EPITalk: Behind the Paper
This stimulating podcast series from the Annals of Epidemiology takes you behind the scenes of groundbreaking articles recently published in the journal. Join Editor-in-Chief, Patrick Sullivan, and journal authors for thought-provoking conversations on the latest findings and developments in epidemiologic and methodologic research.
EPITalk: Behind the Paper
Socioeconomic Disparities, Childhood Obesity, & Adolescents' Mental Health – Insights from the Netherlands
PhD candidate Maria Gueltzow shares the compelling findings from her article, “Childhood obesity's influence on socioeconomic disparities in young adolescents’ mental health,” published in the June 2024 (Vol. 94) issue of Annals of Epidemiology. In this study, the researchers estimate the contribution of the mediating and moderating effects of obesity to the disparity in adolescents’ mental health.
Read the full article here:
https://www.sciencedirect.com/science/article/pii/S1047279724000486?via%3Dihub
Episode Credits:
- Executive Producer: Sabrina Debas
- Technical Producer: Paula Burrows
- Annals of Epidemiology is published by Elsevier.
Hello, you're listening to EPITalk: Behind the Paper, a monthly podcast from the Annals of Epidemiology. I'm Patrick Sullivan, Editor-in-Chief of the journal, and in this series we take you behind the scenes of some of the latest epidemiologic research featured in our journal. Today we're talking to Ms. Gueltzow about her article "Childhood obesity's influence on socioeconomic disparities in young adolescents' mental health." You can find the full article online in the June 2024 issue of the journal at www. annalsofepidemiology. org. Maria Gueltzow is a PhD candidate in Public Health at the Max Planck Institute for Demographic Research in Rostock, Germany, in affiliation with the Erasmus Medical Center, Rotterdam, the Netherlands, which is where she will receive her PhD. Her research focuses on gaining a deeper understanding of subgroup differences in mental health through the use of the potential outcomes framework and genetically informed designs. Ms. Gueltzow, thank you for joining us today.
Maria Gueltzow:Thank you so much for inviting me.
Patrick Sullivan:I want to start out by just seeing if you'll give us some background about the problem you described in the paper. Even in the title, you know, we can see that there's a lot going on here. We've got childhood obesity, we've got socioeconomic disparities, got mental health, and so give us a little background about why this is an important and complex issue to investigate.
Maria Gueltzow:Yes, sure. So, as you mentioned already, there's a lot of factors that play into this project and into this problem that we were trying to get kind of a deeper understanding of. So the general idea was that we were interested in looking at socioeconomic inequalities in mental health among young adolescents, specifically in a Dutch population, and we were specifically interested to gain kind of a bit more insights into possible underlying mechanisms that lead kind of socioeconomic position or that link socioeconomic position to mental health. Now there's a number of factors that likely all contribute or explain this or underlying this, and these could be factors like financial stress of the household, so children who grew up in lower socioeconomic settings, they could have worse mental health because of higher levels of financial stress in the household. It could also be related to mental health problems among the parents. It might be more likely among low socioeconomic settings, which could all lead to kind of the socioeconomic gradient that we see in mental health.
Maria Gueltzow:Now what we were interested in in this paper specifically is we wanted to get a deeper understanding to what extent obesity contributes to the socioeconomic disparities in mental health.
Maria Gueltzow:And now you might ask yourself how obesity plays a role in this, and there could essentially be two things that are happening here, or there are specifically two mechanisms that we were interested in.
Maria Gueltzow:So what could be happening is that obesity is just much more common among certain socioeconomic settings, for example, low socioeconomic settings, and since we know that there is a link between obesity and mental health, that might be a reason why we see kind of more mental health problems among lower socioeconomic settings, which then contribute to these disparities.
