CV Risk: Can Prescription Omega-3 REDUCE-IT?

An Interview With Deepak Bhatt

Interviewer: Melissa K. Walton-Shirley, MD; Interviewee: Deepak L. Bhatt, MD, MPH


November 15, 2018

Melissa K. Walton-Shirley, MD: Hello. I'm Melissa Walton-Shirley for | Medscape Cardiology, with Dr Deepak Bhatt from the Brigham and Women's Hospital. We'll be discussing REDUCE-IT, which studied a highly purified form of fish oil in patients with high triglycerides.[1] This trial generated a flurry of excitement, as it was a very positive trial, but at the same time there's also been some controversy that we'll get into. Thanks for joining me, Dr Bhatt.

Deepak L. Bhatt, MD, MPH: Great to be here.

Walton-Shirley: I really appreciate the opportunity to interview you. Can you give us the top-line findings of REDUCE-IT?

Bhatt: Absolutely. It's a randomized, double-blind trial of icosapent ethyl 4 g/day versus placebo in 8179 patients who had high triglycerides (135-500 mg/dL) and cardiovascular risk. By that, I mean two major camps of patients that we labeled as secondary and primary prevention cohorts.

By secondary prevention, I mean stable coronary artery disease, stable cerebrovascular disease, and stable peripheral artery disease. By primary prevention, in this trial it's specifically patients with diabetes treated with medicine and at least one additional cardiovascular risk factor, which was modeled after the CHARISMA[2] inclusion criteria, with some modifications. This was a rather broad population of secondary prevention and select primary prevention patients.

Very Positive Findings

Walton-Shirley: What was your outcome?

Bhatt: The primary endpoint was a 5-point MACE (major adverse cardiovascular events), which included cardiovascular death, myocardial infarction, stroke, hospitalization for unstable angina, or coronary revascularization procedures. The relative risk was significantly reduced by 25%, with an absolute risk reduction of about 5%, which works out to a number-needed-to-treat of about 21 over the course of an average follow-up of about 5 years. The P value for that was very significant, at less than .001. In fact, there were several zeros—I think six or seven zeros, so it is a very high degree of statistical significance, but also absolute and relative risk reductions that are quite favorable.

Of course, some doctors are less interested in endpoints of hospitalization for unstable angina and coronary revascularization. I'll point out that this is a double-blind, placebo-controlled trial, so there won't be bias in terms of how those endpoints occur. For those people who [are interested in] the triple ischemic endpoint (cardiovascular death, myocardial infarction [MI], stroke), it was significantly reduced by 26%, which again, in terms of relative and absolute risk reduction, is quite good. The number-needed-to-treat for that endpoint works out to be 28 people over a median follow-up of about 5 years. Once again, the result was very statistically significant, with a P value of .000006.

Both the primary endpoint and our key, prespecified, secondary endpoint are highly statistically significant. Then we did what, in clinical trial terms, is a prespecified, hierarchical testing, which is perhaps more important for regulatory purposes. Basically, what that means is that you can keep going down the list of endpoints, but you should stop once a P value is nonsignificant. Anything under that, if it's positive, is hypothesis-generating—it might be right, might be wrong. A purist would say to stop at the point where it's not significant.

We did that, and really the entire list of things that we looked at, down to all-cause mortality, was significant. All-cause mortality wasn't significantly reduced there; it was 13% lower, with a P value of .09—so, a trend. Cardiovascular-related mortality was significantly reduced by 20%. A variety of other endpoints were reduced, including MI (fatal or nonfatal), by 31%; stroke, (fatal or nonfatal), by 28%; hospitalization for unstable angina; urgent or emergent revascularization. Even elective revascularization was significantly reduced.

To help explain that, cardiovascular death reduction, some of our tertiary endpoints, including sudden cardiac death and cardiac arrest, were also significantly reduced. There was a robust reduction in terms of magnitude of effect based on P values across a variety of endpoints, up to and including death from cardiovascular causes.

