Transcript#

This transcript was generated automatically and may contain errors.

Hey there, welcome to the Posit Data Science Hangout. I'm Libby Heron, and this is a recording of our weekly community call that happens every Thursday at 12 p.m. U.S. Eastern Time. If you are not joining us live, you miss out on the amazing chat that's going on. So find the link in the description where you can add our call to your calendar and come hang out with the most supportive, friendly, and funny data community you'll ever experience.

Can't wait to see you there. All right, with that, I am so excited to introduce our featured leader today. Her name is Allie Warren. She's a data science and informatics specialist at Washington State Department of Health. And Allie also gave a virtual talk at Posit Conf about NowCasting, which I absolutely loved. And she got so many questions, she could not answer them all live. So we're so happy to have her here. Allie, it would be amazing if you could introduce yourself, tell us a little bit about what you do and something you like to do for fun.

Thanks so much. Yeah, I'm Allie. I'm a data science and epidemiologist with the Washington State Department of Health. My thoughts are my own. My background is in computer science. And before being in public health, I worked in cancer research, like preclinical on the computational side and like single cell genomics. That's where I got my start in R. And then moved into public health during COVID, helping build like data science pipelines and helping scale for an amount of data that wasn't expected and doing record linkage and things like that. And then more recently, have moved more into doing modeling and analytics, as well as kind of building on training and community practice, helping train our local health jurisdictions in R and building out like a national network talking about AI in public health.

Perfect. And something you like to do for fun. I like to hike and I like to rock climb. And I feel like you have some maybe personal projects around hiking and climbing that we can talk about too, which I'm excited to dig into because we love personal projects here. At least I do. But I wanted to also touch on the topic of your Posit Conf talk, which was now casting. If you could give us a little recap of what you talked about. And by the way, if anybody is registered for Conf or like on the Conf portal, which everybody could do for a for virtual for free, all the talks are still there. So you could still go watch Ali's talk. They won't be up on YouTube for a couple of months yet. But Ali, could you give us the rundown on now casting in public health and why it's important?

NowCasting in public health

Yeah. So now casting is the idea of can we estimate current counts of some reporting data based on what we know about like reporting delays. So this is a big problem in public health because you'll get reporting data that takes a while after someone gets sick. It has to go maybe through testing or it'll go to our local health jurisdiction first or requires additional follow-up and investigation. So there's this long delay between when someone actually gets sick and when we actually know about that case. And it causes this problem where it will always look like cases are declining, even though that's maybe not the case. So I was using now casting to model that delay for pertussis or whooping cough to try and better understand where an outbreak of whooping cough was going and what we expected it to look like.

It was very, very interesting to me because in your talk, you showed some really misleading plots, right? Like it's a plot over time where we're looking at the levels of something like pertussis in the population and it looks like it's going down. If somebody just looked at it at a glance, they would be like cases are going down. We're headed in the right direction. But it was just because there's a delay in collecting the information or processing it or whatever. And eventually that's going to change, which sort of reminds me of the like jobs numbers that get updated two or three months down the road, right? They'll be like, jobs are up. And then two months later, they'll be like, oh, just kidding. In May jobs were actually like way down. And now that we have this info, it's different. It's kind of like that, but much more dire because we're talking about health and outbreaks and stuff like that.

And it causes this problem where it will always look like cases are declining, even though that's maybe not the case.

So now casting is a fantastic topic. We don't do like screen-sharing and presentations and stuff at the Hangout, though it's very, very tempting to do it for this. But you had lots of questions about how that works. And it was very, very interesting. And so I encourage anybody who is curious about this type of stuff, these are great questions for Allie today. And then some other things that we can talk about is public health disease modeling in general, the dynamics of disease transmission. I know a lot of us here have experience with bringing together different data types because you do some data harmonization. And I wanted to ask you some examples of like problems that you see or data challenges that you see and the types of data that you use so we can connect some dots.

Yeah, I think one problem that comes up a lot, I feel like my work is record linkage. It's a lot of different data sources of like, you know, maybe it's test data and sequencing data, you know, back in COVID, we're doing a lot of sequencing and hospitalization data. And you're trying to connect, you know, the same person across all those different data sets. So there's always a use for record linkage or the same person over time, like within our hospitalization records. That comes up a lot. Data quality, you know, and just doing kind of basic QC and things like that, especially during COVID, we would get test data from like so many different sources. Some people are, you know, using more formal reporting systems we've created. Some people are sending emails and using varieties of different forms for that. And so just like harmonizing the different ways people send in data.

