Diabetes management in 2025: A network meta-analysis
McMaster Perspective Extended

Gordon Guyatt, MD, MSc, distinguished professor in the Departments of Medicine and Health Research Methods, Evidence, and Impact at McMaster University, joins Roman Jaeschke, MD, MSc, to discuss a living network meta-analysis of the currently available drug treatments for diabetes mellitus (DM). This interview is a continuation of part 1.

This video offers playback speed control, chapters (list of discussed topics), and subtitles/closed captions (CC). Use the available player controls.

Outline

1. Introduction to network meta-analysis.
2. Key advantages over traditional meta-analysis.
3. Results of meta-analysis of drug treatments for diabetes.
4. Adverse effects and safety.
5. Role of traditional diabetes medications.
6. Living network meta-analysis.

Transcript

Roman Jaeschke, MD, MSc, DPharm: The previous paper was talking about a specific use of one class of drugs in a specific population with renal dysfunction—by the way, I understand, without diabetes, and I will talk to a nephrologist about where it leaves us and at which threshold the SGLT2 should be used. Here the question was much broader and maybe I will ask you, Gordon, again, how did you approach that particular question?

Gordon Guyatt, MD, MSc: Well, there are now lots of different drugs and classes of drugs for treating patients with diabetes. And wouldn’t it be nice, we thought, if one could summarize all the evidence for all these drugs and drug classes for treatment of diabetes in terms of its outcome, its effects, and outcomes that are important to patients? Wouldn’t it be nice to have everything compared to everything else and to not using anything, and find out the relative merits of all these treatments? And so we did it, and it was a big job, as you can imagine—all the randomized trials of all the drugs, putting them together and comparing them to one another. That was the motivation. I am quite personally… I played some role in this, obviously not a leadership role, but I’m quite proud of this, that we actually got together and put together a team that could summarize this massive literature.

Roman Jaeschke: Gordon, you summarized it and I believe you summarized it in the form of something called network meta-analysis. I believe that illustrates the concept. Would you spend a few minutes explaining the network meta-analysis to people who may be less familiar with it?

Gordon Guyatt: Certainly. I think the figure you have on the screen is very helpful for this.

A traditional meta-analysis would look at a single intervention and a single comparator. Each of the lines, the straight lines that you see here, describes studies where a single drug was compared to a single comparator. So for instance, if you take glucagon-like peptide 1 (GLP-1) receptor agonists on the upper left—thank you very much—and compare them to standard treatment, you see that there are a substantial number of paired comparisons. And if you look at the bottom left of the SGLT2 inhibitors, you see as well there is lots of paired comparisons where randomized trials were done looking at SGLT2 inhibitors versus standard treatments. And there’s lots of other straight lines for other paired comparisons. The width of those lines tells us the magnitude, the number of patients, the number of studies, the number of patients who were included in those studies.

The thickest line with the most comparators is dipeptidyl peptidase 4 (DPP-4) inhibitors against standard treatment. After that you have the GLP-1 receptor agonist and the SGLT2 inhibitors, but there’s lots of other paired comparisons. And indeed you see lots of other lines, although they are much thinner, reflecting the fact that there are fewer studies and fewer patients—substantially fewer than in those 3 paired comparisons.

This is what you would call a relatively well-connected network. You see lots of comparisons. For instance, if you take GLP-1 receptor agonists, they’ve been compared not only to standard treatment, but they’ve been compared to SGLT2 inhibitors, they’ve been compared to thiazolidinediones—forgive my pronunciation—they’ve been compared to basal insulin, they’ve been compared to basal bolus insulin, et cetera, et cetera. So there are not only comparisons against standard treatment, although the comparisons of standard treatment dominate things, but there are lots of comparisons of a number of these to a number of other ones.

This describes what we call a well-connected network. The better connected network, the more confident we are about our inferences from our network meta-analysis. So this is a nice network in terms of well connected.

What the enormous advantage of the network meta-analysis is, is that you can combine what we call the direct comparisons, like GLP-1 receptor agonist versus standard treatment, with the indirect comparisons. For instance, if you take a GLP-1 receptor agonist and want to make inferences about its comparison to standard treatment, you can not only look at the direct comparisons, but you can look at the DPP-4. And if, for instance, the GLP-1 receptor agonists are better against standard treatment than the DPP-4s, you gained additional information about the effects.

So one big advantage of the network meta-analysis is that in combining the direct and indirect estimates, we get more information and more accurate and precise estimates of our treatment effect. But an even bigger advantage of the network meta-analysis is that we don’t need direct comparisons. So we can use the… Let’s look at, say, GLP-1 receptor agonists. What has it not been compared to…

Roman Jaeschke: Meglitinides?

