The the big challenge I had that I had a hard time getting through was just to present the presentation of their findings as regression coefficients Yeah. Yeah. Which were very difficult. All over the place, which shows me all
the time. It was I mean, it was very difficult because as we I was Welcome everyone to free associations from the Boston University School of Public Health, the Public Health and Medical Journal Club Podcast for anyone as confused by the latest health study as I am by the fact that public health really isn't in our political discourse to the extent that we've talked about politics a lot in this podcast.
And now, for those of you who are joining us in more or less real time, it's October 2024, and there's a lot of election news, and we're not talking about public health. And I was wondering how we are thinking about that. Yeah. We certainly were talking about public health a lot during the last presidential election, so it is interesting that we are not talking about it almost at all during this cycle. Yeah. I hadn't really thought about that. Do you feel a little bit of a sense
of relief? Yes. A 100%. Yeah. I mean, you could you could interpret this to be to imply that we're we're not in a crisis situation, that people generally don't pay attention to public health when things are going well, so maybe it's a good sign. I've been mixed. I, like, I I I think there was I definitely felt there was a little bit of excitement if anxiety when public health was always being kind of bandied about by politicians, both being slammed by politicians, but also
being boosted by politicians. And now we're we're kinda back to being out of the conversation. Yeah. It's an interesting change. I don't think it was good for us to be involved in the culture wars, so I'm very happy to be out of this. Can I ask how do you define public health? Because to me, there were 2 topics that are related to public health that were being discussed. Okay. Go ahead. Yeah. Health care cost Mhmm. And climate change. I certainly health care cost could could fall under,
a deaf yeah. I I would certainly agree with that. That that makes sense, and that is being discussed, although it's it's not a central issue in this, in this election cycle. Climate change obviously affects public health, but I don't think it's being talked about in a public health context nearly as much as probably we would want
it to be. So I guess my question is, do we want something to be framed as a public health issue, or would we actually benefit from a topic being discussed and addressed without being addressed as a public health issue? Because then that creates a lot of political competition within public health. True. I think you're certainly correct. I mean, I think the the health care side, there seems to be more or less agreement on that. Like, it isn't a contentious issue.
I think climate remains contentious, but not as central not as central to kind of the core themes this time around. But I think you're right that we can define public health really broadly. And then, yes, it's not to say that there's no topics of public health consideration. I think it's just the the contentiousness that was circling these issues in prior years
seems to really have dialed down. Particularly since we're we've just gone through a major hurricane in the United States and are bracing for a second, you would think it would be getting more attention. Yeah. For sure. Anyway, I'm Jessica Liebler from the Department of Environmental Health here at BUSPH, here with Doctor. Matt Fox from the Departments of Epidemiology and Global Health also at BUSPH. Hi, Matt. Good to be back. We are joined by our returning
and amazing guest host, Doctor. Salma Abdalla from the Departments of Global Health and Epidemiology at BUSPH. Welcome, Salma. We're happy to have you back. Thank you, Jessica. Thank you for acknowledging that I'm also at the epi department, unlike Matt. Sorry. You're in the epi department? I switched the order, though. Did you notice? You go epi and global. You go global first. That is correct. That's correct. That's correct. That's correct. That's correct.
Thank you. Selma has pointed out repeatedly that I forget to mention her appointment in epidemiology. Well, we took an unexpected summer hiatus, so a fairly lengthy hiatus that our long term listeners may have have noticed, with the departure of our longtime producer, Nick Guler, from BU. And it's very good to be back, although we will certainly miss Nick and his creativity and steady hand in leading the podcast for so long.
We're debuting a new format where we are splitting the podcast into 2 separate episodes. So you can have smaller snippets of the free associations that you love so much, our dynamic and exciting conversations throughout your busy day instead of in one larger dose, and we are interested to hear what you think of our new format. As a reminder, please give us a rating on Apple Podcasts, Spotify, or wherever you get your podcasts to help public health and epi people find us.
