 All right, we're gonna, we're gonna get, oh, okay. We're gonna definitely get started now. And I need to do this on my screen. Got it. All right, so we are going to talk about a collective pain point for nearly everyone in the country, or no, not the country, well, yes, Norway probably too, but the world. And that is, how are we getting population figures? Where are they coming from? How are we using them? What are the problems that we're having with them? What are some potential solutions to some of those problems? We're going to tie this a little bit back to a primary use case that is in, that for a lot of the functional develop, functional development, new things, for some of the new things that we've been developing, which is micro-planning. So of course, micro-planning being mass immunization campaigns, that kind of stuff where having very accurate, very granular population data is absolutely required. And so that's kind of the larger context. And then we're going to focus on, what exactly those pain points are and look at a couple of country solutions. So it's not just me talking. I am joined by quite a lot of experts in the field. So we have Maria coming to us, coming to us from UNICEF based in Nairobi. She's gonna talk about some of UNICEF projects. Then we have Arthur Haywood and Nora Stoops. Nora's not physically in here right now. Oh, she's right there. You just hide in the front, okay. All right, we're gonna have Arthur and Nora talk about the work that they've been doing in Zambia in collaboration with Grid 3 and some other partners that have some new approaches to getting population figures. Then we have Mania coming to us from Kenya. He's gonna talk about how Kenya has been getting population-based denominators and using those where survey data, or sorry, census data is not available. And then I'm just gonna round it out with some pretty pictures of new features to visualize population data that Byron has made essentially. Okay, so let's talk about the problem. Let's talk about the problem. The problem is that when you look at nearly most countries, you see that you have coverage indicators way over 100%. Who has this? Who's seen this? Yeah, I'm actually looking for confirmation. You guys? Yeah, it's not rhetorical. We've all seen it. It's a huge problem. You see countries that have coverage indicators like 130%, 150%. What does that even mean? Does that mean we immunized some children twice? Double doses? No, I mean, essentially it means that something about the data is wrong, either your numerator or your denominator. Almost always it's, well, there can be a lot of issues with the numerator too, but what we see the biggest problems are the population denominators. They're not, the projection is not accurate. They're either too small or, well, there's sometimes they're too big, but oftentimes they're just too small, right? They did a census 10, 20 years ago, and now they're just doing projections from there, right? Census is a hard to do. They take a lot of time and effort. They take a lot of money. Sometimes they're not, there's many reasons why countries might be, not terribly motivated to conduct census as well, just to kind of glaze over a lot of political ruffling of the feathers. And so it doesn't happen often. So what does that leave us with? We need the population figures. We don't have reliable census data. Census have a couple of other issues. They're, as I said, they're outdated, they're old, but they often don't factor in population mobility changes over time. So, you know, especially in South Sahara and Africa, we have highly mobile populations moving around often, sometimes seasonally. Cities are changing rapidly. Places are becoming increasingly densely populated and our census projections are not representing those well, most of the time. And from those censuses, sometimes we don't have the granularity to actually have actionable data. So the census may not actually go down to say facility catchment areas. You don't have facility populations. Well, clinical services are provided at health facilities. Outreach campaigns are being delivered through oftentimes health facilities. So if we actually don't know how many people are being served by that health facility, it becomes very difficult to actually do some of the major public health campaigns, programs, that we know have impact. Okay, so that's enough for me. I think I would just now hand it over to Maria. Take into the microphone, can you hear me? So Scott has laid out some common challenges or issues with using census-based projections as a source of target population data for program monitoring and things like micro-planning, particularly when censuses have occurred after many years when there's issues of population mobility and also when there's issues of accessibility of the census information oftentimes when it's made available, it's not made available. It's an aggregate form, but in not a very granular or spatially precise form. And so since about a year ago, we started a collaboration with the grid three project with world population from the University of Southampton, World Pop, which is also part of the grid three, the University of Oslo and with UNICEF to first understand what are the practices around target population data availability and usage within health information systems and to explore whether it'd be feasible, acceptable, and most importantly, what are some possible practical uses of innovative sources of population data. And we'll talk about that in a little bit more. There is a data product called the grid three which takes population data, which can be from census or from micro censuses and maps that data where people are in geographic space. And so we wanted to understand what are the opportunities and possible uses for this data. The other area is around capacity strengthening