 My name's Alex Kooke, I'm an associate professor at the School of Public Health, just over an MD1, over in the US. I am currently a normal guy. From the 1st of January, I'll be taking over as a vice-team of research and head of the Baostats domain within the school. So if you're ever interested in doing any collaboration with our school, then from January, come to meet before then, you go to Michael Kern Moss. So I'll be talking about modeling analytics in population health. Firstly, can I just define what is population health? Because not everyone is completely aware of what it means. It's different from health in general. So I do not work with any data that comes from the hospital, because that is not really population health, it's not about the population. That's about patients. Instead, well, we're interested in more people like you, members of the population, who may not be sick yet, about how we can try to prevent sickness and prevent disease. So I work in modeling and analytics. And I guess you guys are data scientists, right? So analytics will be completely, you know more about analytics than I do. But maybe more, I can tell you a little bit about modeling. So what is modeling? By the way, in case you hadn't realized, I'm not from Singapore. Can you guess where I'm from? This might help. So I'm originally from Scotland. Imagine that you're in a bar and you see a man wearing a coat. What would you think when you see a man wearing a coat? You may think, a few things may come to mind. But you might think, hey, this guy probably is quite canny or careful with his money. So I don't want to go rounds with this guy because he might want to go off with his sister. You might think, well, you've seen the film, brave. Scots are brave. Therefore, don't get a fight with this guy because you'll keep on fighting with you. If you know a lot about us, you may know that the Scots have a big problem, have a bad diet, et cetera. But if you really know as well, you might notice some particular adjectives which are blind to Scots that were thrown on it. Or we could be a bit too. Now, all of these are stereotypes of Scots. Stereotypes have an element of truth in them. When you create stereotypes, you are essentially developing a model of the world. You are not perfectly characterizing the world with a stereotype. For example, I am very generous with my money, and I'm not. I love a bad diet, I think. So you're making some approximations when you create this kind of a stereotype. But it can be useful. It's a useful way of understanding the world because actually many Scots do satisfy these characteristics. So when we're doing modeling, we're talking about mathematical modeling, what we're trying to do is basically to take something which is very complex like human beings and represent them in a simple way that more or less captures all of the dynamics of that system. Now, mathematical modeling in the area of population health has got a really, really long history. How long did it go back? 10 years? 20 years? If you go back to the year 1915, in volume one of the British medical journals, one of our top medical journals, you can see this paper by Ronald Ross. So Ronald Ross won the Nobel Prize from medicine for discovering the parasite, which is a positive area, the vector which was stressed out. Now, although he won the Nobel Prize for that, that's not what he considered to be his main scientific achievement of his life. He actually thought of his development of the field of mathematical epidemiology as being the thing which really characterized the success of his career. And if you look back at this paper, the very short scientific paper is just two pages long. On paragraph two, you can actually see one of the world's first mathematical models of a disease. It's set up as a series of ordinary differential equations which characterizes what we know about the epidemiology of that disease. So when we're doing modeling, we are using information, but in a different way from the way that data scientists would usually use information. So when you're looking at analytics for data, then essentially what you're trying to use that for when you're making decisions based on that is you take rich data because you're able to tell you pretty much everything that you can know about that underlying problem. In modeling, we kind of often will have very cracked data. Honestly, data that I have is way worse than the kind of data that's available in the hospital, where you have computer data for all patients of all the cost, and all the tests that you've got done, and so on. The kind of data that we're having access to will be like from an upgrade of a disease with 100 cases in it. So obviously it's much, much smaller than that. So what we want to try to influence policy, and one of the things that our school does is we work very closely with them into health in influencing their policies. What we are trying to do is to capture a different kind of information, rather than information from data, information from our heads. It's using expert knowledge in a sense in order to characterize relationships that may not be explicitly visible in a data set, where it's very small and it's got a little noise in it. And this is often as a cover for weak data. If I have really rich data and I've worked with the National Environment Agency, where we're predicting denki upgrades, and there I've got data on my computer for like 100,000 cases over the last 10 years. That's a very rich data set. I don't need to do any modeling of that, because the data themselves is telling me everything I need to know about it. But when we're out, we're working with some like the disease that Kisha will talk about later. We had a few hundred cases only, and there we really need to account for our knowledge of the system. So what are we talking about in this talk, in my part, so I'll be talking for the first half or so of the seminar, is about diabetes. And I'll be talking about the individual based model that we've developed for diabetes, which is influencing our colleagues in the ministry. So what I'm showing you