Health data analysts are an elusive bunch in the wild. While we see their names periodically show up as middle authors on manuscripts or in the Acknowledgement section they work largely behind the scenes; yet they play a vital role in conducting research that use large data. In this episode we speak with several health data analysts to better understand the role they play in research and, for all the researchers out there, discuss how to make the process as smooth as possible when working with an analyst.
Transcript
Matt Davis:
Most people when they think of research, probably imagine scientists and white lab jackets diligently working at cluttered bench tops to cure cancer. Personally, I like to visualize Will Smith in the movie I Am Legend curing the zombie apocalypse in his home lab.
Matt Davis:
Indeed, there are a lot of people doing bench top research, and a lot of brilliant people working to cure cancer for that matter. However, depending on the type of research, the process can vary quite a bit. With new technology, and our ability to store more information, every year, more and more large data sources become available that allow research to use data that they themselves did not collect. And sometimes use data for research that were initially intended for some other purpose altogether. Some of these new emerging sources of data include disease registries, national health surveys, data from health records, and administrative healthcare data, often refer to as claims data, because the data consists of healthcare service claims, basically billing receipts.
Matt Davis:
To be clear, these data are not simple Excel spreadsheets. They can be massive. Upwards of hundreds of millions of rows, and to analyze them require specific skills. For researchers that use large preexisting data to study health and healthcare, the research process has very much evolved to be a team based activity. While the lead researcher, typically called the principal investigator, might design the study, data analysts do the hard careful work of working with the data. Some projects are so complex it can take a year or more just to create a data file needed to do a study.
Matt Davis:
Health data analysts are an elusive bunch in the wild. While we see their names periodically show up as middle authors on manuscripts, or in the acknowledgement section, they work largely behind the scenes. They don't come from any one specific background. And they spend a lot of time working alone on the computer. Yet they play a vital role in doing research that use large data. In this episode, we're going to dive into the secret life of a health data analyst.
Matt Davis:
I'm Matt Davis.
Donovan Maust:
And I'm Donovan Maust.
Matt Davis:
You're listening to Mining Memory, a podcast devoted to exploring research on Alzheimer's disease and other related dementias.
Matt Davis:
We're joined today by Julie Strominger, Jonathan Martindale and Mohammed Kabeto. Each work on productive research teams here at the University of Michigan that study dementia, using a variety of different types of large pre-existing data. Experience as a data analyst among them, range from being new to the field to more than 20 years of experience. They come from different backgrounds, and each have their own story for how they became a health data analyst.
Matt Davis:
So just to start things off, let's first go around and talk about your backgrounds. Tell us a little about your training before you became an analyst.
Julie Strominger:
Thanks, Matt. I'm Julie, and prior to becoming a data analyst, I received my bachelor's in statistics with a minor in geography, and then I also received my master's in biostatistics. After I graduated, I worked for a research lab studying environmental health for a few years, and I also worked in clinical trials for about a year or so before I made my return back to more academia based research, working at the VA as a data analyst.
Jonathan Martindale:
I'm Jonathan. I started on a track in social sciences research in social psychology. In my undergrad I got involved in a couple different labs, and there was this one mentor who had just really pushed me towards research side and PhD. And so I was heading towards that route, and I finished my undergrad with a degree in psychology and a minor in statistics. And I worked in a lab for a little bit and I realized that the PhD life wasn't for me, but I found that I was really interested in moving on to more advanced pieces of the research. And so what that meant for me was more technical skills. And so I decided to go back to school and I got my master's of health informatics, really focused on programming and data analysis, and then just last year finished that and started working as an analyst full time.
Mohammed Kabeto:
Hi, my name is Mohammed Kabeto. I've been my undergrad at math and then more of applied math. And then I got my master's degree from the University of Michigan School of Public Health. And since then, I'm just working for analyst which is very interesting working for the last over 22 years.
Donovan Maust:
So Muhammad, when you think about getting your master's degree, did you know that a position like an analyst even existed? Or how did you end up getting to this as a career?
Mohammed Kabeto:
Interesting, thank you for that question. Yeah, while I was an undergrad, and then I was just interested in doing an applied math. I had no idea there was a biostat exist. And I know there is a stat because I was also doing some applied math, which is a little bit of more stats, statistical analysis. But I got a chance to participate in a summer program at Harvard University to expose undergrad student to the biostat. And then I got a chance for about six weeks and then to just to work some analysis. That was the first time that I was also exposed to stata. And then there was wonderful individual in that school of public health, which is Professor Louise Ryan, and I think she is right now in Australia back to her country. And so pushed me and then to just go to that route.