Now on to the show. Today, we are going to look at a study considering whether light pollution is associated with Alzheimer's disease. This article led by first author Robin Voigt of Rush Medical College was entitled Outdoor Nighttime Light Exposure, although otherwise known as light pollution, in association with Alzheimer's disease and was published in the Frontiers in Neuroscience in September 2024. Matt, could you take us through what the authors did and what they found? Sure can.
So this is an article in a journal that I don't normally read, and they subscribe to the format that is very similar to Nature's format where they put the methods at the end, which I find frustrating. So I had to spend a bit of time rearranging everything to dig it out in a way that would make sense. But this study was looking at the, as you said, the association between average nighttime light intensity and the prevalence of Alzheimer's disease in the
United States from 2012 to 2018. And the reason they were interested in that was because there is prior research showing that environmental factors, things like light pollution and various other things can affect cognitive health, and conditions like, Alzheimer's disease. Alzheimer's disease, of course, is a a very big problem in the United States
and around the world. And so looking for things that are causally related to changes in, in the prevalence of of Alzheimer's worldwide would give us clues as to how we might go about trying to mitigate the impact. So this was, it seems to me an important, hypothesis to study even if the immediate understanding of why this might be related isn't totally clear, and I'll I'll come back to that when we get into what their proposed mechanisms were.
So I think of this study as, as one of those, like, take advantage of these our new found access to large datasets that can describe, the environment that people are living in and can also track people's medical conditions. So they looked at, Alzheimer's disease prevalence in the Medicare dataset. So so Medicare would cover those over the age of 65, but it also covers
other people, depending on on various things. So it's a publicly financed, health care program and as such, they have data on a very large percentage of of the over 65 population in the United States. It's got information on on chronic conditions, as well as everything is coded through ICD 9 and ICD 10 codes so you can pull out diagnosis of of Alzheimer's. You can also link it to where the person was was living at the time, and so they analyze this data based on where people were living.
They also pulled out some data from some CDC's behavioral risk factor surveillance system, particularly data on obesity, but the main exposure they were interested in was light exposure at night, and that they were able to get from, NASA's VIRS NPP satellite observations. What that stands for, I don't really know, but this is a data set that tells you about, nighttime light exposure, through satellite imagery so that you can code, in the areas where people are living how
much nighttime light exposure. Now you can immediately think about that and say, okay, just because there was nighttime light exposure doesn't necessarily mean somebody was exposed to it. So we can come back to to talking about that. They then pulled all this data together and and connected it so that they could then, again, associate where somebody was living and the amount of nighttime light exposure with their outcomes, these diagnoses of of of Alzheimer's
disease. They had various analyses that they did to try and identify this, but mostly, they they used a linear mixed effects model to account for within state and within country
correlation due to repeated measures. In other words, people who are all living in the same county, the same obviously, the same neighborhood are are theoretically exposed to the same amount of nighttime light exposure, and so knowing you've got 2 people in a data set tells you less than you would if they were, you know, a person in Hawaii and a person in California.
So they wanted to account for that, and effectively what they found was that the more nighttime light exposure that a person was exposed to, the higher the prevalence of Alzheimer's disease. Now you might immediately say, okay. What there are differences in the kinds of places where people are exposed to more nighttime light exposure, that might account for this. So they adjusted for a number of different things, in particular, age, sex, race.
In age, they sort of stratified by old above 65 and below 65, but the the trend was was pretty consistent that both in both age groups, the under 65, over 60 fives, there was increased diagnosis of of Alzheimer's or there's increased Alzheimer's prevalence, I should say, in areas with more nighttime light exposure with the effect being larger in the under 65 population, which might seem counterintuitive, but I think in a lot of ways makes sense because
there's just fewer diagnoses under the age of 65. So any absolute difference in in is just gonna look larger. So that, you know, probably makes sense. Unclear to me whether these are large 0 4. This is a measure of prevalence, so I think that's roughly a 4% difference,
though I'm not totally clear on that. Still, I mean, 4% difference would be presumably meaningful, and it's, because these are continuous measures of light exposure, maybe that's 4% per increased unit of light exposure, but I I couldn't figure out what exactly that is. So it's hard for me to tell whether these are particularly important or meaningful sized effects,
but they do find something. So overall, they find an effect and they suggest that this is likely due to things like sleep disruption, disruptions in circadian rhythms and, any some biomechanical effects related to pro inflammatory factors related to increased night, nighttime light exposure.