That's a lot. And so are you seeing like hospital system data, patient level data? Are you seeing insurance claim data? What are the types of data that you are gathering to put all this information together and make forecasts?

Yeah, so I work with, for the like pertussis, it's, it's kind of like, it's just a reportable condition in Washington. So each case gets reported to the state. So that's just like that level of data on an initial level. Yeah, I work with some hospitalization data, yeah, testing data sometimes. I would like to work with more insurance data, not yet, but yeah, that's definitely of interest.

A typical day and community building

Okay, cool. Well, I would love to start with Natasha's fantastic question, which always gets us a little bit deeper into the context of what people are doing and what they're doing. So I'll read it for her. What does a typical day or week look like for you? What areas do you split your time between? So do you do programming, meetings, planning, stuff like that? Do you do any community building?

Yeah, so my time is split between all three. I'll have like project meetings and planning meetings and maybe like smaller, like one off of like just helping people with like maybe like smaller, like one off of like just helping people with a data science question or like helping people migrate a pipeline or things like that. And then I have my own time to be coding and working on analytics and exploratory data analysis and modeling and maintaining existing models that I work with. And then I also have my kind of community building and training that I do. So we have an internal group called EpiCoders within the organization. And so helping run that and then also working with our local health jurisdictions to do trainings for them. And then also this national network that we started here at DOH to talk about AI and public health.

And that is called PubHealth AI, right? Yes, PubHealth AI. Yep. And I think I have a link to that. So I'm going to put that in the chat for everybody if they are curious about that. And what is your role at PubHealth AI? How did that get going?

Yeah, so I help facilitate that group. It started out, we had a pilot project here working with Google to look at whether we could use AI to classify reporting variables around unintentional drug overdose death reporting. It's a lot of like natural anchor processing where you've got these text heavy data about each death and there's a lot of, there's hundreds of variables that need to be reported to CDC. And so we were trying out whether we could use AI to do some of that. And we were running into challenges of how do you implement AI in public health and what does it work for? And this is a couple of years ago, so this is much earlier models. And we just wanted to like share and talk with other public health organizations about what they were trying to do, what was going well and what wasn't. And so we started just presenting that project to other people we knew in public health and then kind of started drawing more and more people together and be like, okay, what are actual use cases and what are other people trying to do? And how are people dealing with governance and ethical and legal issues? And it has just grown and grown and grown. We actually had someone from Posit who presented at the beginning of this year, Sarah Altman. Awesome. Yeah. And so we just have bi-weekly meetings where it's just, yeah, but usually public health big topics, either a project or a demo or a discussion topic.

And so is it public health organizations all over the United States or is it international? We've got some international too. Yeah. We've got some international, but yeah, public health organizations from like federal to state to local level or tribal groups as well. And then some academics and then some like non-profit organizations. That is amazing. If you are in public health and you are interested in this, everybody go check out the link and take a look at, is there information there about how people can join and how they can join stuff online? Okay, cool. Yeah. So it'll be form. You'll get an email from me about what the organization is and then meeting invites and lots of emails, but yeah. Wonderful.

Audience Q&A: data challenges and coordination

Okay. Well, we have so many questions that just popped up in Slido. We will go through and add them to Slido, but in the meantime, I would love it if Nora could ask her question live. I'm going to highlight it there and we'll help Nora unmute and we can ask about data challenges.

Hello. Chrome just crashed. I'm trying to open up the question again. Well, you know what? I'm going to put it in the chat. So everybody. Honestly, that's my fault for asking so many questions because I think. I was going to say she did ask like three. So with the one that I was thinking about was the challenges of coordinating with other organizations, which I would love to hear about.

Yeah. So for hi, Allie. Hi, my name is Nora. In my last role, I actually worked in pathogen surveillance. So I actually worked as a consultant working in that particular sphere. So I am somewhat familiar with some of the data harmonization issues of working with public health data and trying to figure out how to coordinate and the. Some of the issues, especially around modernization. So particularly, what are some of the data challenges that you have run across to coordinate with other public health boards, be it federal or more local, like county organized county particular. Public health organizations and like, are there initiatives to help standardize and or modernize these approaches, be it from the reporting aspect or even from the having more standardized pipelines across?

