Digoxin in AF and Leveraging the VA System for Data Analysis

; Mintu P. Turakhia, MD


September 25, 2014

This feature requires the newest version of Flash. You can download it here.

Digoxin: Old Drug, New Problems?

Robert A. Harrington, MD: Hi. This is Bob Harrington from Stanford University, on Medscape Cardiology and theheart.org. It's good to be talking with you today. We have an interesting discussion that we're about to have with one of my colleagues from Stanford University: Dr Mintu Turakhia.

One of the things that seems to resonate on this show is, how do we think about the more common things we do in cardiology as evidence emerges over time? This comes up when we think about drugs that were studied a long time ago for which there is no contemporary evidence base, but that have become part of the background of what we all do. There is great interest in how one revisits that over time.

There was a recent paper published in the JACC[1] (Journal of the American College of Cardiology) by Turakhia and colleagues from the VA system, which looked at the question of how we use digoxin in modern-day practice in our patients with atrial fibrillation (AF). Most of the studies of digoxin date back a number of years now, and yet it is a drug that people still struggle with how to use. Part of that has to do with the complexities of digoxin itself in terms of its dosing algorithm—how one thinks about following patients to avoid toxicity, yet at the same time make sure that one is getting a therapeutic effect if you choose to use the drug. It's one of the longest-acting drugs in the cardiovascular armamentarium, but it's a drug where we still have major gaps in our knowledge base.

So, I'm very excited today to have as my guest, Mintu Turakhia. Mintu is Assistant Professor of Medicine here at Stanford University. He is a clinical electrophysiologist and health services researcher. He is director of the electrophysiology program at the VA Palo Alto Health Care System. Mintu, thank you very much for joining us here today on Medscape Cardiology.

Mintu P. Turakhia, MD: Bob, thank you for that great introduction to digoxin, and it's really great to be here.

Dr Harrington: Mintu, these are the kind of studies that capture our community's attention. You and I were talking before we went online about the number of comments on theheart.org about this paper. You certainly have gotten a lot of publicity around it. The lay press picked it up and has referred to it as a "deadly foxglove" and a number of other phrases.

So, let's dive into it with this question: Why did you even tackle this? What were you seeing in clinical practice that made you say, "This is really worth studying?" Because all good observational studies start with a clinician saying, "We ought to know more about this."

Dr Turakhia: That's a great question. There were several layers that got me really intrigued.

Of course, I'm very interested in AF at all levels. But it's an exciting time in the world of AF, because the landscape of how we treat it has changed over the past 10 years. The numbers have grown, and we're facing major questions around the effectiveness of therapies and policy issues in how to deal with AF in large populations. That was one.

As you said, it's questioning some of these old paradigms where we have an experience base—maybe not the evidence base, but the base of collective physicians and clinician experience seems to suggest that this is a reasonable strategy for treatment.

On the other hand, digoxin in particular to me represents one of the ultimate contradictions in medicine. On the one hand, we hammer into every medical student the hallmark findings of digoxin toxicity on an ECG and how it presents. It's actually on the boards of internal medicine, cardiology, and emergency department training.

On the other hand, we have repeatedly convinced ourselves that if managed carefully, digoxin probably safe. Perhaps that's true, but the question we had is, is that really true at the population level?

Dr Harrington: When you were thinking about this (before we get into the study), you were thinking of your training and about taking care of this group of patients. I love your phrase, "one of the great contradictions in medicine." Because you're absolutely right. We talk about digoxin toxicity. Certainly, people like you and I hammer home how to find it on the ECG. Those of us who like medical history also wax on about William Withering and the use of foxglove to treat dropsy.[2]

This is a drug we've known a lot about. But what's been your observation in practice? Were you troubled that maybe people weren't as good at using it as they thought they were?

Dr Turakhia: This study where we looked at outcomes actually follows initial work in which we simply looked at patterns of utilization of various drugs and strategies of rate and rhythm control.[3] This is in the VA system, but also in other healthcare systems. There were a couple of interesting findings. One was that we were seeing a lot more digoxin use in noncardiology settings over the past eight years.

Dr Harrington: By that, you mean primary care settings?

Dr Turakhia: Primary care, general medicine, and family practice. To keep it clean, we were looking at new diagnoses of AF, and even controlled or stratified on the basis of how far people were from a cardiologist to avoid treatment selection about who they see. It was clear that there was more utilization in primary care. This is not an uncommon story for other drugs.

I'm not saying that it shouldn't be used or that one group of providers is not doing the right thing. But there was a gradient of treatment. Even in those groups, the use of digoxin has declined over the past five years, and a lot of studies have shown that. But the initial observation that set up the outcomes question was, is the person whom you see largely dictating the kind of care you get for AF?

