Has the Likelihood of Receiving a Dementia Diagnosis at the End of Life Changed?

In this episode we’ll speak with Dr. Julie Bynum who was the senior author on a recent study that examined how the likelihood of receiving a diagnosis of dementia has changed over the last decade or so. We’ll discuss what this might mean clinically and what researchers that rely on Medicare data should take-away from this finding. We also will talk in general about the various approaches that exist for identification of dementia in Medicare billing data.

Listen to the Podcast

Podcast Transcript

Matt Davis: 

Whenever you use a healthcare service, behind the scene there's a billing record, what many people refer to as a healthcare claim, that documents your information, information about your healthcare provider, what procedures were performed, as well as the charges, and for what diagnoses. Now, keep in mind that diagnoses on healthcare bills are imperfect. A person might receive a wrong diagnosis. On the flip side, a person who has a disease may not receive a diagnosis for a variety of reasons. However, despite the limitations, diagnoses on healthcare bills can shed light into how the diseases are recognized by the medical community at large. 

Dementia has long been thought to be underdiagnosed, but higher levels of public education and more diagnostic and treatment options may be changing how often clinicians document and bill for dementia. Several external factors could also affect how often dementia is identified on healthcare bills, such as the implementation of electronic health records that could make it easier to document diagnoses, and the influence of The National Alzheimer's Project Act, in 2012, that raised considerable awareness. 

Also, although it may be hard to believe, even little things can impact how diagnoses are documented, such as the expansion by Medicare, in 2011, that simply increased the number of allowable diagnoses on a bill. Understanding dementia identification also has important implications for health policy. You see, dementia was added into the risk adjustment strategy known as the hierarchal condition categories used by the Centers for Medicare & Medicaid Services for compensating healthcare plans. 

In other words, dementia is among the conditions that can influence reimbursement. In addition, identification of dementia is important for economic evaluations too, especially the cost of care at the end of life that receives a lot of attention. In this episode, we'll speak with a researcher who examined recent US trends in dementia diagnosis at the end of life ... I'm Matt Davis. 

Donovan Maust: 

And I'm Donovan Maust. 

Matt Davis: 

You're listening to Minding Memory, a podcast devoted to exploring research on Alzheimer's disease and other related dementias We're joined today by Dr. Julie Bynum. Dr. Bynum is the Margaret Terpenning Collegiate Professor of Internal Medicine at the University of Michigan, and the director of the Center to Accelerate Population Research in Alzheimer's, the center for which this very podcast is affiliated. Her research focuses on health system performance for older adults living with dementia who have high needs, as well as the use of administrative data for improvement of healthcare payment and policies. Dr. Bynum has extensive experience using Medicare claims in general for healthcare related studies. She's here today to talk with us about the identification of dementia at the end of life. Julie, welcome back. 

Julie Bynum: 

Hi, Matt. Hi, Donovan. Thank you for having me back. I'm absolutely delighted to be a member of the podcast. 

Matt Davis: 

Dr. Bynum was the senior author on a recent study published in JAMA Health Forum, titled Trends in US Medicare Decedents' Diagnosis of Dementia From 2004 to 2017. The study is attached to this episode if you want to check it out. We'll get into the study in just a minute. Considering your experience with Medicare billing data, or claims, can you tell listeners a bit about the history of research using Medicare data in general? 

Julie Bynum: 

Yeah. That's a good question. You know, here's not many places where we in the United States can get information about the whole population and their health and healthcare. It's really expensive and difficult to collect. After the passage of Medicare, there was a centralized national billing system. As you described so nicely in the introduction, once you have a billing system you have basically a record of each person in that system, and what their diagnoses are, and what healthcare they're getting. As computing abilities improved, and things went from tape to computers and discs and servers, the ability to use the Medicare system's data really expanded for research. 

