#32 Data Science at Doctors without Borders

Hi, I'm Maggie, VP of Curriculum at datacamp, and I know firsthand how overwhelming it can be to keep up with the rapid changes in data and AI. But staying ahead doesn't have to be hard. That's why our curriculum team has curated the best learning content to help you and your team stay ahead, covering everything.

From data and AI literacy for busy.

Executives to cutting edge generative AI for.

Developers, and everything in between.

Whether you're upskilling yourself or your team, we make learning practical, engaging and flexible. It just works. Power your data and AI transformation journey on datacamp.com.

In this episode of Data Framed, a Data Camp podcast, I'll be speaking with Derek Johnson, an epidemiologist with Doctors Without Borders. Derek leverages statistical methods, experimental design and data scientific techniques to investigate the barriers impeding people from accessing healthcare in Lahe Township, Myanmar. If you thought Data Science was all machine learning, SQL databases and convolutional neural nets, this is going to be a wild ride. As to get the data for their baseline health assessments, Derek and his team ride motorcycles into villages in northern Myanmar for weeks on end to perform in person surveys equipped with translators and pens and paper because they can't be guaranteed of electricity. Derek also uses Data Science to research the factors associated with the transmission of hepatitis C between family members and has helped to conduct studies in Uganda, Nepal and all this and more. I'm Hugo Baun Anderson, a Data Scientist at DataCamp and this is Data Framed. Welcome to Data Framed, a weekly datacamp podcast exploring what data science looks like on the ground for working data scientists and what problems it can solve. I'm your host Hugo Bown Anderson. You can follow me on Twitter Hugo Bowne and Data Camp. At Data Camp you can find all our episodes and show notes@datacamp.com Community Podcast hi there Derek, and welcome to Data Framed.

Hi, how are you?

I'm very well and it's great to have you on the show. How are you?

I'm doing great. I'm excited to be to be on the show today.

This is super exciting and I'm in Sydney, Australia at the moment. Whereabouts are you?

So I am speaking to you from Southern Miramar in a place called Daweh.

And are you in an office there or whereabouts are you?

Well, so currently in Dawe. I'm actually in an office. I'm in an office that's kind of attached to a health clinic here in Dawe. I work with Doctors Without Borders and we have been running an HIV clinic out of Dawe for the last 18 years. So that's kind of where I'm speaking to you from for the moment.

Okay, great. So what else are you doing in Myanmar with Doctors Without Borders?

So, so currently we have two projects. So we have the, the HIV clinic in Dawe which has been going, which, which has been operational for about 18 years. But we also recently have kind of opened up a project in Northern Miramar in Nagaland, in a town called Lahe. And by comparison that's, that's only been kind of up and running for about a year now. And we don't necessarily do a lot of HIV up there, but we're kind of basically kind of doing more of like supporting the health infrastructure up there and kind of doing more like general health support. So very, very, very different than what we do in Dalla.

Yeah, absolutely. So these are two very different projects that, that you're involved with. I'd love for you to abstract from that a little bit and let me know or tell us a bit more about your general role at msf. And for listeners out there, we'll be referring to Doctors Without Borders as Doctors Without Borders Medicine Sans Frontier and msf, which is an acronym interchangeably. So hopefully you can, can stick along with that. So Derek. Yes. Tell us more generally about your role at msf.

Yeah, so I, I'm actually currently the, the epidemiologist here. And so basically I help out with a lot of operational research and a lot of monitoring and evaluation. And so in Dawe, where I am right now, we're kind of, we're doing a bit of operational research on Hepatitis C. We are currently trying to kind of treat people for hepatitis C in D way. Recently the price for treatment has dropped considerably from tens of thousands of dollars for treatment down to a few hundred dollars to treat people. And so here we're trying to kind of like sort of scale up the treatment for hepatitis C because it's actually, it's, it's, it's kind of, it's more prevalent here than in other places in the world. But in Lahe I do operational research as well, but I do a lot of monitoring and evaluation. So the projects of like our doctors who go to like township villages, we're kind of just sort of monitoring the routine data that they collect there. But we also, we also have recently kind of done like a big survey, a big baseline health survey that kind of assesses, it's, it's township wide representative and it kind of assesses the health of like, of several different villages there.

Yeah, fantastic. And that's something we're going to get into a lot later in this conversation. But before that, you said you work as an epidemiologist, which of course involves a lot of statistics and thinking about data management, data provenance, all of these types of things. More generally, a lot of, A lot of things which intersect with, with data science. And I'm wondering how you think about data science, how you got into it originally and kind of what your. Your path has been to end up in Myanmar working for msf.