Walton-Shirley: Speaking to the mortality benefit, we saw that meta-analysis in 2010 in the Lancet, including 18 fibrate trials.[3] The events were lower, but mortality was not impacted. So it's not just the triglyceride lowering; it must be how you get there. Can you speak to that?

Bhatt: I think that's a really insightful question. This drug does reduce triglycerides; it's approved for triglyceride reduction in patients with triglycerides over 500 mg/dL (5.6 mmol/L), so it's a prescription medication. What we wanted to test was not the triglyceride-lowering property, but rather the effect on cardiovascular events. Of course, we measured triglycerides, and that was reduced by about 20%, which is consistent with what was known about it. As well, it reduced CRP by around 20%-22%, but that also was known from prior work.

I think each of those effects—a modest effect on triglycerides, a modest effect on inflammatory markers—could account for a proportion of the cardiovascular benefit, but likely a relatively small proportion of the cardiovascular effect if you believe the Mendelian randomization studies, which I happen to. Nice work has been done by folks like Brian Ference.[4] That degree of triglyceride reduction might cause an 8% relative risk reduction in events but certainly not a reduction of 25%.

I think triglycerides are probably part of the story, but I think the much more dominant player is likely EPA itself. EPA levels went up by 360% comparing the two arms over time. I think that might account for the reductions in sudden cardiac death and cardiac arrest. On the other hand, there might have been some other anti-inflammatory or triglyceride-related benefits, maybe even a mild antithrombotic effect, because we did see a trend towards more bleeding—not more fatal bleeding, not more intracranial bleeding, not more gastrointestinal bleeding; but if you lump all sorts of bleeding together, there was a trend, so maybe some mild antithrombotic effect as well.

"Pleiotropy" is a word that gets thrown around a lot. I think there are likely pleiotropic effects here—not only because of the effects on biomarkers, but more importantly, I can't easily think of another drug that reduces cardiovascular death, sudden cardiac death, MI, stroke, emergent revascularization, and also elective revascularization. Why would something reduce elective revascularization and also stroke if there was just one mechanism of action? This is unlikely. I think there are multiple mechanisms of action.

Antithrombotic Effects

Walton-Shirley: Your primary endpoint, like many trials, is loaded with entities that are impacted by thrombosis. I would be interested to know about baseline platelet studies, bleeding times, etc. Can you speak to that?

Bhatt: We did bank blood for biomarker and genetic analyses. We haven't had a chance to do anything other than the baseline, basic biomarkers that I described. We have lots of plans, and many different people have been emailing me with different ideas. There are great ideas out there, but [we have a limited] amount of blood, so we have to figure out what we're going to do.

To do platelet function testing would have been fantastic, but that, in general, needs to be done in real time. We're not going to able to do that now, but we could look at markers of platelet activation, such as P-selectin and so forth, that could provide some insight. We could look at other markers of coagulation and see how they change over time. I actually think that's a really good idea in terms of how to use that valuable blood.

Walton-Shirley: And maybe going forward in other trials. How did you pick the 4-g daily dose? And did you split the dose for tolerance?

Bhatt: It was a dose that, first of all, is approved for triglyceride reduction over 500 mg/dL. Beyond what it might be approved for, the dose was really picked to try to mimic what, in a Western population, might be achieved with EPA supplementation in a Japanese population. We were really trying to mimic serum levels. As you likely know from the epidemiology, people from Japan—and this is true of many coastal areas—tend to eat more fish than Western populations do and they naturally have higher EPA levels. Epidemiologically speaking, they tend to have lower rates of coronary artery disease—not stroke, but coronary artery disease.

Our thought was that if we can mimic those sorts of levels and surpass them pharmacologically, that should provide even more substantial benefit than the epidemiology might suggest and a prior trial, JELIS, had shown.[5] There was a 19% relative risk reduction with 1.8 g EPA—not quite this drug, not quite as highly purified, but still sort of along the same pathway.