and that's something that hopefully we'll have a chance during the Q&A and discussion is to hear from you as we are embarking on this learning journey, what would be helpful for us to share with the community, how can we better document and provide job aids and other resources to share with those that are interested in similar approaches. And then as well as building the evidence base around what are some different implementation strategies for integrating these kind of data and using this data within health management information systems and most importantly around primary healthcare and routine planning. So one thing I wanted to highlight is that census projections are one source of target population data but with the considerations which we just discussed but that there are a number of other sources of information that can be considered. We'll be hearing a bit later from colleagues from Kenya around their use of service delivery data as a target population estimate in coverage monitoring and planning. There's a number of other potential sources for example, the civil registration data if it's comprehensive and complete that's probably gonna be your best most granular source of information but there's often challenges or issues with accessibility of that information and data exchange and also with comprehensiveness and availability as well as other sources of data that we will hear about through some of the experience sharing and these gridded population data. So I just wanted to highlight that but also that each of these data sets has its own considerations to factor in. So why is this gridded population data what's unique about it, what's useful? So one of the key things is that by having a grid so this is a 100 meter by 100 meter grid cell is what the data is made available and for each pixel there'll be an associated population count and one of the benefits of this is that it allows you for a standard format to integrate with other sources of data. So for example, you can take this population data and link it with other things like for example, environmental data around exposure to hazards or rainfall and you can also use it to calculate some things that are geographic in nature. For example, if you're looking at physical accessibility to health services if you combine this population data with other information on topography, environment, travel networks it allows you to measure things like access to care which can be one of the determinants of among possibly others that can be limiting your demand for services or ability to deliver services by having this data also in this granular form you can aggregate it to other meaningful units of analysis. So for example, we heard from Scott about challenges with availability of data for facility catchment areas or for other meaningful units for planning. And so with this data having this flexibility then it allows us to do this kind of aggregation and fill in some gaps in information that again are not traditionally available from the census projections. So with this introduction I'll hand it over to Arthur who's going to share some of the lessons we've learned so far in investigating what are some population data usage practices and opportunities. Is that right? All right, my ears are obviously in the wrong place. I'm gonna talk through my eyes. All right, let me turn my timer on. It's got tell me two minutes before. All right. Okay, we're gonna take you from the global to the specific to a place called, you have a pointer? Chinsali, whoops, no, whoops. Is that the way it's a pointer? I don't know if I get it. If we look in the top right hand corner there you'll see Chinsali which is just about as close to the middle of nowhere as anybody can imagine. It's 12 hours from the capital city. It's a newly created district that has been running DHS since 2014. We got there in 2020 for the project. Can you hear? And there was literally no population data at all. The district didn't have a population because in 2019 they had changed, taken out some facilities and put it in another district. And so for a year they hadn't bothered to consolidate the facilities together to get a district data. Facility level populations, again, same thing. Nobody had bothered. It took about three, four months of constant agitation to get them to allow us. Because that's the other thing, that the facility level populations are put in at national level. And they don't give a shit about the facility. They, it doesn't matter to them whether there's data for facilities and it doesn't matter. And so we'd had no nothing. We had to fight to get facility populations into DHS too. At community level, all of the, every single facility has a community health agent whose part of their job is to do a census every year. So that is there, but there are no guidelines on how to do that facility count. So all of them are different. So it makes it very interesting. There was a very nice Gavi project which is focused on data entry. And one of the things when you get facilities to enter the data is that people start taking an interest in their data instead of just sending it up to the next level, they start actually thinking about it and trying to understand the data that they're collecting. The other thing that the project did which was very good for population, was that it liberated the information officer from being a data entry clerk, which had been, well, every month most of her job was entering data. Then she suddenly had all that data entered for her at facility level. So now she had time to bother about things like population and denominators and indicators. Oh, we've lost the thing on the population. For population, we had zamstatz projections and that was the only thing that was in DHS too, but the estimated populations we didn't have in DHS, it was there, but it