I used to be at Dartmouth where the Dartmouth Atlas started, and that was really where Jack Winberg began taking tapes of billing data and making it into usable files. Really, that has just expanded dramatically. One of the reasons it's so useful is every American over the age of 65 has a record in the Medicare system, so it's been incredibly helpful to have a full population of all the people in that age range. Of course, we don't have it for other age ranges, but for Medicare we've been able to do a tremendous amount of research on the very basis of the fact there's a national insurance plan for that population. 

Matt Davis: 

Is it safe to say, though, that claims should be used for certain things but not everything? Perhaps it might not be the best way to go about studying the epidemiology of a disease. 

Julie Bynum: 

Hey, you're going right to the downside. What about the upsides? 

Matt Davis: 

We'll get there. 

Julie Bynum: 

All right. I think that's really true. If anyone listening to that takes away a message, it's that, no matter what data source you're using, you have to think about what its strengths and what its weaknesses are, and what you can use it for. We know that, for example, not everybody comes in to get all their symptoms diagnosed or treated. That's why epidemiology, which is the study of actually the disease in the population, is a little bit more challenging with the claims data, because maybe people aren't receiving healthcare for any number of reasons. They don't have insurance. Maybe they're scared. Maybe they don't think they have a disease, so they don't come in. That's part of the reasons that claims are imperfect, and you have to be careful about how you interpret and use the data. 

Matt Davis: 

But it is good for measuring the healthcare system. 

Julie Bynum: 

It's really good for measuring the healthcare system, which has been my primary interest across my career. What we observe in the Medicare billing system is the healthcare, the treatments that are delivered to older adults in the United States. As such, that's an important source for understanding what people actually get from the healthcare system, and making some judgments about whether it's lined up with what we think they need, and understanding how we might be able to influence them receiving what they actually need through our payment and policies. 

Donovan Maust: 

So what's sort of the Medicare claims 101 version of identifying people with dementia? 

Julie Bynum: 

When you make a diagnosis with billing data, you're basically looking at what a physician, or a physical therapist, or a hospital has written down on a record as the reason they gave you the treatment that they gave you. There's lots of weedy details about the different kinds of research files and billing files that exist, but basically what we do is we look for the services delivered where we think the diagnosis might have occurred, like in a hospital, or in a clinic, or in a nursing home where somebody might be getting rehabbed. We look at the reasons, the diagnoses for that service delivery. That's how we get diagnoses. That's true for dementia. It's true for diabetes. It's true for restless legs. It's true for any diagnosis that you might be interested in studying. 

Donovan Maust: 

So there are some medications, albeit not very good ones, for dementia. Do people ever use those to find dementia, or do they really just stick with the claims? 

Julie Bynum: 

Yeah. You can use all sorts of different algorithms to identify people, and people use lots of different methods. If you are interested in finding everyone who might possibly have the disease, you could use medications. You can decide to say they can have a visit for the diagnosis or a medication, but if you say they have to have a visit and a medication, that means you're identifying a different group of people. Right? Those are people who have the diagnosis and are treated for it, which is different from the people who have the diagnosis because they've had a visit, or they were treated for it. Those are just some of the weedy details that you've got to be really careful when you're figuring out how to identify your population, or, when you're hearing about a study, to understand who the study really applies to, because some of these choices about the algorithm make real differences on how we interpret the results. 

I mean, like the study that we're talking about today. This is a study of decedents, so keeping in mind what that means and how they're using the healthcare system is really important when you interpret the results from it. 

Matt Davis: 

That's a great segue to get into the study. I think it's safe to say that there's a lot of interest in looking at end of life costs and care, sometimes when things get complicated from a health perspective. Why specifically look at diagnoses among Medicare decedents for something like dementia? 

Julie Bynum: 

The first reason is there's been a tremendous number of studies that examine how we manage this population toward the end of life. The concern being that, until maybe a decade ago, there was worry that we weren't recognizing the diagnosis of dementia as one that means prognosis is poor over the next five years. People were getting treatments that seemed to reflect a lack of sensitivity to what those last few years of life would be like from a quality of life perspective. There's a whole body of work about the use of feeding tubes and other life-prolonging therapies, and frequent transfers across hospitals and nursing homes that really are burdensome to a group of people who might really have limited ability to understand what's happening to them and really be in the last year or two of life. 

