Using Simulation Modelling to Inform Public Health Policies

Points clés à retenir

Exploring Public Health Policy Scenarios

Simulation models provide a structured way to explore “what if” questions, including policies that have not yet been implemented and outcomes that may take many years to emerge. They complement observational studies and trials, with results dependent on the evidence and assumptions built into the model.

Understanding Differences Across Population Groups

Microsimulation represents individuals with different characteristics and risks, allowing researchers to examine how policy effects may vary across population groups. Looking beyond average outcomes is essential for understanding whether interventions could reduce or widen health inequalities.

Representing Complex Behaviours and Environments

Agent-based modelling can represent interactions between individuals, their social networks, and their environments. The NOURISH project illustrated how this approach is being developed to explore food environment policies, dietary behaviour, obesity, and food insecurity in Greater Manchester.

Building Credible and Relevant Models

Model development begins with a clear policy question and an explicit understanding of relevant mechanisms. Stakeholder input, data integration, calibration to observed patterns, and testing alternative assumptions help strengthen models and make their findings useful for decision-making.

Opportunities for Interdisciplinary Contribution

Simulation research requires expertise beyond coding, including behavioural science, epidemiology, qualitative research, nutrition, and policy. Early career professionals can build models, contribute evidence and subject expertise, or develop their skills in critically interpreting modelling studies.

Orientations futures et prochaines étapes

• Start with a focused policy or research question and consider whether simulation modelling could help address it.
• Seek interdisciplinary collaborations and contribute to conceptual mapping, evidence gathering, data analysis, or model development.
• Examine model assumptions, evidence gaps, and the duration of behaviour change, using alternative scenarios to explore uncertainty and identify priorities for further research.
• Progress the NOURISH project by completing the conceptual model, developing and running the agent-based model, and linking it to health and economic impact modelling. No modelling results were available at the time of the webinar.
• Support wider use and adaptation through the planned release of NOURISH code alongside publications, with potential extensions to other policies, geographical areas, and the health and environmental impacts of dietary choices.

Les résumés sont générés par l'IA à partir des transcriptions des réunions.

Les transcriptions sont générées automatiquement. Si vous remarquez une erreur, veuillez envoyer un e-mail à [email protected]

Speaker 1:
think we can start. So good afternoon everyone and welcome to this month’s EASO Early Career Network Learning Hub webinar. Thank you so much for joining us.
Today’s session is entitled Using Simulation Modeling to Inform Public Health Policies and over the next hour we’ll be exploring how simulation modeling can be used to evaluate public health policies, examine inequalities in health outcomes and understand the implications for obesity and population health research. Before we go further, I would like to note that the Novo Nordics Foundation has provided the support to EASO for EECN development activities, including this webinar series, and that they do not have any influence over the content. My name is Beatriz and I’m here with my fellow member Eugenia, who will be co-hosting this session with me. We are both from the Early Career Network board from EASO.
So just some quick notes before we start. Today’s webinar, it’s going to be recorded and the recording alongside any relevant links will be shared during the sessions, will be made available to you afterwards. If you are joining ECN Learning Hub event for the first time, they are monthly online events run by the EASO Early Career Network to promote knowledge sharing and skill development among students and early career professionals interested in obesity.
And for everyone, ECN, it’s completely free to join. And so are this monthly webinar. So feel free if you know someone who are also interested in obesity research, clinical care or policy, please invite them to join to our network too.
Just a few housekeeping points before we begin. This is very informal, friendly session, so please feel free to engage. Do not have to wait until the end. You can pop up your questions in the Q&A parts of the session and we will be keeping an eye on those.
and in the end, me and Eugenia will give them to the speaker so they can be also answered. And after the webinar, a short feedback pop-up will show up. Please do take a minute to fill it in.
It’s very important for us to shape our future webinars, and we would love to hear your thoughts. And before we start, I’m going to hand it up to Eugenia for a few minutes in some updates. Eugenia, for you.

Speaker 2:
Thank you, Bia. So just a few informations about the opportunities that are available for ECN members. We will also share a couple of links with further details in the chat.
So first of all, applications are now open for the ECN Masterclass 2026. And this is a residential course that focuses on hot topics in obesity with a special developed program just for ECN members. This year is going to be from the 7th to the 9th of December in Zagreb, Croatia.
So pretty warm place, hopefully. If you want a break from the cold, you’re free to join us. The application deadline is the 16th of October.
Also, abstract submissions will soon open for our European Congress on Obesity, which this year will take place from the 18th to 21st of May, 2027 in Munich, Germany. There will be also the ECM Best Thesis Award, which will soon open for submissions. So please do consider applying if you’re nearing completion or have recently developed and completed your thesis.
The top three selected candidates will present their thesis work at a dedicated session at ECO and the key winner will receive 500 euros cash prize. So, you know, nice to consider if you are able to apply. And also keep an eye on the IESA website for updates on the application’s openings.
Also, just so you know, we run a regular ECM spotlight feature, which is shared across the IESA website and social media to highlight ECM members and their activity and research and to reach the wider community. So if you would like for your work to be featured, then please get in touch and we’ll be happy to help you with this. And finally, we have an ECM WhatsApp group for quick updates where we also share resources and networking opportunities across Europe.
We would be happy for you to join, it’s no spam. And that’s pretty much it. I should share the links in the chat, but now I’m handing it back to Bia for introducing our guest for today’s webinar.

