In this episode, Lauren talks with Dr. Amy Herring, Professor of Statistical Science and Dean of Natural Sciences at Duke University, to explore how biostatistics shapes real-world public health research. From her early interest in applying math to meaningful problems to her current work across a wide range of health topics, Dr. Herring shares what it really looks like to build a career at the intersection of data and impact.

The views and opinions expressed in this podcast are solely those of the student hosts, guests, and contributors, and do not necessarily reflect the views or opinions of the University of Iowa or the College of Public Health.

Lauren Lavin:

Hello, everybody, and welcome back to Plugged In to Public Health. Today’s episode explores how biostatistics shapes real world public health decisions and what it actually looks like to collaborate across disciplines to solve complex health problems. I’m joined by Dr. Amy Herring, professor of statistical science and the dean of natural sciences at Duke University. Dr. Herring shares how she found her way into biostatistics, what it means to play in everyone’s backyard as a statistician, and how her work spans everything from pregnancy studies to global health research in Tanzania.

Throughout this conversation, we’ll get into what makes collaboration work, why understanding the context behind data matters just as much as the numbers themselves, and how statisticians help translate messy, real-world questions into meaningful answers. I’m Lauren Lavin, and if it’s your first time with us, welcome. We’re a student-run podcast that talks about major issues in public health and how they are relevant to anyone, both in and outside the field of public health.

So let’s get plugged into public health. Plugged Into Public Health is produced and edited by the students of the University of Iowa College of Public Health, and the views and opinions expressed in this podcast are solely those of the student hosts, guests, and contributors. They do not necessarily reflect the views or opinions of the University of Iowa or the College of Public Health.

Thank you so much, Dr. Herring, for being on the podcast today. If you could start off by just introducing yourself. Where you work, what you do, little bit of background, that would be great.

Amy Herring:

Hello, I’m happy to be here. My name is Amy Herring and I’m a professor of statistical science at Duke University and dean of the natural sciences at Duke.

Lauren Lavin:

And how long have you been at Duke?

Amy Herring:

I have been at Duke eight years.

Lauren Lavin:

Okay. Where were you before that?

Amy Herring:

Ah, that’s a tricky question. Before I was at Duke, I was up the road at University of North Carolina at Chapel Hill, our great friends. Except during basketball season.

Lauren Lavin:

And then where did you do your training and education?

Amy Herring:

Yeah, so I got an undergraduate degree in mathematics and in English at University of Mississippi. From there, I went to Harvard School of Public Health, where I got my doctorate in biostatistics.

Lauren Lavin:

So you’ve been a few different places before you ended up at Duke. You are a professor in biostats and you’ve described biostatistics as a chance to play in everyone’s backyard, and I really love that metaphor, so could you start by sharing what originally drew you into this field and how that curiosity has carried you through your career?

Amy Herring:

Sure. I was an undergraduate with a double major and, at some point, I decided that maybe math was the major that was going to pay my salary, and English maybe wasn’t the thing. I loved it, but it wasn’t necessarily what I saw myself doing for a living, but I got really nervous about math, thinking that maybe I would do something and I couldn’t see whether it would help anybody in the near term, and so I was really drawn towards biostatistics, because I felt like, “This is something I can work on a really practical applied problem.” Hopefully, the work I would do would, in a relatively short period of time, have an impact on the health of populations, and I could see what that impact would be. I started looking into graduate programs, and that’s how it all started. That happened around my junior year in college.

Lauren Lavin:

And then what was your PhD training like?

Amy Herring:

That’s a loaded question. Yeah, so it was actually a really great program. I loved Harvard. There was so much going on all the time, you could be at talks all day long and never get anything else done, so there were just a lot of opportunities for learning. It was a really rich environment. We had a cohort of 10 to 15 students a year, usually, in the PhD program, and it was a really great experience, because it was a very collaborative culture. We all worked together on classes. It’s not like we’re trying to beat each other out for things, so it was a really friendly place to be. We took classes for, I don’t know, about two solid years. We had qualifying exams in there, which were stressful, and then the rest of the time was purely research, so I had three years of research.