here are data from some national health surveys which are conducted, have been conducted every six years by the Ministry of Health. This is the three surveys which were conducted in the 90s and the 90s. And what they do for this is that they will recruit 5,000 or 6,000 adults in the population, and then each one of those adults so it doesn't have to see if you got diabetes or not. And so what I'm presenting here is the prevalence of diabetes, DMs for diabetes measures, in working-age adults. So if you just look at these three data points, you think, well, it's a pretty boring plot. You're trying to predict what the future burden of diabetes would be before you think, well, it looks like it's staying constant, maybe going down another level. So I'll show you the data that had been published in the most recent national health survey. So in 2010, we saw that the prevalence of diabetes had risen to around about 13%. Another given number is, can't remember what it was, maybe 13%, maybe 12%. Now, it seems like it's a small amount, but 2% of diabetes, diabetes are very slow disease. You get diabetes usually towards the end of your life. If we had over a six-year time period a rise of 2% in the absolute percentage, that actually is a really significant amount. Now, if we're trying to predict what the future burden of diabetes will be, let's say, in the year 2050, should be well beyond the current government in Singapore, then we obviously cannot just expect that we can take the least raw data and extract it as trends into the future. Because we tried that here and then we did this wrong. So what else could we do? One of the major, the number one driver of diabetes is actually not obesity or overweight, it's age. Although you are, the more likely you are to have developed diabetes throughout your life. So what we could do, if you want to predict the future burden of diabetes is, we've deconstruct the problem, you see, I know a lot about the aging of the population and I can project how the population will age to the year 2030 to the year 2050 and so on. And I know now what is the relationship between age and the prevalence of diabetes speed. Maybe stratified by ethnic world or gender of all. So then I can then project what the demographics will look like in the future, the main driver, and use that to predict what diabetes will look like. Now that will give you a kind of an okay approximation of reality, but still just an approximation. In particular, this is the second most important thing, which is as I mentioned, overweight recently. We know that people are getting more and more overweight. There's another mechanism which may drive future trends. Now here I'm showing you the prevalence of overweight again in adults, again in the National Health Service in the year 2004 and year 2010. So across the board, this is risen, in particular amongst younger adults. We think in particular that there's a transition from for guys between NS and civilian life, because when you're an NS, you're burning off lots of calories, you're also eating lots of calories, then you become a civilian again. You're still eating all the calories, but you're not wearing them off anymore. An exercise of punishment, right? So you don't want to exercise that anymore. So there we see that, it seems like that young males in their twenties are especially overweight, relatives in the past. So this represents basically a fundamental change to the rules. We cannot just expect the prevalence will be the same, but in the age groups, in the future, if more people are overweight, at least. So instead of what we could do would be that we'll forecast the population age structure and we'll also forecast what the obesity prevalence will be like going forward. And so we can combine both of these risk factors together to project the burden of diabetes. So when I'm talking about modeling, that's kind of what I mean. Each of these components is actually quite simple from an analytics perspective, but from a modeling perspective, we're bringing it all together in a more complicated way. So we've been developing a platform within our school called DEMOS, Demographic Epidemiological Model of Singapore. I was really, really proud of that acronym. So what is this? This is an individual level life history model. It combines survival analysis and what we know from statistical demography to simulate individuals at the whole population level how their lifetimes will change over time. You know what I mean, right? We're simulating their lifetimes over time. So within each individual, because we're representing people as individuals, rather than just as a population as a whole, then we can assign characteristics to those individuals that may vary dynamically over time. For example, there'll be a map. And some of these characteristics will be risk factors for disease. So by simulating, by developing this as a simulation model, we can simulate from the past through to the future and then at different time points, we can then project down from the simulation to get a simplified representation. So for example, we could project census tables at different time points. Then check, are we actually reproducing the census that we've observed? We can project down and get national health surveys and say, are we reproducing what the Ministry of Health has measured? So the kind of data that goes into this is small, kind of chaotic, in the sense that we have quite a lot of different data sets that are providing different kinds of information for our model. So we're taking data from national statistics so that the census, which has been up to there every 10 years, we take some of the data from the U of those statistics which since that publishes yearly. So we can use this to get information about things like mortality rate and how it's changing over time for our fertility rate changing over time. And that's that the drivers of the demographics of the population will look like. We also take a cross-sectional health surveys like the ones that the Ministry of Health conduct this would be in the six years so far. This tells us