Name:
Matthias Briel, MD, discusses the principle of intention to treat and ambiguous dropouts.
Description:
Matthias Briel, MD, discusses the principle of intention to treat and ambiguous dropouts.
Thumbnail URL:
https://cadmoremediastorage.blob.core.windows.net/45c8bfe0-c1f6-4e95-87e0-bd1424b2cfdb/thumbnails/45c8bfe0-c1f6-4e95-87e0-bd1424b2cfdb.jpg?sv=2019-02-02&sr=c&sig=7k%2BzVgObvoLXQ0DJCM91nFExzdnqENUaWM2sewSocrQ%3D&st=2024-11-05T06%3A13%3A48Z&se=2024-11-05T10%3A18%3A48Z&sp=r
Duration:
T00H19M25S
Embed URL:
https://stream.cadmore.media/player/45c8bfe0-c1f6-4e95-87e0-bd1424b2cfdb
Content URL:
https://cadmoreoriginalmedia.blob.core.windows.net/45c8bfe0-c1f6-4e95-87e0-bd1424b2cfdb/13944183.mp3?sv=2019-02-02&sr=c&sig=KhEVjHC7HSMexRFJLM9yl%2Bt4rOB6jXPrTjylIGNUNsE%3D&st=2024-11-05T06%3A13%3A48Z&se=2024-11-05T08%3A18%3A48Z&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. Today, we're discussing the principle of intention-to-treat. >> So, intention-to-treat is an important principle because it preserves randomization in a clinical trial. >> This is our guest today, Dr. Matthias Briel. Dr. Briel is one of the authors of a chapter from the Users' Guide to the Medical Literature entitled, The Principles of Intention-to-treat and Ambiguous Dropouts.
>> So, we are randomizing patients to two or more groups in the trial because we want to achieve prognostic balance between groups at the outset of a trial, and so that if we find differences in results in event rates, for example, at the end of the trial between groups then we can attribute these differences actually then to the intervention that was applied in comparison to a control and not to prognostic differences that already existed at the outset.
So, in order to preserve this prognostic balance at the outset throughout the trial, intention-to-treat is necessary. With this principle, we are analyzing the patients in the group to which they were randomized, whether they took their treatment or whether they didn't take their treatment, and this helps us to keep this prognostic balance. >> Amy Thompson, our Associated Editor here at JAMA, continues the conversation with a look at a hypothetical randomized trial from the chapter that studied patients with cerebrovascular disease where two groups are randomized to either surgical intervention or aspirin-only therapy.
>> Dr. Matthias, would you set up the trial for us? Just give an idea about the basics of this hypothetical trial that we'll be using as our example? >> So, in this hypothetical trial, we are randomizing 2,000 patients to two groups. So, the intervention group with cerebrovascular surgery and aspirin and the control group with also 1,000 patients would receive just aspirin. And the outcome of the trial would be the number of strokes then per group.
>> Okay. And so, in our hypothetical example, just in order to illustrate intention-to-treat, we're going to assume that the intervention has no effect. >> Yes, correctly. So, there should be no difference in the end then of the trial between the intervention and the control group. >> So, then let's focus first on the intervention group. We have 1,000 patients randomized to this group. And so, what do we see in the first couple months after randomization just for this group?
>> So, in the intervention group, there are 1,000 patients and usually, patients who are scheduled for cerebrovascular surgery do not have the surgery right away. So, they have to wait, for example, a month or so. And during this first month that they are waiting for their surgery, if there is a constant event rate, then already 100 patients have a stroke before they even have their intervention.
And with this constant event rate, also 100 patients after their surgery have a stroke. So, that makes, in total, for this intervention group 200 patients having a stroke. But if we count only the patients who actually had the intervention, so it's 900 patients, that means that if we analyze the data in per-protocol session, then we would only divide 100 patients who had a stroke over 900 patients who actually had the intervention, which would be a risk of about 11% for having a stroke.
>> And then, let's talk about the control group. So, we have a constant event rate, which we're assuming is about 100 strokes per month in each group of 1,000 patients. So, what happens to the math when we try to compare these groups so we end up with still 1,000 patients in the control group but now only 900 patients who actually received the intervention in the intervention group.