Speaker 1:
Thank you so much, Eugenia. So now it is my pleasure to introduce today’s speaker, Dr. Charlotte Buckley, who is based in the Department of Physiology, Institute of Population Health at the University of Liverpool. She is chief investigator of the NIHR-funded Nourish Project, Nutrition, Obesity and Unhealthy Food Relationships in Food-Insecure Populations, which uses systems modulating approaches, including micro simulation and agent based modeling to explore the relationships between food insecurity, food environments and health outcomes, including obesity. So thank you so much, Charlotte, for joining us. It is up to you now.

Speaker 3:
Great. Thank you. I’ll just share my slides.
OK. Everyone see that, OK? Okay.

Speaker 1:
Yeah, it’s great.

Speaker 3:
So thank you for the introduction. So yeah, today I’m going to be talking about some of the work I’ve been doing over the past few years, looking at using simulation modelling to inform some public health policy related research. So I am currently a senior lecturer in psychology at the University of Liverpool.
But I actually have quite a mixed kind of interdisciplinary background. I spent some time working in an engineering department before I moved here a couple of years ago. So that’s kind of some background to how I’ve ended up doing simulation modelling in a psychology department.
So the learning outcomes that have already been circulated, and kind of essentially what I’m not going to be able to tell you absolutely everything about simulation modelling in the next kind of 30 to 45 minutes. So what I would really like to be able to get across is just to kind of help people understand how simulation modelling might be used to look at potential impacts of policies and inequalities in health outcomes and things like that. I’d also like to give you an idea of some key steps and challenges that are involved in developing simulation models.
And then also right at the end, I’d like to conclude by kind of thinking about where I see opportunities for early career network members to maybe think about simulation modelling in their own research or maybe contribute to simulation modelling. So with those learning outcomes in mind, I’m going to kind of move through four different sections of the webinar today. So I’m going to kind of start from first principles.
I’m assuming some of you may already be familiar with kind of some simulation modeling methods and things like that. But I’m assuming no prior knowledge. So hopefully kind of you’ll be able to understand and follow along regardless of where you’re starting from.
So why would we even want to simulate in the first place? Then I’m going to talk about one simulation method called micro simulation. Then I’m going to build up to a more kind of complex simulation method called agent-based modeling, which is what the Nourish project is about.
So I’ll be using some examples from that. And then finally concluding with some opportunities that I see for ECRs. So to start off with, when thinking about why we might want to do modeling, this is a problem that is used quite a lot when talking about simulation modeling.
And it’s one that I really like. So I just invite you to think about this problem very quickly. So which deal do you think gives you more pizza?
Is it deal one with two 12-inch pizzas or deal two with one 18-inch pizza? Now, these two deals are exactly the same price. This is a deliberately simple example that I like to use.
And most people kind of have an immediate intuition about this problem. And a lot of people will sometimes kind of assume that two pizzas is better than one. But actually, when we kind of formalize that problem, so when, you know, we know how to work out the area of a circle.
And when we formalize this, the 18 inch pizza actually gives you more pizza for the same price. So it’s the better deal. And this is a trivial example.
It’s one from the paper that I’ve cited at the bottom that’s kind of about psychological science. but I think can be sort of more useful quite broadly as well in sort of public health policy research as well. But what it illustrates is that sometimes our intuitive understanding isn’t always correct.
And formally modelling things can force us to specify exactly how we think things work. So obviously, public health questions are more complicated than which deal gives you more pizza. So some of the things we might be interested in public health and public health policy research is kind of these what if questions.
So what if we change food prices? What might happen? What if we change local food environments?
What if we restricted unhealthy food advertising? We’ll be coming back to this one in a little while. What if we increased access to treatment or medications and things?
And then also we might be interested in things like understanding who benefits by how much over what timescale. So could an intervention improve health for everyone? Might an intervention widen inequalities?
So these are some of the complex questions that we’re kind of looking at in public health research. Now, why can’t we just use observational research to give us the answer? And RCTs and observational evidence are obviously hugely valuable.
I’m not suggesting that we get rid of them and don’t use them. But there are some questions that they can’t easily answer. So, for example, policies that don’t exist yet or might be really difficult to kind of randomise people to different conditions.
There are a lot of things where outcomes might unfold over many, many years. We might see different effects between population groups. And there are things that kind of, so RCTs kind of have a place in simulation research in that they can inform what we can put in a simulation.
And simulation can then combine that existing evidence from observational studies to explore kind of possible outcomes, given the conditions that have been observed, to look at what might happen to unobserved things. So what actually is a simulation model? So at its kind of simplest, a simulation model is a simplified computerized representation of some part of the real world.
So we have the real world kind of here on the left. You know, this is like a real picture of the planet. And obviously, we can’t put that into a computer.
That would be really, really complicated. So what we have to do first is simplify. So we simplify the real world into something that can be represented in a computer.
That means once we have that computerized representation, we can change something. So we can run a policy scenario or an intervention scenario. Or, you know, we can run something where we change that computerized representation and say what might happen.
Then we get this simulated outcome. And that’s kind of the, you know, the answer to the what if experiment that I was talking about before. And I guess one important thing to say at this stage is that it isn’t a magical crystal ball.
It’s not like predicting the future. It’s just a way of exploring the potential consequences of the evidence and assumptions that we already have. And that will be something really key as I move through some of the examples in the webinar.
The evidence that we already have is really, really important when developing these models. So what is actually inside a model? I think sometimes they can be a bit of a black box for people and a bit of a mystery.
So on this slide, I thought I would just give some kind of basic ingredients that all models typically tend to have. Almost all models will need some version of these. So first, a population.
So in a public health model, who are we trying to represent? Is it the population of England or is it the population of the US? Or, you know, it could be could be lots of different different populations.
What do we know about the effects or behaviours that we’re trying to represent? So, you know, this is where the evidence part becomes really important. Then rules and relationships.
How do things affect each other? So what does that evidence tell us about, you know, if this goes up, this also goes up. So if we ban food advertising, consumption might decrease.
So how how might those things be happening? And then also we need a representation of time and uncertainty. So how are things changing over time?
You know, how often are people updating their, I don’t know, habits about their eating habits or, you know, their typical daily diets and things? So I’m going to, this is kind of a brief overview of like generic ingredients that tend to go inside models. And I’m going to show you some examples of specifics that will hopefully make this a bit more concrete.
So in the next. Yeah. So in practice, this is essentially a whole bunch of data assumptions, rules and equations that we implement in computer code.
And that is all a simulation model is essentially. But there are different ways of doing that. So in the next section, I’m going to quite briefly go through one simulation method called micro simulation.
And that is a method that’s used quite a lot in public health policy research. So I thought it would be a good starting point. So the main idea behind micro simulation is representing individual people.
So I’ve got a kind of imaginary bunch of people here in the year 2026. And then I’ve got just don’t know if you can see the pointer, but if not, you can see the screen anyway. So the