We aren't at the point with AF that we are with acute coronary syndrome (ACS), coronary disease, and heart failure, where we hammered home the drugs and what people should be on in terms of quality measures. AF is a different beast. It's more heterogeneous. There isn't a one-size-fits-all. But on the other hand, is digoxin really the right drug for initial therapy? That's what we were aiming to evaluate.

Dr Harrington: It makes me reflect. I was on service last week; there are a series of patients that you see now on a general cardiology service, but probably none more frequent than elderly people with preserved left ventricular (LV) function and AF in whom the rate is not well controlled. There is an epidemic of disease among that particular group of patients. It is interesting that a lot of the early work on digoxin was done in people with impaired left ventricular function. Maybe it was to control AF, but maybe it was to treat their LV dysfunction—their systolic dysfunction. But we're in a different era today, and I like the fact that you were asking this in a contemporary data set.

So, let's get into that. You had these observations. You wanted to understand the use of digoxin, and importantly, you wanted to understand the outcomes related to how it was being used. Talk about the methods and the background, what the data set was, et cetera.

VA System: A Victim of Its Own Transparency

Dr Turakhia: The VA healthcare system is unique, because it's different from Medicare data in that it's not just medical claims. Traditionally, a lot of outcomes research in large populations is done looking at medical claims. Claims are, of course, based on the diagnoses or procedures that are coded by the clinicians or by the hospital billing for it, et cetera. A lot can be done with that. What's unique, however, about the VA is that it's a linked system where you have the claims data, but you also have laboratory data and you have prescription data.

For example, the laboratory data allow us (with linkage to vital signs, weight, height, and race) to calculate the glomerular filtration rate (GFR) and to assess the baseline level of kidney function, which was important for this analysis.

In addition, we were able to link this to Medicare and death records. So, the power here is that you're not just capturing the data from within the healthcare system. Of course, you're capturing mortality, which was the outcome assessed. But you're also capturing non-VA hospitalizations or utilizations as well. Those weren't as important for this study. But it's a comprehensive cohort and a very rich way of looking at these questions, where you can basically slice the entire VA population and see what happens.

Dr Harrington: Mintu, I was going to wait until the end to dive into this question, but maybe I'll take the sidebar and have you talk about it now. As you know well because you've been in a VA system for at least part of your clinical life, the VA has come under scathing attacks in the country for inefficient care, problems with access, and a number of issues. Yet, when you see something like this paper, and the description of the data set and the richness of the data available in the VA system, you can't help but scratch your head and say, wait a minute, there's a contrast here.

You have, on the one hand, people screaming about access and inability to do things. On the other hand, you say we have one of the most refined data systems of any healthcare system in this country. Have you thought about that? Reflected upon it? Talked about it?

Dr Turakhia: I have. It's a great observation, Bob, and I think it's an important one. Of course, the primary mission of the VA and the Veterans Health Administration is to provide care to its beneficiaries. But in the process of doing so, they have overhauled their electronic medical record (EMR) system of chart capture—everything since the early 2000s—and have created a really rich cohort of data.

I think that is one of the reasons the VA has the transparency it does. It's not easy necessarily to find out how many patients might be waiting for an appointment, for example, which has been the main issue at hand. But it's this data infrastructure that allowed reporting of that and tracking of it as a measure. So, the infrastructure is not just rich for research; it allows you to create quality measures and other things.

I think it's an important point you bring up, and we really also have to think of these data as generalizable. This isn't a situation where a bunch of veterans are getting digoxin because their doctors are ill equipped to prescribe other drugs, such as beta-blockers. That isn't the case. In fact, when we looked at this in Medicare and other data sets, the prescription of digoxin is declining in all of them, and in the VA, it's very similar. This is not an example of bad care. It's an example of care that's highly generalizable—and as you know, Bob, a lot of VA cooperative studies and other trials have come out of the VA system and have been generalized and validated in non-VA populations.

Dr Harrington: Mintu, I think you're saying something incredibly important: Part of why the VA has been criticized perhaps is because they have a rich, transparent data system that allows them to be measured and criticized. In fact, other systems may escape criticism because they escape detection, because they don't have the data system with the same quality and level of transparency that the VA has. Would that be a fair hypothesis?

Dr Turakhia: I think that's a great point, and you said it really nicely. The Inspector General is saying that people were cooking the books or keeping shadow charts of real waiting lists. Well, I'm not sure that you would even be able to determine in other care systems whether someone was cooking the books, because you may not have that data infrastructure where you can audit the practice and compare what the site is doing vs what the data show.