Maria Gueltzow:Now this is known as mediation. When you talk about methods, or when you talk about it theoretically, it's also called differential exposure or differential prevalence, for example. Now what could also be happening is it could be that obesity is actually equally common among socioeconomic settings, but it could be that the effect of obesity on mental health is just worse among certain settings, and now this could be due to different level of discrimination in the family if a child is obese. So there could be a lot of things going on there, and this mechanism is also known as moderation or interaction. If you talk about it in terms of method or in the theoretical way, it would be called differential impact or differential susceptibility. So this is kind of what we were trying to get a deeper understanding of in this paper.
Patrick Sullivan:So you have these multiple potential exposures that have their own complex relationships and an important public health question here. So, in terms of this sort of range of exposures, this important outcome, can you talk to us a little bit about the study design and the methods that you use to answer the question that you've posed?
Maria Gueltzow:Yeah, of course. So first of all, the cohort study that we used to analyze this problem was the Generation R study, which is a population-based cohort study based in Rotterdam in the Netherlands, in a very specific district, and this study is pretty interesting because in this study pregnant women were recruited and then their children are essentially followed up until adolescence, so up until age 16. And we have the data collection now up until age 13, which is where we measure our outcome. So our outcome measure I mentioned that we're interested in mental health we are looking at internalizing or emotional problems and externalizing or behavioral problems, and these were measured with child behavior checklist 6 to 18, which is kind of a questionnaire that is filled in by one of the caregivers so a parent or another caregiver and it consists of, I think, a number of items. It's about 112 items, I think. So this was our outcome measure.
Maria Gueltzow:So we did our analysis separately for internalizing and externalizing symptoms. Then in terms of our exposure. So we measured socioeconomic position through two separate measures. So firstly we looked at maternal education, which we measured at age five, and then we also looked at household income, which was also measured around age five. Now I mentioned that we're interested in obesity as well. So our kind of mediator in this sense, or like the factor that we're interested in to see how this kind of contributes to the disparities that we're looking at, was measured at age five and we measured obesity actually in an objective way. So we measured the body fat percentage with the DXA measurement, which is called, I think, the dual X-ray absorptiometry. It sounds very complicated- it sounds like a very precise way to measure this.
Maria Gueltzow:It's a very precise way to measure. Yes, so that's what's great about this. So we have a very objective way to measure body fat percentage and get an indication of obesity, yeah, so these were kind of our variables of interest, and then we had to control for a large number of confounders, because this problem is kind of highly complex. Obesity and mental health are very interrelated and so it's also kind of socioeconomic position with these two factors. So, I mentioned, we were interested in understanding these two different mechanisms, so differential impact and differential exposure. So to quantify these, we performed a four-way decomposition analysis, and a four-way decomposition analysis is essentially a method that was developed by Vanderweel, I think, in 2014. And the idea is that you decompose the total effect of something so, for example, of education on mental health into four separate effect components and if you then sum up, so the four effect components are the controlled direct effect and the mediated reference interaction, and then also the pure indirect effect, and if you sum the two of them together, you get the differential exposure and differential impact.
Maria Gueltzow:Now, one of the things that we slightly tweaked in comparison to the original approach is that we were not necessarily interested in actually quantifying the causal effect of socioeconomic position on mental health. But we were rather interested in trying to understand the disparity. So we wanted to understand the socioeconomic disparities in mental health and we wanted to see what would happen if we actually intervene on obesity. So instead of assuming an intervention on socioeconomic position, we wanted to intervene on obesity and see how this could affect the socioeconomic inequalities and mental health. Now for that we had to slightly tweak the approach and we essentially had to just perform a little bit more of a complex analysis. That's what it comes down to. So we did the four-way decomposition with the help of a marginal structural model with inverse probability of treatment, weighting and interventional analogues. So it was just a little bit more complicated than the average for decomposition.
Patrick Sullivan:Thanks for that explanation of the methods, and I think getting at this idea of what it lets you do is to isolate the contribution of obesity.
Patrick Sullivan:And so I think, in the midst of all these methods I'm going to ask you in a minute about the key findings but in the midst of all these methods, I think it's really important as epidemiologists to sort of touch back on the idea that, like why is it important to be able to isolate or to understand particularly this impact of obesity?