Gordon Guyatt: You and I have the same pronunciation problems. I’m glad to hear it. Anyway, let’s say we wanted to compare GLP-1 receptor agonists to these meglitinides and we don’t have any direct comparisons, but both of them have been compared to standard treatment. So we can say, how do GLP-1 receptor agonists do against standard treatment versus the meglitinides? And if for instance, as is the case I’m sure, the GLP-1 receptors have a mortality reduction against standard treatment, but the meglitinides do not, one can infer that the GLP-1 receptors are better than the meglitinides in terms of reducing mortality.

So we have a great additional advantage of the network meta-analysis in that it allows us to make comparisons between interventions that have never been directly compared—through the indirect comparisons. We can combine the direct and the indirect and gain statistical power by doing that. And we can also make inferences about the relative merits through the indirect comparisons of treatments that have never been compared directly.

Roman Jaeschke: Okay. One question here, because all those lines are therefore direct comparisons.

Gordon Guyatt: That’s right. Each of the lines you see.

Roman Jaeschke: What is missing could be construed as indirect comparison. The issue of standard treatment—I take it would be for any particular study—the standard treatment not including the comparator, that’s probably how it could be described. So at some stage…

Gordon Guyatt: And standard treatment differed over time and circumstances and judgement. So the standard… One of the assumptions, or perhaps limitations, of this network meta-analysis is differences in the standard treatment over time and across studies. But we decided, and I think wisely, that since previous standard treatments did not have—and no matter what you used—you did not have any effect on major cardiovascular and renal events, so it didn’t much matter for those outcomes what the standard treatment was. It didn’t do any good. So that is one reason why it is reassuring, or legitimate, to use standard treatment as a comparator, even if the standard treatment differs across the studies.

Roman Jaeschke: So you compared everything to everything, you looked at—I presume—major cardiovascular and renal outcomes. And I think, if you don’t mind, I will go to the table of results.

Gordon Guyatt: Yes.

Roman Jaeschke: Would you take us through it?

Gordon Guyatt: We love these tables, all modesty aside. I helped to develop these tables and what they do is a beautiful thing in my view. They color-code to show you what the best treatments are, and intermediate treatments, and not-so-good treatments. You can see the grey and you can see the colored. The colored are where we have high- or moderate-certainty evidence, and the grey are where we have only low- or very low–quality evidence. So the first thing you can see is that color is where we have more trustworthy estimates and the grey are when we have less trustworthy estimates.

Then the next thing is the darker, the better. So the dark green are the ones that are among the best, the lighter green where we have higher moderate-certainty evidence, and the lighter green are the ones that are still good but they’re not among the best. Looking at this, you can immediately see that there are two classes of drugs with respect to these major mortal and morbid events that are where the real action is, and those are SGLT2 inhibitors and GLP-1 receptor agonists. Looking at this table, it just jumps out at you that that’s where the dark green is, which means number one among the best. And number two, the high- or moderate-certainty evidence.

Roman Jaeschke: So if I can go through the rest of the table, the greens are clearly concentrated in those two. Interestingly, the new class of drugs, nonsteroidal mineralocorticosteroid receptors, are also on the greenish side, which is of interest. But there are also… So looking at this one would say okay, these two stand out, maybe in people who have renal failure or heart failure the SGLT2s are darker or greener. And among people who have nonfatal stroke, the GLP-1s are darker or deeper greens, which would indicate maybe that could be the preferred class.

But there are also side effects. I wonder to what degree those… There is a fairly dramatic cutting down on those important outcomes: all-cause death, cardiovascular death, nonfatal myocardial infarction, heart failure is cut down by one-third, end-stage kidney disease by 40%. I mean, these are fairly dramatic changes. How about the other side? Because there is quite a bit of colors here as well on the harm.

Gordon Guyatt: As a matter of fact, we have high-certainty evidence about a major relative increase—and remember all we’ve talked about earlier in our conversation how what patients are really interested in is absolute effects. In this table we only see relative effects. So we would have to go back to our baseline risk to see how serious the problem is. You can see genital infections happen a lot, but they’re not that serious. So if you know your all-cause death is… If you’re at high risk of mortality and your all-cause death is going down substantially, and your cardiovascular death, and you also happen to have a risk of end-stage kidney disease where the relative effect, as you’ve just pointed out, is almost a 40% relative risk reduction, you are probably ready to say, okay, avoiding those things is much more important than avoiding a genital infection. And even though there’s an appreciable absolute increase in general genital infections, you will go along with it.