Well, so I kind of had a. It's an interesting path to data science. So originally I was actually a biochemistry major way, way, way, way back in undergrad, decades ago. But I kind of, I kind of realized. I got. I guess I kind of realized it very, very early and kind of I got lucky and realized that working in a lab, while very challenging and super rewarding, is not very social. There were internships that I worked in where I would. I used to run kind of like a mass spec machine, a mass spectrometer. And back in the day, these things were huge, like the size of a car. And basically I would just be sitting in a basement and like, just the only people I would see would be, like, occasionally my supervisor, maybe one other person, and that would be for like eight to ten hours a day. And I was like, I can't do this. I have to do something a little more. A little more adventurous. And so I kind of got into public health, but I still wanted to do a lot of science and I still wanted to do a lot of research. And so epidemiology is kind of. It's a, It's a great mix that you take a lot of data. You take a lot and you get to collect a lot of data, which is really fun, but you get to do that kind of in like a. In the field, so to speak. And it doesn't necessarily, like, you don't necessarily have to be going to, like, you know, the northern parts of, like, remote areas to do it. You can, you can do it in a clinic. You can do it with, like, just the data that people collect. And then you analyze it and then you kind of look for patterns or outcomes or how things are associated with one another. And so it's like, it's like puzzle work mixed with, like, being a little bit of Indiana Jones. And it's, it's great. I love it. I'm super, super lucky to be able to do this for sure.

And how did you actually get involved with MSF originally?

Initially, I kind of, I was in grad school and I used to do a lot of HIV work and STDs in general. And I ended up getting very lucky and getting a chance to go kind of to Malawi to look at. Initially I was looking at using a HIV drug called tenofovir as a first line HIV drug. And for people who know about antiretrovirals and HIV drugs, that's how old I am. Tenofovir became a first line drug in 2010 in Malawi. And that's a long time ago. And so I just kind of got into that and I was like, hey, this is really great. I can do data science, I can do my own surveys, collect my own data, analyze it, use a lot of statistics, but then get to travel and get to meet really interesting, really crazy people and kind of like. And it's just, it was really great. And then from there, when I finished school, I was thinking, well, what to do now? And so I kind of met somebody who actually worked on msf, has refugee boats in the Mediterranean, kind of like off of Italy. And he worked on one of the boats helping like refugees, basically. Like the boats that would come from northern Africa with like 2, 3, 500 people trying to get to Europe. MSF has these boats that kind of help basically just, just bring people in to make sure that they're safe, they don't die, they don't drown. And he said it was the most intense experience he's ever had. And I was like, I'll join msf. I'll see if I can like kind of look at the data and things they collect. And it turns out that because MSF has operations all over the world, they need a lot of help with data science because that, that knowledge that they gain helps formulate and help shape their policies. Much in the same way that when you work in a lab, you publish a paper, people are like, oh, this molec molecule is related to another molecule. We're gonna go with this. And that helps shape science. You get to help kind of shape like health policy of sorts, for sure.

And I love that you mentioned, you know, the data collection process is something that you're very passionate about and involved in and that's something we'll get to, particularly with respect to the baseline health assessment you've been doing in, in Lahe Township, because I think, you know, a lot of working data scientists and a lot of our listeners, when they think about data science and data collection, they think, you know, clicks and browser based stuff and stuff that's put in a database as opposed to getting there on the ground into remote areas and conducting the types of surveys you do, that's a little teaser for where we're headed. But before we get there, I think a lot of people have a sense of what Doctors Without Borders does. But I'd love a brief rundown from you on the work that Doctors Without Borders does from your position on the ground.

So Doctors Without Borders basically is a kind of an impartial and neutral NGO that provides healthcare to anybody, particularly in humanitarian crises, be that natural disasters like earthquakes or typhoons, or more like man made disasters like in war zones, they provide humanitarian relief. And so what Doctors Without Borders does is they go to these areas and they basically provide humanitarian aid. And so most of the projects, for example, that I'm part of tend to be very short term. Uh, so before I came to Miramar, I was actually in a refugee camp in northern Uganda doing actually a big household survey on mosquito nets, but in the refugee camp itself. So we were just kind of like living there and doing, and doing work there, but it was short. I was only there for about four months. And so whereas compared to a lot of development work where they can be in an area, for example, like, like Red Cross or USAID or there's plenty of development NGOs that can be in an area for years, for decades. So that's kind of, kind of what MSF does, is provide humanitarian aid to places and crises.

Great. So given that context around what is essentially a mission statement of msf, what are the biggest challenges that the organization faces that data science and analysis and analytics can help to solve?

Basically getting to know, I would say in my experience, just getting to know the context of where a health clinic works. And then how do you, how do you take that knowledge and then, and then inform the policies that like, bring your, your decisions forward. Because MSF works in so many different areas in so many different countries. It's, it's critical that they know about the context that they're working in. And every place is different. So when I was in Uganda, that is completely different than what I'm doing here now at Miramar. And the people who kind of help decide what to do in terms of, like, in terms of projects and in terms of policies, they need as much information as they can get. And so as an epidemiologist, my job is to basically do monitoring and evaluation of some of these projects and then putting together kind of like internal reports that people can read and then be like, oh, okay, like, you know, the, these people that are taking hepatitis C drugs, for example, maybe 25% of them like, don't complete their treatment. What are we gonna do about it? Or in the case of like, like when I was in Uganda, we were looking at passing out mosquito nets to people in the refugee camp. But the problem with that also is that a lot of times people will try, will kind of tend to misuse the nets. So a lot of times you'll use it for like fishing. And I didn't know this, I actually didn't know this until I started working there, until we started collecting the data to look at this. A lot of times people, the nets are really strong and they are like, they're just perfect for catching fish. And they're already in like a little basket type shape and you just kind of throw it in the water and like, and then they're just, they're actually really good for, for fishing, which is a total misuse of the nets. So the, the, the survey that we did there kind of helped us form these educational campaigns. And so we would actually go to different parts of the camp and be like, hey guys, don't go fishing with us. Like, this is, you know, like to, this is more for mosquitoes. And so it really helped get people to understand, like, to use the nuts better and that that's the purpose of the data. So you get the data and that actually helps shape what everybody is going to be doing. And so it's kind of, it's, it's cool in the sense that you actually, if you, if you stick around a project long enough, you actually get to see your results be translated into action. Which is something that I know when I used to work in the States, sometimes you could publish a paper and you, you just never see the results. And then it gets really depressing when you realize that only like three people have cited your paper and you're just like, oh, okay, like what's the point?