Mohammed Kabeto:
And then I also come in here and then do my grad school. And while I was working at the grad school, I was also doing some analysis at ISR, with Dr. Jim Lavkaski, and some analysis. And then that experience really pushed me to just involve in more analysis. And then after I completed, start working with wonderful person, Ken Langa, he hired me, and soon I completed my grad school. And since then and then I was exposed to different discipline, and then that's what pushed me to just continuously enjoy working as analyst.
Matt Davis:
So you have different backgrounds, obviously, which is not unusual I think for people that do this for a living. I was just curious, I guess for Julie and Jonathan in terms of your backgrounds, how much did your initial training and background kind of prepare you to do what you do now? And how much did you just have to learn on the job, so to speak?
Julie Strominger:
Yeah, that's a good question. I think in terms of my background, I would say I did feel very prepared programming wise. The programs that I participated in were pretty heavy SAS programming, also some R programming. I think analysis wise also pretty well prepared, but definitely I think a positive of becoming an analyst is that you are always learning. You're always looking to see what's a new method, what's the best method that we can use to answer a question? And it does require learning on the fly. And so I would say my background prepared me well, but you're always learning. You can't ever know enough.
Jonathan Martindale:
Yeah, I think that's a really good point of this always learning mentality. I came in with very little programming knowledge in the specific language that I use most now, which Julie mentioned as SAS. I did a lot more work in Python and a little bit in stata, and I had some touches with R. I think my training did seem really helpful, but it was really more of around a mentality standpoint of learning to think like a programmer. And once you do that, and once you kind of understand that there's different ways to do things and the general approach you should take, and how to Google the appropriate code and what you need to do for your specific problem. Learning a new language really isn't that hard. It takes a little bit of time, but if you have the basics down of what you need to do theoretically, then getting there and practice is fairly simple.
Matt Davis:
I would think that your background, your master's you described the informatics is probably one of the better things I could imagine. I mean, a lot of people become analysts after doing statistical degrees in statistics or econ, but yours was informatics. I mean, it seems like a pretty good set of skills to develop to do this work.
Jonathan Martindale:
And I focused more on data analysis and stats than I think a lot of my classmates did, knowing that this was the route that I wanted to take. And so that's what led me to a little bit of touch points with stata and SAS, but just not that much. But yeah, it was really a training and thinking I think, more than, specifically this is the program you're going to use, and these are the exact ways you're going to use it.
Matt Davis:
So this is a question for you, Jonathan, because you're relatively new. This is in your first year still or second year?
Jonathan Martindale:
First year. I'm coming up on my one year full-time anniversary year.
Matt Davis:
Okay. So when you go home for the holidays and your nana asks you to describe what you do for work, what do you say? And I don't know if you have a nana, but I guess, how do you explain to a family member what you do for a living?
Jonathan Martindale:
Oh, I usually start by saying that it's kind of boring.
Matt Davis:
Oh no.
Jonathan Martindale:
And I say that, not that it's boring for me, but I think when I describe it and if I were to hear someone describing it, I would think that it's boring. I guess what I say is that I'm working with these people who have different ideas for research and I'm trying to put their ideas into something that's tangible and that we can use to answer some question. And that's a really abstract way to say it. And then I guess if someone asked me about the nitty gritty, I would say, "Well, I'm working with a number of different files and trying to figure out what everything means and how to put it together in a way that makes sense. And so I have this back and forth with the people I'm working with to ask them questions of what exactly they want so that I can program it in a specific way and get them this end product."
Matt Davis:
It's probably just easier to say we're healing dementia, just leave it there.
Jonathan Martindale:
Sure, yeah.
Donovan Maust:
I'll ask this next question of Julie. So first disclaimer is that I have the great pleasure of having Julie be the analyst on my team. So Julie, I might regret saying this and you should forget this question after you answer it for me, but you could easily find work that would probably be higher paying in a non-academic setting. So why haven't you done that? Please don't do that. But why haven't you done that?
Julie Strominger:
I will not. Yeah. I think that the work we do is rewarding on so many more levels than just the amount of money we're paid. We're working with not only a group of people that are intellectually stimulating, it's really interesting to have these conversations about older adults with dementia, and how we can improve their lives. But also just the fact that we are bettering people's lives. That's something that we're all working towards collectively. And I think I have some experience in industry. And while that also is interesting, a lot of times the goal there is to make money or look at always the bottom line.