The interesting thing, the new thing about the environment for me was that I had been at University of Mississippi, which has a lovely college campus field, just like Iowa does. When I went to the School of Public Health in Boston, the public health school is where the medical school is, it’s where the dental school is, it’s where the hospitals are, so I was in the city.

And Harvard has a really big Master’s of Public Health program that is great for working professionals. People would come just to do a year of… You could do it in just a year, so if you were the CEO of a hospital, you just took a year off work. If you were the health minister of Uganda, you just took a year off work, and so there were these amazing public health professionals that were also there at the School of Public Health at the same time. It really felt like going to work and less like going to school, which was a new for me, but a really great experience all around.

Lauren Lavin:

Yeah. It seems like you really soaked up all of those opportunities to learn and let a lot of curiosity drive you during that time.

Amy Herring:

Yeah, it was great.

Lauren Lavin:

I would say that curiosity kind of connects directly to the collaborative style of biostatistic researchers and your style of work. You have worked with such a range of scientists and projects in your experience. What makes a collaboration between statisticians and scientists work really well?

Amy Herring:

Yeah. I think it’s really good when we understand each other’s strengths and what we bring to the table, when you trust the other people, and in that case, it’s just a huge amount of fun. In a lot of ways, it’s a little weird that I would like collaboration, because I came out of elementary school absolutely hating group projects. I thought they were terrible. One of the biggest times I got in trouble when I was in elementary school was complaining about somebody who was on my project team very not politely. I deserved to be in trouble, but it’s completely different at the level where I’m working now, where everybody is really excited to work on a certain project, everybody’s bringing a distinct type of expertise to the table, and everybody knows something you don’t know.

You’re learning from everybody at the table about a project, and to me, that’s just a huge amount of fun. To me, it works best when we have people in different domains, when people are curious, and they’re not afraid to ask questions to say, “Well, I don’t understand what you mean by,” whatever that is, because often, the same word will have different meanings in different fields, and that experience is really worlds away from the group projects that maybe sometimes we don’t look back on fondly from school.

Lauren Lavin:

Do you ever feel a pressure to be an expert in a variety of topics or do you really try to stay focused on the statistical side, or do you have to constantly be learning about whoever it is you’re partnering with their research?

Amy Herring:

Yes to all of it. I think I’m very clear that my expertise is on the statistical side and even certain sub-fields in statistics. Sometimes there could be a statistical question that would come up and I would need to ask a colleague to do some research on my own to try to understand, because it’s outside my special area of expertise, but the more and more you learn about an application area, the more you have an understanding of what kind of statistical methods are needed. I remember when I first started collaborating, I was collaborating at UNC on this really cool study of pregnancy and pregnancy health, and I would go to these project meetings where we would talk about things that weren’t statistics. We would talk about clinical visits, we would talk about… Oh, my gosh. We talked about the freezers. It was like, “I’m going to kill myself if I have to talk about a freezer in one more meeting.”

It got to the point that I would skip some meetings and my collaborator was like, “What’s going on, Amy? We really need you there.” I went back to the meetings and, pretty quickly after that, there was a talk about the freezers and one of the freezers had thawed during a power outage, and then it became a really interesting statistical issue to figure out, “Can we design a little study to see if the quality of the samples that were stored in that freezer had degraded or were the samples still reliable?” I learned that, even though sometimes it’s not a talk about what I’m really interested in, it’s really important to be in the room to hear what the investigators are grappling with. One study I’ve been working with more recently, the SICK study, is part of a collaborative set of studies that are happening in Tanzania, and a critical thing in one of those studies was that, culturally, it wasn’t necessarily viewed as appropriate for someone to have an autopsy, because they wanted to bury the deceased as quickly as possible out of respect.

A lot of the work on that study was around doing these minimally invasive autopsies that could be done more quickly, that wouldn’t desecrate the body, and then trying to understand, “Do we get the same type of information from that autopsy that we might from a more typical one?”

Lauren Lavin:

Yeah, so there’s a lot more to it than just math.