wasn't in DHS. What we discovered was that the population data actually is there. Look on the right-hand side here. Every single facility has got really detailed population by number of households under one year, one to five, blah, blah, blah, blah, everything you want to know. They've also got maps which they draw themselves and they really understand the value of maps. Our challenge was to get that population data and the maps to be used for denominators for program management. So Nora looked at the data, which is always a bad thing to do. If you look at that column that is highlighted there, that is the percentage of children under one as a percentage of the total population. You see some clown at head office who shall be nameless has in order to get the population, the EPI data to fit, they've made estimates of the under one population from 11%, I think is the highest 11.5% down to 4%, which is probably about the most accurate. So somebody at high level has actually manipulated the population to create suitable coverage indicators. Yeah, I say no more. Anything else Nora? Yeah, okay. If you take the Ministry of Health data, you get this. If you take the adjusted, I call it community, that's my mistake here. If you take the adjusted population, you get 76%. If you take the grid three total population, then you get 77%, grid has got this wonderful thing whereby you can calculate the number of people within 60 minutes walking distance of the facility, and then it's 87%. So take your pick, 41% or 87%? What coverage do you want? Yeah, I mean, but this is the kind of issues that we're facing at a district level. This is not a global problem. This is a problem at a facility level. Okay, I mean, where does this come from? And we use this, not really the problem. Human resources, if you tell people what to do, they'll normally do it, but they're not the problem. The problem is organization. And the population is at central, I mean, the management, the control is at central level and they've got no interest in decentralized populations. Grid three is not recognized. So those official data, I mean, sorry, those grid estimates are completely unofficial. We can use the grid data up to, yeah, yeah. You can use the grid data up to facility level, but after that, you can't use it. And there's very little demand. So what did we do? We came with the maps. You can see the maps being, that's what the grid maps look like. They're beautiful, beautiful maps. There's a guy from national level coming down to come and show the people at district level what the maps look like. We started discussing these maps. We started looking at the populations. At every facility, you can see here, oh, you can't see, sorry, there's a map on the wall and the other maps. So we brought in these modern maps and started looking at the population and discussing it with the community. So we've had enormous amount of discussion at facility level. And we learned that this grid map process is incredibly good at local level for planning. But we also learned that we need decentralization that we have to give authority to the districts to enter the data because they're the ones who actually care. We also started to put the data onto DHS-2, alternative populations for example, A&C and antinatal care. And yeah, this is just the start. This is the first step that we've taken to get functional populations at district level. And it's going to be a long way forward. We need the district to function and we need to create a demand for accurate data and my time has run out. You have ears. Mine is a very short presentation and we're just going on. I just want to give you a short introduction and let you know what we did when we got into problem. Because these three bullets are very important. When you have correct data, then the type of action that comes to you is very important. For example, allocation of resources. If you have achieved 100%, then you'll not be given any resources, isn't it? Because you have achieved everything. If there is any money even for donors, you want to be given. So we reached up a situation where we are actually getting all our indicators were 100% and therefore we needed to readjust. So this is just one of the problems we had. I'll give you a bit of our statistics are normally done by the Kenyan National Bureau of Statistics. We do our census every 10 years. So the census we did in two or nine, it gave us a growth rate of 2.9%. So throughout, we'd been using that growth rate 2.9% until we did another census in 2019, which actually showed that our growth rate had gone down, which is normal, isn't it? People are doing family planning and so that's okay. But now it gives us, it made us feel like now all our data, the population we had in 2019 from projections was actually higher than the population that was appearing in 2020. And so what did that bring up? It therefore brought up that the data you have, the achievement you are getting in 2019, if you are getting, like for example, the delivery, the women who deliver in health facilities in 2019 December were around 60%. And those now delivered in health facility in January 2020 were 90%. And there was no miracle that happened in between that. So that is how we, and because now we are talking of 2020, we have the physical census data from the National Bureau of Statistics after the 2019 census. And 2019 we have the projected data. So we had all those problems and everybody was complaining, oh, we are overachieving. And so we decided maybe we could manipulate the data in the system, not in the country because we don't have power, isn't it? To manipulate the population data, but we can manipulate the data in the system so that we consistently get good coverages or reasonably, we don't know what it is now. Maybe reasonable coverages. So just looking at some of the, so