Julie Strominger:
Also, I think that there's a lot of room for creativity in academia. So as Jonathan was mentioning, we're trying to figure out how to answer these questions, how to put these files together. It really takes a lot of thought planning. I think creativity too, in terms of what analysis you're using or how you're thinking about things. And I don't think that necessarily there's a lot of time or room for that in other industries.
Matt Davis:
Let's get into what it's like to be an analyst. So Mohammed, you've been analyst for a long time here at the University of Michigan. So I'm curious, what is a typical day like for you? And to what degree are you working independently versus interacting with other teams of analysts?
Mohammed Kabeto:
When I started working with Dr. Ken Langa and at initial stage, it's just a communication, having a weekly meeting and then that lets me to think every time now when I start working, what is something that I need to do at a week ahead or just some of the schedule that I put for different PIs.
Mohammed Kabeto:
And the typical one is depending on those projects, but there's no day that I did not do without any data managing. The data cleaning part, and then the data preparation part is I would say it's almost every day there, because that's what I think that's most of my time at that. So my typical days, it's also involved that, the analysis part is also involved, and then the meeting, the teamwork, and then I like to work with the team of individual. And as analyst, I'm also involved in a paper center that I work with Dr. Andre Gallecki and then another analyst also, as a team for any particular project. And then I also help that analyst and then do some of the project. So it is a combination of working alone for majority part, and then I do work along analysis.
Donovan Maust:
So Mohammed, for listeners who might not be familiar with what exactly data management entails, could you just give maybe one specific example of what's a data management task that you've done recently?
Mohammed Kabeto:
That's a good question. And then for example for a data that involve HRS, which is health and retirement data, which we know this data has been collected since 1992. And then every six years that they add a new cohort, it has continuously been collected every two years.
Mohammed Kabeto:
So this is a reach data set. Not only that they collected this HRS data set, but there are any other supplemental data. And then also there are HRS data linked with different data, with a census data, with Medicaid data. And then all of those datas are really pulling together. And it depends what the PIs are interested, so pulling all those data set and together, it is a big chunk of time is going, should be unlocated for that. Then, so it needs a lot of time. And then linking this dataset, pulling appropriate variables, even go back and then see corrected if there is any error on them, pulling it back and then constantly connecting, communicating with the PIs to pull those variables, are really a key thing that is challenging but also rewarding, at the end of the project or analysis here.
Donovan Maust:
Thank you. I think for future grant submissions, I want to submit this podcast as an appendix because I think for people who aren't familiar with secondary data or observational data, there's this perception like, "Oh, the data's just there. You can just get going right away with your analysis." And I think people really don't understand just the enormous amount of work that has to happen with these types of data to get them into a usable analyzable format.
Mohammed Kabeto:
One thing that I want to add into that, is there is this thing now of working all this time in a group of three with Andre, Dr. Andre Gallecki as leading and then also an analyst. We try to put a program that could be user friendly for anyone who's going to use it to pull the HRS data set and then making it as just this program is in a working process with working it with SAS. And then we are about 1/10th of the time at now completed and then it'll probably be ready in a year.
Donovan Maust:
Let's go with Jonathan. So for a given project that you've worked on in the past year, what's been the process in terms of actually designing and doing the analysis? How does that happen? What's it look like, the back and forth with the PI on that?
Jonathan Martindale:
Yeah, can I clarify? When you say analysis, do you mean the final stats? Or this entire project of putting it to together?
Donovan Maust:
Let's say the data management part's done. And so thinking more about the final. Getting to the paper, the manuscript.
Jonathan Martindale:
To me, I think Mohammed really said it. Most of my time is spent as data cleaning and management. When I get to the point where I have a final data set, the analysis portion is really quite fast, relatively. And usually, I guess I should say in the couple times I've done it, the PI has a good idea of what they want to see. And so getting to that point, for me it was making sure that I'm doing the right analysis, doing some quick Google searching and making sure that's the right thing. And then it was just, output the numbers to an Excel spreadsheet and send it over.
Julie Strominger:
Something also that I wanted to add on that Jonathan has touched on twice now I think, and is a really good point, is that behind all of us analysts, there is a community of statisticians and analysts that are supporting us on Google and on stack exchange. I can't say how many times, probably at least what, four times a week I'm Googling something just to see what other people have said. There's a famous biostatistician at Vanderbilt, Frank Carroll, he will actually respond to questions on stack exchange, and you can go and search his responses. So very helpful.
Matt Davis:
Sounds like a global community all helping each other out there. On our podcast, we have different people listening and I suspect that given that we're talking to analysts today, that we have some principal investigators listening in. And as you know, research now is a team based thing and people have different perspectives, different objectives and just different backgrounds. So I'm curious, I guess Julie, I mean, what should principal investigators know in terms of how to make the process as smooth as possible when they work with an analyst? And what type of information or guidance do you need to make the process as easy as possible for you?
Julie Strominger:
Yeah, that's a good question. I've been thinking about this for the past couple days and I think number one is having a straightforward research question, I think. And I think that's definitely number one starting with that. And I guess it would probably depend on the team and who's an expert on what data sets, but having an idea obviously of what data we're going to use, maybe some papers that have looked at similar outcomes analyses always very helpful.
Julie Strominger:
And I think as Mohammed touched on clear communication. So meeting every week, discussing the plan, where you're at in the project, where you see it going forward, setting some goals at those meetings, whether it's explicitly stated or not. And then I think Jonathan touched on this too, but understanding that data cleaning takes a long time. Even after doing this for almost eight years, I still try to add 20% onto the data management part of it, even past what I think it's going to take.
Julie Strominger:
So I think it's important for the investigator to understand that that part does take time and it's very important because the quality of your analysis is only as good as the quality of the data that you've constructed.
Mohammed Kabeto:
I would like to add just only one thing that is just like, yeah, she, Julie, had touched all of involving the analyst and the statistical design stage is really important for that analyst to connect the research question with the design and then also emphasize sharing some prior research related to the current one is also really important.
Matt Davis:
So I've had the opportunity to work with analysts over the years, and one thing that I've always tried to do, which I felt like was helpful or was just try to mock up tables and figures for a research article as soon as possible. I found that once I show that to the analyst that's working with me, then it becomes clear. And then all of a sudden the analyst can start working to help me get to that end goal. Do you guys find that helpful?
Mohammed Kabeto:
Oh yeah, that is my question always. Even for the abstract, even not only for the just paper for the abstract and then I'll just ask mock table.
Matt Davis:
It seems a lot easier than trying to describe it when you show what you're imagining. Yeah.
Donovan Maust:
If you were an analyst recruiter, who would you be targeting to recruit? What types of people do you think would make a good analyst?
Jonathan Martindale:
Yeah, I think two qualities come to mind. People who are happy to be detail oriented, and people who are really inquisitive. I guess I might add on quick learners. So someone who wants to keep learning and wants to expand their skill set by figuring out how to do this new thing that they haven't learned how to do before. And then who's also really diligent with making sure that this new thing that they're implementing is actually doing what it's supposed to be doing. Doing these data checks and running that back constantly to make sure that what you're putting out is actually what you want to be putting out.
Julie Strominger:
That's exactly what I was thinking. And I think someone who's not afraid to get into the weeds of, "I'm looking at this ARF package, I need to figure out exactly what the defaults are for each call." You can't just run something without understanding all the other implications that has.
Mohammed Kabeto:
And one thing that I would like to add is as we have been just focusing and then also saying a lot, that the analyst spend time more the data managing and a preparation stage, that is tedious. And then someone who should be patient to work on that is very, very important. And then especially, as Julie and Jonathan, then for someone who graduated or came from biostat or stat thinking that they going to do the analysis part. But a big chunk of time is just a data preparation time. So someone has to be patient enough to work on that, as it's the key. And then whenever that I requested to help them recruit a person, and then that is the one that I really emphasize. The analysis part is just as Jonathan said and then it's going to take this a small amount of time and then they can learn it even through experience, and then just keep in embed the data managing, and then the preparation time is a big chunk of time that I emphasize a lot.
Donovan Maust:
So can I ask a quick follow up? So one could imagine that the skills of an analyst and a data manager in some ways are separate and distinct. And so how important is it, do you think, and in some ways maybe the best use of the analyst's time is not on data management. Do you all think if you could only do analysis, would you do that or do you think to be a good analyst you have to still continue doing some of the management yourself?
Mohammed Kabeto:
I would like to do the data managing part. If I do know the data that I created very well, it is going to be easy for me to do the analysis. And a good example that I would give, is if someone has a data created by someone, that person left the job. If you came and then ask me, can you do this? It is going to take me time to just orient myself to the data that is already created. Sometimes I prefer even to create myself that data again. And so the data part is very, I mean for the analyst it is very important.
