Health Studies User Conference 2026: Parallel session A

Hi everyone. I'll just get started as people join back after lunch. So, um, I'm the chair of this session. Now, we get to hear from researchers using the data that we heard about in the morning. And just to remind everyone, this is parallel session A. So, talks in this session span research using data from Elsa Understanding Society and Health Survey for England, and it's a diverse range of topics. There's another parallel session, um, which centers more around the topic of mental health. Uh, feel free to switch between sessions to go to the talk that you're interested in, and you should have a separate Zoom link for session B. Um, I'd just like to remind the speakers in the call that you have 15 minutes for presentations, followed by 5 minutes for questions. I might chip in when you have 3 minutes or 1 minute left, just to ensure that the program runs to time. The audience can ask questions using the Q&A function at the bottom of the screen, and speakers, if you would like to ask questions, you have to use the chat box, and feel free to do this during the talks. Now, have Amanda Shage, and sorry if I'm not pronouncing your surname correctly, from Queen Mary University of London. Um, hi Amanda. Amanda is a recent career change, embracing academic research. Interest is in salt reduction with the aim to reduce pop- population blood pressure and rates of cardiovascular disease. Her PhD has focused on how to increase the awareness, acceptability, and use of low-sodium salts. She has recently tried her hand at modeling and is working on a dietary health and economic model for low-sodium salt substitution in the UK. So, if >> Yeah, I'm sorry. I shared your slide. Failing right now to find my PowerPoint. I'm so sorry. The test worked at lunchtime, and Yeah. I can see my PowerPoint but I can't see it connected through the screens. Oh. >> You have to All right. >> I'm going to share. >> Okay. >> I've got it on my bar open. But I can't then find it to pull up. I am so sorry. I've even got instructions that I Googled but it's not actually helping me. Um can anybody who has already shared suggest? I Within screens or file or more launch? >> The Yeah, the share button at the bottom. Um I think you need to have that your slides open up. >> Yeah, I have. >> Okay. But you have if you open up the window there is I can see it so yeah, brilliant. You've got it. >> Sure. >> Put it in presenter mode. >> Um >> Oh no, don't share. Sorry. If you close that and just present it like you would just uh giving a presentation. So if you could close that box. >> Yeah, I will. Sorry, you were over the top of it. >> It's okay. Um >> Seamless. >> So at right at the bottom >> No, present teams, no? >> You could also present if one of the icons right at the bottom right hand corner. Yeah, yeah, exactly. Yeah. >> I'm beginning. >> Perfect. Can see it now. Great. >> Um Apologies. An absolutely >> [laughter] >> seamless start like that. >> Yes. >> I want to start properly with two numbers, 153 over 93. That was a blood pressure reading in my own family and it wasn't discovered because anyone felt ill. In fact, it was someone training for an elite sporting event. And following that, we learned that one kidney was functioning very poorly. It was only 21% viability which is the cusp of viability before removal. And that experience has stayed with me and is important as you'll see as we go through the slides. And that's because blood pressure and kidney function can basically quite silent until something is measured. So that's the human reason that this work interests me, but the research reason is very practical. I use the Health Survey England data set because I needed an England specific population blood pressure baseline for modeling salt reduction and potassium enriched lower sodium salt substitutes. And so what I'm looking at today is not new data, it's an application of data from Health Survey England. And it's not normally presented as a reusable age sex treatment input for prevention modeling. And so I'm going to take you through three stories. First, how blood pressure changes across age and sex. Second, why the treatment status matters. And third, why potassium salt modeling needs renal function guardrails and not just broad chronic kidney disease labels. So this slide explains why the work exists. The first I've said is a modeling need and my area of interest is reducing sodium consumption to lower population blood pressure. Increased use of salt substitutes using potassium enriched salt are expected to lower population blood pressure. And so I want to populate a model that has a wide range of health dietary and economic data and create a range of substitution scenarios for potassium salts and see the potential impact on population blood pressure, cardiovascular disease, and chronic kidney disease. So what we're looking at today is just the narrowest snapshot around blood pressure before it's even got into the real proper model. And and for population the model can't just put in the blood pressure um falls or rises. I need to have a proper starting distribution. And so I was looking at a systolic and diastolic um pressures by age and sex. And as it was there, I've extracted treatment status. Um, and then the second need related to my work is an implementation safety need. Potassium-enriched salts are beneficial at a population level, but potassium handling is not the same for everyone. And the subgroup of concern is people less able to excrete potassium, particularly those with more impaired renal function um and taking some medicines. And so the aim is not just to make um a pretty blood pressure chart, the aim is to turn the data into a modeling resource showing on the one hand benefit and on the other safety context. So this is the pipeline from the 2022 Health Survey England. Um, and so the first thing I extracted was blood pressure from age five upwards, which is a great resource. And the kidney markers that are relevant here um are for adults age 35 and over. There were 3,781 nurse visits with at least one paired blood pressure reading for adults and 305 for children between five and 16. And the key technical decision was to keep the systolic and diastolic readings paired, um and that sounds mundane, but it matters. Um, the paired readings let us um derive systolic um blood pressure, diastolic, pulse pressure, mean arterial pressure, and hypertension categories. Um, and so then I mapped those HSC age groups to my modeling that will become the model I've yet to populate fully um to those midpoints and estimated the weight um and used the estimated weight page sex curves. And so this is the core blood pressure resource. And the second half of what I've done is our treatment status and renal safety scenarios. Um, and I've used estimated glomerular filtration rates or EGFRs. But this is the core output. Um this shows systolic and diastolic pressures by age. The males on the left, females on the right, systolic at the top in blue. So we've got a