This transcript has been edited for clarity.
John M. Mandrola, MD: Hi, everyone. This is John Mandrola from theheart.org and Medscape Cardiology. I'm here at the American College of Cardiology meeting and, I'm very happy to have my friend, Andrew Foy, who is associate professor in medicine and public health sciences at Penn State Health Milton S. Hershey Medical Center. Andrew, welcome.
Andrew has a paper at the meeting in which he looks at multimorbidity in clinical trials. I think this is important because one of the biggest things that we do in medicine is to try and take trial evidence and translate it to the person in front of us. This paper looked at comorbidity burden, or in other words, how much like our patient are the patients in the trials.
Andrew, tell us about the general idea behind the study.
Treatment Effect Heterogeneity in Major CV Trials
Andrew J. Foy, MD: Thanks for having me.
The trials that we looked at included ACCORD, which had basically three main studies: blood pressure, lipids, and glucose. We looked at the historical AFFIRM trial, which was rate vs rhythm control. We looked at the BARI 2D trial, which had two studies within it, a revascularization trial and then a glycemia trial. We looked at SCD-HeFT, which had two placebo comparisons, one with an implantable cardioverter-defibrillator (ICD) and then another one with amiodarone.
We looked at the SPRINT trial and we looked at the TIMI IIIB trial, which also had two studies within it. It had invasive vs conservative management for acute coronary syndrome, and it also had tissue plasminogen activator (tPA). Those were the studies that we looked at.
We had patient-level data from these studies, which we got from the NHLBI BioLINCC data repository. With the individual patient-level data from the trials, we classified patients on the basis of their Charlson Comorbidity Index (CCI) scores. Then we essentially redid the original analysis through the CCI-based subgroups and looked for evidence of treatment effect heterogeneity.
I think the main takeaway from the analysis of all these studies is that, generally speaking, the patient populations of the trials tended to be fairly narrow in terms of comorbidity burden, which many people might expect.
One way we characterized that was by looking at the variance of the CCI scores around the median score of trial participants. It generally varied from about 1.5 to 3.25, but in general, most of the trials had a narrow population of patients. The majority of patients had comorbidity scores ≤ 4, which is about a 60-year-old patient with perhaps one or two major comorbidities.
The last finding was that, despite the narrow range of comorbidity burden within the trials, we still found treatment effect heterogeneity based on comorbidity score. It was suggested in about five trials where we found P values < .1 for interaction, and in three studies, the P value was < 0.05.
There was also some other treatment effect heterogeneity on a risk difference scale. Overall, we probably found about eight of what we think are potentially important interactions.
Mandrola: Major cardiology trials that you got open data from enrolled a narrow patient population with little variance. Tell us what exactly you mean by treatment effect heterogeneity, because in a typical paper, we get the overall results and then we get these subgroups. What exactly does treatment heterogeneity mean?
Foy: It means that the summary effect from the trial may differ significantly in a particular subgroup of patients. The way that's traditionally done in randomized clinical trials is this classic one-variable-at-a-time analysis. Oftentimes, there's not much biologic plausibility to why treatment effect heterogeneity would occur based on the presence or absence of these individual variables.
It's really hard to, from a holistic standpoint, apply that to patients who are more complex than having one variable or not. Using a comorbidity score, I think, allows us to more holistically capture patient complexity. You could make a case that it's better for finding treatment effect heterogeneity than the way that we classically do it.
Mandrola: Is the idea that a trial will give you an average result in an average patient, but when you're in clinic…
Foy: Well, it's the average patient in the trial. It may not be the average patient in our clinic. That's the idea with the summary effect.
Mandrola: I've always thought that getting in a trial is a special thing. You have to be a special patient, and then you have special circumstances in the trial. I've always wondered whether that should be factored in when applying the results to the patient. Right?
The Charlson Comorbidity Index
Foy: Yes, but one thing that's really challenging for clinicians is, how do we translate that? If you just look at how the trials are classically presented to us, with Table 1 that has the presence or absence of individual characteristics, it's hard to say anything about who the actual average patient in the trial is. Having a comorbidity score, I think, allows us to look at that in a more holistic way and have a better understanding.