It's an important point, and I think the data here are one of the reasons that the VA can strive to continue to improve its mission. One other thing that's happening on the back end after this digoxin paper came out is that it has the attention of the Administration, because they really want to know how many patients are on digoxin and what might the next steps be, which I'm sure you and I will talk about in this interview. But whether, for example, the infrastructure that we have in the VA and the EMR systems could allow the next wave of cardiology research, which is deimplementation: taking something that everyone has been doing, questioning whether it needs to be done, and then figuring out a way to peel it back.

One example could be the left ventriculogram in angiography. It's highly controversial, but we have other tools for measuring LV function without necessarily giving a big dye load and having a catheter in the arterial circulation for that long. So again, these are just new paradigms to consider as we continue to try to improve care in cardiology.

TREAT-AF: Design and Findings

Dr Harrington: Great sidebar. Let's get back to the matter at hand. Take about a minute, Mintu, and describe the study. What were the methods?

Dr Turakhia: We selected a cohort using national VA data on patients with newly diagnosed AF; "newly diagnosed" was defined as not having a diagnosis of AF coded anywhere in the record in the prior 4 years. Then, we went forward and wanted to take the patients who were seen in a clinic that would handle their AF—cardiology, primary care, women's health, anticoagulation clinic, and other similar clinics—within 90 days.

We then conditioned the cohort on prescription of any medications within the 90 days again to verify that they were, in fact, going to be getting their care at the VA. This was true of the majority of the patients. We then ascertained creatinine and a few other things. The analytical cohort after all the inclusions and exclusions was 122,000 patients with newly diagnosed AF from 2004 to 2008.

Then, what we did was phenotype all the baseline characteristics. We even went as far as measuring such things as how far they were from the primary care doctor's hospital vs the nearest VA primary care center vs the nearest VA cardiology facility, to address potential health services issues and treatment selection.

We assessed all the usual baseline characteristics, of course. We calculated the GFR. We looked at all of the other medications that they were prescribed in that 90-day window. Then, basically we did a survival analysis, using a series of models to evaluate the effect of digoxin on mortality.

Dr Harrington: My compliments, as somebody who certainly has done a lot of clinical trials and read a lot about them; the rigor with which you applied the analytical techniques to these data is impressive. I was particularly impressed by the health services variables that were entered. So, kudos on that.

You had a big cohort of patients that was well characterized, and you had the ability to adjust for a lot of these variables. What did you find?

Dr Turakhia: What we found is that there was an unadjusted association of increased mortality with digoxin. If you look strictly on person-years, the absolute risk difference was not that big. One thing to bear in mind is that among these 122,000 patients, the mean age was 72 years. They were slightly older than an average AF population, reflecting veteran health status. Very few were women: only 1.6%, reflecting the VA. But almost 40% had a GFR that was either at or below 60 mL/min/1.73 m2, or they were on dialysis.

Of these, 23% of patients over that 2004-2008 period got digoxin in the first 90 days, which I'll refer to as "initial" or "early" digoxin therapy. We also looked at adherence to digoxin and persistence, because as you know, Bob, if digoxin was a transient therapy, any kind of intention-to-treat-like analysis may not be telling you the full story.

It turns out that adherence was very good. First, 70% of the patients who got digoxin were on it at one year, which surprised me. Second, the medication possession ratios (MPRs) were very high. The mean MPR was 0.79. That just means 80% of the days of digoxin exposure essentially had pill counts for those days.

Dr Harrington: That's extraordinary, actually.

Comparing Apples With Apples

Dr Turakhia: Yes; I think it tells you that they were taking digoxin for enough time to look at meaningful outcomes. That was important because again, if there were a lot of crossover or short duration of therapy, it's not clear what the analysis would mean.

We had higher unadjusted mortality, and then when we looked at a series of models in the propensity-matched cohort and the adjusted cohort, we basically saw the hazard ratios attenuate and plateau at around 1.2. In the unadjusted models, the hazard ratio was 1.37. After we added age, sex, and race, they were about the same. Once we put everything into the full model, the hazard ratio was about 1.26. If you adjusted for adherence (MPR), it was quite similar.

The interesting thing is that across all strata of kidney function, the hazard ratios were very similar. It didn't matter if you had a normal creatinine (and could in theory clear your digoxin); there was really no interaction.

The next thing we did was to essentially do a full propensity-matched cohort. The interesting thing about the propensity-matched cohort is that we were able to match 93% of the digoxin-treated patients using a very, very strict matching algorithm.

Again, to minimize comparing apples with oranges, we create these propensity scores on the basis of all of the baseline variables, and then tried to pairwise-match these patients (the digoxin-treated and non-digoxin-treated) on the basis of their likelihood or propensity to have gotten digoxin, to try to make this more of an apples-to-apples comparison. With propensity matching, the hazard ratio was 1.21.