And we could talk about whether or not we think those are plausible mechanisms, but overall, they they do suggest that there is a at least a correlation whether or not we think this is a causal relationship, I think we can we can talk about. Great. Thank you so much for that description. Salma, do you wanna jump in and share your thoughts on this paper? Sure. But maybe first, Matt, can you claim the county analysis, county level analysis? What they do they do there?
Yeah. So I definitely avoided this. So part of one of the problems that I have with this study is I can't really determine whether this is a, what we would call, ecologic analysis versus an individual level analysis. An ecologic analysis meaning, we look at the prevalence of Alzheimer's in an area, and we look at the prevalence of some other factor, and we see if they correlate, but we're not saying any individual
has exposure to that individual factor. So in this case, are we saying that individuals exposed to more light are more likely to be, diagnosed with Alzheimer's? Or are we saying that areas with more light have more, Alzheimer's diagnoses, but we're not saying it's the happening at the individual level, in which case, there could be other factors that explain these differences. So you asked about the county versus state level. They they analyze the data by state.
They compare you know, looked at states with higher exposure versus states with lower exposure, and they looked at they went down to the county level. In both cases, though they're just, you know, it's the same idea. Presumably, everybody within a county or within a state has the same exposure level, if you're you're you're doing at that level, which strikes me as as a a bit
too coarse. Certainly, state would be too coarse, but but, within a account even with a county, it isn't clear to me that everybody is exposed to the same level of nighttime light exposure because it's gonna depend a lot on where you live and how much ambient light there is. So, you know, I I can't make perfect sense of what the whether whether those are individual or ecologic, but it just seems like they're just looking on 2 different macro aggregation levels. Yeah. I would
agree with that. I think my understanding is that they were using, and I agree it was it was not it was unfortunately not that clear in this paper. They were using individual level patient records because they were pulling individual level data about comorbid conditions, about race, about age. And so this was an was this individual, did this was this individual diagnosed with Alzheimer's disease?
And then the individual was situated within a county, which was the unit of the ambient nighttime light exposure, the, the the the unit of that exposure. And they took the counties within a given state, and they organized them into, I think, 5 groups. Yeah. Quartiles or quintiles. Quintiles. Think on the basis of, you know, kind of lowest to highest
light exposure. Yeah. And then they used that quintile, that ordinal variable in their exposure model, where I think the whole model was on the individual level, but the exposure sometimes they used an annualized metric, of nighttime light exposure annualized across the year. Sometimes they used it across the whole study period, which was a a 16 year study period. It was a long pretty long 2002 to 2,008? Something like 18. 2010. Yeah. So a long a long study a fairly
long study period. Sometimes they use just one continuous measure to aggregate that. Other times, they use this ordinal variable with the 5 levels Yeah. To reflect the exposure for individuals where they must have had some county level residential exposure data to kind of situate an individual data record within a given county. Yeah. Yeah. And it seemed to me and I guess that was part of the issue for me. It seemed to me that some of the analysis were county level, but some of it was state level
For sure. Exposure as well. I think it was an interesting study Mhmm. Mostly because and I keep wondering about this because this is not something I've read or learned before, anything about before. So to me, it was just interesting to to learn about all those things. And I thought, maybe we should just do more analysis on this topic. It wasn't like, it didn't make me think, okay, this is definitely something I need to worry about.
Like, how much light is in my area that would allow me then to think, is this increasing my, risk for Alzheimer's or not? It wasn't helpful that they only accounted for current residents. To me, that was a big red flag because then I don't know, like, what are you telling me? How long how much exposure do I need?