I think there's a care of people interest in end of life. Sometimes my personal opinion is I think this population is really at special risk because the nature of their disease. They often can't participate in decision-making, and maybe they didn't pre-specify what they wanted, or they might not have the family members or care partners and advocates around them to help them shepherd their last several years of life. I think they're really at risk for the system taking over, sort of doing what the system does, instead of really personalizing and individualizing that choice. I think it's a really important population for us to take a special look, and take special care to make sure we're doing what people really need and want as they progress through the really devastating end stages of this disease. 

Donovan Maust: 

You all looked at diagnoses in the past two years of life. How did you arrive at that specific timeframe to look at? 

Julie Bynum: 

Yeah. Basically, for the people who haven't read the study, which I'm sure is many, we identified people who died in the fee-for-service Medicare population, and then we identified among that group who had a diagnosis of Alzheimer's disease, or any other kind of dementia, using the claims. As long as somebody had data, we could go back years, and years, and years. If they ever had a claim, we could say, "Yes, they had dementia." The reality is, in our healthcare system, and as people present, most people don't obtain a diagnosis 'til very late in their disease. 

What other studies have shown is that going back farther than two years really doesn't pick up more people. Because of the nature of enrollment in insurance, the farther back you go, the smaller and less generalizable you make your population, so we opted to go back to two years. I think the other question is I just made the argument of why dementia end of life, but I think for the particular policy question we were asking, which was has the likelihood of obtaining a diagnosis changed over time? It was important to get a group of people who have as most similar a stage of disease as possible. Right? If we went back 10 years, there could be people who have come forward with the most mild complaint, and then other people who have much more severe disease. 

By really focusing in those last couple years of life, we're focusing on people who, for the most part, have late-stage disease and are relatively similar to each other. It's a study design some people don't like, because it is a follow back from death design. For this particular question of saying what has changed over time from a group of people who lived with this disease at the end of life, it was the right design to go for this question. 

Matt Davis: 

This idea of using decedents I think might be new to some listeners. It's interesting. I think you're hinting at sort of there's a methodological reason for looking at. I mean, if you use something like billing data, you have issues with things changing across time, and health status, and stuff. Looking at people that are at the end of their life presumably are more the same over time and allows you to, sort of. Is that kind of what you're getting at? 

Julie Bynum: 

From the study design, we can't say how all people with Alzheimer's disease experience the last two years of life, necessarily, but we can say, among people who the health system recognizes as having dementia, this is what it looks like. While not everyone who's a decedent who dies with Alzheimer's disease has late-stage disease, certainly someone who has even mild early cognitive loss, and has a tragic car accident and dies, would be in this data set. On average, the majority of people will actually have late-stage disease. That shows by the average age in the study, which was 86. These are older, older people. 

We're trying to avoid having the young people who I would hypothesize, and I think the studies would bear out. There's real differences in who seeks care early in the disease, whether it's economic, educational, cultural. Care seeking is very different earlier in the stage rather than the late-stage disease where people are quite dependent, need care partners. It's highly dependent on the healthcare system longterm care system, so they come to the attention of the system whether they want to or not because of such a high rate of hospitalization and emergency room visits due to the nature of their disease. 

Matt Davis: 

So why don't we get into the findings? 

Julie Bynum: 

Yeah. What we found is that the percent of people dying with a diagnosis of dementia has increased fairly dramatically over about a 13-year period. Back in 2004, about a third of Medicare beneficiaries fee-for-service Medicare died with a diagnosis, meaning they had had a diagnosis at some point in the two years prior to their death. That went up almost to 50%. It was 47% in 2017 had a diagnosis, so that's a 36% increase over that time in the percent of people who, if you said, "Hey, what did grandma die with? Maybe she didn't of, but did she have dementia?" In 2017, 36% more people would say, "Yeah, grandma had dementia," compared to 2004. That's a really big difference. 