Patrick Sullivan:And it's exactly what you said, which is because then you can have a conversation about like how much good could we do for health if we mitigated that? And then obesity is a whole other world where epi and clinical trials and other branches of academic inquiry have given us some tools to address that. So I think the methods are a lot and yet, at the end of the day, they're so important because it lets us tease out something that asks a question about how to improve health and we're going to get to that. So what were some of the key findings? And particularly this idea of the differential exposure to obesity and how that accounted for this total disparity in emotional problems? What was the role of obesity in that.
Maria Gueltzow:Yeah. So talking about key findings maybe, to start off in general, looking at the disparities that we found in our data set, so we do find that they are more internalizing and also externalizing, so emotional and behavioral symptoms among the low educated groups, so among children that grow up in low educated and low income settings. So we do find that there is a disparity between the low and the high educated groups and the low and the high income groups, and we find a point difference of about 0.5 to 1.5 on the scale for the internalizing and externalizing symptoms. So the disparity is there. It's not very large in this setting, which is, I guess, very good news Now in terms of the mechanisms and how obesity contributes to this.
Maria Gueltzow:So what we found is that differential exposure to obesity contributes to the disparities in internalizing symptoms more so than the disparities in externalizing symptoms.
Maria Gueltzow:And specifically, what we find here is that, looking at low versus high educated families or low versus high educated maternal education, we find that if you would set the obesity levels in the low educated groups to that of the high educated groups, the socioeconomic disparity in terms of maternal education would be reduced by about 50%. And when we look at income, we find a reduction of about 14%. So differential exposure to obesity contributes. So this differential prevalence plays a role. Now, when we look at differential impact, we don't find any evidence for differential impact and this is largely driven by the wide confidence interval. So we can't really make any substantive conclusions about the differential impact and how it contributes. Now we also did some gender stratified analysis and it seemed that obesity plays a more important role in explaining the socioeconomic inequalities in mental health among girls more so than boys, and, as I mentioned, it seems to be more important for internalizing than for externalizing symptoms. So I think that's the gist of the findings.
Patrick Sullivan:Great thanks, I just wanted to follow up. You said like there's a one point disparity, I think for education, in terms of the range of this measure. It looks to me like the measures were like four to six or something right, the absolute value, so it's like one point out of four to six. That's sort of the magnitude of it. Is that right? Yeah, I think that's how I could describe it Great. So how did this relate to sort of what was known about this topic before your analysis?
Maria Gueltzow:Yeah. So what's interesting about the study is that this is the first of its kind. At least we're not really aware of a study that has tried to contribute kind of these two mechanisms simultaneously. But if we look at previous literature, overall it seems that the link between obesity and internalizing symptoms is stronger than the link between obesity and externalizing symptoms, which could be a reason why we might find larger disparities and larger contributions for obesity to internalizing symptom disparities.
Maria Gueltzow:When we look at the differential exposure which I mentioned can also be called mediation. So we found a few studies that looked at mediation essentially or kind of inferred that there is mediation present. And there is a finding from a study that kind of showed already that there is some kind of co-occurrence of obesity and mental health and specifically in low socioeconomic settings, which could be an indication for differential exposure. And there's also a few studies that reported an attenuation of the effect of socioeconomic status on mental health after controlling for obesity, which is another indicator for differential exposure. So we essentially confirmed this finding and kind of give more clearly quantified contributions of like how much does obesity actually matter Now in terms of differential impact? There is some evidence out there that obesity has kind of a stronger effect on health outcomes among low socioeconomic settings, but the literature on this is definitely lacking in terms of adolescence and also in terms of mental health. So this would definitely be something that future research could look into, in my opinion.
Patrick Sullivan:Yeah, I think that sounds like a postdoc.
Maria Gueltzow:Yeah, it could be someone's postdoc.
Patrick Sullivan:I'd like to sort of wrap up this discussion of the analysis. Clearly you know part of the strength here is, you know, having this great cohort data and some of the methods you use. What are your thoughts about the other strengths, but also the limitations, of this type of study?