Yeah, and as I stated before, it really seems like data science in, in, in this form is so different to what people think of when they think of tech data science, for example. So I, maybe you could speak to the types of differences that are, I think, dominant in your mind with respect to this.

Yeah, it's, it's, it's different. So like, it's funny because the, the statistics, at its core, the statistics and study design are very, very similar and very much kind of the same. And so which, which I think is like really fascinating that you can take your survival analysis or like your logistic regression or your like cluster designs and you can apply it to like schools in England or you can apply it to like villages in like northern Miramar. But at the same time, the data, I'm getting used to finding messy data and it's smaller data. So while nowadays people are used to having these gigabyte sized data sets with hundreds and thousands and thousands of observations, sometimes here, a lot of times you might only get a couple hundred people. And so you're not actually kind of working with big data. And then when you have problems with your variables, like if you have a lot of missing variables, for example, you might have to start getting creative and do things like imputation or you might have to like drop a question in general, just be like, oh, it didn't quite work out. And so it's very different than using larger data sets.

And even the data collection process is very different. Right. In terms of what you do when you're running surveys, whether it be with pen and paper and then putting them into your spreadsheet or computer program or database later.

Yeah, the. Collecting the data, I love it. That's the fun part, that's the best part of it is that when you actually get to go to the field, like for example in Lahe, we actually got to take motorcycles and go from village to village and it was kind of like off roading, like through the mountains of northern Miramar to go to village to village. And then you do these household based surveys when you get there. And so you're like right there to collect the data and see how it's collected and see what questions work, what questions don't. And it gives you a better feel of where the data comes from. And then it's kind of a, it's nice actually like that you get to kind of actually see the birth of your data. For lack. I mean, that sounds sappy, but it's kind of like how your data is.

Oh, that's a wonderful.

Yeah, like, yeah, probably the most nerdy thing I've ever said.

You understand the data provenance in that sense. And you, you know what all your units are. It's not like you're being handed a CSV or pinging an API where you may not actually have the correct assumptions about, about your data.

Exactly, exactly. Yeah. And it's interesting like I've, I've in msf, I've done surveys where we use electronic means of data collection. So I've used both EPI info on a smartphone and a program called Dharma. Dharma is a, It's a service that's actually started by an ex MSF staff. She worked in the Ebola outbreak, not the most, not the one that's currently going on, but the one a couple years ago and was like, we need better data collection tools. We need like almost real time information about what's going on because things and outbreaks like Ebola change so fast. And so they were actually an epidemiologist and they actually ended up just kind of creating this program that allowed for not only data collection, but it also projected just trends and statistics on a dashboard. And it ended up being, it was really good. But at the same time also like in Lahe, because our teams are out for a week at a time and it was very rural and rugged, we resorted to paper based surveys which are horrible. I mean, I think from here on out I think I'm just going to have to go the electronic data collection route. Like, because with paper based surveys there's no checks and balances. So if somebody puts down like if you're interviewing somebody and they say their age is really like 10 years old, but you put down 100, there's no like little automatic check that will be like beep, you know, like, oh, that doesn't make any sense.

No testing, no, no ability to test your data.

And then also you have to enter the data which takes double the amount of time. So not only you collecting the data, but then you have like get somebody to enter the data which takes forever. So electronic data. I'm glad MSF is adopting smartphone tech for like entering and collecting data. It makes things a lot easier for sure.

But as you said, if you're in a region where you may not have access to electricity for a certain number of days, you know there are only so many battery packs you can carry, right?

Yeah, yeah, that's the, that's the problem. And so we, it was, it was decided finally that because we're spending about a week in the field at a time, so the Lahe survey went over the course of about two months and people would go out for a week, come back for a day or two to rest and then we would go out again for a week. But these are, these are areas with no electricity, no cell phone coverage for the vast majority of it until you start getting to the border of India. And then India has some pretty great cell phone coverage or at least in that, like that region. But you, you really just, you can't charge your battery pack or your phone. And then if you actually drop phone or something happens to your phone, you lose a Whole week of data out of a two week survey process and it'd be very hard to go back and redo it. So we decided to go with, with paper surveys and then we just had like waterproof folders. So it has its pros and cons. I tend to be a little more tech oriented and I was like, but we could have, could have had solar chargers and we could have had like four battery packs. And then, you know, we could have gone out maybe two days at a time. But in the end we ended up doing paper for sure.

We'll jump right back into our interview with Derek Johnson after a short segment. I'm back here for another installment of Insights from Computational Education with Neil Brown, who works as a research fellow in Computing education at King's College London. Hi, Neil.

Hi, Hugo.

