in LDL-C along with a decrease in TG and VLDL-C—especially in those with very high triglycerides at baseline.[27] An increase in LDL-C from a triglyceride-lowering therapy is much more likely to occur when the baseline TG levels of subjects are very high.[28] In ANCHOR, the median baseline TG level was ~260 mg/dL, and in EVOLVE it was >700 mg/dL, and so an increase in LDL-C from a TG-lowering therapy would be more likely to occur in EVOLVE than in ANCHOR. The discrepancy in changes in median and mean TG values (-10% vs -4.3%, respectively) from baseline in the olive oil placebo group in EVOLVE also lends credence to the possibility that such a phenomenon affected those with values near the median more than the rest of the group. Thus, the comparison between EVOLVE and the ANCHOR/REDUCE-IT trials breaks down on multiple levels.
The adverse impact on all atherogenic/inflammatory markers tested in ANCHOR, and confirmed in a separate study with patients of equivalent background therapy and characteristics (REDUCE-IT), is much more reminiscent of a treatment effect than a regression to the mean.
Lastly, apologists have pointed out the decrease in LDL-C in some of the statin-treated subjects in the placebo arm of the MARINE trial, causing a net reduction in LDL-C in this subgroup, despite being given 4 g/d MO. However, the number of patients in this subgroup analysis is small (n=18 in the 4 g/d MO arm). As one FDA reviewer noted when examining the same data,[29]
Whether mineral oil affects statin absorption has not been formally tested to our knowledge. The applicant submitted data regarding patients who were taking concomitant statin therapy in the MARINE trial and who were randomized to the mineral oil group. Only 18 patients in the mineral oil group were taking a statin. The median percent change in LDL-C was -8% in the statin-treated mineral oil group, with large variability (Q1 -36.0%, Q3 +30.8%); the median change was 0% in LDL-C among the 57 patients not taking statins in the mineral oil group. The applicant contends that if mineral oil reduced statin exposure, then LDL-C should have increased after 12 weeks of treatment, not decreased. While the reduction in LDL-C in this group is somewhat reassuring, the small number of statin-treated patients and the large intra-subject variability do not allow definitive conclusions from this subgroup.
Relatively few patients can significantly impact data from a collective small group (especially when considering the median percent change only). Potential reasons for incidental disparities in values from baseline to end of treatment are numerous, such as cessation or reduction in dose of allowed therapies—or the opposite; adding therapies, including supplements, that may affect other prescribed therapies, increasing or decreasing their potency; changes in diet and exercise habits; and/or any of the aforementioned while also taking their statin medication and mineral oil placebo at far removed times of the day. Three or four subjects out of the 18 analyzed that were on statins and in the 4g/d MO placebo group adopting any of the above could completely shift the subgroup’s statistics (the FDA reviewer noted that some of the 18 subjects saw a >30% increase in LDL-C while some saw a >30% decrease in LDL-C). Thus, the subgroup is too small and the variability of data too high to reliably base any conclusions on.
However, later, data were reported on changes in LDL-P and non-HDL-C parameters as determined by analysis of blood samples from all subjects in the MARINE trial, and what was observed appears to confirm, rather than disprove, the adverse MO impact hypothesis:[30]
Increases in LDL-P and non-HDL-C are highly predictive of increased prevalence of ASCVD events, even when LDL-C remains unchanged/low.[31],[32] The above data show that median LDL-P concentration increased by 14.4% and 12.6% (while mean non-HDL-C increased by 19% and 5%) by week 11 in the 4 g/d MO group and 2 g/d IPE group (who concurrently took 2 g/d MO), respectively. Thus, MARINE trial data seem to confirm rather than disprove the adverse MO impact hypothesis.
As an aside, the sponsor chose to report data from the ANCHOR and MARINE studies in the following manner:[33]
Although they do mention “placebo-adjusted,” the clear and obvious message being relayed from the chart (targeting the investment and healthcare communities) is that 4 g/d IPE has a profound impact on atherogenic and inflammatory markers—when in fact, it does not.
For example, hs-CRP levels were lowered insignificantly by ~3 - 4% from baseline in the 4 g/d IPE groups in both studies, yet the slide shows a 36% and 22% reduction in hs-CRP in MARINE and ANCHOR, respectively. This was wholly due to sharp increases in the 4 g/d MO placebo arms, not the result of a CRP-lowering effect of IPE. The graphic also shows that small LDL-P concentration was reduced by 25.6% in MARINE and 13.5% in ANCHOR, but this too was entirely due to increases in the 4 g/d MO placebo arms. In fact, LDL-P actually increased insignificantly in the 4 g/d IPE arms from baseline in both studies. Median apoB was also insignificantly lowered by 3.8% and 2.2% from baseline in the 4 g/d IPE arms of MARINE and ANCHOR, respectively. Yet elsewhere, the sponsor has stated that 4 g/d IPE was shown to significantly reduce apoB by 8.5% (p=0.0019) and 9.3% (p<0.001) in these studies.[34] Once again, this result was predominantly caused by a significant increase in apoB from baseline in the MO-dosed placebo arms, not a significant reduction in IPE arms.
The misleading inference that IPE has a pronounced effect on these biomarkers was further exacerbated by a statement made by the principal investigator of the MARINE trial, quoted in an Amarin Corp. press release:[35]
Increased apo-B and LDL particle concentration may increase cardiovascular disease risk," said Dr. Bays. "In the previously reported results of the MARINE trial, we observed that AMR101 significantly reduced triglycerides and apo-B, without increasing LDL cholesterol, as compared to placebo, in a most challenging patient population having triglyceride levels ≥500 mg/dL. In this follow-up analysis of the MARINE trial, AMR101 [IPE] reduced both total and small LDL particle concentration, which is not only consistent with its known effects in decreasing apo-B and lack of LDL-cholesterol raising in patients with very high triglyceride levels, but also is suggestive of another potentially favorable lipid effect.