Speaker 1:
We can see the pointer.

Speaker 3:
Okay, great. I can’t see the pointer, which is really confusing. So on the left, we’ve got the 2026 population.
And the idea of micro simulation is that we are trying to take the population through time. So to an imaginary version of that same population in the year 2040. So this simulation method represents individuals in a population.
It doesn’t simulate a whole, it doesn’t simulate a population kind of as one entity, it actually tries to represent individuals. So a data set that has different individuals with different characteristics and risks. So those could be, they have different ages or genders or socioeconomic status.
This is one of the ways that we can look at health inequalities, and also risks related to kind of their personal attributes and characteristics. So their life trajectories are simulated over time in this method. And one term that’s used a lot in micro simulation research and kind of one of the key ingredients of micro simulation is this thing called a synthetic population.
And a synthetic population is basically a data set that has individuals that have been constructed so that the population reproduces important characteristics of the real population. So one example of this is a synthetic population for England might use census data from England, so how many people of each age group live in England, how many people of each education level live in England, and it will be constructed. I’m not going to go into detail on the method of how it’s constructed, but it will be constructed so that people generally in the synthetic population are representative of England.
So this is an example from the SEMA project, which is a project that I worked on a few years ago, looking at alcohol policy and health inequalities in the US. And essentially, this project is trying to simulate a population of the US in terms of their alcohol consumption, and then be able to look at how changes to the alcohol consumption might affect their overall health. So we start with a synthetic population.
So where the example I gave before was for a population of England, this one was set up to be representative of the US population. We then kind of take these synthetic individuals on a journey through the simulation. So we start off with a population.
In this case, it was for the year 2000. We then add new people because we know that over that period, the US experienced a lot of inward migration. So lots and lots of new people arrived. So we wanted to represent that.
And also the population model starts at age 18. So when people age each year, you need to add new young people in so that it’s not just an aging population that you’re simulating. So we add new people into the simulation in each year.
We remove people due to death, which is linked to their alcohol consumption, which is based on epidemiological evidence. We then remove individuals due to migrating out. Individuals in this model can change their education because in the US data, education is a key indicator of socioeconomic status.
The individuals in the model can then change their drinking behaviour and they all age by one year. And then the cycle kind of continues round and round so that we’re sort of simulating this representative population. And the key thing here is that we can intervene on this population so we can imagine a policy to change alcohol consumption and look what happens.
So what these models can tell us is how a policy that can change consumption might change mortality from specific diseases, because we’ve got people’s normal drinking behaviour in the simulation here, and then we’ve got people’s deaths linked to the alcohol consumption. So if you intervene and change the alcohol consumption in the model, you can change the mortality related to that. So this is just something I’m not going to go into a huge amount of detail on this slide.
But essentially, we did a simulation experiment on this micro simulation model where we were looking at different versions of alcohol taxation policies. And we were looking specifically at differences between socioeconomic status, which is education in this case. So because individuals in this synthetic population have got different attributes, we can then look at how their results might differ when we put everything together into a simulation model.
And the main thing to take away, so I just wanted to draw your attention to it. It doesn’t matter. The scenarios are just different versions of an alcohol taxation policy.
So changing the price of alcohol by different amounts, essentially. And the scenario four right on the right hand side is where we can see that in our simulation, we found that people with a high school diploma or less seem to reduce their drinking more. So this is changing their mean grams of alcohol consumed per day.
People with a high school diploma or less seem to have a bigger decrease in their consumption than a person with a college degree or more. And that tells us that scenario four doesn’t really matter exactly what scenario four is, but scenario four seems to be telling us that it might have a better impact on health inequalities because it’s more effective at people who are drinking more. So this is the heaviest drinkers are right at the bottom here.
So they reduce their alcohol the most. And you get the biggest differences in the high school, diploma or less group, which actually are the groups that have the biggest alcohol related harm. So one key question for this type of modelling is where this behaviour change comes from.
So in that example I showed you where the kind of the simulation is like going around in a circle, we have this intervention where the policy changes people’s alcohol consumption. In this version of the model, so in this micro simulation example that I’m showing you, that was based on published evidence, so kind of good quality published evidence where people have observed how price changes are associated with reductions in consumption. So those assumptions were built into the model.