That trial was criticized at the time. It was published in the Lancet in 2007. I thought it was a good trial, but it was criticized because it was only Japanese patients, which doesn't invalidate the study, but it might mean it's not generalizable to Western populations. The second critique, which is more legitimate, was that low-intensity statins were used as background therapy instead of higher-intensity statins that are recommended these days for at-risk patients.

The third critique that a lot of people raised, myself included at the time, was that there was no real placebo. It was randomized trial but a so-called open-label trial, so there was no placebo, no mineral oil placebo—there was no placebo at all. Nothing. Some people criticize that sort of trial because bias can creep in. In this particular case, I'm not sure how bias would have crept in, so I think that 19% relative risk reduction is solid.

In our trial, we used a mineral oil placebo. There's been a little chatter about whether the mineral oil placebo can be harming patients, which I think is a far-fetched argument, but I certainly have heard it. For people who just are really bothered by that, I'd say look up the JELIS trial; it had a 19% relative risk reduction, albeit in a different population.

'Active' Placebo and Other Conspiracy Theories

Walton-Shirley: What exactly was the mineral oil placebo?

Bhatt: If we did a trial without a placebo, of course we would have been criticized for that just as the JELIS investigators were appropriately criticized, because that is a legitimate point in trial design. We could have used a sugar pill as the placebo, so the patients would have been randomized to icosapent ethyl or a sugar pill. Obviously, then, the patient would be unblinded because they can say, "This isn't icosapent ethyl, it's a sugar pill."

We had to use something that had the look, the consistency, and the color of the study drug; otherwise, it wouldn't be a truly blinded trial. That's how we got to mineral oil, and it is about 4 cc per day of mineral oil.

Walton-Shirley: It's over-the-counter mineral oil, I guess?

Bhatt: Yes, like you might give to your kids if they're constipated or to older patients. It's just mineral oil and it's an extremely small quantity, so it's hard for me to really buy into it being this toxic substance. As I said, if somebody really is worried about that, the JELIS trial is there.

Within our trial, we did do an analysis where we looked at patients in the drug arm and in the placebo arm, and the drug effect in the placebo versus drug, if the LDL-C from 1 year to baseline increased or if it stayed the same or decreased. In that analysis, the benefit of the drug, statistically speaking, was quite similar. The statistical interaction P value was negative, meaning that there really wasn't something going on.

If mineral oil was really the cause of the large benefit that we saw here—that is, the mineral oil is toxic as opposed to the drug being beneficial—then one would think in that subgroup of patients in placebo who had an LDL-C decrease that there wouldn't have been any benefit of the drug if in fact all the difference between the two arms is toxicity from the drug in LDL-C increase. I don't think that it's some intrinsic property of mineral oil raising LDL-C or interfering with statins that would account for what we saw. If someone doesn't believe what I said—the degree of LDL-C change, a difference of 5 mg/dL, extrapolating from the Cholesterol Trialists' Collaboration[6] over the course of 5 years—it would be up to 3%, 4% (if you want to be very optimistic, 5%) relative risk reduction.

For those folks that are really skeptical, it's not a 25% but a 20% risk reduction. Does that materially change anything? With a good safety and tolerability profile, would someone say, "Oh, it's only 20%; I'm not going to use it"?

If one believes that the difference in the two arms is from the placebo arm, and if you believe that cardiovascular death was reduced in this trial, where that's an adjudicated endpoint, a blinded trial, 99.8% ascertainment of mortality... If you believe that cardiovascular death, as you should, doesn't that then mean the conclusion is mineral oil is raising cardiovascular death? If we believe that, we should stop taping. We should call the FDA and say, "Get that mineral oil off the shelf. It's killing babies and old people."

I understand that trials are critiqued. It doesn't bother me. I think investigators should address concerns that are raised. But of all the concerns that could be raised about the trial, I think this is more noise than any real substance.

Walton-Shirley: When you have such a positive result, though, you come under the microscope. That's just what happens.