It just felt like at a certain point, it's just very general, a type of study that does hopefully would let get other people to do more studies, but not a study that I can actually take conclusively. And I think the other issue also was, it was nighttime exposure to light outside of the house, but we didn't have, we don't have any idea how much light are people exposed to inside their homes and how does that affect their, their risk. So
that was interesting as well. Right. I I very much agree with you. I I think this I mean, I I think the focus of this study was hypothesis generation. Right? And as Matt was saying, kind of using existing data sources to try to to try to leverage existing data sources and see if there might be an indication that there is some association here. I think they made the case that there's a plausible biological hypothesis.
The the big challenge I had that I had a hard time getting through was just to present the presentation of their findings as regression coefficients Yeah. Yeah. Which were very difficult. Right. All over the place, which drives me a little nuts. It was I mean, it was very difficult because as we are talking about where you have individual level health outcome data, aggregated exposure data, the individual level health outcome data appeared to be logistic, like, did purse or or like binary.
Did you have Alzheimer's disease or not? And then their their model was presented as regression coefficients where I couldn't interpret that into it it you know, I it was very difficult as a reader to interpret the the the nature and the extent of
their of their findings. Yep. They did some analyses to contextualize their regression coefficients where they ran models that included more established risk factors for Alzheimer's disease, like atrial fibrillation and kidney disease and depression and things along those lines, and concluded that their night exposure variable was not as strongly associated as some of those existing risk factors. But more strongly associated than others. More right. And I I just I I don't know.
This for me, personally, like, I would not write a paper that used regression coefficients as the and without interpreting it for the reader. I'm I'm a 100% with you. So so and I struggle with this a lot. So you mentioned logistic Mhmm. But then you said binary outcome. I don't think it it was not a logistic model. It was not. It was not. Right. It's a linear Right. Regression. That's fine. Right? It's a linear model of risk. We can do
that with some some caveats. But, effectively, you're just fitting a line to the probability of the outcome over time I'm sorry, not over time, over over sets of groups. If what they actually did, and and again, I I can't find enough documentation to figure this out, is they put in quintiles Yeah. And assume that was linear, then that, you know, coefficient of whatever represents the, you know, the percent change as you move from one quintile to the end Right. To another. That was I think that
I'm not sure that that's true. That was my interpretation also in some of the models. It was the difference between they didn't say if they had, like, a referent quintile and then it was in comparison to the referent or if it was moving 1 quintile up
a step. Yep. And so and and then there were continuous models though, where they were using the continuous measure aggregated either across their whole study period or by year, which would have inferred a different interpretation of their regression coefficient. And so I didn't like, was this an incidence rate ratio? Like, what was this regression coefficient? Difference. A difference. Yeah. So I actually was that was the interesting part for me.
This is the first study that I saw, and maybe I missed it, where they did not have a statistical analysis subheading. Way in the back there, I I I don't know. Had a brief section about it. Maybe it was just data available. So they spoke a lot about the data. Yeah. I didn't see the model. That would actually be very interesting because I was trying to understand what they did do with the model. They were very good at describing the type of data they were using. Statistics.
Okay. They have a 2 paragraphs. Okay. So, linear mixed model was applied to examine the relationship with all data together accounting for the within state or within county correlation due to repeated measures. State and county included x, y, and z variables, it goes on for quite a while. So I I don't know. I also have to say, I take back what I was said before about that 4% number because I realized when I look back in this model, table 1, the regression coefficient is 0.283.
That would if I would interpret that as a percentage, that's a 28% change. That is not I don't think that's what it is. Possible. But then I noticed some of the coefficients, go to 1.03. So that would imply, obviously, if it's a percentage, you can't go above 1, a 100% change. So that's not possible. So maybe these are maybe it's a 0.28 percent difference, which would be I don't know. Seems to be quite small, but I don't
know. I I think the bigger point here is we struggle to understand what this means. We can see that there is a trend. We cannot tell whether or not that trend is large or meaningful even if we believed it to be causal. Yeah. And I think that's problematic. Yeah. I think that 4%, you got it from table 2. So you did not completely Okay. Yeah. Okay. I mean, interestingly in this paper, the authors never translated the regression coefficients.