But one important kind of limitation with this type of method is it’s based on an assumption that only price responsiveness is changing people’s behaviour. And actually, as a behavioural scientist, we might want to consider other things that might change people’s behaviour. So we know that people may change their behavior due to price responsiveness, but also habits, social context, motivation, availability, lots and lots of other things are also important for behavior change.
So we might want to represent those as well. So this is then leading me nicely into talking about agent based modeling, which is a more mechanistic type of model. And I’ll explain what that means in a second.
But before I go into any detail on what ABM is exactly, I just wanted to highlight how this fits in with something that many of you are probably very familiar with. So, of course, when we think about research in obesity, we’re dealing with something really complex. And this Foresight Obesity Systems map is probably one of the best known illustrations of this.
So it was developed to represent these many different interacting factors influencing obesity. So you’ve got kind of individual psychology, food consumption, societal influences and all sorts of other things like food production. And essentially, if we want to understand how a policy works within a system like this, We need methods that can represent these interactions and kind of, you know, we’ve got lines going between many, many different things.
So we’ve got lots and lots of feedback loops. Things aren’t linear. We want a method that can represent that.
This is where agent based modelling comes in. So agent based modelling is a computer modelling method that can represent complex things. And human behaviors are really complex.
So they’re ideally suited to this kind of this type of method. So to illustrate what this method is exactly, we’ve got Agent Allen. So individuals, so micro simulation represents individuals.
So does agent based modeling. The differences in agent based modeling, the individuals that are represented are called agents. So this is Agent Alan.
In our example of what agent-based modelling is, Alan decides on each simulated day how much unhealthy food he’s going to eat. Now, instead of in the micro simulation example, people change their eating behaviour or decide their eating behaviour based on kind of percentage changes and things like that and statistical models. In an agent-based model, Alan might decide on each simulated day how much to eat based on what his friends in the model are doing.
So agent-based modelling can represent social interaction and also based on his environment. So agent-based modelling, the key defining feature of this type of model is that it’s representing people’s environments and also people’s social connections. And those are the things that are kind of generating the behavior in the model.
So agents in the models have characteristics, follow behavioral rules, interact with other people, they interact with the environment. And then the population patterns in behavioral outcomes or obesity emerge over time from these mechanisms. And this is why we call it a mechanism based method.
So to try and make this example a little more concrete, I’m going to now move on to talk about the Nourish project, which is the project that I’m currently leading that is looking at using agent-based modelling to examine food policy and food environment policy specifically. The background to this project is that we know that food environments shape both obesity and food insecurity. And we know that policies to try and tackle the food environment or change obesity might have different effects across different groups in the population.
And as I just went through with the example using Agent Allen, ABM lets us represent these people, environments and mechanisms all together. And our broad question kind of guiding the whole project is how might different food environment policies affect dietary behaviour, obesity and food insecurity? And would these effects differ by food insecurity in Greater Manchester, which is a UK city region?
So the Greater Manchester link is quite important because, as I said a few slides ago, when we build a model, we have something called a synthetic population. This is like the who, who’s being modelled. And it’s nice to represent kind of a real place in an agent based model.
So we have chosen Greater Manchester for our model. and one of the reasons for that is that there’s quite good data available for this region in the UK and also we have some quite engaged policymaker and public health partners from that region who are kind of helping us to shape the model and decide what goes into it and decide what policies get modeled. So this is kind of a broad overview of how the model is being developed.
And you’ll notice that work package three, which is the third one along here, agent-based modeling. So that’s not even the first one. So there’s a lot of stuff that needs to happen before we even build a model or do any sort of computer coding at all.
So the project is kind of moving through these four stages. So the first one is conceptual understanding. So understanding the mechanisms that we think are kind of driving the relationships between food environment policies, what happens to specific elements of the food environment, dietary behaviours, health outcomes, and then use that to, and also use this conceptual modelling and stakeholder engagement to identify priorities for the model. So what should the model even do in the first place is kind of the Work Package 1 aim.
Then we want the model to be grounded in some sort of reality. So So looking at lots of different population data sources to try and inform what the model looks like and the numbers that go into it. Then we’ve got work package three, which is actually building the agent based model.
So sort of putting all of the insights from work package one and two together into this kind of coherent thing that exists in a computer in code form. Then we can run our policy scenarios through it. And then right at the end, we’re doing health and economic impact modelling.