I think you should ask me what that difference means. Well, what does that reflect? Does it mean that, in fact, more people are getting dementia? No, it doesn't mean that. Absolutely not. For everything we've already said, we've said this is about not only people coming forward and getting care, but also the clinical providers recognizing the disease, thinking that there's something to do about it, so they write it down in a medical record and actually bill for it. It sounds kind of silly. You'd think, "Well, don't they just always write down everything?" No, we don't. We write down the main things that we treat. This rise reflects in, likely, our policy, which we can talk more about, I'm sure, of how we record things. 

Also, whether this disease is front of mind. I think that's changed dramatically over this time period. Dementia has become a front of mind, front page of the newspaper kind of disease. Whereas, in the past it really wasn't. This was a disease of people who lived in nursing homes and we tried to pretend didn't exist. We didn't want to see this. Right? This was senility, the word people used. This is what old people got, and we tried to pretend was never going to happen to us. Today, it's a different world. We recognize what this disease means to the individual, what it means to families, what it means to the health system. Actually, there's more science that's saying maybe someday we can actually influence the trajectory of this disease. Until now, we really haven't had a treatment that influences the course of this disease. 

Donovan Maust: 

Just to maybe underscore what thing that you said before, your analysis, what you're not trying to argue is anything about what these people are dying of. Right? It's that they're dying with dementia. 

Julie Bynum: 

Yeah, absolutely. We cannot say on claims data what they're dying of. I mean, I think not with this design. If one were interested, and everybody was hospitalized, it's possible you could look at really detailed analysis of hospital claims to understand among people who died in the hospital. Typically, people look to death records for that. You know, death certificates. In fact, there was a change. One of the things that motivated this study was a number of years back. We noted a change in the approach to billing or to documenting on death certificates and saw that dementia rose as a cause of death. Again, that was not a prevalence incidence change, not a disease change, but a change in the way people think about and document, according to instructions and policy, disease. 

I guess one of the things I really wanted to talk a little bit about is this is not so different from lots of other scenarios. Right? The construct of a disease is influenced by many, many things other than just its pure, true prevalence in the population. Back in the 2000s, maybe you guys are too young to remember this, but there was the rise of restless leg as a syndrome, as a disease that was treated. There's this nice paper by Steven Woloshin and Lisa Schwartz about the media's influence on that, and what public service announcement do, and the drivers behind pharmaceutical companies who now have a drug, so therefore we then have to find the population to treat. 

I think it's not just media and pharmaceutical. That also occurs in our clinical world when we recognize we've been missing the boat on something. Right? Here's an underdiagnosed disease that's incredibly burdensome. We now have a new policy initiative, The National Alzheimer's Act, to really address the burden of this disease. I mean, there's been a wholesale effort to change how we think about and how we address this disease, and that influences who seeks care. It influences what we do in the clinic setting and what we write about in those clinic settings, so this is not unique to dementia. 

I think what we're really trying to do is call out to the public policymakers, researchers, to be aware that there has been this change. Probably culturally, but also in our data that we use to make important judgments about where to invest in the future and how to understand care delivery. That we understand those changes over time. 

Donovan Maust: 

Did you find anything that surprised you, in particular? 

Julie Bynum: 

Whenever you go into a study, when you're a scientist, you have a hypothesis. Right? My guess. I want to answer this question. Right? I want to say, you know what In this case, I think this has changed. You know what? Most of the change should probably be due to this electronic health record in The National Alzheimer's Act and all these changes. I was at equipoise, which is that term of is this going to be right, or is this not going to be right? I'm in the middle. I'm not sure. The biggest surprise is we were right. I mean, that's something. We're not all- 

Donovan Maust: 

Nice to know that then. 