pattern that we would expect there. Um generally rising systolic pressure with age. Um and diastolic pressure we do get this fall from mid life and again that's an expected um outcome. So the familiar um pattern um but it matters that we can see this rather than just have a blood pressure average for adults. Um we need this for modeling. Um and then we can see better where the prevention opportunities sits. And older age groups do have the highest systolic blood pressure and so an intervention that shifts systolic pressure may have very different implications across the age distribution. Okay, and this slide was my sanity check. Um I've already mentioned paired readings um in the data set and HSE are very helpful in the way they um put in the readings. You can take them as raw data or you can take them from a range of averages and I wanted to have the maximum number of viable um inputs for my model and so I just tested whether taking one, two, or three pairs made a difference. And you can see how fantastically close those curves are. So I've chosen to take it just from one you know, a minimum of one reading um in there. I'm not just using one, I'm using a minimum of one which gives me more valid results to put in. And so the next question is whose blood pressure we're looking at? Um and a population um curve is not just looking at one clinical group. It contains people with normal untreated blood pressure and that's the blue line at the bottom. Untreated hypertension is the top red line, the people who are walking around maybe not knowing they've got high blood pressure. Um and then we've got the orange line, which is treated controlled hypertension. And then we've got treated uncontrolled hypertension in the green um near the top. And this matters for modeling because a food policy and intervention is not the same as clinical hypertension management. And a small shift in blood pressure might apply widely. Um but the baseline risk treatment context and residual risk differ between those groups. Um so it's useful for modelers and for thinking about treatment gaps. Okay. But this is where the potassium salt question becomes more than a technical detail because public health guidance has to be safe, simple, and easy to communicate. And currently, World Health Organization frames caution around people with impaired kidney function or compromised potassium excretion. And NICE is broader still. It advises that potassium salts shouldn't be used by older people. Doesn't define what they are. Um people with diabetes, pregnant women, people with kidney disease of all stages, and people taking some antihypertensive drugs. Um so although this might be understandable as public health caution, for modeling, um it's really broad. And the biological concern is much more specific. Who is least able to excrete the additional potassium and therefore at risk of hyperkalemia? And so I've separated the three ideas. Scenario one um being the narrow biological risk, scenario two being the middle, and scenario three being the broad guidelines which are currently there from WHO and NICE. Um and CKD stages three to five is the renal component of those wider precautionary guidelines. Um and this distinction is central to my work. The aim is not to dismiss safety um at all. The aim is to define safety more precisely to protect the people who need um protecting with advanced renal impairment, but not automatically excluding everyone who might benefit from lower sodium and lower blood pressure. So, this chart shows why the scenario framing matters. This matches the slide we've just seen with the three scenarios. And the teal line at the bottom represents the narrow biological high risk group as taken from the HSE data. So, this is the people whose um filtration rate is below 30. Um it's quite small, as you can see, even in the older age groups. The orange line um represents a wider renal caution group um filtration rate below 45. Still modest. And the purple line represents the broader public health caution scenarios that we currently have. The CKD stages three to five. And that group becomes much larger with age. So, the message is not one definition is right or wrong, but that we're answering different questions with them. Um the narrow teal threshold is much closer to the biological risk and the purple is, as I've said, the the precautionary public health guidance. And if we draw the line too broadly, we risk excluding many older adults with the higher systolic blood pressure who would otherwise benefit from the potassium salts. So, this shows the real world um implementation um moving from the bio biological risk to the real world implementation. And the top darker gray bar shows the self-reported doctor-diagnosed chronic kidney disease on the HSE um data set. So, this is the people who are saying, "I know I have chronic kidney disease." And it's really quite small. Um but the objective renal group functions, this is when tests have been done, the blood tests have been sent away, um are much larger. And here, they're the pale gray bar on the bottom, and we've got 45% of adults over 75 objectively with chronic kidney disease, but only 4% of them recognizing that um And so this creates a policy problem. We've got um the others in the middle, so I'll just leave that for the moment. So this creates a policy problem because the advice relies only on people knowing and reporting that they have chronic kidney disease. It's going to miss people with objectively reduced kidney function. But if advice is written very broadly, it may also capture many people who are not in the narrow biological high-risk group, which is the teal color still there. Um And so the issue is not simply who should avoid potassium salts. The issue is that public health in the implementation has to decide how broad the warning should be, how people should recognize whether it applies to them, and whether clinical systems can help target advice more precisely. So this chart really matters. It shows the gap between the known disease, the measured renal function, and broad public health caution. And that gap is where implementation design has to work. Um The warnings reach known chronic kidney disease people currently, but implementation has to account for the measured kidney function. This final chart brings the two parts of my talk together. So on the left is the benefit side. Systolic blood pressure rises with age. Um and this means that older age groups have more to gain or may have more to gain from effective population salt reduction, including greater use of lower sodium potassium-rich salts. But on the right is the safety side. The renal function caution also rises with age. So the same age groups who may benefit um most are also where implementation needs the clearest guardrails. And this is the tension, but it's also the opportunity. We don't have to choose between benefit and safety because we can model it using data like this. Um and this is why the Health Survey England data has been so useful to me. Um it's let me um identify