They just said there's a positive association. It's significant, and then they put the regression coefficients in parentheses. And so it was very it was very challenging, and I think the interpretation of that coefficient depends on what you assume their model structure was, and that was that was not clear. Right? Okay. Yeah. And and again, even if these are percentages, they're not percentages for a model with categories of the of the different groups,
if there are categories. So it's just a linear it's just assuming a linear increase, which may or or may not be right. Yeah. If we were to say that we agree that there's a biologically plausible relationship here Do we agree though? I'm not sure I agree. I'm not I'm not saying I disagree. I'm just saying, like, I I find it's easy to it's easy to
come up with a hypothesis. Mechanism we don't spend enough time thinking through mechanisms, and yet, I think part of the reason we don't spend enough time thinking through mechanisms is because it's very ease like, you could stress and inflammation are like every paper that I read. Okay. This has got multiple different things. So sure, could be. Does that make it plausible? I I I don't know. Is there a better design, was what I was going to
ask. Is there is there a better design that we could suggest? If there's listeners who have an interest in this topic, I think it's interesting. I think the datasets exist. What would we what would we want to see that might be better than this? So we want we want in I think we would want individual measures of the exposure Mhmm. To know whether or not how much people are actually exposed to this. One thing I have to think about is you could think
of this as an instrumental variable analysis. And for the listeners who don't know, that's a approach that gets used a lot in economics, less so in ideology, where we look for variables that are causes of the exposure we're interested in, but they have to meet other conditions in order to be able to estimate the effect of
the exposure on the outcome. So if you say, well, what I really care about is the light that nighttime light that people are actually exposed to, that is certainly gonna be influenced by the nighttime ambient light in your area. And so you'd say that's fine. That works.
But the other thing that an instrumental approach needs to needs to do are there are several things, but one of them is that exposure that that that variable, the nighttime light exposure can have no effects on Alzheimer's disease except through the direct exposure. And I would say night time light exposure is probably also a proxy for lots of other things that could affect Alzheimer's disease. Can I ask you, what would that be? I tried to think about those.
I couldn't think of something. So, it could be a proxy for general, you know, general socioeconomic status. Urbanicity. Things associated with urbanicity. Certainly urbanicity. I don't know how much that of you know, I don't know the relationship there with Alzheimer's disease. But, you know, lots of just differences in in people living in urban areas and rural areas that I would think would relate to both the exposure to the nighttime light and to, Alzheimer's disease. Also, presumably,
night shift work. Yeah. That's the one. The only one I should think about. Greater to the ambient nighttime light. So things like that. And and I think that would be independently related to Alzheimer's. So I I think there's a lot of things you could postulate there. Yeah. Does it matter the night, time light that is within your community versus actually the nighttime light you're exposed to within your home? I I
suspect it matters a lot. I mean, if you have if you live in a home, in which you have light, blocking shades and you go to bed and don't experience it, or, you know, you're indoors and in places where you're not experiencing that, maybe not. On the other hand, as you pointed out, indoor light is also an issue. So if there's, you know, less outdoor light, but you've got more indoor light, that may be equally as disruptive to sleep, and therefore, that mechanism, if
it's true, would also be activated. Yeah. There were a lot of interesting factors here. Yeah. Yeah. No. No. So thank you. Thank you. I just I think I appreciate that it made me think. I could also say topic there. I think so maybe, yeah, it's it's it's a great place for people to study more. Yeah. Mhmm. And then we can read those studies. I would agree. Yeah. I agree too. I think this would be a really interesting kind of
experience for students. I think thinking through misclassification in regards to this data source would be a neat exercise. Agreed. Agreed. Okay. Thank you so much for joining us for that interesting conversation about light pollution and Alzheimer's disease. That's the end of our program for now. If you have any feedback on this or any other episode, you can reach us at BUSPH's website. We wanna thank our new producer, the amazing and amusing Michael Sanders at BUSPH for
sound producing and editing. Thanks for joining us. We hope you enjoyed it, and we hope you join us for our next episode.