Julie Bynum: 

Yeah. We're not always right. I think that it was so clearly right that the change was much bigger than I expected. The time period of the change, it was not a steady increase over time, year-over-year increases. What we saw was this bump up in the middle years when at least three things happened. I think Matt discussed these in the introduction, but we were having a lot more discussions. The NAPA, The National Alzheimer's Plan Act, passed. We also were having rapid expansion of electronic health records, particularly in hospitals in this period, which is important because that's where we've seen the billing of this diagnosis really increase. Once you have an electronic health record, it's much easier to code and carry forward diagnoses than when everything was hand-written. Right? 

Then we also see Medicare increasing the number of slots. It sounds so mundane, but they increased the number of slots of diagnoses. I now can say not just the top 5, the top 25, or whatever those numbers are. I can now list many, many more diagnoses that were maybe farther down my list. All those things sort of happened around the same time. We saw the slope, the increase in the rate of diagnosis, being put on records really directly happened in that time period, as opposed to a gradual increase over time. 

Matt Davis: 

Out of all those things, though, incentives around payment, when did that kind of start? 

Julie Bynum: 

Yeah. I've been talking about culture. 

Matt Davis: 

[inaudible 00:24:34] 

Julie Bynum: 

Yeah, cultural influences. If this was all culture and our attitude towards disease, we would be seeing steady increase over time. Right? Until we maximize. Until we entirely get rid of underdiagnosis. But that's not what we saw. We saw this very time limited increase, and then this sort of increase flattens out. 

Incentives. So why are there incentives to bill more diagnoses? Well, it turns out that health systems have these scores for their sort of over illness of the population they care for, so it's to the advantage of the health system to have us bill as many diagnoses as possible. That's not even about just the individual disease. Right? Just to diagnose as many as possible. 

Back in the day, when I first started many, many years ago, there were differential payment for psychiatric diagnoses versus medical diagnoses. You'd have a real incentive not to bill the dementia diagnoses that would've fallen under a psychiatric diagnosis code, because you would've been paid less. That incentive is actually gone long ago. The new incentive coming up, and one of the things that really drove me to want to do this study, was that now in managed care, Medicare Advantage, dementia is part of the risk adjustment for their payment models. 

Now, this study that we're reporting on here isn't in managed care in Medicare Advantage. It's in fee-for-service. It was important for me to establish a baseline, what's going on in the population, so that when we look at the influence of incentives on what people get and the care they receive under Medicare Advantage, we at least know the context from which we're coming. We try to account for that and not attribute everything we see to only the financial incentive, because there are multiple things going on here. 

Matt Davis: 

Many of our listeners probably aren't healthcare providers, so I have a really basic question about how diagnoses make it in the bill. You're both practicing. Both Donovan and Julie are both physicians. I'm just curious. That process of when you treat a patient, somewhere in the record you click the diagnosis. Is that exactly what shows up on the bill, or is there a level beyond that, in terms of billing and coding, that could potentially change things? 

Julie Bynum: 

Yeah. It sort of depends on setting a little bit. Let me talk about the outpatients setting first. In the outpatient setting, basically what I bill will end up on the record. It's kind of funny how that works. I could have to write in a diagnosis. When I first started, we had little check boxes on a sheet of paper, and I could check the boxes. The common diagnoses were there, but if I wanted something else I had to write it in by hand. That's a little disincentive to use a less common diagnosis. 

Then we went to the electronic health record. You can actually type in a diagnosis, and search, and get a much more precise diagnosis. Then you click those. Then it'll be automatically offered up to you, so in subsequent visits you can reclick them again if they're important. One of the interesting things when you talk about incentives, I already mentioned the health system incentive to get us to bill certain kinds of diagnoses. Certain diagnoses have higher weight according to HCCs, so then you start getting health systems offering you the higher weighted diagnoses, which is pretty fascinating. Right? I could use this code or that code in my mind for the same disease. They seem really similar, but this one has a little bit more incentive for payment. Very sophisticated electronic health records can start offering that up. That's on the outpatient side. 