Maria Gueltzow:Yeah. So I think one of the main strengths is that this is kind of the first of its kind, as I just mentioned. I think, like I said, I hope there's other people that are working on this and that there might be other papers coming out that at least use the same kind of methodological approach. So this is one of the strengths of our papers, for sure. I also mentioned that we have an objective BMI measure, which I think is quite rare and I think is definitely one of the strengths of the paper as well. So we don't have as many problems with kind of self-reporting BMI, especially among children, where it might be the parent that reports BMI, whereas there might be some sort of bias in kind of recall bias, and just the way that it's reported might not be fully accurate. Just the way that it's reported might not be fully accurate. Then with our study that we use. So the Generation R study is a really great study but not really representative of the underlying population, so we have a little bit of a selection bias. So the Generation R study is in general a little more highly educated than the underlying sample, which is something to take into account, to take into account and then because the generation of our study has been going on for so many years I think now it's about 15 years roundabout. There's, of course, a little bit of issue with attrition, especially if, specifically, children who grew up in lower socioeconomic settings are more likely to drop out. I think this could really affect our study. So this is something to keep in mind. And then, in terms of the differential impact, we had some issues with getting enough certainty to really quantify it. So there seems to be some sample size issues. So we had a sample of about 5000 individuals. We might need a few more to really quantify differential impact.
Maria Gueltzow:And then, lastly, this is a causal study, so we have to also think about kind of the underlying causal inference assumptions and I will not go into too much detail about this, but one of them being the assumption of unmeasured confounding. So we controlled for quite a large number of confounders. But since obesity, mental health and socioeconomic position are so interrelated with each other, I think there's a chance that we might have missed some or that we had some factors we couldn't control for just because of data availability. So, for example, we couldn't control for paternal BMI, which would have been an interesting factor to control for. Now, um there are a few other assumptions, but I wanted to mention one more, and that's the consistency assumption, or the assumption of well-defined interventions.
Maria Gueltzow:So now what we are inherently doing is we are assuming a hypothetical intervention on obesity, where we essentially reduce obesity, for example, among the low socioeconomic groups, right. But this of course raises the question of how do we actually achieve this reduction in obesity, and I think there's a lot of discussion to be had about this, of whether obesity in itself could be defined as a well-defined intervention or if it would come down more to, maybe, lifestyle changes. But then you always need to kind of take this into context with how could we tackle socioeconomic inequalities, and I think this is a discussion I could talk about for a very long time, but I would just say, maybe have a read at the paper if you're interested, and there's also a few studies I can recommend that would be really interesting to read. I think there's a study by Schwartz et al from maybe 2015 and a study by Kaufmann from 2019, which really go into this kind of discussion of well-defined interventions, which is very interesting, I think.
Patrick Sullivan:Great. So we're going to turn now to a section of the podcast we call Behind the Paper, and this is really meant to think a little bit more about how we work and learn and collaborate as epidemiologists and to understand what factors make it possible for us to do this work or maybe influence how we do it or how we see it or explain it. So I'll just start by asking you know what was not necessarily from a technical basis, because there are a lot of like technical pieces to your analysis but, just like as a researcher, what was challenging or frustrating about conducting this particular analysis and what parts of it were kind of enjoyable, you know, joyful for you, and what parts were that like things that you kind of had to slog through.
Maria Gueltzow:Mm, hmm, yeah, maybe starting with kind of the negative, or something that really frustrated me is that the method that we use now is quite computationally intensive, so it just takes a very long time to run the analysis, which I don't know if this would be a technical issue that you are referring to, but yeah, it's a little bit tricky. So I think one set of analysis in the end took about five hours to run, if I remember correctly, and just to get the main results ready. I think we had to run about four sets of that. So then it's very frustrating if you start running your analysis and then you realize that you forgot to include this variable or like you saved it wrong and then it just didn't get saved, even though you were running it overnight. So I think this is something I was getting really frustrated with, but it's kind of yeah- were you running it like on a mainframe or on a local machine?