Amy Herring:

Yeah. For sure, for sure.

Lauren Lavin:

What areas do you say that you’re an expert in as far as statistics sub-fields?

Amy Herring:

Statistics sub-fields. I think a lot of the projects that I work on, having common data that are correlated. By that, I mean that I’m measuring multiple different outcomes on the same person. I might be looking at your academic performance in English and your academic performance in math, and maybe there’s some positive correlation between those things, because good students might tend to be good students, or I might be looking at something like your dietary patterns, and I might be looking at those over time. How much do you change the type of food you eat as you get older, as you go through substantial life changes, like going off to college or having a baby? Or something like that. One general theme is that I’m looking at variables that are related in some way, so we would call that correlated data or longitudinal data.

Lauren Lavin:

Yeah. In the example of a student being good at both math and reading, how do you handle that type of problem with the data?

Amy Herring:

Yeah. If we have data that where we’re measuring similar things on a person, what we would not want to do is assume those observations are independent. If I see that you’re a really good student in biology, then probably in my head I’m thinking, “She’s also likely to be a really good student in chemistry and maybe to be above average in physics and in other subjects,” and we’ll have to account for the correlation and the data. There are a variety of ways we can do that. One way that you may hear people talking about is through a random effect or a subject specific parameter that we introduce in the model that tells us how related these items are within an individual.

Lauren Lavin:

I’m glad we have people like you to help us with that. When you’re thinking about collaborators, how do you go from their scientific question to developing a statistical approach that addresses that with them?

Amy Herring:

Yeah, that’s a great question. As I work in teaching classes and educating our students at Duke, whether they’re undergrads or master’s students or PhD students, I think that’s actually one of the hardest things to learn, because you’re sitting there talking to somebody about something that, on paper, has nothing to do with math, and then now I need to take that and turn it into a model or some sort of framework that I can use to answer a question. There might be back and forth with the individual, because maybe I can’t get at exactly the question they posed to me, because the study design doesn’t allow it or they haven’t collected exactly the data that we might want, so there can be a little back and forth where I say, “Okay, if we take this modeling strategy, I can answer this question. How close is that to what you want?”

We’ll do a little back and forth to try to understand what the data sources are, if the data sources aren’t adequate to answer the question they want, “Hey, can we go get some data? Is it easy, an easy thing, to collect some more data where we can really answer the exact question they’re interested in? Or is it a situation in which we can answer something that’ll tell us about the question that they’re interested in, but it might not be a direct answer to that question?” An example of that might be the old studies, the old epidemiologic studies they were doing in smoking and lung cancer.

The gold standard-type study to see if smoking causes lung cancer, one might say we would do a randomized study where we take a group of people and we tell half of them to go smoke and half of them not to smoke, and we follow them all their lives and see who gets cancer and who doesn’t. That’s obviously completely unethical, so we have to use different strategies to answer that question.

Lauren Lavin:

An example of a collaboration that you’ve recently done is the sepsis characterization in Kilimanjaro, or the SICK. Is that what you call it, SICK study? Could you walk us through what the original goals of that project were and how those goals shifted once you ran into some unexpected challenges?

Amy Herring:

Yeah. SICK is a really cool study and the principal investigator of that study is Matt Rubach, who is in the Department of Medicine at Duke and also has an appointment at Kilimanjaro Christian Medical Center in Tanzania. He treats a lot of infectious disease patients and sees a lot of patients with sepsis, which sepsis occurs when your body has an infection, and it’s basically an overwhelming reaction to that infection that can lead to this overload that can actually be fatal. In fact, a fair amount of the world’s deaths annually have sepsis as a contributing cause, so there’ve been a lot of studies of sepsis because it’s hard to treat, the mortality rate is high, but these studies have been carried out in developed countries for the most part. There’ve been big studies in the US or in Europe, and through those studies, they’ve identified some tentative subtypes of sepsis, and they think the different subtypes could be responsive to different types of treatments.