So, Neil, a couple of questions that I hear all the time are, can I program? Am I any good at it? And is it all too technical for me? For example, what are your thoughts on these types of questions?

A lot of people are interested in whether we can predict success in programming ahead of time. So universities tend to be interested in this kind of research because if you can predict who's likely to drop out of your course, you might sort of not want to take that person on the course. Careers advisors tend to like this kind of thing too. You know, you find that certain people say if they like maths or if they're good at maths, they're good at programming, then it's a pattern that they can recommend to people.

So what are the outcomes of these research studies?

So essentially, like I say, lots of people have looked into it and just none of them have ever really found any strong links where they can predict success in programming based on other factors, whether that's sort of achievement in other subjects, whether it's gender, all sorts of different things. A couple have found some weak associations with things like maths, but they often fail to factor out general academic achievement. So if you're good in one subject, you do just tend to be better at other subjects overall. So you have to factor that out if you want to look at pairings between specific subjects.

So wait, what you're telling me is that there's nothing that can reliably predict success at programming?

No, it doesn't look like it. And sometimes you sort of feel that the researchers doing the studies are a bit disappointed and they maybe see if they can find some a sort of statistically significant result to put in their paper. But I think actually if you just spin that round it's actually quite a positive message. So there's nothing we've found which we can reliably predict ahead of time whether you're good at programming. So it doesn't matter if you're good at maths or bad at maths. It doesn't matter whether you think of yourself as arty or whether you don't think of yourself as arty. We've no idea whether you'll be good at programming or not. So why don't we just get everyone to try programming and then we'll actually see if they're good at it.

Yeah.

That's awesome. So I still think there is a general impression out there that some people get programming while others don't, although we don't know who actually be good at it. Is it right to think that some people are good at it while others flounder?

Yeah. So if you've ever done deprogramming teaching, you may recognize this. You often feel that some people in the class just get it and they're like, racing ahead and they're fine, and other people are just really stumped and they just don't seem to get to grips with it at all. And of course, there's always a performance difference in any subject. So whether, like, say, with maths, art, or whatever, you always have some who are better at it than others. But there is this general opinion that computing is really split between these two groups of get it and don't get it. So there was a study looking into this where they were interested to see whether computing does or doesn't have a bimodal distribution of marks. So what they did was they came up with some mark distributions and they gave them to computing teachers and said, do you think this is bimodal? And they actually fed them a bunch of unimodal distributions. But when the computing teachers thought it was from a computing class, they were more likely to say this is bimodal than when they thought it was from a class from another subject.

Wow. So it seems it's actually in our minds.

Yeah, it seems like it might be. And so not only that result, but if the teacher themselves thought that it was split between some people get it and some people don't, they were more likely to say that the distribution was bimodal. So it seems like it really may be in the minds of the teachers.

Wow, great. So a takeaway is that potentially anyone can program and there's no separation between those who magically get it and those who don't. I love it. That's Worth bearing in mind. Thanks, Neil and I look forward to chatting again soon.

Yep, speak to you soon.

After that interlude, it's time to jump back into our chat with Derek Johnson. And you very kindly sent me through the study protocol for this by way the baseline health assessment in Lahore Township. And there was. So I was, my eyes were stuck to this, this PDF while I was reading it. There's so many interesting things in there, particularly with respect to the data, data collection process. And one thing that stood out to me was that it actually said that the local Nagar dialect is a non written language. So I presume you had translators there who were speaking Naga to the locals and then writing, writing down the data in a different language or can you just give me the rundown on that?

Yeah, no, it's, it's, it's, it's interesting and it actually comes up a lot when we do surveys in remote areas, particularly areas with several different languages or different dialects. What kind of ended up happening is we recruited, we had about 34 people for this study between the drivers and the data collectors, about, about 12 data collectors and the rest were drivers because we had to carry supplies on motorbikes. So we actually needed quite a bit of motorbikes that just carry like tents and to car, carry food and to carry water and things. But so the data collectors themselves were actually recruited like locally and they, they speak Burmese like, they speak, they speak Miramar, but when you get to like parts of Naga land, they actually, they, they kind of like what you were saying. They only they, they speak like a Naga, Nagamese, which is like a, an umbrella term for a lot of like local village languages. And a lot of these languages don't, are not written down. So the paper surveys themselves were translated into, into Burmese and the people we hire, they spoke like Burmese, but they also spoke like their own village dialect. And so they'd be speaking their own dialect but then translating it back to Burmese. And so like, luckily usually that causes a lot of problems in surveys, particularly if you're acting about complex things or behavioral questions. But this survey was a lot of health questions and so a lot of it was things like have you been coughing for over two weeks? Yes. No. Is there blood in your cough? Yes or no? So we didn't have to worry too much about the translational differences. But, but that does become a problem. That definitely does become quite a problem if you start asking about more like sensitive questions. So if you start asking more about like in a lot of refugee camps, if you start asking people about like why, why did they, why did they flee? Like why did they run away? That's actually quite a different question. And it's open ended. And that requires that you actually have the proper translator or people speaking the same language as the person you're interviewing. Otherwise your data gets to be a little funky and not representative what you're doing.