when we looked at the population, the registered births were 1.1, these are in millions, then you see we looked at the KHS, the KHS reported was 1.1, which is consistent. And then BCG was now around 1.3 million. And then we looked at OPV, which are also getting at birth, at that, and then the census was 1.1. It looks almost not so badly off. But it's still within. So we had to think of how to harmonize there and the one denominator. And so we decided actually to use the delivery or the data that we collect in DHS, which has been consistent for a number of years. What followed this is actually a complex mathematical procedure. And even though I was very good in mathematics in primary school, this type of math I wouldn't like to venture into it. But actually these are some of the factors that we looked at like population sensors, the state of the data quality that you want to use into that model. And then if there are any missed opportunities, you calculate supposing they all came, what would it be? These are just the characteristics that we put in. We also put in some social cultural issues in some counties, some regions, people don't actually go to health facilities. So you can't just give them the same percentage as the others. And then we also talked of access to health facilities like all these people are getting 100%. Yet we know very well, due to the roads and everything, there is no way people would have all gone to the hospital for that. And then we also looked at the routine data for the last five years, including the reporting rates and then the sensors and then previous age categories. These are the factors that we used to come up with what we think is the correct denominators for the age groups. So I said, this is areas we looked at. The rest of the mathematics, I think I would really be very careful not to go there. But at least for the sake of my friend Nora, whom everybody likes mentioning, we had two opportunities to use BCG data or Pentavalent. But according to Nora during lunchtime, she prefers Penta1, but she will tell us why. So using all these, we abstracted that from the system. We looked at the Penta1 and then added some percentages just for those and then we were also to do all the manipulation of what we were actually looking at. The data I was telling you and just removing some of the things like twins and triplets. There are some people who are delivering many children at home also, they are very fertile. Not necessarily fertile, they get some drugs because they have been infertile and then once in a while they get a boom. So those can also affect our data and that. And then you have the surviving infants. So it comes up with a very long formula that we use that I was trying to avoid. We are the N and everything. What we need to tell you is that we had a problem from the sensors, which is the real data. And then we had a problem of the data that we had been projecting. And therefore, and our patient, our normal service delivery data were consistent with the projected data because we used to get good percentages. But now we're inconsistent with the real data. So we were faced, do you believe? Do you believe the service delivery data or do you believe in this? And if you believe in that then, so we thought, let's believe our service delivery data and just for them, use this data and put it in the system so that we continue getting consistent percentages. I think that is what I just wanted to let you know. And the details we can discuss for those who are very good in mathematics how we did it, but these are the parameters we used. So we just, we are all the subtractions that I really thought we shouldn't go into it. All these averages subtracted just and what, what, all these are very complicated factors that makes our DHS look smart in terms of coverages. But there is a sentence back home. Like, for example, I can say I'm very popular here but then they'll tell you but on the ground is very different. So now the ground is what could actually, what is the real data on the ground? Maybe we still have to think what is the correct population data? And so if the service delivery data have been consistent all these years, is it true that DHS has also been cheating as all these years? So what do we say? Those are just the final questions and I think I would like to leave it at that. All right, last little bit and we will have some time for Q and A, which is exciting. I know that Kala probably has a lot of questions. So we'll get to that. All right, so, you know, we're hearing about these, these implementations, these various projects, Arthur, Grid 3, UNICEF, various countries like Kenya doing different things. And what we're trying to do here at Oslo is we're trying to follow them as closely as we can. And we want to make sure that we're backing up these projects with actual, practically useful functionality, right? Abilities to calculate indicators, different types of analytics tools. And the Grid 3 stuff that Arthur was talking about was really, really useful for us. It really gave us a very practical look at these are the kinds of analytics that are happening at facility level, for example, right? He showed us there's a map, they're hand drawing maps at facilities. And so let's try to get that into DHS too. Can we support that, right? And a couple of things that we've done. This is all review, these have been mentioned before, but just to kind of put them into the context of this presentation. The first thing that we've added back in last year was this facility profile. Has anyone actually used this facility profile functionality? No, that's crickets, that's tough. The really cool thing about the facility profile is that you can click on any facility, you can put any kind of key stats you want on this facility profile, right? So you could put under five