Yeah, OK, big question. OK. But I will say, like, Nora represents a big chunk of the questions that came in on Slido that were a lot of them were like data quality. How do you put these things together? This is my point. This is my old role before I was on the job market. So it's like a lot of people can relate. Yeah. So I think one challenge within Washington in the data harmonization level is we have like our local health jurisdictions, which are kind of counties, but a little bit different, have a really different sizes and have very different resources. So, you know, there's everything like King County includes Seattle. They have like a lot of people and their own great data people and are doing a lot of great work there and have a lot more resources. And then we have like very small counties. And so like you have small numbers where it's like, oh, we can't even, you know, maybe share the number of people for a case in those counties or things like that. And maybe they don't have as many resources. So it's challenges there of like, how do you provide, you know, for doing public reporting or things like that? How do you balance those different things of smaller units versus like and different needs and things like that? So that's one big challenge. I think the other work like directly with people from the local health jurisdictions makes a big difference for like federal reporting side of things. There's been efforts to do more data modernization and it's been a big effort over the past five years. And COVID definitely changed things in terms of like building out new systems and moving to like cloud and trying to do more standardization. Public health is having some challenges obviously right now and funding and things like that. So what will continue is always hard to know. But yeah, I think there's definitely more efforts that's noticing. I think there's a lot of like work groups and community groups to like talk between states. Yeah. It's advantage of like there's for everything that I'm trying to do. There's all these other groups that are probably trying to do the same thing. And if I can just talk with them, you can learn what works and what didn't. But it's also like hard to stay connected. So I've probably found getting people conferences and like having these like commutes of practice really helped with that. Yeah. And there are like data standards that exist in health rate that can help. Although like insert XKCD, now we have 15 standards coming here. Is that a problem that you face as well? Yeah, that definitely happens. Yeah. For a lot of like, there's like, you know, cold categories of like reportable conditions that are very

standardized in how they're reported and what data is provided doesn't mean it's actually happening the same way in each state. And there are like different data delays and things like that. And so I think it's always important to be actually talking with the people who own each data set, because there will be data quality issues that, you know, you don't realize just looking at it. But yeah, there's definitely efforts to standardize things.

All right. Let's ask another question here that might be in the same vein. So many questions on Slido. Oh my gosh, they're going crazy. Give me one second while I filter through. As people vote for them, they like change position in Slido. So I'll be reading one and then all of a sudden it'll be like boop, boop, boop, boop and move around. But there was one that was anonymous. I'm going to put it at the top so it stops moving. And it says, how do you handle issues about whether data is trustworthy enough to release? What are your thoughts on concerns about political pushback when the numbers change as you get closer and closer to the truth? And I don't know if this is something that you handle at all, but it does touch on NowCasting a little bit. Like if numbers go out and then a week later or a month later, they're different. Is there any way that you handle that trustworthy part? I feel like NowCasting might be the answer.

Yeah, I think that was like part of my interest in NowCasting was because some of the, yeah, you power port, it looks misleading. I even notice I'll like look at something and I'll be like, oh, look, numbers are declining. And then I'm like, I have to remind myself that I know that this is reporting data that will be adjusted. So yeah, I do like NowCasting for that. I think though with any like modeling, there's probably a lot more hesitancy and caution because yeah, we know that the harm or the problems that can come from putting out an incorrect model, even though if you're trying to do your best and communicating that, it is going out to a wide audience who aren't familiar maybe with the modeling efforts and things like that. So being able to, yeah, probably things happen slower where you're trying to do a lot more evaluation before things go out publicly and really communicate that uncertainty as it exists is always good.

I think, yeah, during COVID, yeah, this was obviously a big topic of conversation of, you know, early models of like, okay, there's just not that much information and what does it mean to put out a model? I think things like how you phrase things of like, okay, this is a scenario production, not a forecast of like, there's kind of different wording in terms of like what certainty people might be assuming from each of those things. And so, yeah, it's a lot about the communications side of things.

Oh, the assumed certainty is scary, right? Like you always want to caveat everything with the fact that all of these numbers are somewhat uncertain, even though you are a public health organization, you only have the data that you have and it's only as accurate as it is when it gets to you and not everything gets reported.

Oh, the assumed certainty is scary, right? Like you always want to caveat everything with the fact that all of these numbers are somewhat uncertain, even though you are a public health organization, you only have the data that you have and it's only as accurate as it is when it gets to you and not everything gets reported.