Absolutely. So now I'd like to kind of step back a bit and talk about the baseline health assessment project in LaHaye Township as a whole. So maybe you can give us the rundown as to the motivation behind it and how it played out in practice. Yeah, and what, what is it?

So when MSF decides to open up kind of a new project, they first do what's called like an exploratory mission. So they'll have a couple doctors and a couple logistics people. They'll go to an area and they'll do like a quick assessment to see if there's like any sort of like glaring health needs. So in Lahe, for example, they went up there and they found that the access to healthcare was incredibly poor. Most people couldn't access a health clinic if they wanted to. And at the same time there was a lot of just basic infectious diseases that were going untreated. So there was this idea to okay, now that we know that there's this general need, we need to be more specific. And so usually what happens is after an exploratory mission they'll do kind of like a baseline health assessment, which is a much more formal scientific health assessment of like the needs in a particular area. And that's kind of where the epidemiology kind of comes into play because you end up actually kind of designing a survey to collect the data for that area and then how to collect and then what type of information you want to collect. So the baseline health assessment had a couple different parts. So there's like your basic demographics, there's health seeking behavior, like where do you go if you are sick, how often do you go to a doctor? But then there was also assessments on malnutrition for children. So we assess the nutritional status of children under 5 years old and then we also try to assess vaccination status. And so there's a couple different components to the survey was pretty long, but it's a very cursory, very kind of like general overview. So one of the things we found in this for example, is that there's a lot of respiratory illness and we, because this is all like just self reported, just things that they, they mentioned we try to get to like what do they think it is, but people don't go to a hospital and there's, there's just no way to really tell what it is unless you go to a doctor. So we ask all these questions like, about it, but in the end it all becomes kind of still like fairly, fairly basic information captured in a very specific, like, kind of scientific way. And so for example, for this particular baseline survey, the villages in Naga, like in Lahe, they're very, they're spread out and in the mountains. So it's very clustered. So a village might be an hour and a half drive away from the next village, but really it would only be about like 30km away. It just takes you a long time because you're literally driving on like a dirt path. And so the problem is, is like within the Lahe township there's about 107 villages. Like how, how do you sample enough villages to represent Lahe township? And then within those villages, how do you sample enough houses to make sure you're actually capturing the information in the village? So you end up with this like two stage cluster design to your survey, which is pretty, pretty neat actually. And it's, it's kind of a, it's, it's different than health surveys I've done before. So like in Uganda, it's a big refugee camp and it was split into, well, where we were in northern Uganda, the refugee camp was split into six different parts. And we almost pretty much did just sort of like random sampling in each of the parts. So we didn't do as much of a clustered survey design. It was a little easier to do. So that's kind of how the data collection process happens with that. And it was kind of neat because it actually, we kind of mixed a little bit of old tech and new tech to this. So the who, the World Health Organization, has recommendations for cluster designs, dating, how to sample it, dating all the way back to like the 70s where if you, you basically when you get to a village and you're trying to randomly select villages, you throw a pen up in the air and then wherever it lands, where that pen is pointing, that's the house you go to. Because they didn't have like all this like, like smartphones and you know, all this tech. And now what we did is it's kind of nuts that you can actually get satellite imagery of different villages in Lahay Township. And so we put that into QGIs, drew the borders around the village and randomly dropped points into that picture and they were like, okay, that point is close to this house. That's the house you're going to go to. And so we kind of, instead of doing the pen method from back in the 70s, we actually did this like GIS way of selecting the households within the village. So it was kind of, it was kind of cool actually. It was, it was pretty neat. I got to work on some, some GIS work, so. Which was pretty good.

That's really cool. So how many people are there in Lahe Township and how many villages and how many people did you end up interviewing over? What, what time scale? I mean, I've just asked you four questions actually, so that's, that's far too much. But I'm just getting kind of trying to get a general idea of quantities here.

So it's actually. So Lahe Township is one of three parts of the Nagaland area that's in Miramar. The majority of Nagaland is actually over the Indian border, but there's three, three sections within like the Miramar side and, and Lahe Township is one of them. And Lahe township itself, it's, it's in the mountains and there's not a lot of people that live there. There's about, it's about120,000 people. And Lahe Township or Lahe town is kind of the center of it and that's about 3, 000 people. And then there's about 107 villages that are in the township. And these villages will move every couple of years. So what? Something we actually had to do that we were told to do beforehand is when we would select a village, we actually had to send a team out there to make sure that village was still there. And it happened Once we selected 30 villages to represent this township. It's called ground truthing. But you just go out there and you're like, okay, like does this village exist? Is there enough houses to do the survey? It happened once where a village actually kind of like relocated like a 10, 15 km away because they had water problems. So the stream that was kind of providing water to this village dried up a couple like a year or two ago. And so everybody sort of just migrated pretty much to find water. And so that's kind of kind of the context of Lahe is that it's mountainous and it's sparsely populated. The baseline survey itself took about two months. We did it in 30 villages and we chose those villages based on population size or as close to population size as we could get. So we had basically like kind of the size of Potla village in the Miramar census. And then the chance of being selected in the village, like the village to be selected was weighted on this size. And so for example, Lahe town itself was chosen twice because it's, it's like up 3,000 people, whereas like the next biggest town we went to had about 700 people. And so we chose the 30 villages and then within each of those villages we did 30 households.