So we’re then linking with other types of modelling. So we’re actually linking with a micro simulation model to estimate the potential impacts of our model policies on population health, health inequalities, economic outcomes. And I’ll give a little bit more detail on that in a minute.
So step one is conceptual modelling, and it can it can feel a little bit overwhelming. it can feel a little bit like this when you when you first start so um like what does the model actually need to represent and which parts should be included like obviously we can’t put the entire foresight obesity map into an agent-based model that would be insane so we need to decide which parts should be included so instead of instead of kind of doing this we actually follow a structured process. So this part of the project is being led by Hazel Squires at the University of Sheffield, who’s done a lot of work on this kind of type of conceptual modelling.
So how do you decide what goes into a computer model? And it kind of follows this, it’s sometimes called participatory systems mapping. So what we do initially is review the literature.
What things do we think are important in the relationship between people’s food environments and people’s behavior we’ve then had two stakeholder workshops so far where we’ve kind of presented this evidence to stakeholders and essentially moved loads of post-it notes around on pieces of paper and drawn lots and lots of arrows between things to try and represent what you know how we think things are working come up with an initial conceptual model have some further stakeholder input refine the model so it’s a kind of cycle cyclic process where we kind of start with a version of what we think the system looks like and then we’re kind of iterating that and refining it and there are some key questions when we’re doing this right so what are we trying to represent what are the mechanisms that are important what policy questions should we be able to answer what can we leave out is a really important question as well. So when we’ve done all of that, we end up with something that looks like this. And I am not going to go through every single box and arrow in the whole systems map. But this is what our systems map ends up looking like in the end.
And it looks really complicated. And I don’t expect you to be able to follow every arrow. But what I do want to highlight is the movement from so we’ve got the food environment on the left so these are things like access to healthy foods availability of healthy foods the price of hfss foods so high fat salt and sugar foods and then these are influencing people’s observable attributes so people’s kind of individual characteristics so things like food insecurity is a really central thing in the middle here, because what we’re interested in ultimately is whether policies might affect people differently depending on food insecurity.
And then we’ve grounded this in a behavioral science theory called ComBee theory. So ComBee is capability, opportunity, motivation, and we’re using those as a predictor of behavior. So ultimately, we’ve kind of organized everything that we’ve found in the literature and from our stakeholders and people in Greater Manchester who are interested in public health policy.
And they’ve said, okay, we want you to simulate this policy. We’ve then prioritised based on expertise and published literature, what the most important things are in the pathway between that policy and the effect. And that’s what this ends up looking like.
So then we have their capability, opportunity, motivation will then influence their fruit and vegetable consumption or their high fat, salt and sugar food consumption. And I think one thing to highlight is the value of, you know, we haven’t even written any computer code at all at this stage. but this is a valuable exercise in itself because it forces us to say what causes what and how is this all working?
And that is then what the computer model has to calculate. So ultimately, all of these arrows need to be a piece of code in the final model. Then we have step two, which is to populate the model with some data.
And here, I’m just giving few examples. So we had like all of these things around the agent here are boxes on the diagram that I showed on the slide before. And this is just showing how we’re populating the model with this data. So we’ve got food insecurity.
So people can have a one, two or three, which gives their food insecurity status and that’s based on the Family Resources Survey and Food and You Too, which are two UK surveys. Then they have some things like usual consumption of fruit and vegetables, that comes from the National Diet and Nutrition Survey, that’s linked to other kind of personal attributes so that it would be linked to their socio-demographic factors in that data set. Then we’re using some more specialist surveys and data sets.
So we’ve got some data from Emma Boyland who has looked at looking at a person’s in the moment craving for high fat, salt and sugar foods linked to their high fat, salt and sugar food advertising exposure. So we’re integrating kind of some more specialist data to give us some of the things that aren’t necessarily measured in population surveys. And then we have a whole bunch of socio-demographic indicators as well.
So they’re coming from the Health Survey for England. And these are things that are really important when we start to want to look at inequalities and differences between people. So we’ve got age, sex, ethnicity, IMD, which is Index of Multiple Deprivation, LSOA, which is Lower super output area which is a geographic area so where they live because we need to be able to link in the model to their food environment and then other things about their household like household size income occupation and these pieces of information all need to be integrated so the simulated people have got plausible combinations of these characteristics and when we simulate them through the model and their behaviours change and things about them change, we can then kind of sum it all up at the end and say, okay, what’s the difference between this group and this group in that policy outcome?
