Dennis P. Scanlon, PhD: Is it time for us to redesign, or to think differently about, the cardiovascular outcome trials? There’s a lot of issues at stake here. There’s power issues. There’s issues, as you’ve mentioned, Dr Bloomgarden, about maybe not having enough patients to be able to conduct a randomized controlled trial. There’s timing issues—how long do we follow? What are the endpoints that we capture? And other things such as, how much data do we collect retrospectively? I guess maybe a general question is, what are your thoughts on what the future of studies needs to be, both to be feasible from a cost perspective but convincing from a scientific perspective?
Zachary Bloomgarden, MD: There are now over 400 million persons in the world with diabetes, and in 2040 it will be over 650 million people. They’re going to get not only heart disease, but kidney disease, and eye disease, and all of the complications—neuropathy, amputations. And so, as hugely expensive as these trials would be, doing trials that are designed to look at different strategies—let’s say the non-hypoglycemia strategy of SGLT2 (sodium-glucose co-transporter 2) inhibitors and GLP-1s (glucagon-like peptide-1s) added to metformin versus the hypoglycemia strategy of sulfonylureas—designing such a study, and spending the huge amounts of money on it, and willing to wait a decade for results will really have a huge impact for hundreds of millions of people in decades hence. And so, I hope that we can look at this encouraging new stuff and say, “OK, let’s not just focus on the most ill 10% of patients, but let’s try to figure out how all people developing diabetes should be treated going forward.”
Silvio Inzucchi, MD: A comment on Dr Bloomgarden’s thoughts is that diabetes trials, long-term, are really hard to do, and that’s because there’s a progression of hyperglycemia, or gradual deterioration in beta cell function that makes it very difficult to keep people only on 1 drug. It’s not clean like that. When you started talking about 2 different strategies, I immediately thought of BARI 2D, which was a study that was published a number of years ago that tried that very approach. This was metformin plus a TZD (thiazolidinedione)— rosiglitazone (which is not commonly prescribed anymore), versus sulfonylurea and insulin. And we thought that was a beautiful design—stable coronary disease patients, insulin provision, and the standard approach versus insulin sensitization, which, at that time, was the more novel approach. And we all anticipated there would be a benefit in the insulin sensitization approach.
As it turned out, though, cardiovascular events and overall mortality were absolutely neutral in those 2 arms. Now, when you look at the study, it was a challenging study to conduct because there was a lot of what I call “cross contamination” between the groups. We have a patient on insulin sensitizer who has an A1C of 8.5%. What are you going to do? Well, in 2004 or 2005, when the study was conducted, you’d put the patient on insulin. All right. Now, you’re no longer comparing 2 clean approaches anymore. Maybe it’s a little easier with, now, 12 drug classes as opposed to 5 drug classes that we had back then. But I think we really have to think carefully about what’s going to happen to these patients in these trials if their glucose levels deteriorate. We still want to protect them from retinopathy, nephropathy, and neuropathy, don’t we?
Kenneth Snow, MD, MBA: And I think, certainly, there’s no risk that the analysis of big data (no matter how good it is), is ever going to replace the role of a randomized clinical trial to answer a specific question. But what we do know is that there are certain situations where a randomized controlled trial just doesn’t work because the population is relatively homogenous to answer the question that you want answered. So then, you’re stuck with the question of, “Well, can’t I expand this into other populations? Do I need to do a full other randomized controlled trial to answer that question or not?”
So, that’s 1 possible role. And the other, as Dr Bloomgarden referred to earlier, is that in a condition where the incidence or the prevalence is so small, that a randomized controlled trial is just not practical. And yet, across the entire database of available healthcare databases that’s either existing or that should exist or could exist, could you then maybe not prove the question you’re asking, but at least have such supportive evidence that you now have a direction of what makes the most sense to do in the absence of a randomized controlled trial? And that’s also true of studies with duration. We are being asked, in randomized clinical trials, to answer a question within a duration that, at length, is going to be 3, 4 years before other factors—the cost of the study, the complexity of the study, a loss of patients dropping out of the study. It just becomes too difficult to do them. So, that’s a place where that data can really be helpful.
Zachary Bloomgarden, MD: I think we’re moving in to cardiology. The cardiologists have faced this problem over and over with bare-metal stents versus drug-eluting stents, and bypass versus stents, and so on. And so, the technology changes, and the potential treatments change. Over the coming decade, more new classes will develop, but, nevertheless, there are some important questions that it would be nice to address by big clinical trials in real populations.
Robert Gabbay, MD, PhD, FACP: I totally agree, and I think we’ll always need to do those. But I think, increasingly, just out of the reality (and you can hear this in this discussion), there are so many questions and we know there are just not going to be studies to answer all of them, or they’re going to take 5 to 10 years. As Zachary said, there are all these people with diabetes. What do we do in the meantime?
I think as big data becomes more sophisticated, the analytic techniques become more sophisticated, and studies are done more accurately, we’re going to have to rely on that kind of data to answer some of the questions that there are unlikely to be clinical trials on. I think the challenge in that world is that we’ve seen those kinds of studies that are done well and those that aren’t done well. And to the casual observer, or to many providers that are just reading an abstract, they’re not going to know whether the study was done appropriately or not. And the danger is that kind of data, then, sways clinical care. I think a better arbitration of study technique for big data analysis will really help move the field forward.