clearly the likely blood pressure benefit and identify where renal caution might sit and test what happens when guidance is drawn narrowly or broadly. And so, the message I want to leave you with from this chart is simple. Targeted modeling helps protect both population benefit and renal safety. So, I want to leave you these three things. First, the Health Survey England um can be repurposed as an applied practical modeling um input. Second, blood pressure distributions are not just averages. Um they show where the population health gains um from sodium reduction may occur, and treatment status curves show the clean clinically meaningful subgroups underneath the overall curve. And third, potassium-enriched salt implementation needs targeted renal guardrails. The current guidance is understandably broad, but modeling can separate the advanced renal function risk from the broader precautionary wording. So, in one sentence, Health Survey England um can help us estimate who may benefit, who may be missed, and who may genuinely need protection. Um thank you. I'd really love to hear your thoughts um on this um and particularly how [clears throat] or if this is worth sharing more widely. Do I write a paper with this in, or is this just too narrow, too dull? Um and secondly, how would other people handle this balance between the tension of the precise renal function thresholds and the current broader public health safety wording? Thank you. >> Thank you so much, Amanda, and thank you for sticking to time. So, um if anyone had any uh questions for Amanda, please put it in the Q&A box or the chat um including any feedback that you might have on on work um which is currently part of her PhD. I found it was a a great example of showing um the benefit of having a health survey including the self-reported diagnosis of chronic kidney disease and showing that when you collect biological measurements, there's a massive discrepancy. So, only a few people actually report kidney disease. But, the biological measurements show that a lot more people have it. And I just wondered why why you think that there might be this discrepancy in the population? >> I think it is somewhat surprising and HSC have picked us up themselves in their report. And I've read the headline before I'd kind of dived into um the data. And I think it is just where we're saying chronic kidney disease exists and taking any abnormality in the glomer- such a hard word to say the glomerular filtration rate. And so, really that sits normally around 90. And so, I think HSC has taken again quite a broad ooh, when is something not quite right? Um and so, we've got this wide discrepancy. But, it does indicate that people knew that they had that, then they should be going back. It's one of those you're looking for repeated readings, length of um readings. Um and so, it's people who clearly are not knowing there might be anything wrong currently. >> Right, yeah. Really interesting. Um so, we've got a question from Susan Walker. Um So they say it's definitely not too dull what you're doing. Um Are you indicating that an individual risk can be calculated from routine health reports? >> Yeah, maybe you can help me understand that question, Linda. An individual risk can be calculated >> Maybe like a Uh could it mean like a risk score or something from the uh from the reports, yeah? >> Yeah, it clearly So this is taken real data and I really like surveillance data and the modeling I'm doing is all based on surveillance data. Whereas and I'm new to this, I've understood that models often use different literature data sources, which obviously come from reality, but I really like the solidity. Um and so yes, if you're seeing something in your own readings that don't fit with those sort of normal expected ones, then that should be an indication that something is wrong, but clearly it's not a clinical diagnosis, but might be a flag to someone to go and get it checked by someone who will have proper insight and take the multiple readings that you need and so on. I hope that answers the question. >> Yeah. Great. So thanks. Um we've got one more question question. Um So this is from Lorena. I think a paper explaining details of how it was used for modeling would be very helpful and perhaps you can give other examples of data in HSC that can be used for modeling as well in your experience. Like what would you use next for other application? Okay. Maybe not necessarily really known. >> Yeah. Um so I'm really early on into my PhD but thought I'd start with some modeling cuz I've never done that before, so why not terrify myself? Um and I must admit I did terrify myself. The HSC data I found so hard to download, so it's been said a couple of times how easy it is and I'm like, oh. Um but I did get there. So, what I've also want to look at more is the urine markers, and there's spot urine markers but for people interested in blood pressure and sodium intake, and this is all linked. Um, they are of interest. 24-hour urine samples are the gold standard to show us what urine samples are really showing, and these are spot ones, but I do want to use them. But I can't talk more widely, and I have got So, for the model I'm populating, I think I think I've got about things, and this is only all that's going in is the blood pressure, even though I talked about the other things today. That's because once I got into it, I thought, "Oh, this is interesting, and I want to look." And so, I think, is there a paper here? Um, but yeah, I have not done anything else with HSE. >> Great. Um, so, lots to do going forward. >> [laughter] >> Um, I've got another question for blood pressure seed. What is the next steps for this research? Are you planning to use these BP curves in a full model of the health benefits and risk of potassium-enriched salt substitutes? They're quite quite similar to the What's next? >> Um, so, I've chosen So, this figure model that I'm populating is a model that's been created by um a different team altogether, and I'm going to populate it with Originally, I thought UK data, but in fact, it's going to be English data, and that's just because that's where I can find the comparative data. It's got about 20 inputs, and while I started to try and do the whole UK, I can't find the right things that line up. And I've only extracted 1 year because I'm looking at 1 year, so I'm I'm looking at 2022 for all the data that's going in. Um, and so, the next step is that yes, that is my blood pressure data that's going in. It's got standard deviation and whatever So, I've set it to the modeling age points. Um, and and it may become my chronic kidney disease data. It wasn't my chronic kidney disease data. I went through a really complex thing with the UK data service with the I can't remember quite what they're called and took data from all the kidney dialysis sites, but I actually think this might be better and so I need to go back and compare that or maybe combine it in some way of doing that. >> Great. Yeah, as I think that I think you did answer the next question which was why did you choose HES 22 rather than multiple years data? So I think >> Yeah. >> answer that if you don't have anything to add to that. Okay, great. Yeah, HES 22 has the EGFR measurements which is obviously consistent with the the topics that you're looking at and also does 21 just so you know. Okay, brilliant. Thank you so much and I wish you all the best for the rest of your PhD. >> Thank you so much. Thank you. Thanks for the questions everyone. >> Okay, so next we have Harry Tattan Birch. Dr. Harry Tattan Birch is a senior research fellow and statistician in the Department of Behavioral Science and Health. The primary focus of his research is the impact of novel nicotine products such as e-cigarettes, heated tobacco, and nicotine pouches on cigarette smoking and public health and his talk today is called cigarette dependence is greatest in smokers who start young for a pre-cross-sectional study 1998 to 2022. So >> Thanks very much, Linda. Uh yeah, so I'm Dr. Harry Tattan Birch. So I'm from the Tobacco and Alcohol Research Group at UCL. Um and I just like to say thank you to everyone who works on the Health Survey for England um for collecting and making this data um available to me because it's it's great to have a resource that's been running for so long so that we can look across uh what's happening over time and across different cohorts. Um, and yeah, um, I'm looking at, um, the link between when someone starts smoking and how dependent they become on smoking throughout the rest of their lives, um, using Health Survey for England data. Um, and just to give you some background on, oh, one second. On the topic, um, most people start smoking, um, in childhood or adolescence. They have the the first puff of a cigarette, um, before the age of 25. And of those who who try a cigarette, um, we know that cigarettes are incredibly dependence forming, meaning that many will be go go on to become dependent and then struggle to to later quit, with uh, 2/3 of people who try a cigarette going on to smoke daily for at least some point in their lives. And this is why, um, a lot of policy is focused around preventing people at young ages from, um, starting to smoke, um, because then you can hopefully prevent, um, smoking in adulthood, um, given that very few people start smoking past age of 25. And one of the ways that policy does this is through age of sale restrictions, where you ban the sale of, uh, tobacco products to anyone under a certain age. Um, so for instance, the UK used to have an age of 16, uh, a minimum age of sale, but it got increased to 18 in 2000, uh, in 2007. And some countries have gone further, with the US introducing Tobacco 21 in uh, 2019, um, aligning with their, um, minimum age of sale of alcohol. Um, but the UK has, as you you might know, uh, gone further recently with the introduction of the Tobacco and Vapes Act, uh, which has recently received Royal Assent. And what this does is introduced um a smoke-free generation policy. Um and it this will kick in in January next year. And what this does is essentially raise the age of sale by um 1 year every year. Um which means that anyone born after 2008 will never be able to be legally sold cigarettes because every every year that they age into um the age of sale will increase by 1 year away from them. Um and the idea of this is to create a a generation of people who no longer smoke. Um and a challenge a historic challenge to age of sale policies is that um the benefits might be limited if the policy simply delays smoking rather than preventing people from smoking from starting smoking at all. Um and the example people point to is Japan who have had a minimum age of sale of 20 um for over 100 years. And as a result um that you see that people start smoking much later in Japan than in other countries. Um yet in adulthood they still have very high rates of smoking among men. Um and because the health harms of smoking accumulate over like decades and decades of use um people would say, "Well, just delaying the start by a couple of years might not necessarily be uh lead to that big population health benefits." Um but a pushback against that would be that maybe later uptake in itself would be better um through other mechanisms. Um so for instance, maybe delaying when someone starts smoking uh will make them less dependent throughout their life because uh smoking at a young age might be a sensitive period uh where people are more likely to um become dependent. And that is what we wanted to look at in this study. So, the research question was, do smokers who start at a younger age have higher markers of cigarette dependence into adulthood uh than those who start later? And as I said, we used the data from the Health Survey for England from 1998 to 2022, uh which is a yearly cross-sectional survey, and all the participants are interviewed and then invited to a nurse visit a few days later, where they give biological measurements. And what we're interested in here is the saliva sample that they give. Um because that allows us to capture one of the core measures um of dependence. The participants in the study were adults aged 25 and up who currently sm- who reported currently smoking cigarettes. Um so, this is important. We're looking retrospectively at their age of starting among people a sample of people who are currently smoking. We're not looking prospectively from childhood into adulthood. Um so, a total of 26,000 smokers were interviewed and 14,000 provided a saliva sample. The core exposure that we looked at was simply the question, um "How old were you when you started to smoke cigarettes?" And respondents could give any age in years. Um and we analyzed um because we have such a big sample here, we analyzed it both as a continuous variable looking at like the linear decline with every additional year delay in starting. And we also looked at each individual year of age separately as a categorical variable um to see how well that linear fit mapped onto the underlying data. Um and we had three core outcomes um which giving us giving us an indication of cigarette dependence. The first is uh saliva cotinine. So, that's the concentration of cotinine in someone's saliva. And cotinine is the main metabolite of nicotine. So, it's the thing that nicotine breaks down to in the body after it's been consumed. Um and what this gives us is a really precise and widely validated marker of how much nicotine someone's taken in over the past few days. Um which is really useful now because some measures of dependence have become a um don't have the same meaning over time because what we see is at a population level, people are smoking fewer cigarettes as cigarettes have become more expensive. Yet, the amount of nicotine that they um are getting from each cigarette is actually going up. So, if you look at the average amount of nicotine that people are getting, um it's relatively stable over time. Because people are essentially um smoking their cigarettes more rigorously to get as much nicotine that they as they can out of it. Um Our second measure was self-reported cigarettes per day. Um and our third measure was self-reported um smoking soon after waking. So, this is the question, how soon after waking do you usually smoke your first cigarette? Uh and we dichotomized it at 15 minutes within 15 minutes or after 15 minutes. And this is a really important marker because it tells you after a period of abstinence when someone's been asleep and they've not been able to have a cigarette, how long do they do they wait before the cravings to have a cigarette overwhelm them? Um and if someone's having it basically as soon as they wake up, that's a really good indication that they're um heavily dependent on cigarettes. And it's very predictive of whether someone will struggle to try and quit. So, we analyzed uh the data using log-linear for the continuous measures and log-binomial regression for the binary before and after covariate adjustment. And the core estimate was the relative percentage change in each outcome for every 3-year delay in starting smoking. Um and I'll just go straight into the results. Um on the horizontal axis here, you've got the age of starting smoking. And on the vertical axis, you've got the three different measures of cigarette dependence. And what we can see is there's a downward-sloping curve, whereby the later someone starts, the lower the level of dependence on cigarettes. So, it's a 12% decline for every 3-year delay uh for saliva cotinine and cigarettes per day and 21% for every 3-year delay for smoking soon after waking. But an obvious issue with these results is these are unadjusted for any covariate. So, someone might say, is this really about age of starting or could it be caused due to other differences between people, other confounding factors? Like, for instance, we know people who come from more disadvantaged backgrounds are more likely to start smoking at a young age and they're also more likely to be dependent on cigarettes. So, that could induce an association um unless we account for it. But when we do account for some of the measured covariates, so current age, sex, housing tenure, social grade, education, and survey year, we still see um strong associations with just slight attenuation um suggesting that it's robust at least to these measured covariates. Um just to summarize these results, um we see that in England, people who start smoking earlier in life become more dependent than those who start later. And the effect sizes are really, really large. So, just to put it in perspective, someone who starts smoking at age 10 has double the nicotine intake and double the cigarettes per day in adulthood that compared to if they had started smoking at age 24. And the associations found across all birth cohorts, uh cuz we looked across all different birth cohorts, is also found across all different current ages at time of survey. Um across all survey years. Um it's a really it was a robust association. Okay, so what explains that association between age of starting and dependence? And I've picked out three things here. The first two are causal explanations and the third is maybe it's not causal at all, maybe it's due to confounding. The first is the idea of a sensitive period that I mentioned earlier. Uh which is the idea that people who start smoking earlier in the process of neurological development may be especially susceptible to dependence. And this is supported by a number of findings including experimental animal models where rodents who are exposed to nicotine early in adolescence have lasting changes to their cholinergic system in regions of the brain related to dependence. Um and there's also previous research showing that um people who start smoking younger are more like more rapidly transition to daily smoking, also an indicator of dependence. The second explanation is simply that people who start smoking younger will have at any given age in adulthood been smoking for longer um leading to more pronounced prolonged reinforcement of smoking behavior. And then the third explanation is maybe these um associations are not causal at all. So, there's a a big risk here in this study of collider selection bias because we're selecting only people who are currently smoking cigarettes. Whereas, there's a whole host of people who would have started smoking say age 10 and then quit um before the time point in which um the survey was measured. Um so, that leads to a collider selection bias. But, actually that bias would lead us to underestimate, not overestimate, the association between age of starting and dependence. Um I did some simulations to sort of dig into that. Um and another uh risk on the other side is the risk of unmeasured and residual confounding. So, we we adjusted for um quite a few measures, but they were only met measured at the survey time point. They weren't measured at uh during childhood and adolescence. Like, prior to when someone starts smoking, which would be the ideal for causal inference. Um and those might be the most important factors. Um so, it'd be interesting to try and replicate these findings um using longitudinal studies. And I'm looking into using the British birth cohorts now um to dig into this a bit more. So, the conclusions for policy um is that policies that aim to discourage you smoking, including tobacco 21 and smoke-free generation policies, may have two potential benefits. The first is if they reduce initiation, so the absolute number of people who will start smoking across their life. And the second is even among the people who do start, if policies delay the time in which they start, then if these results are causal at least, this would suggest that they'll be less dependent throughout the rest of their lives, which will make it easier for them to then quit smoking later. But of course, policies need to be evaluated directly, especially something like the smoke-free generation policy that's not been introduced anywhere yet. It's never been evaluated, and there are risks of some unintended consequences. Um so our team will be looking into trying to evaluate what what actually happens after this policy comes in. Um but you can find out more here, and I'd like to thank all my co-authors on the study. And if you're interested in doing any sort of analyses using the smoking data in the Health Survey for England or from the British birth cohorts, then let me know. I'd also like to flag there's a really useful paper by Liam Wright, um who uh provides harmonization of the different but the smoking measures across the different birth cohorts. Uh and thank you. I'd like Let me know if you've got any questions. >> Thank you, Harry. That was brilliant. Um really great to see analysis on the whole range of the Health Survey for England that's been collected, and those biological and self-report measures of um smoking dependence. Uh so we've got quite a lot of questions that have come through. So uh I'll just take them in order. Um so the first is from Lorena. Did the data capture any participant with periods of smoking, quitting, smoking again? Uh it could also be interesting, if available, to analyze differences. So if there were periods without smoking, stopping and starting later. There was that information about quitting and starting and quit attempts, I guess. >> Yeah, yeah, definitely, and that's um so we don't have those measures in the current study because it's that retrospective analysis of people who are currently smoking and we don't have the like a detailed measures of um all the transitions throughout someone's life. So, periods when they weren't smoking, periods when they were. Um which is why I think it would be really interesting to look at it um prospectively and not not just looking at dependence, but looking at um the ultimate consequence of dependence, which is making it more difficult for someone to stop smoking. Um so, looking at um cohorts of people um who start smoking at a younger or older age and then seeing whether those who start smoking younger um are more like uh find it more difficult to quit or and their their periods of abstinence are shorter than people who um start later. >> Great. Thank you. Um another question from Evelyn. Thanks for a compelling talk. The findings are really strong signal. Following the mention of residual confounding, is there any consideration in calculating an E value to quantify exactly how strong an unmeasured unadjusted variable would need to be to completely explain away your findings? >> Um so, I have some mixed views on E values in general, but um if one were to calculate a sort of E value effect um for this, um it would be massive. It would have to be a massive um confounder to be able to explain it. But, having said that, we've also not adjusted for a ton of things that could have strong associations with age of starting and could have strong associations with dependence. Um but yeah, um we've we've done a bunch of different uh sort of um looked into um how all these different um biases might affect the findings. But yeah, I do think it's um this These results combined with some of the previous studies. I didn't actually get to go into it, but I'll just mention it here. There's also a a neat twin study where they've got discordant twins, one who starts started smoking younger and one that started smoking later. And they found that the twin who started smoking earlier would became more dependent on smoking um throughout their lives. Which is sort of um that accounts for a lot of the shared environment and the genetic factors. Um But yeah. I think that that is the big risk of this analysis is the residual and unmeasured confounding. >> Right. Um another question from plus side. One of the most interesting aspects of your finding is the potential role of a developmental sensitive period. Which of the explanations newer developmental vulnerability during adolescence or longer duration of smoking exposure you discuss do you think might have the greatest impact? >> I think um I think the sensitive period is probably the thing that's um driving most of the effect. But I'm also I guess they're not mutually exclusive. I imagine some of the effect is due to amount of time that someone's been smoking and some of it is due to um adolescence being a sensitive period. That would be my view. >> Okay, great. Thank you. So I think we've got time for the two questions uh that are left. So this is from Martha. Would it be possible to use the same data set to look at people who successfully quit smoking? For example, is there correlation between how did how old were they when they started smoking versus how long until they quit smoking? Or it's a cohort cohort cohort of current smokers and former smokers different in terms of age at which they started. >> Yeah, I know that's a great point And um that that is the the sort of thing that I think a longitudinal data set would get best at. But we can also you can also look at that using the cross-sectional / retrospective reporting um because age of starting is captured in um ex-smokers. Um there are just a few issues with that in terms of um people sometimes don't report being an ex-smoker if they only had a short period of smoking. Um So then you sometimes miss some of the people who might have been the least dependent and just smoked for a few years and then stopped and then they don't even ever report having smoked. Which is an issue that we've seen before. Whereas you can sort of capture that if you've got a longitudinal study where you're you're like, "Wait, well, you might say now that you're not you've never smoked, but we can see uh 20 years ago you reported that you were smoking." >> Uh yeah. Yeah. So So the final final question from Laura, "I don't doubt the sensitive period interpretation, but that is there interpretation that both age of uptake and intensity of use are both measures of something like something else like attraction to smoking?" >> Yeah, I think that would be um we don't we haven't really captured sort of um the confounders that we've measured haven't got things like someone's general sort of risk-seeking behaviors or >> Mhm. >> Yeah, which I think would be one of the core confounders here. Where some people um might be more likely to try things and that might also lead them to become more dependent. I'm not sure. Yeah, again, I think that that's why going forward having some sort of longitudinal data on this would be really useful. >> Okay, brilliant. Thank you so much for answering all the questions and maybe see you back with the data on the British birth cohorts at a different time on this. Um, thank you. So, next. >> Thanks, Linda. >> Um, yeah, next we have Baowen. Hi, Baowen. So, Baowen Ji is a lecturer in quantitative methods and social epidemiology in the Department of Epidemiology and Public Health UCL. She leads the work theme of the Equalize ESRC Centre for Life Course Health Equity. Her research has concentrated on paid and unpaid work and social determinants of health within a life course epidemiological framework. So, feel free to share your slides and start. >> All right. I'm in presenting mode. Am I in presenting mode? >> Yeah, you are. Yeah. >> Uh, on my side it's not so >> to present uh your your slides, yeah. >> So, people are okay seeing the slides now, right? >> I can see your note view rather than the presenter's view. >> Sorry about that. How to do the hide presenter view? And then How about now? Can you see the slides okay? >> I can now, yeah. Perfect. >> Okay, without notes. >> Perfect. >> Okay, thank you so much. Thank you, Linda, for the intro. Uh, so I'm going to talk about flexible working, which is quite different from the topics uh has covered in this session. So, we look at the influence of flexible working employment on well-being outcomes among people uh with long-standing illness or chronic disease. So, as um Linda mentioned before, I'm I'm based at this uh ESRC center called Equalize. So, we are doing life course research uh uh, really looking at the solutions to reduce health equity. So, I also want to use this opinion, um, this opportunity to say a little bit more about Equalise. So, uh, in Equalise, we have this, uh, ecosystem. So, Equalise, uh, we are a big group of researchers from six different universities. So, the hub is at UCL. Uh, we also have University of Strathclyde, University of Glasgow, City, St. George's, University of Essex, and the University of Toulouse. So, we we are using mixed methods doing this cross-disciplinary research. We We also have seven different government partners at the national, devolved, and local level. And we also have 13 third sectors. Um, so, they're either advocacy or community partners. So, in Equalise, all our research are co-produced with our stakeholders. So, the aim of Equalise is to achieve actionable solutions to reduce the UK's widening health inequalities. And within Equalise, we have four different research themes. So, the first theme is about learning environment. And second theme is about work. So, in the work theme, which I'm currently leading, uh, we're looking at employment and also job quality. And this flexible working, uh, research is part of this work theme research. The other theme is the care theme. So, we look at both, uh, child care and and paid care. And the last theme is the place theme. So, this place theme is a really kind of, uh, a cross-cutting theme. So, we look at how learning, work, and care differs in different places and how the relationship on health really differ by different places. So, this is about Equalise. And then I'm going to talk about this piece of work about flexible working. So, flexible working is increasingly common across industrialized countries, and flexible working is about when, where, and for how long individuals undertake work-related tasks. There are three main types of flexible working arrangements. The first type is reduced work hours, for example, um part-time job. The other type is this uh flexible time schedule, for example, some people they choose to work during term time only. And the last type of flexible working arrangement is working from home, like many of us today are really working from home. So, in terms of influence of flexible working on people's health and well-being, uh there's some evidence shows that for both men and women who are using this kind of reduced hours flexible working, they report lower levels of chronic stress. Well, for the results of uh flexi-time schedule or work from home, the results on health and well-being are quite mixed. Some find positive effects, some find negative effects, and some find no effect. And flexible working may be particularly important for individuals with long-standing illnesses or chronic disease, given its potential to really support the employment retention and well-being of people in this group. However, there's no existing research has specifically examined flexible working's effect in this group of people. So, you may wondering why we are interested in people with long-standing illnesses. So, this figure is uh from ONS, so it shows the number of people aged between 16 to 64, so they're working age, but they are currently not working and not looking for jobs due to long-term uh sick days. So, here you can see there's a clear difference. Um so, the whole graph shows the number of um not working and not for looking for jobs due to long-term illness uh from 2000 to 2025, but we really see a kind of a sharp increase since the pandemic. So, uh for example, from 2020, we really see a increasing number of people that are not working and and not looking for jobs due to this long-term sickness. So, then we think about what are the potential solutions this. So, we don't want people like who have this long-term sickness, and then we really force them to work and then further sacrifice their their health condition. So, that's why we think maybe like flexible working is the potential solution, and that's why we really want to look at whether flexible working and can help people with long-standing illness to keep healthy and also stay in the labor force. And that's why we want to do this piece of research. So, the aim of this research is to provide up-to-date evidence on whether flexible working can support employees with long-standing illness or chronic disease. Uh we are using quantitative data, so wave two to wave 15 Understanding Society data. And just to say a little bit more about Understanding Society. So, Understanding Society is also known as this uh UK pay household longitudinal study, so UKHLS. It's a longitudinal survey of about 40,000 households uh at baseline in the UK, so it's very large, and it's also nationally representative. Um the first wave started in 2009, and then every year the same people have been like followed up. And the most recent wave is wave 15, so um in year 2023. So, it's a very uh useful data set to answer these kind of research question. So, in terms of the the sample size, we have about 12,000 employees um with long-standing illnesses or diagnosed chronic disease at baseline. So, this is our um study sample because to use flexible working, you need to be employed. So, that's why we focus on people who are working at baseline and they also have long-standing illnesses or chronic disease at baseline. And then among this group of people, we estimate whether the uptake of flexible working influence their work exit and their well-being or not. And the method we use is the fixed effect method. So, the fixed effect is comparing people to themselves over time. And just to say a little bit more about uh fixed effects. So, this kind of fixed effect models can estimate how the changes in the uptake of flexible working influence their employment and well-being by really differentiating out these time-invariant characteristics. For example, like um education, personality, or early life characteristics. These kind of like time-fixed characteristics can be uh automatically accounted by the fixed effect modeling. So, in the fixed effect uh model, we can we only need to adjust the full-time equivalent rates. So, in this uh fixed effect modeling, our exposure is a binary one. So, yes or no take of flexible working. And then the outcomes uh first outcome is employment. So, again binary, working or not working. And then we also look at several uh well-being outcomes. So, uh GHQ measured uh psychological distress. And then SF-12, this mental health component measuring this um mental health functioning, and also the SF12 physical functioning component measuring the level of physical health. We also look at two uh satisfaction uh outcomes. One is about this overall life satisfaction. The other is about people's satisfaction with their leisure time. And in this fixed effect modeling, we we use a one-year lag to make sure that the the exposure happened 1 year before the outcome. Um and the model we adjust the full-time wearing covariates. Uh at the moment, I only adjust the full-time wearing age, full-time wearing number of children, and full-time wearing marital status. So, um to begin with some descriptive results, so here it shows how people use different