population. You could put BCG coverage, right? And all you have to do is click on it. It brings up this little panel here, pictures of whatever, and you get instant stats. So you don't have to dig around through, you know, other analytics to try to pull each one of these out. Each one of these is a separate bar chart or pivot table. It's a, hopefully the goal, even though you're not using it yet, is that it's a very practical tool, just a quick snapshots as you move around facilities. The other one is that we added population data from WorldPop and Grid 3 via the Google Earth Engine. And so this is really exciting because now we're all that population data that Arthur was talking about, that these folks who really know how to do it way better than, well, this is probably the best. We're able to make that publicly available for everyone, right, via the Google Earth Engine. So you can just turn on these layers and you can get that super granular population data for any countries in which they have produced it, which I think they've produced it for nearly, Maria, have most countries now? Yeah, all of Africa, so yeah. So that's available to everyone using DHIs too. The next couple of things is we brought in, you know, if you go back to Arthur's map, you know, they've drawn households at the facility level. They know where people physically live and what do we need to do? Well, we need to know where people physically live too and the good thing is that Google has a publicly available layer called Google structures or buildings and we're able to pull that in. So Google has some kind of machine learning where they're actually able to identify the outlines of structures and, you know, we can then lay over population data and we can say, these are where, this is how many people live there and this is exactly where they're living, right? So if we're doing outreach campaigns, door-to-door things, then planning, you know, super low level micro-planning, this is practically useful data. And you can see that they've done, yeah, quite a lot. They're still working on it, it's constantly improving by 64% of Africa, 516. Sometimes it does think that a particularly square tree is a building, but they're working on, yeah, all that kind of stuff, so. The next thing, and this was probably the biggest achievement and kudos to Bjorn and the whole team who worked on this, actually made it, you know, brought it to real life, is we are able to now have multiple geometries associated with a single org unit. And not just two, you can have three, you can have four, there's actually no practical limit, essentially. But this means that you can have a point for a facility and then also draw the catchment area. And then we can turn on those buildings layers, we can turn on those population layers, we can turn on thematic layers, different kinds of data, and we can put them all over each other, have data triangulation right there on the dash, on the map, that's the goal. So you know how many people live there, you know where they live, how many structures there are, and you have say like coverage indicators on top of that. So you know the particular areas that are bad, you know. You know, how do these catchment areas come? Well, it's, you know, different models, different methodologies. I hear that, I know that Coyt is on the call, he's leading up Crosscut, and they will be presenting for the app competition. Crosscut is an organization that uses, that will draw the catchment areas. They use things like road conditions, driving time, walking time, terrain, river crossing, land covers to draw the catchment area. So it's not like someone is physically going out in the field and walking around the perimeter of the catchment area. They're doing this, you know, in an intelligent way, using the available data that they have. And then what they allow you to do, at least Crosscut allows you to do, is you can go into their own software and you can manually change. So you can draw the catchment area to be what it is. But that requires folks to go out of DHIS too, which is fine, and go to Crosscut. But we're thinking about how can we bring that kind of functionality into DHIS too, as well in the future. And then also Grid3, as Arthur pointed out, Grid3 is another fantastic organization that's able to draw these kinds of catchments and then put them into DHIS too. In fact, Grid3 wrote the JIRA tickets that gave us the requirements for putting this kind of functionality into DHIS too, which is great. All right, I think, oh, and sorry, just how I always have to promote it. Everything's available offline on your cell phone, on the dashboard. So there's no reason that you can't have this when you go out to a rural health facility, right? Yeah, all right, yeah, this is just talking about the Google Earth Engine importer. Let's take a time, I'll skip over it so that we have a few minutes for questions. So questions for any of our presenters. I guess you can clap too if you need like a break, yeah. So maybe, are there any questions? Yeah, Colin, thank you. The author was talking about the district must be allowed to set their own population estimates. But he should remember we had that in daytime before 1996. And because each sub-district were getting money according to their population, all of them wanted a big population as possible. It was obvious to everybody, we added it together, it didn't add up. So even some countries like South Africa, there is just the political decision that the census data and the projections from the statistics of South Africa will have to be used. But even if you look beyond that, you can't allow too much freedom at the local level. It's got to add up, right? At the same time, I do also understand that yes, it's a problem that the few people that often deal with population dynamics at the statistical, you