>> Exactly. In the control group, there are 1,000 patients that received only aspirin who are not scheduled for surgery and with this constant event rate of stroke, there is also 100 patients having a stroke in the first month after randomization and another 100 patients having a stroke in the second month after randomization. So, total number of patients here having a stroke is 200 patients over 1,000 patients who were allocated to this control group meaning the risk for a stroke would be 20% then in the control group.
And if we compare then the two groups, so the 11% risk for stroke from the intervention group and the 20% risk in the control group, we see that there is a clearly lower risk than in the intervention group which is false because we said already, we assume that there is no difference, no treatment benefit, from the intervention. It's just because we artificially excluded the 100 patients in the intervention arm that had not yet [inaudible] the intervention.
>> So, in this chapter, there is I think a different aspect of intention-to-treat that I would like to touch on. You know, it doesn't just help researchers avoid incorrect interpretations of their data. I think that as clinicians we really want to assume that our patients are going to adhere fully to their treatments to maximize the benefits, but we also know that noncompliance is very real and not uncommon.
So, I think that the intention-to-treat principle in this way also improves our interpretation of the data for a more real-world population. >> Exactly. So, with the intention-to-treat principle, we are really looking at a population level. So, it's on average, including all the patients randomized to the specific group, whether they are compliant with the treatment or not, they are considered in the analysis.
That means it is very relevant for policymakers, for example, who need to have precise data on a population level. It might be more useful than for clinicians to actually have data for a subpopulation of the total that was actually fully compliant with treatment. For example, if you have a patient in front of you in your office or in the hospital who you know that this patient will be completely adherent to your instructions and to treatment instructions.
And so, for this patient, the relative risk or the benefit that you can explain from the results of a randomized controlled trial that was analyzed in an intention-to-treat fashion might be not as accurate and maybe not very convincing because the treatment, in fact, is basically diluted by the fact that there is a certain proportion of patients who were not compliant with treatment. The treatment benefit for this individual patient who is very compliant with treatment, this benefit might be much larger and that is more relevant than in this specific case.
And in order to come up with estimates for this very compliant subpopulation then we can do a per-protocol analysis. >> Just to make it clear, can you define a per-protocol analysis for us? >> So, a per-protocol analysis is an analysis in which investigators include only patients that fully follow the protocol of the trial. That means they were not lost to followup so they were available for each visit until the end of the trial and who also followed the instructions of their treatment arm.
So, when they were randomized to the intervention arm that they fully adhered to the interventional treatment or when they were randomized to the control arm that they took their control treatment as instructed. >> There was a very interesting example in the book of a trial in which two groups were discussed and they were analyzed in this per-protocol fashion with the people who were highly compliant with the treatment versus those who weren't. And I found it fascinating that there were significant benefits for the highly compliant patients no matter if they were randomized to the medication intervention or placebo.
Can you speak about that? >> Yes. That is a general phenomenon that patients who are fully compliant basically represent a healthier subgroup of patients who have, in general, healthier behavior and, therefore, also have a better prognosis then in the end. So, no matter whether they are in the intervention or in the control arm, patients who are compliant with therapy often live healthier and, therefore, have a better prognosis.
>> And so, you mentioned doing a per-protocol analysis of a subpopulation in order to really explore the best estimate or what the maximum effect would be for a highly compliant group of patients. The chapter mentions this issue as a possible limitation of the intention-to-treat principle. And the piece of advice in the chapter is that the safest sort of approach for a clinician interpreting that study would be to assume that for a highly compliant group the best way to potentially apply the results to that highly compliant patient would be to assume that the benefit seen in the study is actually an underestimate.
>> In a trial that was analyzed based on the intention-to-treat principle, the results are basically then an underestimate of the benefit a patient would have if he or she was fully compliant with treatment. Therefore, it is sometimes helpful if you then also analyze the data in a per-protocol fashion and then come up with a more appropriate estimate for patients who are fully compliant with the treatment.
