Composite Endpoints: Why Use Them?
Robert A. Harrington, MD: I am Bob Harrington from Stanford University. I am here in Washington, DC, with David Cohen from the University of Missouri-Kansas City and Saint Luke's Mid America Heart Institute.
At the Transcatheter Cardiovascular Therapeutics (TCT) meeting, we have heard a lot of trial results. I am often asked about composite endpoints. Why do we use composite endpoints? What are some of the challenges with using composite endpoints? I would also like to discuss a very interesting study[1] that your group recently published on how patients and clinicians may view the components of composite endpoints differently.
David J. Cohen, MD: It is a great topic, because composite endpoints are ubiquitous in our trials. Almost no trial uses pure mortality as an endpoint.
Dr Harrington: Why do we use composite endpoints? Why don't we just pick one outcome and say that is what we are looking for an effect on?
Dr Cohen: The biggest reason for using composite endpoints is efficiency, because it gives you more endpoints. If the right conditions occur, composite endpoints improve the efficiency of the trial. You could enroll fewer patients in a shorter observation time, for example, but many things have to break right for that to work.
Dr Harrington: That is what we all worry about. We all say, "We need more events." We want to reduce the sample size, and we can reduce sample size in part by having more events. So we increase the efficiency by, for example, not having to follow people as long. What has to break right?
Dr Cohen: Ideally, all of the component endpoints have to be reduced to a similar extent and to the same degree, in the same direction.
Dr Harrington: That sounds simple, but it is not.
Dr Cohen: It isn't, and it is often violated, and there are challenges. But if everything breaks right, you have a very consistent effect. You have a similar relative risk reduction.
Dr Harrington: If you have 20% risk reduction in death, myocardial infarction (MI), and stroke, it's easy.
The Challenge of Endpoint Inequity
Dr Cohen: Right, everything is clean and easy. You understand exactly what is going on. What usually happens in these trials is that they are dominated by one component or another, and it is usually the least important of the different components. The worst-case scenario is when the dominant component goes in a different direction from the other components, and then you have a real challenge in interpretation.
Dr Harrington: For example, it would not be unusual to see revascularization go in a different direction from MI.
Dr Cohen: Right. The most apt example in our field of interventional cardiology is the SYNTAX trial,[2] a composite of death, MI, stroke, and revascularization. Revascularization is dramatically reduced by bypass surgery, and stroke is increased. We have a challenge because those were the only endpoints that differed at 1 year. How do you interpret that? What is more important? How do you weigh the different magnitudes?
Dr Harrington: Let's talk about how one thinks about the analysis. In a traditional time-to-event analysis, it is strictly the time to that first event. It doesn't matter whether a nonfatal event occurs first, even if a fatal event occurs later. You count the first event in the primary composite.
In the secondary analysis, when you are looking at all the components, you count the other events, and you might even do a hierarchical analysis, according to what is the most important. But that is a problem, because we are essentially saying in a time-to-event analysis that all these events have equal weight.
Dr Cohen: Right. Most people will say that they don't have equal weight, but we don't know what the right weights are, so "equal weight" seems less arbitrary than anything else. It is still an arbitrary assumption, however, and is clearly not right.
Dr Harrington: From a mathematics perspective, it is easiest to say that they are equal. You reach one, and you are done. However, it is important, and I always caution people, "Make sure that you are looking at all the individual components, and make sure that you have counted everything. If you prevent MI, did you prevent MI because the patient died? You have to make sure that you are looking at all of the different events."
Weighting, Rating, and Trading Off Endpoints
Dr Harrington: Let's talk about weighing the components. Traditionally, we have not given them any weight, but people have tried over the years. Years ago at Duke, we (the TAMI Study Group) were very interested in understanding the importance of a stroke relative to a heart attack. What is the importance of an intracranial hemorrhage relative to dying?
Dr Cohen: At that time, you were doing some trials of a drug that had overt trade-offs. You studied tissue plasminogen activator vs streptokinase, and you found lower mortality but higher intracranial hemorrhage.
Dr Harrington: We could count that because one half of the patients with intracranial hemorrhage died, and the other half had a disabling stroke.
Dr Cohen: Such a big disability is very relevant. In interventional cardiology these days, the most common event is bleeding, which goes in one direction and ischemic events go in another direction. This occurs because our strong antithrombotic therapies almost always increase bleeding. Those challenges exist, and that is what we tried to address in the study.
Dr Harrington: I was fascinated by that because this balance, and the trade-offs of ischemia and bleeding, are something we have worked on for two decades. The US Food and Drug Administration (FDA) struggles with how to define net clinical benefit. Do you look at a composite endpoint that includes bleeding? Do you look at different ways of weighing it, or only the most severe bleeding or most severe ischemia events?