Bhatt: You are absolutely right about that. One other thing I'll mention, too, is there might be a bit of regression to the mean, because to get into the trial, the LDL had to be under 100 mg/dL (2.6 mmol/L). That is, we didn't want to take people whose LDL is 180 mg/dL (4.7 mmol/L) and treat them with a new drug, and then the trial is really positive and then everyone would ask why we didn't give them a generic statin first. That would have been a legitimate critique, so we made sure, on entry, that everyone had an LDL less than 100.

But what can happen when there's that sort of entry criteria—it's the same with blood pressure trials, too—you have to have this or below to get in, and when you measure it again, by chance it might be above that value. It's quite possible that, over time, the LDL drifted up actually in both the drug arm and the placebo arm.

In the ODYSSEY trial,[6] for example, the LDL-C drifted up over time in the alirocumab arm. That's a little complex because we downtitrated the drug based on low LDL-C, so I won't make much of that arm. In the placebo arm of alirocumab, also over time you'll see the LDL-C drifting up. Unless you think the placebo we used in that trial was also doing something bad, this is an issue that, really, it's what you alluded to—the trial results are so strong that people are trying to find particular critiques.

I think the results really are robust across a variety of different endpoints and across a variety of different subgroups. The data were independently validated by our DSMB statistician. She took the raw data and derived the primary, secondary, and prespecified tertiary efficacy endpoints herself and matched those of the initial analysis.

Independent Analysis

Walton-Shirley: That flies in the face of the conspiracy theory that the company was somehow involved in the result.

Bhatt: Right. Amarin was the company that sponsored the trial. They've certainly been involved in the trial's execution as well as in the data analysis. A company statistician listed as a coauthor in the New England Journal of Medicine, Lisa Jiao, a PhD-level statistician, did indeed do the analyses herself. She's a good person; she wouldn't cook the books. Even for conspiracy theorists who think, Oh, a company employee is not going to provide an honest answer—those folks are really conspiracy theorists. I mean, do they really think people are going to jeopardize their livelihood and go to jail by making up numbers?

But for people who really do believe that sort of thing, our independent DSMB statistician, independent of the academic steering committee as well, had come up with the same numbers. I think the overall findings are really quite robust, and as we look at different things, might there be some shift in this number or that number? To a trivial extent, maybe. That often happens in trials. Here, we're talking about for the primary endpoint, a P value of .00000001.

That is such a high degree of statistical significance. I don't say the number just because it's a nice P value, but it allows for lots of different sensitivity analyses. You can, for example, throw out all of the patients who had an LDL-C increase, as I did, and the results still remain robust. You can throw out patients who had diarrhea, perhaps, from the mineral oil. There was a slight imbalance there that was statistically significant, but even if you take those patients out, the results remain quite robust. You can look across different subgroups and there is a consistency of benefit.

I think the trial results are what they are. Does that mean we've answered every question out there? Absolutely not. Our goal was to get the main results out for the medical community in a timeframe that would be useful to them. There are lots of legitimate questions that we already had and more that have been raised by the medical community in the months or maybe years to come.

We're going to answer these with different analyses from the clinical trial data, hopefully complemented by biomarker and genetic analyses. We have also planned and are talking to a number of academic groups about a cost-effectiveness analysis, which is also important. My guess is it'd be pretty cost-effective, with a number needed to treat of 21, but we're going to do it more formally.

I think a major analysis that we have planned but haven't had time to complete is looking at recurrent events and total events, because what we've reported is the classic, conservative time to first event. If the patient gets hospitalized for unstable angina, that's what's in the primary endpoint. If that patient subsequently has an MI, that's not counted in the primary endpoint. As was reported here, for example, from ODYSSEY Outcomes, when you look at the recurrent events, you get the full impact of what atherosclerosis does and potentially a greater idea of the true benefit of a drug intervention.[7] We've got that planned and we'll see what that shows.

There's a lot more coming, and I hope to be able to, on behalf of our steering committee, answer a lot of the questions out there.

Walton-Shirley: I really appreciate it. You've certainly answered a lot of questions today, and as usual, we really appreciate your insight. Thank you so much for joining us at Medscape |, and thank you for joining us as well.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.