This was super exciting. What we see in non-highly developed countries is that people with sepsis look pretty different. In the US, for example, sepsis tends to be a disease of older adults, they tend to have comorbidities of older adulthood, so they might have heart disease, they might have diabetes, they might be obese, but in Sub-Saharan Africa, if we look at a population, the patients presenting with sepsis, they tend to be younger, they tend to have different organisms causing the infection. They have higher rates of HIV in the population, so they look really different, and when they started using treatments developed in developed countries on the patient population in Africa, in some cases, they actually saw that the treatments were harmful. That caused the researchers to step back and say, “Whoa, these subtypes may not apply in Africa in our setting in the same way they do in Philadelphia,” which is where one of the big studies was carried out.

The purpose of the SICK study was to derive subtypes specifically for the Tanzanian population that they were seeing in the emergency rooms and in the hospitals, and then compare those to the subtypes that had been derived in different populations in the US and Europe. What was hard about that was like, “How similar is similar for a subtype and how different is different?” And it’s super hard to measure. We would see differences, for example, in rates of HIV, but we knew that there’d be big differences, because the rate of HIV in our population was really high and the rate of HIV in the US population is like well under 1%. We saw that people in the US were older. Of course, we knew they would be, because we knew the source population was going to be older.

We saw different characteristics of disease and it became really hard for us to figure out, “Is a fever difference of one degree a big difference between groups or not such a big difference between groups?” That was something that was hard for us to answer and led us to try to develop new statistical methods.

Lauren Lavin:

What kind of methods then did you come up with to try to deal with those problems in the data really?

Amy Herring:

Yeah, that’s an ongoing project actually. It’s something that certainly raises more questions than answers at this point, which has been really exciting for us. We had a PhD student at Duke, Alex Dombowsky who was really interested in clustering methods, and what he’d arrived was a way to center a clustering solution on prior data. That was one approach we used, was to think about, “Okay, we can think about the clusters we got in the US population, for example, as a starting point,” but then we’d really like our data in Tanzania to inform the clusters we see at the end of the day. That was one strategy we’ve used. We’ve also been working on writing grants to explore different strategies to get clusters that are easier to interpret, so that then they’re easier to compare. Yeah, it’s been a whole lot of fun.

What’s been really fun about that is that it’s been a team effort. We brought in PhD students and some postdocs in statistics, but also Matt Rubach is on our team, or my original investigator. Another physician at Duke, Deng Madut, is also involved; David Dunson at Duke and Statistics. We have a long list of people now and we’re gathering more every day as we see more about the data and see really interesting problems to solve.

Lauren Lavin:

Yeah. How often do you feel like you use methods that have already been established versus creating new methods to address problems?

Amy Herring:

Yeah, that’s a great question. If we could answer this question with a T-test, I would do it and be done there. Simple, easy, super interpretable. We definitely don’t want to get big tools out if we have a really reliable, studied, easy tool that we can apply to the data. I think, for the most part, a lot of times, in a collaboration, there is something out there that works or it works enough. Maybe we have a good understanding, maybe the data aren’t exactly normal, but they look pretty darn good, and I have a big sample and I’m pretty confident that what we found is right. I have ways to assess whether my inferences are reliable or not. In this case, it was a Wild West and we had some simple methods we could use, but no really tried-and-true ways to make comparisons.

I would say it’s the exception rather than the rule in a collaboration where you have to build something brand new and, often, there’s either something that’s like a great fit or something that I can tweak a little bit and make it work, or maybe the case on a recent project we had, where we were looking at damaged neurons with Joel Meyer, an investigator at Duke. They had data that were damaged scores from zero to six, basically integer scores, and they don’t look at all normal, because yay for us, most neurons aren’t damaged. There’s a lot of zeros, no damage, and not that many sixes, so terribly highly damaged.

The team had some methods that worked pretty well, but we’re thinking of, “We can make your method more powerful. We can make it easier to detect differences when they do exist if we make some tweaks.” Sometimes there’s a case like that where there’s nothing wrong with what they’re doing, we can just improve upon it if we put a little effort into it.

