This presentation was delivered as part of a workshop session on the application of adaptive design principles to clinical trials in speech-language pathology. JoAnne Robbins, a swallowing researcher, discusses challenges in her past and current clinical trials, while Sharon Yeatts, a biostatistician, provides her thoughts on potential opportunities and pitfalls in integrating adaptive designs to this work. For an overview of adaptive designs, see Adaptive Trial Designs for the Development of Treatment Parameters.
The following is a transcript of the presentation video, edited for clarity.
JoAnne Robbins, University of Wisconsin – Madison:
So where we’ve been and where we’re going. So, Jeri Logemann had really wanted to do a big clinical trial in the 1990s. The NIH said they had some money that they wanted speech pathology and ASHA to apply for. And so I agreed that I’d share this particular study in swallowing, which was very, very cumbersome and large.
We had two parts to this study. The first, we called it the short-term part. We looked at patients with dementia or Parkinson’s disease who were known aspirators. When we saw them aspirate with video fluoroscopy, if they aspirated on thin liquid we presented them with three different interventions in randomized order. Chin tuck. Nectar-thick liquids. And honey-thick liquids. Because these were the most common treatments at the time being done by the field. I’m not ever going to tell you that’s a good criteria to use in the future.
Part two was then to have patients who aspirated on thin liquids to use one of these interventions for three months. And that would be at random.
So the primary outcome of the first short-term part was: Did they minimize aspiration, these dementia and/or Parkinson’s patients, by one of these three interventions? Chin tuck, nectar-thick or honey thick liquids in the short-term, in the fluoro suite. And over three months of being assigned to one of these interventions, did they get pneumonia? Or did they not get pneumonia?
I want to just back up for one quick minute to say that the NIH applauded this study. They kept telling us through the years, and it went from five years 10 ten years, that they really supported the similarity — or our attempts to replicate what was going on in the field. Again, not really good justification for doing a study. I think we and our representatives at the NIH learned a whole lot.
So, what we found out to begin with was that the materials weren’t even in the field that we needed to do a randomized controlled trial. We had to design bariums, imaging agents that would allow us to compare data in a multi-site randomized clinical trial. Because everybody was doing their own thing with their own materials. That was our first year that was put into that part of the study, which wasn’t even collecting data yet. Big mistake.
And then our findings were very interesting, I do have to say. We found that, we did show there were differential effects of the interventions in the fluoro suite on immediate elimination of aspiration. That is, the thickest, the honey-thick material eliminated or diminished aspiration best in the fluoro suite.
In the three month drink, one of these conditions that you’re randomized to, either chin tuck, honey-thick or nectar thick, honey-thick was not the most effective. In fact, liquid categories, when we collapsed everything, we found that the incidence was higher for pneumonia for honey than for nectar. And, as we looked at our adverse outcomes data and our hospitalization data, because remember these were aspirators, all of these people. We found that those who were randomized to drink the honey-thick liquid for three months, which was shown to be the best quick fix in the fluoro suite, ended up making patients sicker over a longer period of time.
And we had no sham groups.
So there were many drawbacks to this study. And I think the biggest message that I had was: Really work with knowledgeable people who can help you design a clinical trial that can get done in a reasonable period of time. And that you’re coming in at the right level of methods that you need to conduct so that it’s realistic.
Like I said, this was a ten year study. It had been funded for five, and it took ten. And I learned an awful lot.
I’m now going forward. And those results are published. I mean, they’re in the Annals of Internal Medicine. They’re in one of our finest journals in speech pathology. But it was really tough, and there were lots of bumps.
We did try to look at the data periodically. Not me, we were blinded. But our statistical analysis group called MS from Washington, great people, would look at it periodically to see when we could stop.
We could never stop because we never had enough differences among our groups. So it was a lot of pain. None of us got pneumonia doing the study, but it was really uncomfortable.
I’m now moving into the area of exercise. And we’re very, very excited. We do have a quantified type of exercise that we’re doing as an intervention with patients. We’ve done it, short studies, small cohorts with stroke patients.