information about things that prevalence of risk factors and prevalence of disease in different groups in the population. You'll note that there's a lot of uncertainty on those. You look at it, the common principles are very wide. And that's because when you start cutting them up by different groups in the population, then you find the sample sizes are quite small. We also combine this with data from our own pro-mark studies at the school ones. So a pro-mark study is where you invest a lot of money recruiting people into a study and it's like millions of dollars for this. And then you measure a lot of them at baseline and then you basically wait for them to die. And you can use what you measure at baseline to get your information about what the risk factors are for that or for different diseases. So like if you get cancer, then your cancer will be notified to the registry of diseases over at MOH. And so we can then relate our cohort data to subsequent disease. Now you can imagine that not many people get cancer at a typical one-year period. So we have to recruit a large number of people based on it. So our cohorts are around 100,000 people from across the world. Now for some of these, we actually follow them up and we get them at a second time point. And so we can use this to understand how disease changes in the same group of people over time. So we have to merge all of these kind of data together. So I'm going to give you kind of a simplified representation of the model. So you get an idea of how we do it with modeling approaches. So this will be populated with arrows and boxes. So we'll start off with demographic variables. So we take the most important demographic variables of age, gender, and ethnic group. We know there's differences, for example, in ethnic groups for disease. So the prevalence of diabetes in India and Malay is much higher than it is for Chinese. For ethnic Europeans, we are at the lowest genetic risk for diabetes. Yay. A few people in the audience might be happy, but most of you don't. So this, because these are important risk factors for pretty much everything that goes on after this, we'll use these and we'll feed them into a model of the demographics of the population. So this model then creates, as output, trajectories for individuals, which gives you the time in which to get birth or maybe the time in which to die, and also migration in and out of the economy. Demographic variables also feed into a model of BMI. I'll tell you a little bit more about that in a subsequent slide. It essentially gives us a BMI trajectory, which accounts for typical changes over your lifetime. And this BMI trajectory together with demographics feeds into a model which gives us as output incidents of identities and also prevalence, instances in new cases, prevalence of total cases over your life. And because diabetes is a known risk factor for death, as well as other diseases, so we put this one back into the demographic model. A really great example to remove this spot. So we have a genetic model as well, but actually it's just for fun because we've got a lot of genetic stuff over in our school. It doesn't contribute anything, so you pretend it's not there. But essentially that gives us an extra influence in the diabetes model. And because we want to be able to project economic burden of disease, then we also want to work out who in the population has diabetes and are they working or not? And what is the contribution to the labor force? So we have built another model, which is of individuals moving between being out of work in work, which gives us a work trajectory, and we combine this with diabetes incidents to get a measure of the amount of diabetes in the workforce, as well as in the population as a whole. When you want to influence policy, what we find is that one of the most important things is to be able to go to a cost because Minister of Health is not really in charge. Minister of Finance is the one that's in charge. So if they want to go to say that something is going to be a burden to the health of the country, MOF is going to say, I don't care. If it's a burden to the economics of the country, MOF will say, okay, we should do something about it. So the demographic model, I'll just talk a little bit about that. So it is of the various aspects of your life. So for example, we have mortality rates, which we don't change over time. We basically built, using what we want to call out, we built a model which projects what that will look like over time. Fertility rates have been going down even within the two decades that we've got models set up for. So the feet fertility is a bit later and then overall fertility is lower. The data that's since that published is actually not very good for estimating this. So what we'll give out is a fertility rate in people aged 15 to 19, 20 to 24, 25 to 29. But if you think about it, probably no people aged 25 and aged 29, they're quite different fertility rates. Not a lot of people aged 25 are having kids. Fair number of people aged 29 are. So it changes by a lot even within those kind of bins that they will use for their aggregations. The hardest thing for the demographic model was actually to model migration. This is very sensitive topic. So there's not a lot of good data on this. We know from other settings that there's a typical migration pattern for age, so people tend to migrate if they are aged less than five or aged around about late 20s to late 20s. So basically, when you're working age and you have kids, but the kids are not yet at school, then migration can happen. Once your kids start school, you don't want to migrate anymore. So we find this in other countries and we were able to reverse engineer it for Singapore by looking at the change in the age pyramids that are published by SingSend. Again, we've done this for a different ethnic groups, but it all looks very similar to the one. So what this allows us to do then is to project how the profile of the population will change over time. So if you go back to about 1970s, it really wasn't age pyramids. So age pyramids are so-called because they look like pyramids in some countries. By the 1990s, the base of pyramids is moved up near this flat part of the bottom. When we can see that by the year 2050, we're looking at more like an age-care part of pyramids. So I'm going to go bear shape right now because you have too many more people. Okay, so we have a rebuild model of BMI, which I alluded to earlier. So what we're doing here is we have data from our cohort studies, which tell us about individuals at baseline and at one round of follow-up. So we have the BMI at two time points. And by looking at the patterns from across the whole cohort, we're able to build a longitudinal model which is individual-level. So we use a hierarchical model where we've got hyper-counters which govern the variability between individuals. What this characterizes is a very typical profile for weights. So on average, people who go on weight are about 18, 20s. And they kind of keep on being all the way for some time. They start to lose weight again when they get about 50 or so. So we can characterize that in this model. This lets us project what the future burden of overweight and obesity will be by looking at changes in the younger stage group that we've observed in the sense. So Singapore will start to look more and more like Scotland over time. So the final model that I want to talk about within the overall framework is our model of diabetes. So looking at the size of the audience, there's probably one or two people who have diabetes in the room. But there's probably more people in the room who have diabetes but don't know about it. So quite often you'll have diabetes but you just don't know about it. You're not going to be testing for it. So it's not like flu. If you've got flu, then you know you've got it because you've got horrible symptoms of it. But as you play around about 50% of people who have diabetes in Singapore don't know they have diabetes. One of the major problems when we're trying to reduce diabetes in the population. So we have our cohort where we have only two time points and for many people they'll have started off without diabetes but the second time we find that they've got diabetes. We don't know when they develop diabetes during that time window. So what we've done is we've developed a model of potentially cryptic unobserved diabetes which depends on a bunch of risk factors like age and ethnic group and so on. And we fitted this using a combination of Markov team on Karlo, an important sampling based on the PMID that we had before and the demographics. So by tracking the population's aging and ill health we are able to forecast what the burden of the disease will be. So for example, we can look at what we think the profile for overweight, obese, and overweight will be. And I've used that to project what we think the future prevalence of diabetes will be. You may have read about in the news that we're expecting to have a million diabetics. Maybe I read this because I'll let you know. It's about my work so I'll find out how I'm going to do that. So our projections of one million diabetics have actually been fed through to the Ministry of Health and we're signing to when Singapore declared war diabetes last year. So I'm a bit of a warm-up. Fire the first shot in the diabetes world. I think this is really unfortunate. I don't think we're going to win this war, but never mind. I've been recorded, right? The thing about me about not winning the war is let's cancel that one out of here. Okay, now the purpose of being one is not really because we want to just project the numbers. We want to go to look at what the burden of this will be and how can we try to change things so that the burden is less? So what we can do with this simulation model is to look at the effect of different interventions, what would the downstream benefits supposed to be? So if we are, so we don't really care about diabetes, diabetes itself doesn't really matter. We care about not diabetes, but the complications that come from that. So if you have diabetes, if you're a high risk of diabetes, so heart disease, et cetera, okay? So when we are thinking about interventions, these will be interventions which either focus on the risk factors for diabetes or diabetes itself. So what are risk factors? Overweight. So an intervention which tries to reduce weight in the population will target the risk factor, lower diabetes, and therefore lower diabetes. What else could you do? You could have an intervention which focuses on the disease itself. So screening people to try to identify if they have diabetes and get them on to medication. That won't change the risk factors, but it means that these will have lower risk. So I'll talk about now two such interventions. So one is weight loss programs. So one of the ministries asking us, how can you try to assess how much, if they give us a weight loss program and they can describe what we think the effect of it will be in terms of reducing weight. Can we then say, what will that be in reducing diabetes? So for this, we're gonna project under no changes what we think the future burden of diabetes and some of the complications. So for example, here for AMI, that's heart attacks and strokes. So we think that by the year 2050, we're gonna have basically about a three-fold rise in the number of heart attacks and strokes. It's a really good time to be a doctor. Not a radiologist, but any other kind of doctor. So MOH will come to us and say, okay, we want to build our program. We think that we'll roll out a population level and it will reduce the weight by some amount. Can we want to do it? That's what we say. Well, we need to transform this. I'm telling you, it's very vague, it's stupid words. I want to get numbers. I want to be able to describe this in a more mathematical way. So we'll come up with an algorithmic way to describe it. So for example, we say, let's assume all adults in the year 2018 who have a PMI, which is medium-weight strokes, and then because I've worked with NUHS, NUHS controls this part of the country. So we can actually then break things down by the different regional