I will say, on the outpatient side, if I order a blood test, there has to be a diagnosis code that motivates that blood test. I can't just say, "I'm going to order a glucose because the sun's shining." Right? I have to say, "I'm ordering a glucose because I suspect diabetes," or, "This person's having frequent urination," so there is an issue of you have to have diagnoses tied to any other kinds of things you're ordered. 

On the hospital side, all those things I just said are true, but there is an extra laying of coding and billing people really reviewing, because there's an extra layer of how those actually end up on the bill. Happens in the outpatient side too, but much more in the hospital side, which relates to this paper. Right? What we saw, the biggest increase was in the hospital setting. Somewhere along the message got out that we want this. This diagnosis is important to have there, and you saw the biggest increase on the hospital side. 

Matt Davis: 

That's fascinating that, in electronic medical records, that there might be situations where you type in a diagnosis, and it's like, "Wouldn't you prefer to do this one?" I didn't know that. 

Julie Bynum: 

Yeah, it's true. I have to say the other thing, and this is important for dementia, is that people, patients, their family members, see those bills and diagnoses. That actually can have an influence on what a clinician does. I say that particularly in the case of dementia, because there are some communities this is very stigmatized. They don't want that on their record. If you're a physician, and somebody comes back to you and says, "That's wrong. Don't put that on my record," that can really influence you and affect the way about how you handle this diagnosis in the future. 

There are some things that are just more sensitive than others. Right? I mean, I think the history of HIV is along those lines. There are just diagnoses that are more sensitive. In some scenarios, dementia is a sensitive diagnosis as well. 

Matt Davis: 

So you mentioned that you looked at different settings. Were those different places that the person got the first diagnosis of dementia? How did you do that? Tell us more about that. 

Julie Bynum: 

We mostly put this toward understanding where the claims were going to show up for research. People get their healthcare in multiple settings simultaneously. They're in the outpatient setting. They're in the hospital. They're getting hospice. They're getting home care services. The first thing we wanted to see is just how much does this diagnosis get billed in all those places? The sort of professional services, which is the physicians, the visits, are the places that you'd most commonly see it, and that's been consistent all along. 

We've seen major rises in the hospital setting and in the hospice setting in the diagnosis being used. One of the areas we were concerned is this issue of misdiagnosis, like somebody puts the wrong diagnosis. You might see that if you only ever seen one. Right? When you see a diagnosis in only one setting, maybe they don't quite have it right. The classic one there is delirium. Somebody's in the hospital and they get delirious. Could that be mistakenly billed as dementia? That can be confusing clinically. That was the other thing we looked at was in the case where it was only billed in one place. We've seen that drop off dramatically. Really, when the diagnosis is being billed, it's being billed across multiple different settings today compared to maybe 20 years ago. 

Donovan Maust: 

I could imagine that over the period of time. What? I think, 2004 to 2017 that you all looked at, there were possibly significant changes in what end of life care looked like. I wonder sort of how did you think about that or account for that, and then is there any reason to think that that influenced the likelihood of finding a dementia diagnosis? 

Julie Bynum: 

Yeah, Donovan. That's a really important question. It was a real motivator for being on this study, because I already said at the beginning. One of the things that drove this field is concerns about what kind of care this population's getting at the end of life and how it's affecting the quality. You might think, you might hope, that if people are more likely to recognize the disease, they'll take it into account and have really important conversations with the person and their family about their goals of care and what they want done. You might expect them to get less aggressive care at the end of life, less care in the ICU, fewer of these treatments that are meant to prolong life in the short term. I don't know if that term makes sense, but putting people on a mechanical ventilator is really prolonging life in the short term. Right? You have pneumonia today. It's going to keep them alive or bridge this, but if somebody has a disease limiting their life over months to years, it doesn't necessarily help that disease. Right? 