So step three is also really important. So calibrate to real world patterns. So one thing that we really, really want to make sure of is that the model can reproduce important patterns that we see in the real world.
And we know, for example, that we have a very strong socioeconomic gradient in obesity prevalence. So here, this is illustrated really clearly here. So we’ve got, this is prevalence of obesity.
One, which is the top line, is the most deprived decile in England. And then 10 at the bottom is the least deprived decile. So this is something that we need our model to represent because if our model, when it’s kind of representing people’s behaviours and attributes, if our model can represent this, we might be more likely to trust the policy experiments that come out of it because the model can generate the kind of population patterns that we know exist.
So we know this deprivation decile exists and our model should be able to reproduce it. The final stage is to project some health outcomes. So I said a few slides ago that the final stage of this project is to link to an existing health economic model.
So we’re going to be kind of projecting behavioral changes that the model is estimating kind of through those mechanisms in the big crazy systems map, that ultimately will get projected into a change in energy intake, which then gets changed into changes in BMI, changes in health conditions, health rate, equality of life, ability to work. And then we can look at differences between population groups in these outcomes. One key thing just to quickly note is that because this is an agent-based model, where a micro simulation might just stop at the health outcomes and you know health related quality of life ability to work because it’s an agent-based model and it involves this interaction we’ve got these feedback loops so some of the longer term outcomes will feed back into the system so this kind of hopefully allows us to get a more realistic representation and also kind of you know, factor in things that might be unexpected because they’re happening over longer periods, we might be able to simulate things that we might not have otherwise thought of. So kind of the complexities of how ability to work might then feed back into other things like a person’s food insecurity.
Okay, so that’s it in terms of a little tour of some of the different types of simulation methods. And I just wanted to quickly conclude over the next minute or two with just some ideas I have for how ECRs might be able to get involved in kind of the different opportunities for being involved in simulation modeling. So I think there’s kind of three options.
And I think maybe I didn’t want people to just sit here and be like, oh, well, I don’t coding. And I never want to code a model. So this isn’t for me, because I think that’s actually only one, only one area that that you could kind of go into.
So obviously, for people that are interested in developing technical skills, building models is an obvious one, right? So you know, learning how to code up a model in R or Python, if you like statistics and coding, that’s a good option. but I also think hopefully I’ve demonstrated you know I haven’t actually talked about a lot of the real technical ins and outs of how you would do modeling and that was kind of on purpose because I think one thing that gets forgotten about a little bit sometimes is that it’s not it’s not just a technical exercise so simulation projects need input from lots and lots of different disciplines.
Most simulation projects I’ve worked on have been really interdisciplinary. So people from behavioral science, epidemiology, qualitative research, people are increasingly understanding the value of qualitative research and kind of informing what decision rules might feed into a model. And also we need input from people who understand policy because we want to code our policy scenarios in a way that makes the most sense and is going to be the most valuable.
So these contributions could be things like defining questions, thinking about mechanisms, collecting evidence. So I think that there’s a lot of opportunity for lots of disciplines to kind of increasingly collaborate on simulation projects. And then I think the third and final one is use and interpret models.
So, you know, when you come across a model in the literature, maybe you might become a bit more of a critical user. So you might think, oh, like, what does their systems map look like? How have they actually mapped out all the mechanisms and evidence?
Like, what have they actually done here? And I think everyone hopefully can become a more critical user of these methods. And my final slide, why I think this matters for early career network members.
So the key concept is that simulation models allow us to turn evidence and assumptions into these structured kind of things that live in the computer as what if policy experiments. Why it matters? Well, I don’t think I need to convince you that obesity is shaped by lots and lots of complicated stuff.
So interacting behavioural, social, environmental influences. I’d add commercial influences to that as well. And lots of relevant policies are really difficult to measure and observe consequences and have potentially quite long term effects.
So simulation can be really useful for that. What you can take forward. So you can engage in simulation modeling in different ways.
So you could learn to build a model, collaborate in their development, or just become a more informed, critical user and interpreter. And I think the key thing that I’d like you to kind of take forward and take from this is that you need to start with the policy or research question. It’s not just like, oh, I want to build a model of this thing because I think it’s interesting.
It’s like, why? Like, what do we want the model to do? And then when you start from that point, you can start to think about whether setting it out as a simulation might help to connect up some evidence or test some scenarios or do something that an existing method can’t. So I think I’ll leave it there.
I’m very happy to take questions. I’ve seen the chat, but I’ve lost it on my screen. So I might just stop sharing or minimize this so I can see.

Speaker 1:
You can keep it. We’re going to ask you the questions also. So for me, if you want to keep it, maybe it’s easier to, if someone has a question, you can go directly to some slide. But thank you very much, Charlotte. It was a really great webinar. really great presentation we do have a few questions um eugenia if you have some also

Speaker 2:
that you would like to to say something first about the i think it’s uh we can give priority to the questions because i’ve seen we have a couple well more than a couple actually might more come with more answers coming as well so i i think we can start with those so the first one is from jelena i hope i said the name right question for later this question was actually sent quite early in the webinar how did ai change simulations also any specific

Speaker 3:
program used and preferred thank you uh good question so i was at i was at a simulation conference only a few weeks ago where there were many many sessions talking about ai and simulation I think one thing that has definitely changed is, you know, the kind of bar or barrier to coding is much lower now because AI can kind of help with code in a way that, you know, you used to have to. When I first learned agent based modeling, somebody actually gave me a physical textbook and I had to use that to figure out what I was doing. Like that, that’s no longer a thing, right? We can just go on chat GPT. Oh, can you write me some code to do this?
I do think that that’s a really, really good thing because it means that people who aren’t kind of confident programmers can still engage. I do think that we should be careful. So one of the things being discussed at the conference was like about AI isn’t quite ready to do it from scratch.
And AI, we know, has inbuilt biases and issues like, you know, with the data that’s being fed into it. So we can’t just blindly be like, hello, AI, can you give me an agent-based model? Because all of that will have those biases built into it.
So I think we can absolutely use it as a tool to help, but it’s definitely not at the stage yet where it can just run off and do this all by itself.

Speaker 1:
Just to complement this, because the question was like a two-part and there was also another person who asked the same thing. So which program do you use for simulation modeling? Like, for example, R or Python or something else. There was also someone who talked about Vensim, like systems modeling with Vensim, something like this.

Speaker 3:
Okay, yeah. So microsimulation, I tended to use R. So R is quite good for, so we wrote a package for the SEMA project that I mentioned. We wrote an R package that kind of has all of the functions that do the different parts of the simulation.
And that’s really, really good. Agent-based modeling is a little more complicated. So we actually use, so it’s a toolbox called Repast.
And it’s like an agent-based modeling library. And it originally, so I originally learned it in C++. But that is talking about like barriers to use.
There’s a really big problem in agent-based models written in C++ because nobody can use them. So we have now switched to Python, but still. So Repass brought out this agent-based modeling toolbox in Python fairly recently.
So we use that now. And the advantage of that is you can do something called object-oriented programming. and that lets you kind of section out your model in a way that’s a bit more neat and tidy and easy to kind of build that complexity in rather than R which kind of wants everything in a sequence and when it gets too complicated it’s quite hard to manage.

Speaker 2:
That’s great, thank you. And we have a question about, I think there are two questions which are quite similar, so I’ll read them together. One is, can agent-based modelling account for biological differences across the population?
And another, which I think can go with it, is, as we know, obesity is a chronic and erupting disease, and the effects of behaviour change are often difficult to maintain. In general, do you know if popular modeling approaches use across the research reflect that reality? Do you have an example?
So how about biological differences and behavior change?

Speaker 3:
Yeah, biological differences is a really, really good one. I don’t know of anyone doing it. I would really like to.
If whoever asked that question wants to chat about that, then I would be very interested. I think in theory it can, right? Because in theory, you just set up your agent attributes and that kind of, you know, they could have, I don’t know, FTO gene stuff in there and that could reflect the biological, you know, the distribution across the population.
And I think that’s super interesting when we start to think about GLP-1s as well. You know, when we think about how biological stuff might interact with behaviors in an agent-based model, be really interesting. So yes, it can be done, but I don’t think anybody’s done it.
So maybe somebody could. That’d be really cool.

Speaker 2:
And following from that, we have… Oh, sorry, sorry, keep going. And there was another one on…

Speaker 3:
Yeah, so the second part of that question about longitudinal evidence, I think this is a real challenge and limitation with simulation modelling because one thing that I hope came across was that we have to decide everything that goes into the model, right? So we decide how long the behavior change lasts for. And we can only do that, you know, if there’s a clinical trial that shows that it lasts for a year, then that’s what we program.
But one really, really good thing about simulation modeling, I think, in answer to kind of this specific question is we can then say, OK, well, what if that’s not true? Like, what if we change the length of the behavior change to three months or we change it to one year? So then you can present policymakers and stakeholders with, oh, we actually don’t know.
But if this is true, then this is what might happen. And if that is true, then that’s what might happen. And kind of having those conversations with people about what the truth might look like and therefore what the model is showing can be really interesting and useful as well.
But also suggest where we need to collect evidence from RCTs and things as well.