health service which is in charge of the different parts. So we can use that to project what is the burden of different diseases in the different RHSs. And not just to say, this is just now, we can actually also project into the future because we know for a new housing estate is going to be developed by the year 2030. For example, the 10-day estate around the world is your own. That will probably, it will be up and running by 2030. And we know what the typical age profile is for new estates, right? Again, potentially like young married couples, maybe with young kids. So we can, from what we've learned about from past new times, we can then project what the evolution of new estates like Tengel will be and use that to project what the burden of disease will be each of these parts of the country. Oh yeah, and then, because it's individual levels, we actually, we can zoom all the way down to like blocks and say, this is like, what we expect prevalence should be for different diseases in the block. And that's used for our NUHS, but they're like the screening programs which blocks should be targeted for screening. I'm talking a little bit too long, so I'm going to, I'm going to head over to you in a second. So, I hope I spoke to you, so thanks. Thank you so much. You're welcome. Thank you. Thank you. Thank you. Thank you. Thank you. So, so I was going to talk about smoking a bit, I won't because I've spoken too long about diabetes. So what I'll do now is I'm going to hand over to Keisha to talk a little bit about some of the work that she's doing on Zika from the Zika outbreak of last year. So it's a very different flavor. Diabetes is a very slow disease. Zika, very quick. So we had an outbreak in just a couple of months and it caused a lot of panic and it required lots of overtime for our team. So Keisha will talk a little bit about that. Okay, I guess just now Alex is just sharing about something that's like non-communicable disease. But you must be familiar with what happened in Singapore last year. We had the Zika outbreak and it pretty much the whole world knew about it. CDC in America was talking all about it. We had travel bans or travel restrictions to Singapore. So exactly when it was published, it was published on 27th of August in Straight Science. That is our most recognized journals as academics. You get it the straight times, you make it big, really. All right, so, and you must be familiar with this mosquito. I think irritated y'all quite a bit. This is the Agus egypti, also responsible for transmitting dengue. So the same mosquito transmits Zika in Singapore as well. So the very next day, suddenly they tell you 41 cases, oh my goodness, but actually not really because it's from the retrospective testing of cases. So what we did was that with the data given to us by MLH, we found that the first known case actually dates back to 31st of July. So it's not just like, oh, suddenly ministries hiding information. It's just that nobody knew Zika was actually really spreading at that point in time. And when they found out, it dates back all the way to 31st of July. So Ministry of Health actually brought in together a group of people. So we call ourselves the Singapore Zika Study Group and the work they're presenting today is actually really a collaboration between everyone that is here. So researchers at the School of Public Health, mainly us, we were helping the ministry just answering two basic questions such as how quickly was the outbreak going to grow? As it was unfolding without interventions and how widely dispersed was the Zika outbreak when it was happening itself? So answering the first part, we had case data given to us by the Ministry of Health day by day, they were coming in. And if you know that an individual is infected with Zika, you know that this individual can go on to infecting mosquitoes. So Zika does not really transmit much from individual to individual, but it transmits faster through a factor. And this factor is a mosquito. And this mosquito can go on biting you. And we found out that on average, an infected individuals would lead to the infection of about three people. And this is actually at the construction site. So I'm not sure if you're familiar with where the outbreak was happening. It was somewhere in our junior area. So that was what we saw at the construction site. So this was the case data as it was coming in. And if there were no interventions, we found that about 90% of the population would be infected. Yeah, but then again, having said that this is, if the scenario was the same in the rest of the island, like it was at the construction site, but we know that at the construction site, you have like puddles of stagnant water. So there was actually an avenue for mosquitoes to start breeding and then biting human beings around the construction site. So NEA and National Environment Agency and the Ministry of Health, they panicked a little, but it immediately activated that intensive control measures at that area. So vector control measures that were targeted at getting rid of breeding sites as well as isolating cases. And they managed to sort of bring down the transmissibility of Zika. So now we're seeing about one infected individual could potentially lead to the infection of about one case. And if you're familiar with the epidemiology, then you would know that this would lead to no longer, the disease no longer is very further. So we found that if the intervention was successful and less than 10% of the population was now den-infected. So the next thing that we did was to look at how widely dispersed was the Zika outbreak in Singapore. We must have been familiar with SARS. I think we all talk about SARS all the time. When you have an infected case with SARS, you do contact tracing. You ask the individual, okay, who have you contacted and then who are the closest contacts you have? But you can't actually do this for Zika because you cannot contact trace mosquitoes. You cannot say, A, who you bite today? Tomorrow who you bite? But actually from the pattern in space and time, so from the case data, we knew where