Dr Harrington: It's quite striking. I'm going to ask you three questions as we get near the end. Me, as a trialist, I have to ask you, as a health services researcher, to convince me that the confounding doesn't bias this. Do we need a randomized trial? This is not a randomized trial. It's as good an observational analysis as you can get, but it's not a randomized trial. You want to tackle that one first?

Dr Turakhia: We can't with certainty know. We never will. There will be confounders that we've not identified. We couldn't measure LV function in 122,000 patients. We don't have those data. This is not a controlled cohort; this is a clinical cohort.

Of course, you have to deal with time-varying confounders as well. So, maybe there are reasons that inform discontinuation of digoxin or resumption of it in the interim, and that somehow affected things.

The common criticism we have received—and it's a fair point—is, "I don't use digoxin unless I've tried A, B, and C and they don't work. So, you really should be comparing digoxin as a fourth-line therapy to see what the result is." That's a fair point, but what we wanted to do is to look at early or initial therapy to avoid all the nuances of time-varying confounders and things like that. That's where we saw this effect.

Dr Harrington: You're showing it in the way people are using it.

Dr Turakhia: At least one way they use it. The other thing we did, Bob, is we performed a sensitivity analysis where we plotted the hazard ratio of a confounder—a theoretical confounder, such as frailty. That's always the one that gets mentioned, even with coronary artery bypass graft (CABG) vs percutaneous coronary intervention (PCI), for example. Then, we looked at what the differential prevalence would have to be in the digoxin group and the non-digoxin group. For example, how much frailty would you have to see in the digoxin group and not be present in the untreated group, and what would the hazard have to be to explain that?

What we showed is it's improbable that a set of confounders or a single confounder could capture all of that. For example, if frailty were present in 5% of the untreated patients and 20% of the digoxin-treated patients, again looking at initial or first-line therapy, then frailty would only be able to explain the observed difference. That's the upper limit of the confidence interval if frailty itself was unmeasured and had a hazard of 2.4. So, yes, it's possible that there are unmeasured confounders.

Dr Harrington: You guys have looked pretty hard and have exposed the data to pretty rigorous analysis.

Dr Turakhia: We looked hard, and we think it's improbable. I think the take-home point is: Given all the other options that we have to treat these patients, should we be using digoxin as early or initial therapy? I'm not saying that every patient should have their digoxin stopped. I don't want to be an alarmist, and I tried to go out of my way to convey that in the press—but of course, the headline about fatal foxglove is sometimes more tantalizing.

I think digoxin does have a role. It may have a role as adjunctive therapy. It may have a role if other things fail—but that's not what we looked at. We're really looking at, is digoxin the right thing for early therapy?

Dr Harrington: That's going to be the question I want you to leave us with, Mintu. What is your advice to clinicians? Docs are reading this, and they're saying a couple of things. One: "Should I stop digoxin in my patients who are currently on it?" I think you have helped us with some of that. Two: "Gee, should I use it at all?" And, three: "Well, wait a minute, is there a population that I could get away with using it in? Because I think there is."

Dr Turakhia: Great question. So, should I stop digoxin in my patients who are on it? I think that's a tough one. I think it really depends on what their comorbidities are. There is a lot of heated controversy in the particular area of digoxin in heart failure.

As you know, Bob (but I'm not sure that everyone knows), the DIG trial was in patients with reduced systolic function and normal sinus rhythm.[4] The DIG trial had nothing to do with AF.

Dr Harrington: Right.

Dr Turakhia: Everything on heart failure and AF comes from other observational studies or substudies, such as the one from AFFIRM.[5] It may be reasonable to use it in those situations. I think one should look at what the other options are and what patients have tried. Have they been on other rate-controlling agents? Beta-blockers, even calcium-channel blockers, have a reasonable role and a pretty good safety track record. Is that something that may be more tolerable to the patient? If so, then it's worth a try. In patients who are really complex—multiple comorbidities, brittle heart failure—I think that's a really tough one to unpeel. I think careful management may be the best option.

In my practice, I tend to avoid digoxin unless I've exhausted most of the other options; I will try various rate-controlling agents among my rate-control patients before I land on digoxin. It's important to recognize that digoxin works in AF with vagal enhancement, and even a small amount of exercise is going to overcome that.

So, the other question is, does digoxin actually work?

Dr Harrington: That's a great summary and advice for clinicians' real, practical considerations. We've been discussing the TREAT-AF Study with the primary author, Mintu Turakhia, from the Palo Alto VA Health Care System and Stanford University.

Mintu, I want to thank you for joining us here today on Medscape Cardiology and theheart.org. This has been a terrific discussion, and we look forward to seeing how this plays out in the medical community and seeing what kind of analyses you follow with. Thanks for joining me here today.

Dr Turakhia: Bob, thank you. It's been a lot of fun and, of course, a privilege to be on the show with you.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: