Dr. Lloyd Edwards is Professor and Chair of Biostatistics at the University of Alabama at Birmingham. Dr. Edwards has an extensive background in collaborating with researchers in a broad range of areas in biomedical research, including cardiovascular disease, cystic fibrosis, cancer, aging, pediatrics, and minority health. His primary area of applied statistical research relates to the analysis of longitudinal data. Specifically, his statistical research includes derivation of techniques for computation of power, control of Type I error, and measuring model fit in linear and generalized linear mixed models.
+ Full Transcript
Rosemary Pennington: Odds are, as you’ve read research or news stories on the efficacy of particular medications, trends in aging, or issues related to minority health, you’ve likely encountered the work of researchers specializing in biostatistics.
The University of Alabama Birmingham, or UAB’s Biostatistics Department, explains that those interested in the subject study, “how data, population studies, and health, intersect”. That intersection is the focus of this episode of Stats and Stories, where we explore the statistics behind the stories and the stories behind the statistics.
I’m Rosemary Pennington. Stats and Stories is a production of Miami University’s Departments of Statistics and Media, Journalism, and Film, as well as the American Statistical Association. Joining me in the studio are regular panelists, John Bailer, Chair of Miami Statistics Department, and Richard Campbell of Media, Journalism, and Film.
Our guest today is Lloyd Edwards. Edwards is the Chair of Biostatistics at the University of Alabama Birmingham’s School of Public Health. Lloyd, thank you so much for being here today.
Lloyd Edwards: Thank you for having me as your guest.
Pennington: As I was reading the introduction, I shared the way your department, sort of, explains what people who study biostatistics do. How do you – when people ask you what it is you do – how do you explain what you do?
Edwards: There’s different explanations, but often times when I’m explaining it to laypeople, I just tell them it’s the application of statistics through biomedical data, because you have to try to keep it simple enough for people to understand. We know it’s a lot more complex than that.
John Bailer: So, Lloyd, how did you become a biostatistician? What attracted you to the field?
Edwards: I stumbled into it.
[LAUGHTER]
Edwards: Like most people do. I was a math major in undergraduate, physics and economics minor, and I asked my professor when I was graduating, what I wanted to go to a field that I could easily go from academia to industry if I had to, and he said statistics was it. So, I originally went into math stat, and then I discovered biostatistics from a friend. My friend told me that you didn’t need to have any biology background; I knew it was the field for me.
[LAUGHTER]
Bailer: So, did you go immediately into work as a biostatistician in an academic setting when you were done, or what were kind of the twist and turns as you got to where you are now?
Edwards: Well, when I got a master’s in mathematical statistics from the University of Maryland College Park, I went and worked for three and a half years, and that kind of set my vision on things. I worked for TRW defense systems doing anti-submarine warfare, and I realized I really didn’t want to apply my mathematical skills to, yes, a warfare. So, when I got my PhD in biostatistics, I knew that I wanted to apply my skills to health-related stuff. I wanted to have a positive impact. So, I got my degree at the University of North Carolina Chapel Hill, and I stayed for 25 years.
Bailer: What a great place to go for graduate study.
[LAUGHTER]
Edwards: Sounds a little biased to me.
Pennington: We know, John.
[LAUGHTER]
Edwards: And it was a great place. Fabulous place to be. The exposure you get. The problems that you get to see. The people that you meet. And if I hadn’t gone there, I don’t know if I would have met you, John.
Bailer: That’d been a loss for me.
Richard Campbell: Lloyd, I have a question just about some terms. So, you’ve been involved in longitudinal studies and longitudinal data? Can you talk a little bit for those of us who don’t really know what that is? Can you talk a little bit about what longitudinal data is for a layperson?
Edwards: Longitudinal data is repeated measures on the same variable outcome over time. In today’s society, in today’s technology, we now can collect, and store, and keep track of vast amount of data, and part of that is following individuals over time. How repeatedly you may want to do it. You can follow them for years, and the question is, how do you now extract information from the individual and also the population, where repeated measures and longitudinal data – new longitudinal data analysis techniques – allow you to estimate trajectories for both the population, the group, like gender, and race, but also individuals in that group. And that’s a newfound thing that’s found in all sorts of applications everywhere.
Campbell: Can you talk a little bit about variables or factors that alter that kind of data over time? You mentioned specifically, maybe gender and race, how our attitudes might change over time?
Edwards: Well, it’s interesting, because when I said outcome over time, just say lung function, you may measure that every day over time, however, but we don’t necessarily think about what we consider static variables changing over time. In today’s society, you know, we have gender that changes over time. Race can change over time. If you’re from a mixed-race couple, you could claim one heritage in one year and another heritage in another year. If you were doing a pharmaceutical company study, let’s say set the gender and somebody changes their gender over time, those are some of the more complicated things that can, then, affect how you’re looking at that outcome.
Bailer: So, Lloyd, one of the things that I think is probably maybe the most familiar example when I’m asked to give examples for longitudinal data would be like the growth curves that people see when they take their kids to pediatricians, sort of, plotting a child’s position and height and weight on those curves. Can you think of other examples that would be kind of common and familiar?
Edwards: The new technology we have with these devices are opening up different types of data like, continuous-time glucose measurements, or 24-hour ambulatory blood pressure measurements, where you’re continually taking measurements on an individual, on a specific day, for 24 hours. I’m involved in this study where we’re proposing to take continuous-time glucose measurements and analyzing it, and that’s complicated. We’re talking about taking measurements every minute over at 24-hour period. So, you have a massive, large amount of data that’s very, very complicated. I looked at one of those trajectories, and it’s challenging, to say the least. So, you have a number of things that we’re doing in daily. You think about this Fit Bit technology we have that you can count your steps every day, you can plot that out. So, we have a vast amount now of opportunities to do repeated measures or longitudinal studies.
Pennington: Lloyd, how important is work like this for public health programs or schools or how helpful is this compared to, maybe, a one-shot study that might look at one moment in time, but having this continuous data available to you?
Edwards: The problem is, one shot in time is just that. It’s a biased look. It may not show you the trajectory over time. I had a study doing weight loss, where weight loss will decrease over a short amount of time, then it goes flat, where people don’t lose any more weight. Well, if you don’t look at that entire trajectory, you presume that if someone stays on that weight loss plan, they will just continue to lose weight, and that’s just not the case. I mean, we’ve seen that in reality. So, one shot in time is informative, but it can give you a wrong picture of how something develops over time. You know, every disease has a time course involved with it, well, most of them do. So, it’s better to know what that time course is like and that process is like.
Campbell: So, I have a question, sort of related to the example that John gave but something we talked about. So, I have a grandson who’s five, who’s always been on 50th percentile for weight, and 90th percentile for height, can you predict anything about how tall this guy’s going to be if he stays there? And those have been pretty consistent, because we get that report every few months. John, you can weigh in on this too.
[LAUGHTER]
Edwards: Yeah, we have the equations for that. When it comes to growth, height, and weight height charts, we pretty much have that down pat. They have all types of instances. For example, my son, my youngest son, both my sons are 6’6”.
[LAUGHTER]
Edwards: Those growth charts meant nothing to me, because they were both off the charts. There are areas even for us, where they’re so rare that you can’t chart it well. In your son’s case, 50th percentile weight, he must be pretty slim.
Campbell: Yes, he is.
Edwards: They have those trajectories pretty well. So, yes.
Pennington: I think what Richard wants to know is, is his grandson going to grow up and be a big NBA star and take care of his grandpa in his old age.
[LAUGHTER]
Edwards: You know, it’s interesting with boys, because you have growth spurts. I know I had it and both my sons had it, and you just don’t know. The complication with genetics and everything else. Both my boys are from the eighth, to the ninth, to the tenth grade, sprouted like six inches. And I remember they used to be so tired. They complained about their joints hurting, and I just laughed to myself and say, that’s Mother Nature. You grow that big, that fast, you know your ligaments has to stretch. Your body is metamorphosis. [LAUGHTER] But yes, your son should be fine, and again, with today’s technology, we give them a few steroid shots, he’ll be good.
Bailer: Hey, I want to follow up on some of those internet of things, new equipment, new measurements that you describe like the glucose measurements over time and other kind of continuous monitoring data. Were these part of an outcome looking at something like diabetes or some other types of disease, and how were you going to use these types of continuous measurements over time, for doing that analysis?
Edwards: The one where I’m involved in trying to propose is actually looking at pregnant women and how their glucose is affected during the day, because they have some complications, if they do have diabetes-related or high glucose in that regard. So, we’re going to use it to try to monitor how their glucose or diabetes may be affected. And you could imagine that if we have the trajectory right and measuring right, that if it’s doing continuously, that there could be alerts or alarms if things start getting high. Instead of waiting until there’s danger and you’re going to the emergency room, you could have an alert that says, listen, you’re about to go too high on this glucose; take your insulin, or intervene in some type of way. Same thing with 24-hour ambulatory blood pressure measurements. If one knows that the blood pressure is about to spike, you can do that if you’re continuously monitoring. Otherwise, you have to wait until you’ve collected all the data, wait until the next week when somebody’s downloaded the data, say oh, you had a spike. You really don’t want to wait for that.
Bailer: It’s too late.
Pennington: You’re listening to Stats and Stories and today we’re talking biostatistics with UAB’s Lloyd Edwards. Now Lloyd, health news tends to be one of those things that gets covered a lot, particularly on broadcast news, but also on online and newspapers. What is the most frustrating thing for you when you’re reading news stories about things related to biostatistics? What do you think journalists get wrong most often?
Edwards: Oh, my goodness.
[LAUGHTER]
Edwards: Because they have to try to cannibalize what we’re saying. They just tend to just not get it right. And plus, the people who are interpreting it or giving it, the information, aren’t scientists themselves. So, half the time we hear such wrong things, and it gets exasperating, but it’s a similar thing we face as statisticians with respect to manuscripts, writing manuscripts. Oftentimes, when we’re writing with a subject matter expert, they don’t allow us to expand on the analytical technique that we’re doing, and John knows oftentimes, they’ll try to hold us to one paragraph to explain something very complex. Oftentimes, people shy away from the complexity. There’s layers to things, but people – the news want to report black and white situation, where there’s a lot of gray area when it comes to stuff like that. For example people will say that smoking causes cancer, however, there’s really not a causal relationship there. You can say smoking is highly related to cancer but there’s people who smoke who don’t get cancer. Lung cancer that’s what we’re talking about, and that’s kind of how the interpretation can be wrong from journalists or anybody else.
Bailer: Yeah, your comment about shying away from complexity really resonates, and I think we often see that, in terms of kind of just the size of an affect without of any sense of uncertainty. What I think of is almost this overly strong sense of precision of what’s being reported, as opposed to sweeping under the rug some of the nuance. So, do you have certain strategies when you try to convey the uncertainty or variability or kind of nuance to a conclusion when you’re working with investigators or when you’re thinking about reporting this to the general public?
Edwards: Oftentimes, I try to tell them to interpret somethings cautiously. You get a p-value of 0.049, when the cut off is 0.05, you know, they running to the bank with it. They’re saying, we have a significant result, where we’re saying, no, not quite. You need to interpret it correctly. The other thing that gets me is that, oftentimes, people don’t understand, these are averages. These are means with some type of spread on them. Some type of range of values. But they’ll talk about a median or mean, but there’s a spread that people can follow it within. So, it gets frustrating to not be able to talk about those nuances on a certain interpretation.
Campbell: So, this is a – your comment about journalists and looking at some of your abstracts – I always try to imagine as a former journalist and a journalism professor, how would I explain this? So, some of your abstracts you talk about a linear mixed model?
Edwards: Yes.
Campbell: Could you talk a little bit about what that is, and explain that to a journalist?
Edwards: The linear mixed model is basically an advanced analysis of variance technique that allows us to relax a number of complicating or restricting assumptions on a model. It is a model that contains both fixed effects and random effects. When I say fixed effects in terms of modeling, is talk about fixed effects would be something like gender or race. And then you have the random effects would be the individuals and the times that the individuals may have, because any individual may have different observations on different times. The linear mixed model allows us to incorporate all those effects in, but it’s a very complex model, because you’re modeling two things. You’re modeling both the mean portion of the curve and you’re modeling the co-variance. How these variables co-vary with each other and how they’re correlated. So, it is a fairly complex procedure that is hot; man, it’s hot!
Campbell: I actually understood that.
Edwards: This is making my career right now.
[LAUGHTER]
Campbell: Thank you, I actually understood that.
Bailer: And that ties directly to all your interests in longitudinal data?
Edwards: Yes, sir. It does.
Bailer: So, that’s that whole repeated measures things and worrying about correlations and cross measurements within an individual.
Edwards: Yes, and one of the things I that, that issue of variability, as you well know, John, as a statistician, I wouldn’t have a job if it wasn’t for variables.
[LAUGHTER]
Bailer: Oh, Amen! That’s job security.
Edwards: That job security. We embrace it. And I don’t want that variability on the golf course, but real life? Yeah, let it vary. And trying to explain that variability impacts a lot of things, and that’s complex, and we do it with complex language. The language that we generally do these types of models with and develop them with is matrix theory and matrix language. Well, not everybody has that type of background. So, it can get complex. For example, you can’t know what we mean by a co-variance matrix, if you don’t know what a matrix is.
Bailer: Yeah, one of the things I’m curious about, what’s been the most interesting problem that you’ve worked on over your career?
Edwards: Interesting enough, it was the application of the linear mixed model to premature infant data. It was an eye-opening experience for me, but because we were trying to get the mixed model out in the real word, and there was a great reluctance for a number of entities to embrace it.
Bailer: Well, what was the outcome, Lloyd?
Edwards: The outcome was sleep wake states for premature infants.
Bailer: What does that mean? Sleep wake status.
Edwards: When a child is asleep, they would make certain measurements in terms of let’s say, Rapid Eye Movement, and whether they had active sleep or REM sleep. I don’t have the explicit definition around here but –
Bailer: No worries.
Pennington: But were you comparing them to non-premature babies to sort of see if there’s a difference between them?
Edwards: They were comparing premature infants from one cohort to another, one-time cohort, because technology, even for them, changed over time. So, they wanted to see where the one cohort that was studied earlier – with their techniques – was different than another cohort that came later, where the technology had supposedly gotten better. And it was interesting, because in trying to publish the paper, we had a lot of resistance, and the resistance was from the person who had a technique that they didn’t want us to use. But they did not want us to get the mixed model out there, because it would have wiped their technique out. So, we battled with that for a little while and then finally overcame. But it’s a classic case also of how these types of things are discovered. Here I am as working on it from the theoretical side. One of my professors gave a talk, my senior professor Ronald Hamm, gave a talk, and one of the nurses sat in there, and she knew nothing about mixed models or anything like that, but she knew she had repeated measures data, and she came to us, she said, can you do that thing to my day?
[LAUGHTER]
Edwards: And we said, yes, we can.
Pennington: Lloyd, I’m going to go back to the journalist question, because you clearly are frustrated with journalistic coverage of medical stuff. I, as a former science and medical reporter, would also say, I am also frustrated. But what advice would you have for a reporter who’s working on a deadline, who doesn't have a lot of space in their copy or their broadcast story to go as in-depth as they could - or as I think a lot of us would like them to do – what advice would you have for a journalist to be able to communicate the complexity of something in a short amount of time when it comes to health data?
Edwards: To work with that person and ask them explicitly, because that’s something we train on also. If the journalist were to tell us, could you explain this in three or four sentences, as to a layperson, then we could give it to them. But oftentimes, they’re asking us questions, we’re responding from our professional side, and we’re giving them the full Monty.
[LAUGHTER]
Edwards: But if they told us they need a three or four sentences just to summarize something, we would give it to them. So, if journalist would just work with us in that regard and be more explicit. Sometimes you have the feeling that they’re working in a gotcha type of scenario just to see if you can say something provocative that they could take the run with it.
Bailer: Well, I’m wondering from the stats side of this is just the willingness of the stat community to be such collaborators, and also the willingness to kind of, not do the technical work and kind of do this type of outreach, and extension, and partnership. Do you think there are barriers for that?
Edwards: Yes, it is, John, because we’re in the field of introverts.
[LAUGHTER]
Bailer: Yeah, yeah, I was thinking that’s how I’d describe you, Lloyd.
[LAUGHTER]
Bailer: John’s not very introverted either, I would just like to state, for the record.
Edwards: I’m an introvert. I’m forced to be an extrovert, but I’m an introvert. Half the time, we don’t want to talk to people.
[LAUGHTER]
Bailer: Amen, brother.
Edwards: If I had my dithers I would sit in my office eight hours and not talk to anyone. Another funny thing, for example, I’m trying to get my department more visibility. So, to try to do that I’m asking all my faculty to do a taping to put up on our website. Man, it’s taking me a year to try to get somebody to step up. None of them wants to be taped. So, I’ve had to pull a rank and say, OK, this is how the schedule is going to go; we are introverts, and we just – it’s something that we need to get away from, because we’re the ones that should be – can be the best at explaining our conclusions and what we’re doing. So, we have to work at that.
Campbell: So, you talked a little bit there about being a department chair, and I’ve been a department chair, John’s a department chair now. Talk a little bit about, when you’re recruiting graduate students, who are you looking for? What kind of background? How do you make those kinds of decisions?
Edwards: Well, biostatistics is a mathematical discipline, but we’re part art and part science. So, number one, 99 percent of our recruits are math majors or math minors, because they have to be able to get past the mathematical side. But we’re also looking at other things, do they have balance? Can they communicate, because if they can’t communicate, that’s a central part of what we do. Do they have a sense of wanting to do exactly the type of work that we do, and that’s applications type of work? That’s kind of dirty work. It’s rewarding, and it actually pays pretty well too. But as a math – and you got to understand the types that you’re talking about. We’re talking about people with math backgrounds that are introverts. Some of those other aspects, they don’t care about. They don’t care about – necessarily – the communication, because you’ve sent your whole life not doing that to a group of people. So, we look for characteristics that they can grow. We generally talk to all of them so we can get a sense, because nowadays, it’s not enough to be like an Einstein if we can’t communicate with anybody.
Pennington: Well, Lloyd, we’re going to have to leave it there. Thank you so much for being here today.
Edwards: Are we finished already?
Pennington: We are.
[LAUGHTER]
Pennington: We got to stick to that time.
Edwards: Oh, what happened? That went pretty quickly. OK, thank you guys!
Pennington: Thank you so much. Thanks for being here today. Stats and Stories is a partnership between Miami University’s departments of Statistics and Media, Journalism, and Film, and the American Statistical Association.
You can follow us on Twitter, Apple Podcast, or other places you find podcasts. If you’d like to share your thoughts on the program, send your email to StatsAndStories@MiamiOH.edu or check out our website, statsandstories.net and be sure to listen for future editions of Stats and Stories, where we discuss