the cases were getting infected. Where people live and work and when they were getting infected or when they started sharing symptoms, you can sort of project how the cases would be like. So this is how we represented the 455 confirmed case up to the end of November. And you can see how the disease was sort of like spreading around people here. So Zika is actually a disease that is asymptomatic. So majority of the cases actually do not feel it. So you may actually be infected with Zika but because you do not know that you're infected, you go to work. So you're actually able to transmit the disease at your workplace as well. So from here we noticed that you can, what our model could do was to not only just say, who infected whom, but where were they infected? Because we had the case of home and work address. So that's what this model is doing here. So for, could you click next on the next slide? Thank you. So for example, individual number four was infected at work by a cluster that was seated at his workplace. So this is in Algeria and this cluster was seated by someone who one of his co-workers. So his co-workers infected first, infected the mosquitoes around him and then led to the infection of him. He then can go on to seat clusters at his home and work because he does not know his sake. So he goes back home to woodlands where he's now able to let the mosquitoes around him bite and then you have other cases that spring off from there. So both at his home and his workplace. So this is how we see the mosquito cannot actually fly from our unit to woodlands. We know that the mosquitoes, the mosquitoes can fly at about 150 meters and that's the furthest but it is the human that brings the disease around. So this is how we see the serial rise part of the island being affected with Seeker. So you say, oh, there's a cluster in Algeria. Next thing you know, the cluster is in Bichon. Next thing you know, the cluster is in Sarangoon. So why is that so? It's because humans move from places to places. What we also notice is that the secondary cases can go on to seat further clusters. So they go and bring it to their workplace and their homes and et cetera. And you know that now about a large number of people are actually infecting people at home and infecting people at work. And large number of people are getting infected at home and work. So in the past, MOH usually just collects home data. They say that, oh, here's the residential address you can sort of target such infectious disease but in this case, we say no, you can't. And factor control measures that were targeted at home only because you know at your house, they'll fumigate, they will tell you, let me check your pots and your pots to see if there's water stagnant, water dead. You realize that that does not contribute to most of the infection. In fact, you need to also look at what happens at the workplace. So what we did was to evaluate, is it better to just look at the home alone? Like if we assume that Zika could only transmit at home versus if you now say that it can be transmitted at home and work. And we did some modeling with that. We found that the model that looked at both home and work was able to better explain the disease itself. So the time between the infection was about five days and the median distance about 400. That's pretty much the rough size of how people move between classes. Yeah, so when you're studying vector-borne diseases it's actually really important to understand that spatial information is very, very important. So that's what we did there. We did some spatial temporal modeling. And we do not exactly know, as we mentioned here, we look at how cases are. We don't have the exact place where individuals are infected. But the modeling acts as a lens for us to see, okay, this is the potential route in which the cases being infected by this way. So yeah, I guess that concludes our part because I think we're running out of time. Okay, so I'll just give you a bit of a sense of what population health is about. How it's different from clinical work. And also about what modeling involves. So it involves taking data which are often not very rich. Here, how many data points do we have? Either a few hundred or one, one hour break. So we're looking at data where actually the size is so much smaller than you would be able to get if you were getting data speaking to the web or you're getting it from an existing database. Okay, so I hope you don't mind. I'm gonna take advantage of this to do a little bit of advertising. I've got a job I need to fill. So within my research channel and EHS, we're looking at our research fellow who ideally have a PhD as to this day, as to the end of the day, sort of where they need to field. And their main role will be to develop an online panel to go some funding from the Ministry of Health in the last set of web calls. And so I kind of personally need to hire some for this. Also, I was speaking to one of my colleagues, one who has all the data on 100,000 dengue cases over the last decade, as well as like a million records about mosquito brain. And so she's looking to hire a data scientist over at the National Environment Agency at EHI. This is an image of someone who's got stats of computational biology or bio stats or data analytics background. So if you're interested in either of these two jobs, contact me to the first one or raise your hand for the second one. So instead, find an eight-kitchen watch every time. Okay, so we, together with MIT, so NUS, we have, we organized this thing called a data ton, which is basically hackathon with actual data. And we had one just, no, mid of this year, sometime in end of June, where we got a group of people came down together to just hack on data over two days. If you're interested, look out for it. We will be having one in June next year. So the webpage is still up. You can go and check it out. And we hope to see you there. Thank you. It's free of charge. All right, thank you. I like the kitchen, pretty fascinating stuff. Actually, we still have some minutes, so we're going to... Oh, I still got the showcase.