What we found is that ... Well, let's back up. For everybody, care at the end of life has changed. We've seen a big increase in the use of hospice, for example. The population with Alzheimer's disease just parallels what's happened in the rest of the population. There's been a shift toward less use of the ICU in a terminal hospitalization. The last time somebody's hospitalized, they're less likely to be placed in an ICU. Overall, we've seen more deaths occur outside of the hospital. That's been shown a number of times. In the case of Alzheimer's disease, those deaths are split between in longterm nursing homes and in the community. We've seen that. 

In terms of life prolonging therapies that I mentioned, it's a mixed bag. We really haven't seen much difference in the use of mechanical ventilation and dialysis, but we have seen drops in the use of feeding tubes. There is, again, another body of literature in quality of care effort around the use of feeding tubes. It likely made that influence more than a general attitude toward end of life care. 

Bottom line, one of the hopes might be that by recognizing Alzheimer's disease, the patients, the families, the doctors, the nurses, would sort of rally around and come away with a less aggressive treatment of care in the 2017 period compared to the 2004. We did not see that uniquely happening in Alzheimer's disease. Really, what's happening to people with Alzheimer's disease is very similar. It's happened to everybody else who's dying at older ages in the United States. 

Matt Davis: 

I'm assuming that we don't necessarily think that the prevalence of dementia is necessarily changing over time. I guess, from your perspective, what should people take away from the study? 

Julie Bynum: 

Yeah. That's a good question. Where do we want to push on helping people with this disease? I think the big takeaway here is, yes, the community of medical and health service providers are recognizing this disease better than they ever have. That's a good thing. You know, you can't improve care of something that you're not recognizing. Underdiagnosis seems to be declining, at least in late-stage disease, and that's good news. 

I think, for the researchers who are listening, the big message here is really remembering when you're designing your studies, and thinking about it, that there are many things influencing what we see in administrative data that aren't purely about the disease. This study really highlights it. Right? The changes that occurred are not about something happening in the community and increased rates of the disease. It is totally about what's going on, and how we will for it, and how much people think about the diagnosis. We're not seeing any other differences there. 

My hope is that, now that we're moving toward better diagnosis, or at least more frequent diagnosis, that we can then focus on actually improving that care delivery. It's a long haul to get people better care. Just how we bill it alone doesn't reflect that quality. I think that's the next place for us to go. 

Matt Davis: 

This is probably a little bit too much in the weeds, but sometimes researchers, when they construct a cohort, they're trying to identify people with a certain onset of a disease. Sometimes, for things that are rare-ish, they have to go over multiple years to do that. Do you think that if we take sort of what you showed in the diagnosis of dementia over time among decedents, that's sort of a broader implication that it's being recognized in general? Do you think that researchers should kind of take pause in terms of just considering that when they think about what years they're going to put together to do a study with? 

Julie Bynum: 

Yeah. Lots of studies try to compare now to some years before. We've gotten so much better because we are doing something. We see what we compare now to 10 years ago, but you really have to think about the people identified now are different from the people who were identified 10 years ago. That's why it was so important for me to get this study on the books before we do that next wave of studies, so that we understand how different the population is or is not over time. 

In this case, the people diagnosed with dementia actually, if you look at their age, comorbidity scores, gender makeup, they're not that much different than they were in 2004. That's why I think this paper argues that we're not diagnosing a different set of people. We're just diagnosing more of the people that have always been diagnosed, if that makes sense. 

It could've been that, say, there was a community. Maybe the younger community of people with cognitive impairment were coming forward, or maybe a certain racial or ethnic group were really coming forward, and the real effort's being made in underdiagnosis there. We don't really see that in the data. The underlying characteristics of the population identified have not changed very much. I think your point is important that whenever we do things looking at changes over time, you always have to be really careful on your interpretation and your methods so that we don't pass along an inaccurate message to our community. 