And the difference between the two, so the results analyzed following the intention-to-treat principle or in a per-protocol fashion, is larger when the nonadherence to treatment is very substantial. If there are not many nonadherers in a trial, then the per-protocol analysis and the intention-to-treat analysis usually give very similar results. >> I'd like to touch on one more issue that is brought up in the chapter and that is the misleading use of intention-to-treat.
Can you discuss this for our listeners about how this principle might be misleadingly applied or mentioned when a study is published? >> Intention-to-treat has been identified as an important feature for a valid randomized controlled trial. So, the analysis, as I explained earlier, yields results that are unbiased if the prognostic balance at the outset of a trial is kept throughout until the final analysis.
And, therefore, a lot of reporting guidelines pick this up and recommend it that intention-to-treat should be mentioned in the analysis section and that that would be then a quality criterion basically for a randomized controlled trial. Investigations have often followed these reporting guidelines and then just used this term without often completely following this principle. They thought they would do it, but they did something completely different or even just slightly different, but they still called it intention-to-treat without really explaining what exactly they did and which patients they finally included in the analysis and which they did not.
And this leaves a lot of room for uncertainty then for readers of a trial publication. And investigators even use the new terms like modified intention-to-treat. So, if investigators not explain really in detail which patients were included in the analysis and how investigators treated patients that were lost to followup, for example, in the trial in the analysis, then it is very difficult to understand what the investigators actually did.
>> And CONSORT addressed this in a statement about five or six years ago that recommended different language in order to be absolutely clear. What is that recommended language to look for in a trial to be able to assume that investigators faithfully applied the principle of intention-to-treat? >> Yeah, that was actually a very good update of the CONSORT guidelines that mentioned explicitly that the term intention-to-treat was used very often without any further explanation what exactly the investigators did in their analysis.
So, this CONSORT update in 2010 actually recommends against using the term but instead, the investigators should explain what they did in terms of whether they analyzed the patients in the groups to which the patients were randomized, whether they took the treatment or not. But the essential for intention-to-treat is to keep the randomized groups from the randomization process until the analysis.
>> As opposed to intention-to-treat, which would then include those participants who were lost to followup or did not fully adhere to their treatment? >> Exactly. So, in an intention-to-treat analysis, you would include all patients in the groups to which they were randomized. However, to follow the intention-to-treat principle, you don't necessarily have to include all patients in the analysis, meaning patients who were lost to follow up, for example, for whom you don't know the outcome, do not necessarily have to be included in the analysis to call it intention-to-treat.
The point is that the two concepts are often mixed. So, some researchers are fully convinced that it is only intention-to-treat if all the patients are considered in the analysis. So, also including patients that were lost to followup for which investigators then have to assume some outcomes so they impute data for these patients. But this is just a method to handle missing outcome data, basically.
So, patients lost to followup. For the intention-to-treat principle, this is not directly relevant. For the intention-to-treat principle, it's only relevant that patients are analyzed in the groups as they were randomized. So, whether they changed their treatment or did not take their treatment at all, that does not matter, they are still analyzed. But lost to followup is a different issue. These are two different concepts.
>> In the chapter, it talks about the ambiguity that can arise when referring to participants of trials who drop out. Where does that ambiguity come from? >> So, the ambiguity from the term dropout can arise for two different reasons. So, one reason is that a patient is actually lost to followup so that he or she just moved away and was no longer available for a follow-up assessment, or the patient was just not adhering to treatment and, therefore, is called a dropout by the investigators.
So, again, these are two different phenomena and it is often not clear to the reader of an article whether an investigator means a patient lost to followup or just a non-compliant patient when the investigator uses the term dropout. >> But by the intention-to-treat principle, that second example that you mentioned, patients who are termed dropouts because they don't fully adhere to their treatment, would have to be included in the analysis? >> Yes. Exactly.
>> Dr. Briel is an Assistant Professor at the Basel Institute for Clinical Epidemiology and Biostatistics, part of the Department for Clinical Research at the University of Basel in Switzerland. He is also a remote faculty member at McMaster University in Canada. More information on this topic is available in the Users' Guide to the Medical Literature and on our website JAMAevidence.com where you can listen to our entire roster of podcasts.
Thanks so much for listening and we'll be back soon with another episode of JAMAevidence.