Maria Gueltzow:I was in Rotterdam at the time so I was using I think it's called a server. Yeah, it's where you could at least run things in parallel, so it sped up things a little bit. But yeah, there were definitely some computational challenges, just in terms of the space that I needed to run this.
Patrick Sullivan:There were different challenges too to running. I worked at a time when we submitted big jobs, like big surveillance data jobs, to mainframe SAS and so we'd get in a queue and then we would dial up. We'd have to have a dial-up mode to see where you were in the queue and you'd watch for a few hours and then it would start running and immediately stop, which usually meant I left a semicolon out of SAS someplace. So you have to find that fix that put it back in the queue. So in some ways we do a lot of stuff on our desktops now, obviously, but like this is a complicated enough analysis that you're using that sort of server structure to do it, so that's definitely adds a piece. Conversely, what was sort of most joyful for you, most engaging or like enjoyable about it, just like personally or your scientific self?
Maria Gueltzow:I wrote this paper when I was actually on a research stay in Rotterdam. So my second author, Joost Oude Gröniger, I worked with him and I worked very closely with him on this paper, which was actually very interesting because I have never worked with him before and I think we had a very interesting collaboration, kind of engaging in all these causal discussions, so I think it was very that was very enjoyable for me and then also so I mentioned that we used a little bit more of a complex approach to deal with some of the issues that I mentioned earlier. But it took me a while to figure out how to actually apply this forward decomposition if we're interested in explaining disparities instead of looking at causal effects, disparities instead of looking at causal effects and this was something that I think it was equally challenging or frustrating and enjoyable, because it just took me a while to figure out the solution, which I didn't figure it out. It was just in a paper somewhere. I just had to find that paper. So that was definitely an interesting process to go through.
Patrick Sullivan:Great. So this is really exciting work that you've done, and I wonder, as you move forward towards your graduation, what you hope to accomplish in the long term of your public health career. What's your big "hy, like what do you want to get done in the world with these methods?
Maria Gueltzow:Yeah, that's-that's a very big question to ask. I think, ultimately something that I've been starting to tell people.
Maria Gueltzow:So maybe it will become my mission, maybe not, maybe I will change my mind, but I think I would definitely like to bring these more complex methods more into public health, because I feel like in epidemiology people are kind of more and more using kind of causal inference methods and more complex methods and machine learning and so on.
Maria Gueltzow:But I have a feeling that in public health this is still something that's really to come, even though a lot of these more complicated methods, they will actually give us the answers to the questions that we're actually asking. So, for example, with this paper here, I've been talking about hypothetical interventions and I think when we think about public health, ultimately we're trying to improve health, right, which would kind of come down to what can we recommend to improve public health or like reduce problems that we have in public health? So I think these methods are great and I think ultimately I want to try to kind of make them a little bit more accessible and have people not be so scared about these methods, which I used to be one of them so I understand.
Maria Gueltzow:So I think that's my mission.
Patrick Sullivan:Yeah, I think there's two pieces to that sort of bridge. I'm so glad you mentioned the bridge to impact, because these methods are complex and we can spend a half hour podcast talking about the methods, but in the end, the question is the one that you identified, which is you know, how does this give us information that's actionable, that lets us be more targeted or more impactful in the things that we do to improve health? So, I think, as a sort of co-career goal of like learning more about these methods but also their impact and there are some folks who in their careers, like to live in the methodologic bubble and that's a place to live but I think there's also a model for having one foot in sort of complex methods but also asking at each step, like you know, how does this lead to improving health? And I think, for my own experience, is that you get that second part by the time that you spend with people who are running programs for these kids. Or you know school systems that are trying to intervene in structural ways, and it's a lot of times it's those conversations with people who may know more than you know about the intervention side, or you know the lives of the people whose health you're trying to improve.