Tools and skills in public health data science

There were a lot of questions about skill sets needed for inside of public health, breaking into public health, things like that. So, I think that this like kind of we're at the halfway mark is a great place to ask about some tools as well. So, I know that what you work on probably part of your products that you produce are dashboards, public health dashboards of some kind, but I figured I'd ask what tools you use. Are you all using R or Python? Are you deploying Posit Connect? Do you have things that are public or is this just for consumption inside of your health organization? And then also, I will follow on with that in a minute because I know that one of your co-workers won a Posit Impact Award as well. And I can add a link in the chat to those so everybody can see them.

Yeah. So, I work primarily in R, some Python as well. And I feel like, yeah, the balance between organization is mostly probably R, some Python, some SQL obviously used as well, and maybe bits of SAS still remaining. And we do use Posit Connect. It's a newer tool and we're kind of building out a larger cloud environment. We also have Databricks. We use Power BI for most of our dashboarding for like public facing dashboards. Everyone's all have some internal Shiny dashboards, but nothing public facing for that.

Yeah. Amazing. That's a really broad list, right? Like if you're looking at getting a job in public health, not that these tools are across the board, but just some of the things that you could be familiar with. So, Databricks for data storage, SQL for manipulating and retrieving data. I heard Posit Connect and Shiny for internal dashboards, Power BI for external dashboards, and then both R and Python. Do you have things that you work more in R in and more in Python in as far as the types or tasks that you're doing?

Yeah. Most of my day-to-day work is, I would say, in R. I'm probably farther on that side. It's definitely like kind of the more probably common language in public health and a lot of the like analytics and modeling. There's a lot of great existing R packages that I take advantage of. So, and probably more familiarity with that. So, if I'm like helping another team like write a pipeline, I'm probably more likely to do that in R. But then there's some things that maybe kind of more on the data engineering side and others in my team who are further on that side are more likely to be working in Python and like large-scale scaling and take advantage of like PySpark and Databricks for like more parallelization can be quite useful.

Sweet. Thank you. Okay. I hope that that gave everybody some context on tools around public health, but you know, this is the Washington Department of Health specifically. So, take all of that, you know, with a grain of salt. I have never seen so many questions in Slido in my entire life. And Allie, you had so many questions when you gave your virtual talk at PositConf as well. There must be something about Allie that really, really produces questions. So, what I'm going to try to do is consolidate some of these questions into one and nominate someone to ask because there was one from Belinda that was about some of the backgrounds of like your colleagues and your team and whether or not there are open positions in your organization and then whether or not you're looking for people with both public health and data science experience.

I know that you're probably not doing the interviewing and the job hiring decisions. I think that your boss is here today. Not that I want to put Paul on the spot, but I think Paul is here today. So, if Paul wants to raise his hand and jump in at any time, I will let him choose whether or not he wants to do that. But could you give us some context around the breadth of experience and skills inside of your team at large or maybe how many people are on your team, different things that they do?

Yeah. So, my team is like a kind of data science specific team mixture of like data science and data engineering. What are we? Eight, nine people. I don't know the exact number. And we, our team started during COVID. It was like definitely building out that additional data science capacity and how do you scale to data needs that COVID brought about. And so, I think my team is more heavily on the like computer science background side of things and maybe with a little bit less of the direct public health versus like other teams in our organization that are further on the public health side. Obviously, always collaborating and having both those skills and experiences quite useful. I didn't have a public health background. Yeah, I was working in cancer research. So, different side of things, but still building pipelines and similar kind of data modeling and analytics tools. We are unfortunately in a hiring freeze right now. It's obviously a, yeah, challenging time in public health. So, not hiring at the moment, but always good to keep an eye out.

That is okay. I think that's kind of across the market right now. We're in a little bit of a rough patch with hiring, but I will say again, if you see jobs and you want to share them, or you're looking for one, head to the Discord server to the jobs and open to work tab. There are actually people on the server who have gotten jobs there from job postings, which makes me so, so, so happy because it's a lot easier to connect there with people than it is, you know, randomly on LinkedIn. And you have to be like, I'm that person from the hangout. Remember, just get on the Discord server. It's so much easier.

Okay. Belinda, thank you for asking that question. Give us a little bit more information about the team. I'm going to head back to Slido here and see if we can find a question that is a good representation of a few questions that we can put together. Let's see. We had some AI questions, more data quality questions.

NLP in public health

I, there was an anonymous question about NLP that I think is great because you had mentioned that you get some big text-based data about things sometimes. And it says, I would love to understand a bit more about NLP use cases inside of public health.

