Name:
Gordon Guyatt, MD, discusses network meta-analysis.
Description:
Gordon Guyatt, MD, discusses network meta-analysis.
Thumbnail URL:
https://cadmoremediastorage.blob.core.windows.net/0668b047-beb6-4725-bc0a-bb65fec56d4b/thumbnails/0668b047-beb6-4725-bc0a-bb65fec56d4b.jpg?sv=2019-02-02&sr=c&sig=fNywf1hY9xKCGYhLZjc%2BAjzN6DL9Ru9j7yB%2BAqJeHqI%3D&st=2024-12-22T05%3A59%3A30Z&se=2024-12-22T10%3A04%3A30Z&sp=r
Duration:
T00H09M03S
Embed URL:
https://stream.cadmore.media/player/0668b047-beb6-4725-bc0a-bb65fec56d4b
Content URL:
https://cadmoreoriginalmedia.blob.core.windows.net/0668b047-beb6-4725-bc0a-bb65fec56d4b/14293655.mp3?sv=2019-02-02&sr=c&sig=Q4VFY%2BfhqcyvlrfxbIk1ABjxI2q9EstGSV8KM3cSn6g%3D&st=2024-12-22T05%3A59%3A31Z&se=2024-12-22T08%3A04%3A31Z&sp=r
Upload Date:
2022-02-28T00:00:00.0000000
Transcript:
Language: EN.
Segment:0 .
>> Hello, and welcome to JAMAevidence, our monthly podcast focused on core issues in evidence-based medicine. I'm Dr. Demetrios Kyriacou from Northwestern University and Senior Editor for JAMA. Today, we're speaking with Dr. Gordon Guyatt from McMaster University in Ontario, Canada, Editor of The User's Guide to the Medical Literature. Dr. Guyatt is going to talk with us about network meta-analyses. Dr. Guyatt, what is a network meta-analysis, and how does it different from a regular meta-analysis?
>> Well, a conventional meta-analysis compares two different options for treatment. Sometimes an active drug versus a placebo or control, sometimes one treatment versus another. And what the meta-analysis does is it pools data across studies to arrive at a single best estimate for each outcome. So aspirin versus placebo for preventing cardiovascular events, and you get a point estimate that pools across studies.
Aspirin also increases bleeding, so you would get a pooled estimate across studies, one for the reduction of cardiovascular events and another for an increase in bleeding. What has happened in recent years is that for many conditions, there are now a number of treatments available. And when there are a number of treatments available, typically all of them have not been compared against one another. And even if they were all compared against one another, it would be a little confusing to try and sort out which is the best or which is the better.
And that's where network meta-analysis has come along in the last decade to address that. So network meta-analysis builds on the conventional meta-analysis that just compares two treatments, but it compares all the options available. So just to make it simple, if we have four treatments, A, B, C, and D, it compares A to B, C, and D; B to C and D; and C to D. So all possible comparisons. That's -- when you have four, it's relatively small.
If you have a dozen, it gets to be a large number of comparisons. So it's -- first thing is we now have everything compared against everything else. And I mentioned that often, A and B might not have been directly compared. What the network meta-analysis does, it nevertheless generates a best estimate of A versus B, but using indirect comparisons. So if you're interested in A versus B, but you don't have a direct comparison, but A has been compared to C and B has also been compared to C, that provides an indirect estimate.
So bottom line of a network, it advances on conventional meta-analysis by allowing a simultaneous comparison of all the available treatments using both direct and indirect evidence. >> What are the major limitations of a network meta-analysis? >> Well, major limitations are like a conventional meta-analysis, limitations in the underlying data. So if there are very few comparisons of one agent versus another and also limited indirect evidence, you end up with very imprecise estimates with very wide confidence intervals.
Sometimes, also, the data is inconsistent from one study to another. And we look for explanations of that inconsistency, but sometimes we cannot find it. Sometimes, we may have studies that are adequate in number, but the studies may not have been well done, and there may be high risk of bias. So one set of limitations of either a conventional or a network meta-analysis is limitation of the data that informs your estimates.
Network meta-analysis, I would also say that it's only a decade old, statistically sophisticated. And some of the rough edges are still being worked out as to the best way of summarizing the data, which is also something of a limitation. But the statisticians and methodologists are working on it, and I think that should be sorted out in the next few years. >> What are some of the aspects of a network meta-analysis that you look for to determine if it is a valid analysis?
>> That's the quality of the process of the network meta-analysis. And a number of characteristics are ones that we would also apply to a conventional meta-analysis. So you want explicit eligibility criteria that are also appropriate. So you need the patients to be similar enough across all the studies. You would like to see the interventions administered optimally. You would need to see a risk of bias assessment -- risk of bias for randomized trials being concealment of randomization, blinding, and completeness of follow-up.
And you would like these judgments to be made in a reproducible fashion. So in other words, you want two people to be able to agree that a study is eligible, two people to be able to agree whether there was low or high risk of bias on the basis of concealment, blinding, and completeness of follow-up. In addition, what you would like to see, in particular with a network meta-analysis, is for each of the paired comparisons, you would like the people doing the work to tell you how trustworthy they are.
In other words, what the quality of the evidence is, or at least provide the information that would allow someone to make those judgments. >> So the understanding of results from a network meta-analysis are somewhat similar to a regular meta-analysis. >> Yes. So in a regular meta-analysis, the comparison is A versus B, and you get your best estimate of effect. You get your confidence interval, and you get a rating of the quality or certainty of the evidence.
That A versus B appears in just the same way in a network meta-analysis, but in addition, you get A versus C, D, E, F, and G, as well as all the other paired comparisons. So each paired comparison gives you exactly the information that you would get from a conventional meta-analysis; point estimates, confidence intervals, ratings of the quality or certainty of the evidence. >> So clinicians can apply the results of a network meta-analysis to patient care in a similar way as a regular meta-analysis.
>> Yes. It's a little complicated when you have large numbers of treatments. It certainly has the potential application in exactly the same way. One may be a little overwhelmed by the results, and clinicians may need help in interpreting it. But the essential information is the same sort of information with the same applicability. >> Is there anything else you would like JAMAevidence listeners to know about network meta-analyses?
>> As I said, it's only been really around for a decade, and it is evolving. And at the moment, most of what has been done up to now has not been optimal in providing one crucial piece of information. They will tell you the estimates of the effect for all the paired comparisons. They may even give you rankings of the treatment. But what they will not do is to tell you how trustworthy those estimates are.
And some of the estimates may be high-quality evidence that's trustworthy, others may be low-quality, untrustworthy. Things are evolving so that in the future, those ratings of certainty or quality of evidence will be available. But at the moment, unfortunately, a lot of the network meta-analyses out there are limited in their usefulness because they don't tell you which data you can trust, which estimates you can trust, and which estimates you cannot. But as I say, more and more, those quality ratings are going to be available.
>> Dr. Guyatt, thank you for giving us your insights and advice about network meta-analyses. This has been Demetrios Kyriacou with JAMAevidence speaking with Dr. Gordon Guyatt. For more information regarding network meta-analyses, see Chapter 24 of The User's Guide to the Medical Literature. To hear more of our podcast, go to jamaevidence.com.