Mandrola: Let's talk about the CCI and comorbidity burden. What exactly is it? Why is it important? How is it different from just adding up a bunch of patient characteristics?
Foy: The CCI has been around for a long time. It's probably the most highly validated comorbidity score. It is a significant predictor of mortality, morbidity, and hospitalization. The predictive ability of this score has withstood the test of time in many ways. Even if you don't have all the variables of the score present, the relative weights of the individual covariates relative to one another retain their weight. The score is quite flexible in that way, and you can apply it in these clinical trial populations.
Mandrola: One of the really disruptive things — you gave this lecture at the Kentucky Chapter of the ACC on comorbidity burden — that I had always thought and we learned from the statin trials, is that the higher the average risk of the patient, the more likely that relative risk reduction will result in an absolute risk reduction. The sicker the patient, the more they're more likely to benefit, but some of the findings that you find are not that. It's like the opposite.
Foy: I think we predicted that, but I don't think most people truly appreciate that. A great example was the SCD-HeFT trial. We know SCD-HeFT is one of the historic trials showing a benefit of ICDs for primary prevention of sudden cardiac death and of death in general.
You probably presume that the patients that were in that trial reflect the patients that you see generally in clinical care, but in the SCD-HeFT trial, we found a significant signal where it was actually patients in the first quartile of comorbidity score who had the most significant benefit. For people out there, that score was ≤ 1.
Mandrola: The healthiest patients.
Foy: They all had the comorbidity of having systolic heart failure, so this would have been a patient in their 40s with no other risk factors other than cardiomyopathy. Also within SCD-HeFT, patients in the fourth quartile of the comorbidity score had no benefit whatsoever from the ICD. The hazard ratio was 1.0. Those patients were three times more likely to die.
Mandrola: Tell us simply why the youngest, healthiest patient would stand to benefit from an intervention, whereas an older, higher-risk patient who's more likely to have an adverse outcome and to have the effect — sudden death — wouldn't they benefit more?
Foy: It has to do with a consideration of all the domains that are involved in determining treatment effect heterogeneity. Those four domains are outcome risk, which is the outcomes that we'd be interested in. For example, in SCD-HeFT, that was sudden cardiac death and how that causes death in general. Then there's competing risk, which are events that occur and don't allow for the outcome of interest to occur. That would just be death from other causes in a trial like SCD-HeFT.
Then there's also treatment-related benefit and treatment-related harm. I don't think for the purposes of SCD-HeFT , for ICD, there's too much treatment-related harm. Even though patients in the highest quartile of CCI scores had the highest risk of dying and also the highest risk of dying from sudden cardiac death, they also had a much higher risk of dying from other things. The ICD can't do anything about that.
Under those circumstances, the benefit of the ICD is washed out. We should also keep in mind that in SCD-HeFT, in particular, that fourth quartile of patients had a CCI score ≥ 5. I would argue that's probably the average patient in whom we put ICDs in practice.
Mandrola: I'd also want to note that there have been other papers on that. There's a paper by Fishbein and colleagues that looked at the 6-minute walk test, showing that patients who did very poor in the 6-minute walk test didn't benefit.
This is a hypothesis-generating study, so we don't change what we do now, but what are the implications for this kind of work looking at comorbidity index from these trials?
Foy: First, it lends credence to the idea that we may have issues with external validity of our contemporary clinical trials. The notion that these results may not be representative for the patients that we think they're representative for, I think that's the first thing.
Two, I think it shows that comorbidity matters, and it has an interaction with treatment effect. Because patients with higher comorbidity tend to be excluded from clinical trials, we need to be very careful about assuming that they also benefit from things that we find to be useful in contemporary clinical trials.
I think the third is that trying to more systematically capture these variables would be a really huge benefit to the evidence-based medicine community and to people and physicians, in general, who are trying to clinically translate results from RCTs.
Trial funding comes from many places and many independent bodies, but if there could be some sort of general consensus around what variables are going to always be collected so we can always assess this, I think that would be a massive benefit to the field.
Mandrola: Excellent. Thanks for coming on. I really appreciate it.