Speaker 2:
Okay, I should carry on because we have a lot of questions coming in and they’re all very interesting. So, the next one is, so basically we would have to make like a projection of what we’re like for the model, kind of like we want to go to that end, but we don’t know how we’ll go there. That’s one question that’s been asked about how to use this kind of models.

Speaker 3:
Sorry, what? I think I missed the first part.

Speaker 2:
So the question is, we need to put the outcome that we would like to obtain, so like a projection of what it would look like, but not like kind of like what we want to aim for, that kind of projection, but we’re not quite sure how to get there.

Speaker 3:
Yeah, so I think that can be the starting point, right? So we want a model that can tell us, I don’t know, fruit and veg consumption differences by food security status. So then we say, OK, policymakers, what sort of policies are you thinking about?
We’re thinking about trying to get fruit and veg into low income communities. So then we say, OK, so maybe that means more shops exist in that community. Does that then mean people purchase more fruit and veg?
But oh, no, if people haven’t got very much disposable income, they might struggle. So what are the other? So then like that’s it’s kind of the starting point for thinking about the mechanisms, if that makes sense.

Speaker 2:
okay and then we have for step two populate the model with data does the model require individual level data what if we only have aggregated ones with no linkage between

Speaker 3:
different data sources yeah good question so we tend to use individual level data because it kind of gives us a really good starting point for that synthetic population where ultimately we need individuals represented however there are kind of synthetic data generation techniques that i don’t personally use but people do where they will take aggregate statistics and be like okay the mean and standard deviation of this characteristic in this population is these numbers and generate a like a fake data set that meets those requirements so people do do that but it’s much easier and it’s easier to get correlated attributes between people. So, you know, if you have a data set where a person actually has their age, gender and fruit and veg consumption, it’s easier to use that than try and like make up those correlations.

Speaker 1:
Now a question from Stanislava that is also more into the ECN early career professionals. So she says, thank you very much for this fascinating presentation. And she would like to know that this type of work could be part, for example, as a PhD project, or would it be more suited for a wider scale team, as it sounds very complex in scale? And are there any current opportunities for research scientists, dieticians to get involved in informing this research?

Speaker 3:
So I think the PhD question is a good one. I think, yes, it absolutely can. And I do have a PhD student working on an agent based model. But one of the kind of, I guess one of the main things is scope, right? so the model I’ve presented is a big three-year project with like quite a big team and that’s why it’s quite big and complicated with lots of things going on so I think for a PhD project it would be about um trying to kind of narrow the scope and I think there are absolutely opportunities for dietitians to get involved um i guess it’s more like i don’t know trying to identify people that are working on these types of projects and i think at least me personally i’m always happy to kind of have more in the modeling world we kind of call like domain expertise so it’s like you know people that know stuff that we don’t and we can like extract things from their brain and try and put them into the model so yes but practically speaking it sort of depends on how many people are working on these projects and but i think generally they’re quite receptive to collaboration

Speaker 1:
so i think we have question maybe for one more more question uh more for the future so there’s a person that said thank you for the presentation and they was wondering about the next

Speaker 3:
steps of the nourish projects yeah so i yeah i should have mentioned that so we we’re currently somewhere in between stage one and three so we have almost finished the conceptual model and we’re moving into developing the abm the project has got over a year left so we haven’t got any kind of modeling results yet but so that’s like the short-term next steps is where like finishing the project and actually running the model to get the results in terms of like longer-term next steps i’m hoping that this model can be extended in lots of different ways to different policies in different areas outside greater manchester and we’re planning to release the code like open source alongside publications so other people can use and adapt it as well. So one area that I’m thinking about at the moment is expanding it into looking at environmental impacts, so like looking at meat and dairy consumption and the health and environment impacts of that. So I think there’s lots of potential extensions to this.

Speaker 2:
Thank you. We have a question about the library’s name of program, but honestly, guys, I would say if you have more detailed questions like that, I think we can email Charlotte. She’d be more than happy to provide the link because we don’t have much time left, I’m afraid. But I think we can conclude here.

Speaker 1:
Yeah. So I would just say thank you for everyone who has joined. Thank you so much, Charlotte. This was a very fantastic session with a lot of questions and very interesting, interested people in this topic.
So it is a topic that is gaining a lot of interest. So just to everyone very quickly, please do not forget to for the feedback form. And we do hope to see you on our next month e-learning hub.
Have a great rest of the day and see you guys.

Speaker 2:
See you, everyone. Thank you for joining.

Speaker 3:
Merci.