Lauren Lavin:

I think that’s really interesting, how you just point out that sometimes there’s a method that fits pretty well, but with some tweaks. I think that’s where statisticians, especially if you’re a PhD student listening to this, I think can be really helpful in your own research.

Amy Herring:

Yeah, sure.

Lauren Lavin:

As you reflect on the SICK project now, what opportunities are you most excited to see come out of that study going forward?

Amy Herring:

Yeah, so I think what we’re super excited about… It’s a two-phase study, so in phase one, we analyzed some data from the first part of the study, we’ve derived clusters, and we have phase-two data coming in now where we can validate those clusters. I’m really excited to see how well they validate. If we do find clusters, then I think the next step would be to study those in a little more detail, so back to the drawing board. It’s not a point where we would be able to say, “Here’s a new treatment that targets cluster three,” or something. What we would do at this point is say, “Cluster three looks really interesting. We haven’t seen clusters like that previously.” Looking at the factors that are related to this cluster, maybe that gives the physicians some ideas about what kinds of treatments they could try. That, to me, is this super exciting part.

One thing that we couldn’t do with the study that’s actually exciting, we couldn’t do what we wanted to do, was that we did want to link the clusters to a mortality outcome and see, for example, for some clusters, higher mortality clusters than others, and that’s actually been really hard, because our mortality rate is low. Yay for a low mortality rate, that’s great, but from a statistical power perspective, we have a little less power than we thought we might have to look at differences in mortality, because of better clinical care and we also think because of the features of the patients that we’re seeing in the emergency room, they’re just maybe not as sick as some of the patients that were in the US studies, they’re doing better. Doing better is a good thing.

Lauren Lavin:

That makes sense. These types of stories and conversations I think are really helpful for students and even faculty members. I know that you’ve mentored a lot of students and junior scholars over the course of your career, so what advice do you have them about finding meaning and creativity in collaborative work, especially when the path forward isn’t always straight?

Amy Herring:

Yeah, it’s never straight. I remember many times in grad school where I thought I’d never get out, when I would just sit there forever spinning my wheels. One thing I think is useful is really digging into a project. The very first student I worked with on these projects in Tanzania was Kelly Moran. Kelly is now at Los Alamos National Laboratory doing super cool work, but when she was a PhD student, she had an opportunity, through Duke’s Global Health Institute, to become a global health doctoral scholar. What that did for Kelly, if she hadn’t had money, it would’ve paid her tuition, but she actually had a fellowship from the Department of Energy, which was really great. Keep your eyes open for cool fellowships like that. But she got money from the Duke Global Health Institute to visit the sites in Tanzania.

Kelly got to go to Tanzania, got to see, meet the investigative team, got to take part in seeing how data were collected, how data were processed. We were both going to go back together to visit the team, and then COVID hit, so we didn’t get to do that, but it was a great experience for her. When I think back, even before I went to grad school, that summer between undergrad and grad school, I had this internship with the Department of Agriculture. I had the opportunity to participate in some data collection on the Cotton Objective Yield Survey. That means you go out into a cotton field in July in Mississippi and you have to wear long pants and long sleeve shirt, because of pesticides, and it’s hot and you’re cranky. I learned all sorts of super interesting lessons that I didn’t realize at the time, but then I got into grad school and learned things about, “Measurement error. Yeah, digit preference.”

That’s what was happening at 4:00 PM when I’d been out there nine hours, and I’d look at a cotton plant and say, “Yeah, 50 cotton bowls.” As opposed to counting them one by one. I think anytime you can actually participate in the data collection process or see how things work. I got to learn what happened when they drove out to a farm and the farmer said, “No, you can’t come on my land to count my cotton bowls,” even if that’s what the government drew as our sample. I got to see how they pivoted to carry out the survey to pick another plot and things like that. I’d say go in the field, get your hands dirty, do some community outreach event where you get to talk to people about the kind of work to see what their questions and concerns are. I think that can be really rewarding versus sitting in the office with your laptop, maybe not seeing any impact or seeing a bigger picture on a day-to-day basis.

Lauren Lavin:

Yeah. It definitely adds a narrative to the work that you’re doing, which I think is what helps create some meaning.