know, central level might not be, in fact, as professionals, they often don't want to go down to the lower level because their models become so uncertain, right? So they are afraid of what the other professionals are gonna say about them and they don't wanna go there. So I do see you have that kind of major challenge, but I'm just saying still the population, when you look at the national DHIS II or any other database using population estimates, they need to add up and hang together. I wanted to raise one other issue and that is, and I see this coming in country of the country. Some of you might remember the US, where the Trump administration tried to do various things to the latest US census in order to depress the number of likely illegal immigrants. Remember, the census is supposed to be everybody, whether you're legal or illegal or whatever, right? But they were trying to actually manipulate the rules. And we have a case now for Sierra Leone, the midterm census, where the World Bank pulled out the day before the census and saying they are not prepared enough. They pulled all the resources, but the government still went ahead. The population of the capital free town has dropped to half of what it was in 2015. And I asked people, you know, have somebody new part of all free town without anybody noticing, because I've never heard of any single capital anywhere, not even during World War II, where the population got cut in half in seven years. Problem is, of course, here it's political and everybody is convinced that the current government has depressed the number in free town because it's an opposition stronghold. So I'm just saying, we are, when we come to population dynamics, we're also dealing with this potential of political interference, right? For things that has nothing to do with health. And the dilemma is then, are we then allowed to have different numbers for health than what is used for elections or for budget allocations, et cetera? I'm just saying it's becoming a challenge for us as professionals. How do we navigate this difficult political terrain? Thank you. Yep, that's a very, very good point. Just a couple of things to respond to that. I think the important thing that we're presenting here is that we are supporting alternatives. We are supporting various pathways or kind of representations of population in DHIs too. But we are not hard coding anything. It's always up for countries to choose, to decide which one works for them. So Kenya has different approach than Zambia, who has different approach than Sierra Leone. And it's always like, we just need to support all the options. But you're right, it is extremely political and good luck with that. Like lead three will be officially recognized as a high quality source. Since they have just come on, they've just sort of become available. And I know they are working with several statistical offices like in Sierra Leone, right? But still, their methodology is not proven yet. And I would say the way to prove some of this is to go down a- It's well-published. It's a well-published methodology. It's not like they're making it up. I mean, it's, and it is verified, they do quality-controlled measures. Several offices who have been doing censuses in their way for the last 50 years are not likely to take a method coming from the university in Europe and say, oh, what you're doing is far, far better than what we are doing. It will take time before such a method becomes really recognized as an alternative source. Very good points. I agree with you. Put it, put it in there and use it for triangulation discussions. Yeah. I have a question. Comment, yeah, please. So when you are talking about relationship data, it's not a HHS2 issue. So it's, we, for example, maybe we use estimation. The last census is from 2003. And there was earthquake and the shock inside the population. So we made some estimation. And then, hey, so the estimation doesn't present the reality, what we did. We did, we choose to work at the community level. And with the community health workers, each community health workers can cover 1,000 families. And each community health workers make the census inside this area of work. So unfortunately, we don't have enough resources to cover all the country with the community health workers. So with the census done by the community health workers, we can give target to the facilities because the community health workers is linked to the facilities inside the community. So it's a local strategy to collect the estimation data. And we try to make it heavy. It's working. So we want more resources to have more community health workers and to give target to the facilities. Yeah, thanks. That's really interesting. I haven't heard of a strategy like that before. There's a question or comment in the back. Hi, my name is Lauren. I work for MSF. My team is thrilled about the multiple polygon shape point for Oregon. And as an MSF employee, we obviously don't serve an entire population of any one catchment area. And these population maps are great. I love Google Earth Engine. My question is if there are any plans to be able to have something like a raster file in an Oregon unit layer that we can visualize in the map app so that we can micro plan a map for ourselves. Do we have plans for that, Bjorn? Something called, you might have seen external layers which are raster layers, but they are sort of bad images when you import them. So do you want to extract data from them as well? Yeah, so you need a mapping server to host them, yeah. So, yeah, but maybe we should discuss this in the analytics session. We have an analytics, yeah, experts lounge where you can just come up and talk at them for a while if you need to. Okay, I think that's all we have time for. Thank you so much for joining. Hopefully it was a stimulating experience and stuff.