Yeah, so this pilot project was with our drug, initial drug overdose reporting data. And so data comes in from like coroner reports and medical examiners and scene investigations and police reports. And so it's all this text data. People in that team then compile all that combo of data into these kind of narratives that are a few paragraphs long and kind of have a centralized format, but are very varied to the varied kind of data that they're taking in. And so they're very like, yeah, text narrative with some format to like, you know, how information is broken up in terms of like medical history and scene reporting and investigation and like information about methods of use or items paraphernalia found or bystanders present and things like that. And then from those text narratives, they then extract these like 300 variables that get reported to the CDC to kind of look for trends in initial drug overdose deaths, things again around use and patterns of use and things like that. But the very complex variables are subtle things like was a bystander present at the time? And that's like a complex definition. And each of them kind of have these like long definitions or big, you know, whole books about how each variable should be classified and much past work on how they've done that. And so we're trying to see whether we could use AI to train, do that classification. These were much, much earlier models. So the small pilot. And so it didn't continue, but that was the goal.

Thank you. Yeah. It doesn't matter that it didn't continue. So many things in data science start and stop and don't continue. It's like 80 to 90% of the stuff that happens.

Personal projects and hobby data science

We had a question that had just popped in from Lauren H. Let's see if we can find Lauren to unmute Lauren Hagan, I think. And I'll put it in the chat as well.

Hello. All right. Hi, I'm Lauren. And I was wondering if you have any projects outside of your work as part of the Washington Health State Department.

Yeah. So like hobby projects, my hobbies are like mostly hiking, backpacking, and climbing. And so I like to use my huge eyes to just assist those projects. So one thing is that in Washington for backpacking, most of the permits have to be put out in the like spring. And so I need to plan when I want to do various hikes. Washington has a really great website called the Washington Trails Association. I love it. I would always recommend it if people are planning any hiking in Washington and people like write trip reports about their hike and like various features of it. It's so great. It's widely used. So I took all of those trip report data and used it to like estimate better when hikes are more likely to be like snow free and less buggy. And if you like have like fall foliage, things like that. So I use that for my planning for putting in permits for hiking.

And then I guess for climbing, I got in a conversation with someone about GeoGuessr and they were claiming that they could do a similar thing for climbing routes, that they could see a picture of a climbing route and guess where in the world it was. And that was a debate we were having. And so I made a little just Shiny app of that, taking pictures of climbing routes and geolocating them. And then you could like choose on a map where you thought they were and score it, which was fun. They were surprisingly good at it. They were right. They were right. That is such a niche thing. I don't think that I would be able to do that. If it was Boba tea locations in San Antonio, I could probably do it.

But yes, that sounds so fun. And I don't know if you have like public links to those things, but if you don't at the moment. Okay. Okay. I was going to say if you do at some point and you ever wanted to share, we would love to give them the hug of death by all clicking them at one time.

Data fragmentation across government levels

Okay. So let's head back over to this. There was one that was about fragmentation. I think it was from Jay, Jay Lee. I'm going to highlight this in the chat. Let's see if we can unmute, help Jay Lee unmute.

I'll copy. Yeah. Hi, thanks, Allie. Hi. Yeah. I was wondering how you deal with like fragmentation just in terms of like your data sources coming from different places at different levels of in terms of like your data sources coming from different places at different levels of government or health care, how you work with like one data source might come from the federal government, another might come from the state, another you might have to go to all of the different counties or public health areas that you were describing and like sometimes those are in the same format, sometimes they're not, and how you work with all of those issues.

Yeah. A common problem, yeah, getting data from different labs, different counties, things

like that, trying to establish data standards to like decide how you want your data to look, encouraging people to use them if you can, but often it's just like a lot of, yeah, not manual but laborious work to standardize things and then sometimes that's just how it goes. Also people will use different ID systems for the same topic. So there's a lot of data, like, yeah, tables that I have to translate between different systems of, you know, different names for the same things.

So it's, yeah, it's a lot of just having those things. It's, yeah, often a pretty kind of manual process to kind of collect all that information. You would ideally like everyone to use the same thing, but that's not really realistic. So it's often just trying to get a handle of the data of like, who do you, who can you find that can tell you more about how this person treated their data and what the differences are and then trying to join them.

Definitely like, yeah, record linkage where, yeah, I've done different approaches to that, it's non-problemistic or sometimes been interesting with neural network approach to that at one point, but yes, it's a common problem. It's a big, a big question. I don't know if anyone ever has a final answer to it. It's just like, yes, this is a problem and we all deal with it and it's like a case by case thing, right, of how we're going to deal with, like Lauren mentioned, standardization. There's also problems with missing this all over the place where you're having to make decisions on a case by case basis about how you're going to handle it.