So that's around 900. Are you surveyed in, in total?

Yeah, about about 900 houses. The average household size came out to be about, about seven people. The like inner quartile range was between five and nine people per household. They do, there's a common pract like intergenerational living, like kind of joint families. So you'll have like, like younger children with the mom and the dad and then you'll have the grandparents and sometimes you also have like aunts and uncles that live in a house as well. So household sizes can be actually quite big. So it ended up being a little over 5,000 people that would represent. We didn't interview all 5,000. You just took one one member of the household and then just asked questions about everybody to that one person. That way not everybody had to be present.

How do you choose which member of the household?

It actually came down to like a little bit of like cultural acceptance. So usually it was like kind of like the, the father of the household or like kind of like the male figure. A lot of times though people, people do a lot of agricultural work and so if it was kind of in the middle of the day, a lot of people would already be out in the field. So what ended up actually happening more often than not was a lot of times we, we did get a lot of like female household heads that would answer for the survey. We also got a lot of grandparents that kind of answered for the survey. The only requirement really to kind of be considered ahead of household was that you had to be over 18 and you had to have been able to answer questions for everybody in the household. So that was the only inclusion criteria to be kind of deemed the head person to answer the survey.

How are the insights that again from this baseline health assessment turned into actionables and deliverables for msf?

Currently actually, because we found a lot of respiratory illness with this survey, we're actually helping to put in a what's called like a gene expert machine in La Hay Township Hospital. And so it's a. To help to test for tuberculosis. So the kind of like the old school way of Testing for TB is you hack up your lungs and like, and then you spit your sputum onto a slide and then you stain it and then you put on a slide and you have to have like a trained health professional to kind of like look at that slide and be like, oh, that's your microbacteria. Like, you know, you have tb, which is not very sensitive or specific. It's got a sensitivity, specificity of in the 50s or 60s. So it's pretty crap. But with this technology, the GeneXpert machine, it's highly accurate and it basically can give you test results in a couple hours. But it's expensive and it requires a bit of a constant electrical supply. And we wouldn't have put it in unless we knew that there was a high amount of, of respiratory illness within the area.

And is this type of actionable developed from the insights gained from the health assessment, similar to other projects at msf. For example, your work in Uganda in.

The refugee camp, it's pretty similar. So a lot of the operational research MSF does, it's all geared towards doing actionable results. So in Uganda, we, one of the surveys we did was on mosquito net use because there's a lot of malaria at the time and malaria is cyclical. So when, when the rainy season kind of came about, that's when you would see spikes in malaria. And so we did this assessment on bed net use right before the rainy season. Be like, hey, does everybody have bed nets or do, how do you use bed nets? And then the results from that kind of led to a mass mosquito net distribution and also the educational campaign to make sure you're using mosquito nets. Right. So we ended up actually, a couple months after survey, ended up actually passing out a whole bunch of mosquito nets. It ended up being close to like 13, 14,000 mosquito nets for the area, which is, it's a great way to see your data in action. And it's, it's, it's great. That's probably like the number one thing I like about doing epidemiology for MSF is that you get to, if you're lucky and you stick around long enough, you get to see your results turn into something actionable.

So, Derek, something I've been thinking a lot about recently is how we don't necessarily have good models for, or we haven't settled on, you know, a global model for how data scientists, statisticians, statistical modelers are embedded in organizational structures and businesses. So I'm wondering how, you know, data science is embedded in, in the organizational Structure of MSF data has always kind.

Of been there in forms of like, kind of like internal reports, but the operational research aspect of it is actually kind of somewhat new. And it's, it's, it's funny because you can kind of look at like in the early days the MSF never really kind of like published a lot of like scientific data. But then all of a sudden there's actually kind of became like two big data data hubs in msf. So there's episodes in Paris, which is like a big repository of a lot of the data that MSF collects. And they also provide a lot of support when it comes to designing various studies just in terms of like study design. And then there's the Mason unit in England which does the same thing. The Epizons tends to focus a lot on the French speaking countries and the Mason unit tends to take a lot of the English speaking countries. But these are two big data repositories. So MSF has actually become quite, quite serious in the last five to seven years, more and more like seven years on how data is pretty much the best way to inform your policy decisions. It's hard to argue with the numbers and before, especially with a lot of the work that MSF does. A lot of it can be a little controversial and a lot of it can be quite risky. So when you have the hard science and the hard numbers and data there, that adds a lot of weight to your policy decisions. For for example, with the Ebola crisis that happened a couple years ago, it was very, very important to have kind of real time data of where cases were clustering because that's where you would send your health workers, not just your doctors but like educational people. So like your health promoters to be like oh like if somebody is bleeding, like don't, you know, bring them to a doctor. But like don't, don't like touch the blood and like it's a good way to prevent the transmission of, of Ebola. But you have to do that fairly fast because like Ebola like the, the incubation time is only a handful of days and then the more like it, the, the mortality like rate of it, it actually would kill people in, in about two weeks time. So you had to act really fast. And so the data that you collected was the best way to, to inform your decisions. Otherwise you'd just be arguing over what you heard from people in the village at the time and it would take, take forever to do something.

We'll jump right back into our chat with Derek after a short segment. Now it's time for a segment called Blog Post of the Week. I'm here with Emily Robinson, a data scientist on the growth team here at datacat. Hi Emily.