And it is device-facilitated in terms of the Madison Oral Exercise Device that gives patients feedback or, shall we say, performance information. I just want to show it to you. It has four sensors in the mouth on a mouthpiece. You press your tongue against any or all of the sensors, depending on what version of the protocol you use. You’re given performance in terms of the target value you’re shooting for. It’s based on sports medicine principles. And we’re moving forward. It’s very fun. The patients love it. We have compliance. They don’t want to stop. They do 30 repetitions, three times a day, three times a week. Eight weeks.
So I do want to say that with this exercise intervention, which we had patients perform for eight weeks. As I said, three times a day, 30 repetitions, three times a week for eight weeks. Because we did our cohort study on stroke patients, who are a fragile group, and we didn’t want to extend it too far knowing that we’d have attrition over time. And the data that we show here with anterior and posterior tongue are stroke patients, are the bars. With the mean where the circle is. And hyphenated line are normal age-matched controls. You can see in the front of the tongue that at four weeks the stroke patients really increased strength by 63 percent. And in the next four weeks by 37 percent in the front of the tongue. And similar kinds of findings in the posterior tongue over the eight week protocol.
The thing is they’re still getting better at eight weeks, but we had designed the study and had the funding to only do the eight weeks. And, as you know, to do research takes money. So we now have been doing this eight week protocol, which, did I should say, not just strengthen the swallow, but you can see the data here that shows diminished aspiration-penetration scale scores. So the swallow is doing better over time as they get stronger over those eight weeks, which is really very wonderful. And we have significant differences there.
The VA just funded our new clinic. And it’s unusual to get VA funding for a clinic, but it’s a model demonstration clinic. And in the first year, we had maybe, we started late. As everything in the VA, you wait for funding and has to go through so many hoops. So we had about 20 patients in our first year. And we did show a reduction in the incidence of aspiration pneumonia with this eight week intervention by 62 percent. And a 90-some percent reduction in hospital admissions. So we’re getting some really good outcomes in terms of health status.
And in designing the future to do a larger clinical trial, we have issues that have to be taken on. And I think adapting this strategy is going to be useful in terms of really honing in on a variety of specifics.
And I just thought maybe, Sharon, you would have something you might add to our thinking. Thank you.
Sharon Yeatts, Medical University of South Carolina:
Well, I don’t know how much I can add to your thinking. One of the issues still, I think, for adaptive design from a statistical perspective is that, as the slide very nicely shows, it’s never just one thing that we’re trying to figure out at a time; right?
And it could very easily be the case that maybe three times a day, three days a week is okay. Maybe five times a day, two days a week does just as well. Maybe all of these things interact with each other. And so, and, of course, it’s difficult to get, fit all of those pieces into the same design.
You can always do them sequentially and think about it that way. But then you’re not going to get at this interaction piece.
So, for instance, if you start by saying, I want to know how many times a day is appropriate? When you move to the next step, you’re going, and say, okay, now I want to know how many days a week do they have to do that? You’re going to get the right answer to the question that you asked, which is if three times a day is right, how many days a week should I do? Not, if I’m looking at the whole space of how many times a day and how many days a week, what is the optimal combination?
There are designs for that that allow to you look at multiple predictors at the same time. But, of course, with each of these new, with each of the questions that you ask comes another level of complexity that has to be addressed. And, as you said, doing a standard clinical trial is often complicated enough. So sometimes incorporating all of these things can really make your head hurt.
I work in an area where adaptive designs right now is sort of the hot topic. It’s a buzzword. And everybody wants to do an adaptive design even if there’s no adaptive question. We think adaptive designs are exciting, and that’s what I want to do.
And adaptive designs, they can make getting to the answer more efficient. But there’s a lot of that thinking, in some cases you may be able to say, I really don’t think there’s an interaction. I think whatever, regardless of whether I looked at the whole combination space or whether I took each of these pieces individually, I’m going to get to the same answer. Doing it sequentially is much easier. But you do run that risk that you’re sort of missing the optimal combination.
I think often we forget to think about the fact that a lot of the information, I don’t know a lot, some of the information that we need to answer these questions we can get from the untreated folks. We think it’s just not fair to study them and not give them something. But in your case it’s really important to know when do they stop getting better? Where do they plateau? How can we find that information out?
And I don’t know if that’s something that maybe could be done through a registry since you have the clinic. But, again, those are not, they’re not going to be the untreated patients; right? You’re not going to just let these folks go without anything.