Matt Davis:
I just want to pick up on something. That idea of being detail oriented. I see that as so important for what you guys do. I remember just the smallest thing can mess up an entire data set. And sometimes I like to joke that, we talk about this thing we call the laugh test. It's sort of like when you do your first pass and you're looking at simple things, it doesn't make any sense whatsoever. Because I've had data files sometimes provided to me where the analyst kind of wraps things up and says, "Here's your data file", and there's like a 2% prevalence of a condition that I know is 25%. And I'm like, "that can't be right." And it came down to a really simple thing. So I see that it's just so easy to mess up data. I don't know if the public really understands that. And especially if you're working under time constraints and you've got to stay focused, those little things really do matter.
Matt Davis:
And I was just thinking, I mean a couple of you in particular, have done this for a long time with the same team, how important trust is I suppose. Once you have that relationship with the investigative team that you're working with, and just how important that is. Once you have that kind of bidirectional trust in terms of that you know you're working on an important good project and at the same time they're trusting that you're being careful and diligent in all your work.
Matt Davis:
Last question, we can start off with Julie and kind of go from there. Just a general question, I guess we're curious, what do you like about your job? And what are the challenges?
Julie Strominger:
I enjoy the team that I work with. I think that I've been working with them now for five years, and it is really nice to have that trust that you just talked about having been built. I think that is, especially when you start out on a team, that is such an important thing to think about.
Julie Strominger:
And I think it's extremely rewarding. Like I mentioned earlier, we're all working together to improve the lives of older adults with dementia, that's the end goal. I personally enjoy data management also quite a bit. So it's kind of like a puzzle. So I love when I have blocks of time that I can just get into the data, start looking at things, does it make sense? And I guess also the challenges develop as I gain more experience in this area because I feel like recently we, Donovan and I, have been talking a lot, sometimes I guess about just, we're using secondary data to answer questions.
Julie Strominger:
And so the challenges really relate to using secondary data to answer these questions. There are inherent biases that come with the fact that we're using secondary data. For example, there's research suggesting that older adults in rural areas are diagnosed with dementia later, relative to urban areas.
Julie Strominger:
And then I would say just an add-on, I think in particular, I would also say an ongoing challenge is estimating the amount of time it does take to clean data. Because there are always issues that come up that you aren't expecting. And I think sometimes I personally can get really attached to timelines. And so just having some flexibility in my mind that there are going to be issues that come up, always add 20% to the estimated amount of time I think it'll take.
Mohammed Kabeto:
For me, and the interesting thing is just teamwork and then working with Ken Langa for the last 22 years, and then for him to just bring a new team that I work with, and then it's really rewarding. And then also working with the different discipline. It could be dementia, it could be a stroke, it could be cancer, and all this different discipline that I work with. It is just rewarding and then vary and then I really enjoy working on it.
Mohammed Kabeto:
The challenging part is in though working with the different discipline or the different PI, the challenging part is managing the time. Which project that you're going to just work on, and then which team that you're going to just complete first. And then all this, it's challenging. And the other one is just like to understand the subject matter. And then I would like to do as much as I can, if it is dementia, and then the path of physiology part of it. And that would help me to do my job is I tried as much as I can to understand it, and then sometimes I start and then not completing it. So that is the challenging part that I would really want to pick it up and then learn more about.
Jonathan Martindale:
I'm just going to build on what Julie and Mohammad said. My favorite part is this kind of puzzle piece of the project where you're trying to put different data sets together and figure out how they work, and what your end product should be and how to get there. And I think it's really fun when you have these different projects or PIs who have different ideas of what they want to do. And so you just get these different perspectives of what's interesting, and how you can do different things with the same data. It keeps it fresh for me.
Jonathan Martindale:
And then I guess for difficulties, it's kind of similar to what Mohammed said on knowing your subject matter. I recently started a project that's not in dementia, and I realized that even over the last year of working, like Matt, you mentioned the last test. When you transfer to something new, what's your bar for the last test? You don't know what should be there, so you have to start looking into it more and asking more questions of what should I be seeing? And then that might get you communicating more with the PI, showing them intermediate numbers like, these are the number of people that I'm identifying with rheumatoid arthritis, for example, and does this match your expectation roughly or is something wrong here?
Matt Davis:
Well, this has been great. Mohammed, Julie. Jonathan, thanks so much for joining us.
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 data products we've created for dementia research.
Matt Davis:
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/podcast.
Matt Davis:
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 NIH or the University of Michigan. Thanks for joining us and we'll be back soon.
Listen to more Minding Memory podcasts - a part of the Michigan Medicine Podcast Network.