Hi Hugo. Thanks for having me on. I want to share a great post from Mikhail Popov on the difference between data scientists and data analysts. These titles are common in industry, but there's not a clear, agreed upon definition of each and how they differ from each other. So Mikhail put out a call on Twitter to ask his followers what they thought. He got over 30 responses, and this post summarizes the main themes.

So what themes did Mikhail see emerge in the responses?

Well, one is that data scientists are data analysts who can code. This viewpoint takes into account that graphical user interfaces, or GUIs such as Excel limit what you're able to do. When you program in a language like R or Python, you greatly expand the breadth of problems you can solve.

So I like this distinction a great deal and think it's a good one to keep in mind. It actually reminds me of Hadley Wickham's talk you can't do data science in a gui, right?

Another view is that the ability to do machine learning is a differentiator. While not all data science projects require machine learning, being a data scientist does require that skill set.

Ah, this is an interesting one that I think is valid, but that I also disagree with, particularly as I don't always think the distinction between machine learning and statistical modeling is all that useful. If anything, I'd personally reframe this distinction as data scientists understand enough mathematics to be able to build robust models, whether they be called statistical models or machine learning models. But maybe that's semantics. What other distinctions did Mikael see emerge?

Well, finally, instead of focusing on data analysts versus data scientists, one can distinguish between Type A and Type B data scientists. A stands for analysis. Type A data scientists have strong statistics skill and the ability to work with messy data and communicate results. B stands for build Type B Data scientists have very strong coding skills and focus on putting machine learning models such as recommendation systems into production.

Yeah, I like this different formulation of the question and actually had a great discussion about the differences between Type A and Type B data scientists with you, I think on the podcast when you appeared last year and maybe even with Robert Chang. He's definitely, definitely written about it. Anyway, what then are some practical considerations for a company when choosing a title?

Great question. This was also addressed in this post and full disclosure, I'm the one who applied with these thoughts, at least for the New York and San Francisco tech scene Data scientist roles command a higher salary. You also may lose talents if you have the data analyst title. People care about titles, and data scientists is generally seen as more prestigious. If someone finds a similar role at another company that has the data scientist title, they may prefer that role.

That's interesting. So how are different companies handling this?

Many big tech companies, including Spotify and Facebook, are shifting away from data analyst titles to data scientists. Lyft actually just wrote a post about their decision to do this. They also changed their data scientists to research scientists. In some cases, they'll add a specialty after the data scientist title. For example, Airbnb currently has openings for data scientists, analytics, data scientist algorithms, data scientist inference, and more. This way you can have data scientists as a big umbrella term, just as software engineer is, while specifying in the title and job description what aspect of data science they'll work on.

Oh, this has been a great discussion, Emily. What should our listeners take away as the core conclusion?

I really liked Miguel's closing thoughts. He emphasizes that while this shows there's no one agreed upon definition, having that is not necessarily the goal anyway. Rather it's to make sure to be specific when writing a job description and not assume the title is self explanatory. I would also add it's good for job seekers to remember to cast their net widely and not just look for positions with data scientists as a title. You may find roles you're interested in that are called product analyst, business analyst, machine learning engineer, or more.

Thanks Emily, and I look forward to speaking again soon.

Thanks Hugo.

Time to get straight back into our chat with Derek Johnson. So something you mentioned earlier was that you see more and more abilities for tech to be used in the work MSF does and the type of surveys you've been doing. What else is there in the future of data science at msf which isn't there or hasn't been discussed in this conversation?

Well, so something that I kind of want to see be used a little more, particularly when it comes to health promotion, and something that I'm kind of playing with a little bit is network analysis. So like social network analysis, particularly for health promotion, in addition to outbreak outbreak epidemiology, which is that's kind of what most people think about when they think about like a network analysis and infectious disease. But it's also a great way to find who in a community has the most influence and who do you really want to target with like your health promotion behaviors. So if you're trying to kind of get people to like, like something fun and something silly. Like, like, brush your teeth more. Who do you really want to talk to the most? Like, do you want to talk to the children? Do you want to talk to the moms? Or maybe, maybe it's the grandparents that happen to have, like, the most sway. And by doing like, a network analysis, you can see who has the most connections and then you can see the strengths of those connections. And rather than like, target an entire village and be like, hey, everybody, brush your teeth and have like a big, like, rock concert on it, you can just target, like, a handful of people, knowing that they would spread the message to a good chunk of the community.

So it almost hurts me to say this, but what we're looking for are influences, right? This is influencer culture.

Yeah, yeah, yeah, yeah, exactly. It's kind of. It's. Instead of taking like, like the influencers of, like, fashion and stuff, you can kind of give it like a, like a little bit of a health bed and be like, all right, like, maybe, maybe if this, like, the cool guy, like, were to brush his teeth in public, maybe he has the biggest influence. But, like, it's getting to, like, know who has these influences in a way that it's really hard to do with traditional data collection methods. So you can find, like, with a lot of, you know, like, odds ratios and P values and just you can find the strength and magnitude of associations for like, one variable and another, but you can't really tell, like, okay, like, that's, that's just the relation between those things. But, like, what outside of it influences those factors as well.