Amputations, it’s a small increase in relative effect, and amputations are rare. So the absolute increase in amputations… Amputation is pretty nasty, but the absolute increase… The relative increase is small, the baseline risk is small, so that doesn’t happen too much.

Ketoacidosis. Unfortunately, the incidence is pretty low, [but there’s] doubling the risk. But once again, ketoacidosis can be easily treated.

I could go through all the outcomes in this. For SGLT2 inhibitors and GLP-1 receptor agonists those adverse effects are either relatively uncommon, well relatively uncommon, and—other than perhaps amputation—things that can be easily managed. So although yes, those adverse effects are there, relative to the importance of the benefits that patients gain they’re not that compelling, at least to me.

Roman Jaeschke: If I could add from the previous paper. We were talking about SGLT2s. They looked at the genital infection and the increase was in absolute terms about 27 per 1000. The diabetic ketoacidosis (DKA) was 4 per 1000 per 5 years, actually, so fairly uncommon but something to clearly keep in mind. And in those previous meta-analyses, the lower limb amputation, the estimate was 2 per 1000 per 5 years and it was actually not statistically significant. Here you are on the border of statistical significance, so maybe there was 1 or 2 more studies added either here or there to switch it from crossing the threshold of no difference to not.

But let me look through this whole table and please try to convince me that the changes in the overall strategies of management are not as profound as, “Start with one of those two drugs and use the other if you need it.” I mean, all the others are grey. What’s in your mind—and I will ask this question to an endocrinologist as well, or people specializing in diabetes—what is the place of all these other drugs which are very commonly used? Metformin, sulfonylureas, basal insulin, bolus insulin—what’s your take on it?

Gordon Guyatt: We had a long time when we were treating diabetes where nothing—except for maybe the angiotensin-converting enzyme (ACE) inhibitors in proteinuria—nothing influenced the major outcomes, the major cardiovascular outcomes or renal outcomes, nothing. So we were giving these things out to lower blood glucose without any impact on the major cardiovascular or renal events, and that’s what these results show. When we put it all together, there’s still no benefit from these things.

On the other hand, there could be… I think there is a reason for lowering blood glucose. The reason for lowering blood glucose is not to prevent these major events. Lowering blood glucose, as far as we can tell, does not prevent the major events. On the other hand, we do see people coming in with hyperosmolarity, very sick, they may be extremely sick, they’re at risk of dying, and that’s because they’ve got glycosuria, they get volume-depleted, and they end up in these hyperosmolar states. It may be that the glycosuria also predisposes to urinary tract infections.

So there are reasons for lowering the blood glucose and the SGLT2 inhibitors in particular are not very good. They’re about the worst we have of the things we give out in terms of lowering blood glucose. So you’re certainly not going to give these drugs out, the other drugs out for lowering for major events, but you may be thinking of giving them for lowering the blood glucose. And pick somebody, take somebody who is young, relatively young, type 2 diabetic, say, they’re 50 years old and they don’t have hypertension and they have no other risk factors, but they don’t like that they’re subject to urinary tract infections and they don’t like running to the washroom every 15 minutes because of their glycosuria. Metformin might be the right drug to start with in such individuals.

Anyway, the bottom line is, I wouldn’t throw them out, but just realize that you’re giving them for one and only one reason: to lower the blood glucose and the possible issues specific to lowering blood glucose, which isn’t preventing the major events.

Roman Jaeschke: Alright, your views are explained and I hope they will be of use to our listeners. They are to me.

I reserve the final judgement until the next two recordings with an endocrinologist and a nephrologist. But in the meantime, I want to thank you for combining and putting this evidence together. I think it was a gigantic work. I’m not sure how you accomplish looking through those thousands of papers and combining them this way, but from the user perspective I want to thank you for it.

Gordon Guyatt: As I said, particularly that network meta-analysis of all the diabetes drugs, which we now have turned into—it’s my Chinese colleagues who are leading that, and particularly a guy named Sheyu Li—it’s now a living network meta-analysis.

Roman Jaeschke: And where can we find it?

Gordon Guyatt: Well, we’ll update it. You will find it when enough changes that there are messages, we will put it out and you will find it. And hopefully once again, the BMJ will publish it. I suspect they will.

Roman Jaeschke: Gordon, thank you very much and thank you for giving us your time and expertise.

Gordon Guyatt: It’s been a pleasure, Roman.