Investigator bias in industry-sponsored clinical trials has far-reaching implications.[36]
We find the reporting of these data by the sponsor and commentary by the principal investigator to be irresponsible. It seems that the sponsor has been relying on the adverse impact of mineral oil on placebo group subjects to make misleading claims as to the efficacy of IPE therapy. The REDUCE-IT trial results may unfortunately prove to be the culmination of this tendency.
[4] Agoritsas T, Merglen A, Shah N, et al. Adjusted Analyses in Studies Addressing Therapy and Harm
Users’ Guides to the Medical Literature. JAMA. 2017.
[5] Katherine S. Button, John P. A. Ioannidis, Claire Mokrysz, et al. Power failure: why small sample size undermines the reliability of neuroscience. Nature. 2013.
[7] Yusuf S, Wittes J, Probstfield J, Tyroler HA. Analysis and Interpretation of Treatment Effects in Subgroups of Patients in Randomized Clinical Trials. JAMA. 1991.
[8] Sleight P. Debate: Subgroup analyses in clinical trials: fun to look at - but don't believe them!. Curr Control Trials Cardiovasc Med. 2000;1(1):25-27.
[9] Yokoyama M, Origasa H, Matsuzaki M, et al. Effects of eicosapentaenoic acid on major coronary events in hypercholesterolaemic patients (JELIS): a randomised open-label, blinded endpoint analysis. The Lancet. 2007.
[10] Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias. Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA. 1995.
[11] David Moher, Ba' Pham, Alison Jones, et al. Does quality of reports of randomised trials affect estimates of intervention efficacy reported in meta-analyses? The Lancet. 1998.
[12] J. BEYER-WESTENDORF, H. BÜLLER. External and internal validity of open label or double-blind trials in oral anticoagulation: better, worse or just different? jth. 2011.
[13] Jüni P, Altman DG, Egger M. Systematic reviews in health care: Assessing the quality of controlled clinical trials. BMJ. 2001.
[14] Dariush Moza?arian. JELIS, ?sh oil, and cardiac events. The Lancet. 2007.
[15] Providing Clinical Evidence of Effectiveness for Human Drug and Biolog
[16] JOHN H. LEE, JAMES H. O’KEEFE, CARL J. LAVIE, et al. Omega-3 Fatty Acids for Cardioprotection. Mayo Cin Proc. 2008.
[17] The ORIGIN Trial Investigators. n–3 Fatty Acids and Cardiovascular Outcomes in Patients with Dysglycemia. N Eng J Med. 2012.
[18] Aung T, Halsey J, Kromhout D, et al. Associations of Omega-3 Fatty Acid Supplement Use With Cardiovascular Disease Risks: Meta-analysis of 10 Trials Involving 77 917 Individuals. JAMA Cardiol. 2018.
[20] R. Marchioli. Treatment with n-3 polyunsaturated fatty acids after myocardial infarction: results of GISSI-prevenzione trial. European Heart Journal. 2001.
[21] Rezaei Y, Gholami-Fesharaki M, Dehghani MR et al. Statin Antiarrhythmic Effect on Atrial Fibrillation in Statin-Naive Patients Undergoing Cardiac Surgery: A Meta-Analysis of Randomized Controlled Trials. J Cardiovasc Pharmacol Ther. 2016.
[23] Effects of eicosapentaenoic acid on major coronary events in hypercholesterolaemic patients (JELIS): a randomised open-label, blinded endpoint anal... - PubMed - NCBI
[24] Kukull WA, Ganguli M. Generalizability: the trees, the forest, and the low-hanging fruit. Neurology. 2012.
[26] Kastelein, John J.P. et al. Omega-3 free fatty acids for the treatment of severe hypertriglyceridemia: The EpanoVa fOr Lowering Very high triglyceridEs (EVOLVE) trial. Journal of Clinical Lipidology. 2013.
[27] Jan Oscarsson, Eva Hurt-Camejo. Omega-3 fatty acids eicosapentaenoic acid and docosahexaenoic acid and their mechanisms of action on apolipoprotein B-containing lipoproteins in humans: a review. Lipids in Health and Disease. 2017.
[28] Feingold KR, Grunfeld C. Triglyceride Lowering Drugs. Endotext. 2018.
[31] Otvos JD, Mora S, Shalaurova I, Greenland P, Mackey RH, Goff DC. Clinical implications of discordance between low-density lipoprotein cholesterol and particle number. J Clin Lipidol. 2011.
[32] Mora S, Buring JE, Ridker PM. Discordance of low-density lipoprotein (LDL) cholesterol with alternative LDL-related measures and future coronary events. Circulation. 2014.
[34] Christie M.Ballantyne, Rene A.Braeckman, Harold E.Bays, et al. Effects of icosapent ethyl on lipoprotein particle concentration and size in statin-treated patients with persistent high triglycerides (the ANCHOR Study). Journal of Clinical Lipidology. 2014.
[35] Amarin's Phase 3 MARINE Study Results Presented at American Heart Association's Scientific Sessions 2011 | Amarin Corporation plc
[36] Ahn Rosa, Woodbridge Alexandra, Abraham Ann, et al. Financial ties of principal investigators and randomized controlled trial outcomes: cross sectional study. BMJ. 2017.
600.png)
600.png)
_resize.png)
600.png)
620.png)
640.png)
720.png)
600.png)
600.png)
_resize.png)
600.png)
620.png)
640.png)
720.png)