Amy Herring:

Yeah.

Lauren Lavin:

That’s really great, tangible advice. Did you ever get to go to Tanzania after that first trip was canceled?

Amy Herring:

I have not been to Tanzania. We were hoping maybe to have a conference there as a closeout, but we’ll see. Yeah, still on the to-do list.

Lauren Lavin:

Thinking a little bit more broadly, how do you see the role of biostatistics evolving over the next decade, as global health challenges and data sets become more complex because we just have so much more of them, but the infrastructure really isn’t there to support that?

Amy Herring:

Yeah, that’s a great question. I’ll maybe talk about something that Iowa is known for, and it is the least sexy statistics course out there is experimental design. The statistician was super, super important course, especially now, because what we’re seeing are lots of data sets that are massive data sets, electronic medical records data or data harnessed from the internet. There’s a great paper by Xiao-Li Meng, who was a dean and used to be a department chair of statistics at Harvard, on the big data paradox, where we might have a lot of data, but they may not give us a lot of information, because there’s certain biases in the data. If I go to a grocery store in Durham and ask people about the reputations of Duke University and the University of North Carolina at Chapel Hill, I’ll get a really different answer than I would get if I went to a grocery store in Chapel Hill and asked the same question.

With a lot of these data sources, with the internet, we don’t know who’s going to certain sites, we don’t know who’s engaging with those sites very well, and you have to be very careful to get high quality data in those settings. Electronical medical records data, that depends on who has money to go to the hospital, who has resources to get to the hospital, who is able to take certain types of treatments or not, based on facts around their life. I think understanding careful experimental design is something that’s really key, because people want to make causal inferences about the relationship between an exposure and a disease or another outcome. They don’t want those correlations that they’re detecting to be spurious in some way.

Lauren Lavin:

Absolutely. That’s a good thing to be thinking about going forward, quality over quantity, right?

Amy Herring:

Yep.

Lauren Lavin:

To close this out, if you could clear up one big misconception the public or even other scientists have about biostatistics, what would you want them to understand?

Amy Herring:

Big misconception. First, I would say statistics doesn’t have to be scary, so I’m sorry if you’ve been traumatized in a statistics course, but we’d love to teach you one. I think there are a lot of good statistics courses out there and there’s been a lot of effort in making statistics a lot more fun. The other thing I would clear up is that, even though I might have a doctorate in biostatistics, I know nothing, mom and dad, about what medical treatment you should be getting. Please ask your doctor, not your daughter. I’m not an expert. I’m not that kind of doctor. But sometimes people will ask for medical advice, because they know you’ve collaborated in an area, and it’s, “Let me go talk to my doctor friend,” because… I don’t know how to advise you during your pregnancy-

Lauren Lavin:

I love that.

Amy Herring:

… a date.

Lauren Lavin:

That’s great.

Amy Herring:

Yep.

Lauren Lavin:

Thank you so much for taking time out of your day to chat with me. I really appreciate it. I learned a lot. I think this is a great conversation, so thank you so much, Dr. Herring.

Amy Herring:

Yeah, thank you. Great to talk to you, Lauren.

Lauren Lavin:

That’s it for our episode this week. Big thank you to Dr. Amy Herring for joining us today. In this conversation, we heard how biostatistics is far more than just numbers. It’s about asking better questions, understanding the context behind data, and working across disciplines to solve problems that don’t come with clean answers. From designing studies in global health settings to navigating messy real-world data, this episode highlights how curiosity, collaboration, and adaptability are central to meaningful public health work.

This episode was hosted and written by Lauren Lavin and edited and produced by Lauren Lavin. You can learn more about the University of Iowa College of Public Health on Facebook. Our podcast is available on Spotify, Apple Podcasts, and SoundCloud. If you enjoyed this episode and would like to help support the podcast, please share it with your colleagues, friends, or anyone interested in public health. Have a suggestion for our team? You can reach us at cph-gradambassador@uiowa.edu. This episode is brought to you by the University of Iowa College of Public Health. Until next week, stay healthy, stay curious, and take care.