For sure. And I like the idea that you're thinking about network theory and network analysis in this respect because as you were saying, network analysis is thought about a lot in terms of infectious epidemiology, but we know that it has a huge role to play in even non infectious epidemiology. You know, I think one of the, one of the common examples is people don't contract obesity from each other, but if you have a network and you're connected to more people with obesity there, you've got a higher likelihood of having it yourself at some point. Standpoint.

Oh, exactly.

Yeah.

That's actually a really great, great example of it is that, yeah, like, obesity and dietary habits are greatly influenced by your friends and network. I can definitely speak for experience. I used to, to smoke cigarettes for about, like, six years, seven years, and all my friends smoke cigarettes. And it wasn't really until I kind of like, moved from. I grew up in Boston, but moved From Boston to Philadelphia, just got like an entirely new friend circle where nobody smoked and everybody was healthy. And I was like, huh, maybe I should quit smoking. But I mean, if I never, if I never changed my friend circle, I, I'd still be smoking two times a day.

So we haven't talked much about, you know, the technical, technical stuff that, that we love so much. You, you did mention that for sampling, you do a two stage cluster sampling methodology. Also been discussing network analysis, but, but I'm wondering what's one of your favorite data sciencey techniques or methodologies? Just something you love to do when playing with data.

Oh, that's a good, that's a great question. That's really good. I don't want to sound lame and kind of basic, but logistic regression, I know that's like super simplistic, but like odds ratios are great because everybody understands them, they're easy to calculate. And you'd be surprised how much data you can get with a binary response. You just like, you know, like, do you like mangoes? Yes. No, like, do you use condoms? Yes. No, like, you know, do you, are you an injection drug user? Like, did you go to a doctor last week? Yes. No. But like, you know, like anything that's kind of like binary. And then by extension you can do multivariate logistic regression type, though I'm not too blanking on the term here. But basically you can also do like odds of things with more than, more than a binary set. You can do it with like different levels of categories as well to get different odds. But that, that works out really well actually when you not only like collect and analyze data, but when you present it. So if data science is really gonna kind of like lead the way in helping to change health policy, in this case, you have to be able to kind of communicate your results to people who might not be as minded. So while I can get into like coding about like R and show, you know, all this like data management and things, and it's great and it like, I know the project coordinator for la, for example, that doesn't like it at all. Like, and unless you can like kind of put it in a graph, like their attention span kind of wanes after like three minutes. And so yeah, logistic regression is probably a great way to communicate a lot of results in my, in my experience.

I agree. Yeah, I agree completely. And as you know, I've said this time and time again on, on the podcast, but for people who are non technical, you can show them that a 10 increase in this feature results in this probability in, in the outcome. And in that sense it's interpretable, which gives you massive gains.

Yeah, exactly. And it's, and that's. That, that, that like, kind of what you just said is like a great way to just explain it. It's, it's short, easy to understand. You know, just something like that. Like the odds of this risk factor increasing your risk of contracting HIV are such and such. However, behind the scenes is the way you sample, the way you collect that data to make sure your analysis is correct. And that gets to be quite complicated. For example, the two stage cluster design thing using randomly assigned GPS points, that's somewhat complicated and it takes a little bit to do. But in the end you kind of like, you try to distill it down to things everybody can kind of digest.

So, Derek, my last question for you is, do you have a final call to action for all our listeners out there?

Yeah. Yeah, I guess, I guess I'm a little different than a lot of people that come on this podcast in the sense that I don't always use massive data sets or I'm not always kind of like crunching out numbers and very clean data. But at the same time, like, I think, I think people should get into data science and get into kind of more, more field work out there. There's definitely a place for people to get out of the office, collect their own data, analyze it, and then actually have like actionable results. And so don't let the fear of sitting in a cubicle somewhere just pumping out code dissuade you from doing data science. Data science is for everybody.

Fantastic. Derek, it's been such a pleasure having you on the show.

This is great. I had a real fun time. Really good.

Thanks for joining. Our conversation with Derek about his work with Doctors Without Borders, we saw the unique challenges encountered on the ground by Derek and his colleagues in performing baseline health assessments in Laha Township, Myanmar. These range from small sample sizes to the need for robust experimental design to basic challenges in the data collection process which we don't really think about in tech. Practical challenges such as lack of electricity, forcing surveys to be conducted with pen and paper, the unique problem of working in non written languages and riding motorcycles to nomadic villages that have moved by the time you get there. We also saw that insights from such baseline health assessments are turned into actionables, a paradigm of this being the. Because with this survey they found a lot of respiratory illness, they're putting in GeneXpert machines at Lahe Township hospitals which will help to test for tuberculosis. Make sure to check out our next episode, a conversation with Max Kuhn about data science in the pharmaceutical industry in general and at Pfizer in particular. Max was Senior Director of Non Clinical Statistics at Pfizer for over 10 years, is now a software engineer at R Studio, creator of the Carrot package for machine learning in R, a data camp instructor, and an open source beast. He's also hilarious. What do data science and statistics at Pfizer even look like? What is non clinical statistics? And what are the biggest challenges in pharma that data science can help to solve? All this and more next week. I'm your host Hugo Bown Anderson. You can follow me on Twitter HugoBown and Datacamp. At Datacamp you can find all our episodes and show notes@datacamp.com community podcast it.