Sizing Up Differences of Opinion
Dr Harrington: You took it a step further. God forbid, you actually went and asked the patients what they thought! Tell us what you did.
Dr Cohen: We asked patients and doctors. I shouldn't take too much credit for this study.[1] It was designed by my colleagues at Mid America Heart Institute (or, as it was called at the time, the Heart Institute): Josh Stolker, who is now an interventionalist at Saint Louis University, and Paul Chan. They had the idea of trying to understand how patients weigh these different endpoints.
They set up a station in our outpatient clinic to administer very structured surveys to 785 patients who were just sitting there waiting for their appointments. Like most places, they have downtime while waiting.
Dr Harrington: You might as well engage them in the research process.
Dr Cohen: Exactly. Five different endpoints make up the typical composites for lipid-lowering or revascularization trials. These are death, stroke, MI, repeat revascularization, and hospitalization from unstable angina.
The patients were asked to take 25 "spending weights" (eg, 25 pennies or 25 dollars) and allocate them across those five endpoints.
Dr Harrington: They had to spend the entire 25 and no more? It's the classic way of doing this kind of research.
Dr Cohen: Exactly. The order of the surveys was varied to keep everything balanced. It was very rigorously done. The survey was completed by 785 patients who had coronary disease and were candidates for these therapies. They also did a survey of investigators of revascularization trials and lipid-lowering trials. They gave the investigators the same set of surveys.
There were several findings. First, the clinicians and the patients didn't agree. That was perhaps the most important finding. The patients rated stroke and MI a little worse than death. Death was in the middle, and revascularization and hospitalization were rated as the least important.
The clinicians rated death the most important by far. Stroke and MI were intermediate, and then revascularization and hospitalization were least important. Within the patient group, there were differences—individual to individual, and within certain subgroups. One of the most noteworthy was among the elderly. The older patients (65-74, 75-84, and > 85 years) in the study rated stroke and MI substantially worse than death. Death was not so important to them. The revascularization endpoint was the least important for older patients, whereas the younger patients rated death as being more important than stroke or MI.
So, there were differences between clinicians and patients, and differences within patients.
Dr Harrington: We did a study[3] years ago with one of the platelet glycoprotein IIb/IIIa inhibitors. In one of the studies in which we had both stroke and coronary patients, we asked the patients about trade-offs. We asked, "Would you accept bleeding if you could avoid another heart attack or stroke?" Not surprisingly, patients who had had a stroke did not want to have another stroke and were willing to accept the trade-off of bleeding, whereas patients who had had an MI weren't as willing to accept the trade-off of bleeding.
This tells us that we have to think differently. It is interesting from a PCORI perspective, and we should consider this more.
Endpoints and Decision-Making
Dr Cohen: We absolutely have to consider it. It seems to me that the patients are the arbiters here. They are having the events. The clinicians know more about the events, but the patients are having them, so their opinions have to be respected the most. It is very interesting.
Jeff Shuren from the FDA gave a talk on the first day of TCT about how things are being modified at the FDA, and he mentioned that it is a priority for the FDA to be able to approve devices that the agency doesn't think has the appropriate risk/benefit ratio. For example, if the safety profile is somewhat dubious, but patients think it is an acceptable trade-off, the FDA is willing to consider approving it, which is a very new stance for the FDA.
Dr Harrington: That is a fundamental shift. What it says to me is that we have to do the kind of research that your group did to quantify that. We have to rigorously define what those trade-offs are.
Dr Cohen: We still have a lot more work to do. This was a specific set of trade-offs and a specific set of values, and you can imagine using this information in a patient-centered decision-making model.
Dr Harrington: I have been fascinated by the truly informed consent that you have used and studied that asked people about trade-offs—trading off some re-stenosis for some bleeding, for example.
Dr Cohen: We are trying to understand what is important for each individual patient. This would allow a clinician at the bedside to assess the patient's preferences and then use those preferences to direct what you do.
Dr Harrington: As we do more and larger clinical trials that are embedded into the point of care, using data coming from the electronic medical record, we need to spend time thinking about how we record what the patients think.
Dr Cohen: That is exactly right.
Dr Harrington: This has been a terrific discussion. The whole issue of composite endpoints continues to be an important one. For the researcher, it is a critical issue. For interpreters of the research, it is critical. I love the fact that we are now engaging patients in trying to understand how they value these endpoints.
Dr Cohen: I completely agree. Hopefully I will be able to come back and share some more findings as we evolve this research further.
Dr Harrington: Thanks for joining us, and thank you for listening.
© 2014 WebMD, LLC
Cite this: Is a Stroke Worse Than Death? Depends on Who You Ask - Medscape - Nov 14, 2014.
Comments