Using Health Claims, Registries, and EHR Data to Measure Multimorbidity

Join Dr. Sarah Berry and Dr. Sandra Shi from Harvard Medical School as they discuss how to use the wealth of existing databases in research on multimorbidity, as well as the pros and cons of each. Find out which data type is best suited to your specific research questions.

To view a transcript click here then select the transcript tab. 

Transcript

Sarah Berry, MD, MPH:  Hi, we're happy to have you here today. I'm Dr. Sarah Berry, an associate professor in medicine at Harvard Medical School and Hebrew Senior Life. And I'm here today to talk about Data Measures and Measurements. I'll be joined for - delighted to have Dr. Sandra Shi. Dr. Shi is an instructor at Harvard Medical School and an assistant scientist at the Marcus Institute for Aging Research here at Hebrew Senior Life. Sandra, thank you for joining us today. 

Sandra Shi, MD: That's my pleasure. 

Sarah Berry, MD, MPH:  I appreciate, and again, learning from you and in sort of this area about how do we measure multimorbidity. There's a lot of different databases out there, and how could a researcher or a clinician choose a measure that's right for them? 

Sandra Shi, MD: That's a terrific question, and I'll be honest, I feel like every day I'm still learning about new data sources and new measures for multimorbidity.

So the first thing I'll say [00:01:00] absolutely is that I don't think there is such thing as like the one right measure or data source. It depends so much on the research question and the topic, and I think what really, you know, clinicians or researchers on what they're looking for and the question that they're really trying to answer. And also the resources that they have available to them. So depending on those key topics or distinctions will really vary what ends up sort of being the best measure and also the best data set for them to use.

Sarah Berry, MD, MPH:  So there's no easy answer. There's no perfect or right data set, and there's no one set of measures you really have to dig a little deeper into what is the research question and, and what's available. 

What are some of the tradeoffs that researchers might face when they're balancing between some of these real world data sets versus data that's measured, maybe [00:02:00] prospectively through a trial? How do they balance that?

Sandra Shi, MD: You hit the nail on the head between, what I kind of bucket as the real world data, which sometimes people refer to, things like electronic health records or insurance claims data. But broadly speaking, I think of real world data is data that's just being generated from routine care as it happens. The major pro of this is that it's going to be very comprehensive and it reflects the practice in reality by clinicians and health systems as a whole.

And the other nice thing about it, typically speaking because it's just generated from routine clinical encounters, you have a high volume of data and so you'll have lots of patient encounters and lots of patients. And so for things that may be particularly rare or rare events, or rare serious, you know, adverse events or side effects, these are things that might [00:03:00] be more feasible to study in this sort of real world data just because of the, amount of data that is collected.

Sarah Berry, MD, MPH:  I am guessing there might be differences in generalizability too, right? Some of our sicker patients or those with the sort of highest burden of multimorbidity, maybe we are better able to capture them in in real world data, or is, what do you think?

Sandra Shi, MD: It's interesting. We have to think about how the health system sort of behaves as a whole. And so sometimes there's concern that among the more healthy people who actually maybe don't have that much multimorbidity, that there might be a tendency to sort of upcode those people. So for example, if there's a relatively healthy, older adult that only has hypertension, then they might be more likely to capture some other things because the clinician is sort of looking for things to kind of bill for in some ways.

But at the same time with multimorbidity, oftentimes when folks are very frail or maybe there are a lot of things going on, that might be [00:04:00] a situation where some chronic conditions, such as hypertension or hyperlipidemia might be undercoded because a patient might have comorbidities that are more morbid or they sort of take more precedence in the clinician's mind, and so they might not be capturing some of those chronic conditions with as much detail. And so that's definitely something to consider as well when working with the real world or claims data. 

Sarah Berry, MD, MPH:  Yeah, I had not considered that, how it could go both ways for those with a low burden of multimorbidity and also those with a high burden of multimorbidity.

I'm guessing then that the, the methods that you're using to measure multimorbidity also vary quite a bit, depending on the database and depending on the research question. Can you talk a little bit about that as well?

Sandra Shi, MD: Absolutely. So regardless of whether it's real world, you know, large claims data, or even from survey data or registry data, [00:05:00] ultimately it can be very daunting because there's such a high volume of data that's collected and available.

Broadly speaking, there are three ways that Multimorbidity is assessed. The most simple one that I think is the one most people think of is just simply counting the number of comorbidities that a person has, which is really easy; doesn't require any sort of fancy statistical methodology. And so it's very easy to implement.

But a downside is that this may be too crude, and I think most clinicians would say just because somebody, for example, has hypertension and hyperlipidemia, that counts as maybe two comorbidities, but it's not the same as somebody has advanced lung cancer and liver cirrhosis. And so I think a common criticism of doing a simple count is, yes, it's easy, but perhaps it's a little bit too crude.

And so sometimes people will want to weight things and here again is where the study [00:06:00] question comes into consideration because there are indices that researchers have developed that weigh based on functional impact or, or mortality. And so depending on the research question, it may make sense to try to weight these comorbidities differently.

And there's been a lot of really talented researchers doing excellent work, developing different validated indices. And so depending on the specific research question in mind, it might make more sense to use one versus another. Just to name a few that might be familiar to the audience. If you've ever heard of an Elixhauser [Comorbidity]Index or a Charlson [Comorbidity] Index, these are examples of indices that are perhaps been waited for a particular thing such as mortality.

And then finally with novel statistical methods, the last type of multimorbidity tool that we're seeing now is things that employ complex latent class analysis or cluster analyses. [00:07:00] These are really more advanced statistical techniques. The advantage being, there're very empiric and data driven, but that also means it's very reliant on the data set itself.

And so for those methods. My, my general advice is, phone or friend and probably have a, a bio technician supervising that analysis because you wanna make sure at the end of the day that you feel comfortable interpreting and understanding the results that are being generated by the analysis. 

Sarah Berry, MD, MPH:  Sounds like good advice. So tell me any other, words of wisdom or any advice for young investigators that are interested in measuring Multimorbidity.

Sandra Shi, MD: I would say don't be daunted by it. I think when you first start searching for how to measure multimorbidity, especially in EHR or claims based measures, even just try to do a brief literature search or a PubMed search, you see all of these options and it looks very complicated [00:08:00] very quickly.

But ultimately, if you really focus on what is it that the research question is asking and what resources do you have available, you can start to narrow down the data sources that would be helpful for your question and how you wanna measure it pretty quickly. And so definitely keeping a laser focus on the question that you're really trying to answer is probably the best way to start approaching the different methods and data sources available.

Sarah Berry, MD, MPH:  Would you agree? I feel like sometimes in aging research we let the, the sort of, the enemy of perfect, you know, get in the way. You know, we we're so looking for this sort of perfect measure that sometimes there, there is not, not one or, not a perfect data set anyway. Do you also see that too? I, I worry sometimes people,let that hinder them from getting started.

Sandra Shi, MD: Oh, that's a great point. Absolutely. Unfortunately, just like there's no perfect method, there's no perfect data source and so sometimes there does have to be a little bit of a compromise and maybe, you know, a particular [00:09:00] measure is not available exactly in quite the way that you're hoping to measure that.

But absolutely, you know, feasibility is incredibly important and definitely not letting perfect be the enemy of good and published, I think. It's so important, especially for young researchers who are often very eager to get the ball rolling. 

Sarah Berry, MD, MPH:  Well, great. It's been terrific to talk with you today, Sandra. I certainly learned a lot and in looking through your, your PowerPoints and encourage those, listening to this to follow up on that as well. Thanks so much for your time. 

Sandra Shi, MD: Thank you so much Sarah. Have a great day.