You hear me. Whenever I speak, I give grand rounds, or give anything on this. You'll see, whenever I do a claims based study, I always talk about people diagnosed with dementia. I don't talk about people with dementia, because there is an enormous population of people with dementia who are not yet diagnosed living today in our communities. I always challenge people. If you ever hear me mess that up when I'm using administrative data, these are the people who've come forward and been identified, so they are diagnosed with the disease, but there's certainly a much bigger community out there. 

Donovan Maust: 

We've covered a ton of ground. I'm curious for, say, researchers who are interested in kind of moving into this space, getting into using Medicare claims for dementia. Any additional pearls? Anything else we haven't touched on that you think would be really important from the get go? 

Julie Bynum: 

Yeah. Number one, don't start from scratch. There's a ton out there. There are published papers. We have recently published papers. We advance science by doing repetition and being able to recreate what somebody else did. If we're all using different methods, we don't know that our results are comparable. Definitely reach out to any of us. Reach out to any investigators you know, the published literature. Don't start from square one. There's been a lot done. 

A really important message, particularly to the non-clinicians who use this kind of data, is being deeply aware of where the data comes from. That sounds so obvious. Right? It's just obvious you should know where your data comes from, but if you're doing a survey in the field, you've seen the questions. You can read the questions. You know how they're delivered. The administrative claims are kind of a black box for a lot of people. They don't understand the process of how a person comes to a clinic, chooses a doctor who may or may not make a right or wrong diagnosis. All of those subtle things that influence that process of symptom to actually recognition and diagnosis of disease is complex. You can't just take the data at its face. Learn it. Spend some time understanding the process through which the data you are using for your research came to be. You've got to be accurate and really have the most impact on the field. 

One thing I would say for researchers, too, is that in the space of health policy, healthcare delivery, we have different kinds of people doing that research. We have physicians like me and Donovan. We have people like you, Matt, who are data scientists who have some clinical background, but not in this area at all. Then we have wonderful economists using these claims data. To my mind, because of the intricacies and the nuance of the data, the clinical process, the challenge of understanding cost data, the best research is multidisciplinary research. We need to put the knowledge and the experience of the clinical provider alongside the really good scientist, or alongside the really good economist, to really get the best methods with the best questions most accurately interpreted. 

I think this is an area that requires that more than others, because the stakes are high dollar-wise. We're talking about Medicare billing, and Medicare expenditure, and people's life policy. The stakes are really high. A lot of the knowledge is siloed, so getting that real interdisciplinary work I think is where we're going to advance the field most substantially. It's also the most fun. 

Donovan Maust: 

This has been great. Julie, thanks so much for joining us. 

Julie Bynum: 

Thank you for having me. 

Matt Davis: 

If you enjoyed our discussion today, please consider subscribing to our podcast. Other episodes can be found on Apple Podcasts, Spotify, and SoundCloud, as well as directly from us at capra.med.umich.edu, where a full transcript of this episode is also available. On our website you'll also find links to our seminar series and the data products we've created for dementia research. Music and engineering for this podcast was provided by Dan Langa. More information available at www.DanLanga.com. Minding Memory is part of the Michigan Medicine Podcast Network. Find more shows at UofMHealth.org/Podcasts. Support for this podcast comes from the National Institute on Aging at the National Institute of Health, as well as the Institute for Healthcare Policy and Innovation at the University of Michigan. The views expressed in this podcast do not necessarily represent the views of the NAH or the University of Michigan. Thanks for joining us, and we'll be back soon. 

 

Article Referenced in this Episode

Davis MA, Chang CH, Simonton S, Bynum JPW. Trends in US Medicare Decedents' Diagnosis of Dementia From 2004 to 2017. JAMA Health Forum. 2022 Apr 1;3(4):e220346. doi: 10.1001/jamahealthforum.2022.0346. PMID: 35977316

Resources

The Bynum-Standard 1-Year Algorithm for identifying Alzheimer’s Disease and Related Dementias (ADRD) in Medicare Claims data.

Looking for more Minding Memory podcasts?