types of flexible working arrangements by the demographic groups. Um so, we look at three different types of uh flexible working first. Uh so, those showing green bar are the reduced hours, and then yellow bar are the flexitime schedule, and then blue bar are the uh working from home, and the pink bar are using any of these three flexible working arrangement. So, first to look at gender, and then we we find a very clear gender difference. So, women, they are much more likely to use reduced hours uh arrangement compared to men. And then in terms of flexitime or working from home, we didn't see much gender difference there. And we'll also look at partnership status. So, we find that people who are married or separated or widowed, they are slightly more likely to use reduced hours flexible working arrangement, as well as other um two types of arrangement compared to those who are single or cohabiting. I think this is maybe kind of reflecting an age effect because for this descriptive one which is a cross tab between marital status and the percentage of using flexible working. So in the fixed effect we will adjust up for this age effect. And in terms of the number of children we find that those who have a child they are more likely to use all three types of flexible working arrangement compared to those without any child. And then we also look at whether there's any differences by whether there's illness is limiting or not and we find that those who report their illness are limiting they are slightly more likely to use all three types of flexible working arrangement. So this is just a pure descriptive to give you a sense of how different groups of people using flexible working arrangement. And then we move on to the fixed effect results. So first we look at the influence of work exit. So here it shows the odds ratio of work exit. Odds ratio of one means there's no effect and actually we are looking at the the the gaps between the bar and then this reference line. For example when we look at both men and women together and odds ratio for reduced hours is about .5 so that suggests people who are in this group and then they are using reduced hours they are 50% less likely to exit from work compared to those who are not using reduced hours arrangement. And then for flexi time the the effect size is similar to reduced hours. And then for working from home, uh and the odds ratio is about 0.8, so people using working from home, they are 20% less likely to exit from work. And then we look at men and women separately. So, the effect size are slightly stronger for women than men because that the gap between the the odds ratio and um the reference line is bigger uh for women than for men than for men. So, but the difference is not that big. So, we see some difference for working from home. And uh also for reduced hours compared to men, but not too much for flexitime. And then we look at the influence on health and well-being. So, to start with uh men only, uh we didn't see much effect except for the influence on GHQ. So, a higher level of GHQ means more psychological distress. And this negative coefficient suggests that men who are using reduced hours, they have lower level psychological distress. So, using uh reduced hours is good for men's uh mental health. But we didn't see much effect for physical health or for satisfaction. And then we didn't see any effect for uh flexitime or working from home for men. And then when looking at women, we find a few more significant results. So, uh first for reduced hours, we find that women who are using reduced hours, they have better mental health functioning. So, a positive coefficient for this SF-12 means a better mental health functioning. So, this result kind of like um consistent with what we find uh for men. And then interestingly, looking at flexitime, we find a kind of like a negative effect on women's mental health. So, women who are using flexitime schedule, they have a higher level psychological distress and a lower level of mental health functioning. So, using flexitime is bad for women's mental health. And this is what we find. And then also, we find women who are using flexitime, they also have lower satisfaction with their leisure time. So, this is quite interesting and then we think maybe because women who are using flexitime schedule, they try to really combine everything together because they're still working for the same amount of hours, like working full-time, but they try to squeeze everything into a very short period of time. So, that's why maybe they have poor mental health and the lower satisfaction with their leisure time. And then for working from home, we didn't see much influence for women's health and well-being. And then I did a uh some sensitivity analysis. So, the standard fixed effect models uh presume that the effects of variables are symmetric. So, the effect of increasing a variable is the same as the effect of decreasing that variable, but in the opposite direction. So, basically, standard fixed effect models uh assume that an a change from zero to one is the same as the change from one to zero, but in the opposite direction. So, in this sensitivity analysis, I use these uh uh symmetric uh fixed effect models. Um so, basically, um I allow the two directions have different effects. So, I kind of decomposite the fixed effect into a positive change, so zero to one, and also a negative change, one to zero. And then find out from from most of the analysis, they are just the symmetric, so there's no difference between one to zero or zero to one change. There's only one exception. Uh and I find that so the take up of reduced hours, so zero to one change, has a slightly stronger association with lowering the risk of exit from work than uh stop using of reduced hours, so one to zero change. But still over picture. >> Uh just uh what what exactly there? >> Okay. So actually I can't really see you, so I just don't know like the timing. Um >> That would be >> Yeah. I I just uh quickly summarize the key findings. So um we find that among employees with uh long-standing illness or diagnosed chronic disease, uptake of any flexible working arrangement is associated with a lower likelihood of work exit, particularly for reduced hours and the flexi hours arrangement. And the effect size on work exit is slightly stronger for women. Reduced hours uh arrangement is associated with better mental health outcomes for both men and women. Um but flexi time arrangement is associated with worse mental health and satisfaction outcomes for women. We also test the pre-pandemic waves and we find that working from home is associated with worse mental health. So our next step is really to look at the characteristics of working from home in in pre-pandemic waves to really understand the reason why. And we also try to do some subgroup analysis, for example, look at whether union is limiting or not or look at different domains of health conditions. And finally just to acknowledge my co-authors, Beijing, Constance, and and and thank you so much.