Patrick Sullivan:So I'm really glad that you mentioned like bringing together like these kinds of methods with that, you know, connection with where the action is going to happen, like is it at the school level, is it at the family level, is it the kid level? So that's great. So I want to end just talking about mentorship, because when we talk about being in a degree program, your mentor, or your team of mentors, is such a huge determinant of, like, how fast you get through it and how impactful it is and the quality of your life. So can you talk a little bit about one or more mentors who've been, like, really key for you in the process of doing this work and why they were great mentors? What about their mentorship was great for you?
Maria Gueltzow:Yeah, so I have a team I would call it a team of three supervisors that kind of helped me through this whole PhD process and kind of getting through it somewhat intact, yeah. So I think all of them they really helped me a lot, starting with one of my promoters I don't know if you use the word promoter, too in the US.
Patrick Sullivan:I was going to say we don't use the term promoters, but maybe we should, so tell us what a promoter is.
Maria Gueltzow:Yeah, so promoter is usually a professor that will be officially on your dissertation, but then you can also have co-promoters, or I would call it a supervisor, which is more like someone you talk to kind of on a weekly basis, whereas the professors you might talk to on a more like a monthly basis.
Maria Gueltzow:But yeah, so I want to talk about my two promoters, very strong in statistics and very critical, which is very helpful to kind of improve and kind of get more insights and see where I can make some changes, and the other one is very supportive. So it was a very good combination and in terms of my supervisor, I think he probably helped me the most, also because I had kind of more contact with him. I had like a weekly back and forth with him and he's essentially the person that really brought me into this field of causal inference and not being scared of causal inference, which I think I'm really grateful for, and I think it kind of brought me to where I am today at like kind of this intersection of methods and how does this translate to like the bigger picture.
Patrick Sullivan:it's really interesting how you talk about the team and I think, for those who may be listening, who are mentors, thinking about these different roles.
Patrick Sullivan:You know these different roles that are that the earlier career people that we work with need, some of which are around strategy and some are of which are around just like how to get through this experience, and some of those other needs might be quite technical about how to make this code work or how to interpret that, and so I think in some settings, in some situations, it ends up as being in one, those needs being in one person, but I think the ability to talk about you know what's needed because you've done a really challenging thing, and so we need some academic supports and sometimes we need some like interpersonal supports for that, and so thinking about those two roles is interesting. So I'm going to move to a last question here, which is you are about to graduate with your doctoral program, and congratulations. What advice do you have for students who are at that point writing? What have you learned that you think might be helpful to other students who are at this place in their journey?
Maria Gueltzow:Yeah, I think it depends a little bit where you write your dissertation, so I think it might differ.
Maria Gueltzow:So for me my dissertation is cumulative, so I've been kind of doing research, writing papers, and then a few months ago I started writing the dissertation and I think one kind of advice that I can give is that at least for me, this felt like this last chore that I have to do, like writing everything down and making it come together. But I actually think that it was very insightful and I would advise everyone to kind of try and have fun with it a little bit, because I think that's kind of your chance to talk a little bit more about, like I said, the bigger picture. What are the bigger issues here? What are kind of the challenges that I encountered that I never had the space to talk about? So I think it's kind of a little bit of a mindset shift, which that's something that helped me a lot to see kind of more the positive instead of being like, okay, I just have to get this over with and then this was a really enjoyable process for me actually.
Patrick Sullivan:Great. Well, congratulations on where you are in your career, congratulations on the manuscript and on your upcoming graduation. Which bring us to the end of another episode. Thank you again, Ms. Gueltzow, for joining us today. It's such a pleasure to have you on the podcast.
Maria Gueltzow:Thank you so much for inviting me again. It was a pleasure.
Patrick Sullivan:I'm your host, Patrick Sullivan. Thanks for tuning in to this episode and see you next time on EPI Talk, brought to you by Annals of Epidemiology, the official journal of the American College of Epidemiology. For a transcript of this podcast or to read the article featured on this episode and more from the journal, you can visit us online at www. annalsofepidemiology. org.