Presenting uncertainty in data analysis

Thank you so much, Jay, for your fantastic question. I had another one that was about linkage from Liz Eisenhower, who has been on the Hangout before and who is a long time Data Science Hangout community member too. Liz, did you want to follow on with your question? Because it was a little bit more about incorporating uncertainty than just the problems, the challenges.

Right, right. Exactly. Yeah. I was, thank you. I was asking about kind of what it looks like when you're presenting the results of an analysis that's had all these kind of duplicates that you've had to maybe come up with rules to de-duplicate the individual observations that come from different sources. And maybe there's other, you know, different data cleaning that you're doing. And then at the end, when you're presenting that, what does it look like? How do you try to, or how can you kind of present uncertainty in that, in that work?

Yeah. Oh, man. Big questions. No, these are good questions I have to think more about. Yeah, I think often it's, yeah, just like on the communication side and visualization side of how do you incorporate uncertainty into your visualizations, I think is an interesting thing. I, yes, and it's a lot of testing it and getting feedback of do people know what this means? Things like that.

And I think, yeah, being honest about the problems with data and your caveats there, it depends, you know, how much direct communication you actually have with the person you're trying to communicate those. And I use a dashboard and it's just like, this is going to go out to the public. And you know, they're, I don't get to talk with them directly. And so you're just hoping that the visualization stuff and you have your data notes, but also often no one reads those. And so you kind of have to accept that fact. And then sometimes it's a more direct communication of like sharing a report and you have it face to face. And so you can tell someone about the problems with it and communicate that and the limitations.

But yeah, I think, yeah, I think there's an interesting kind of like the visualization side of things of what you can do and how you can try and convey that uncertainty. Just I admit it's probably earlier on this side where I'm just trying to get those uncertainties into our visualizations. And that's where like, now I think it's been interesting of like, how do we at least let people know that we have this uncertainty instead of maybe it just being like a note or like a gray bar over an area.

All right. Fantastic. Yeah, those are all hard questions, but I like the question of like communicating it versus just dealing with it, because those are two different things.

Sensitive data redaction and public data resources

All right. I have a question in Slido from Tamalu. If you would like to unmute and ask your question live, that would be great. And I will highlight it in Slido and put it in the chat.

So I want to ask, what kind of models do you use for personally identifiable data redaction and other sensitive data redaction? Yeah, I actually don't work that much on that. Yeah, I'm either doing internal work where I'm directly working with the PII and, you know, data is iterated, but you can share just internally. And so there's, you know, more versatility about what can be shared. Or it's on the public-facing side where things are fully aggregated. We have like rules around, you know, values under 10 not being reported, things like that. So it's very strict of like limitations on the data that's being reported there. So I think I tend to fall on either two sides of that, where it's either fully aggregated, you know, not any PII, or I'm fully working with PII and it's just for internal use. But yeah, I don't have as much experience with the redaction side of things.

That is totally okay. That was a great question, Tamalou. And if anybody does work in public health and does work in redaction or censoring of data, let us know in the chat. Let us know what you do and what models you use and how you take care, because it is super important.

And we had a question that was also to the group. I'm going to highlight it and address this to everybody, because it says, for folks interested in public health, what are some good projects to work on? And I think that regardless of what field you're aiming for, whether it's public health or not, everyone is going to have some good ideas on types of projects that you can do and the ways that you can share your work. So we'll open this up to everybody. Please share in the chat.

And this is an anonymous question, but I will stick it here. Maybe we can make a thread. But Allie, I was wondering, do you have any advice for this person? If someone is interested in getting into public health, are there some good projects they might be able to work on with public data that they could find? Yeah. So there's a lot of public data that gets put out, you know, where you can find ours, you can look at the Y2H website or Open Data Portal, but then, yeah, CDC is also a great source for this. In terms of really open projects that are, you know, high-level complexity, but anyone can participate in, there's COVIDNet and FluNet for predictions and forecasting around those. It's fully set up to execute repos. You can see the data that you get to use and what the forecasting data has to look like for that. But yeah, they're interesting forecasting challenges, really challenging problems. But it's a nice way to, if you just want to start working with the data and try out and see what it would look like to do, it's very well set up to work with.

Cool. I'm going to add the link to COVIDNet in there and let me see if I can find FluNet as well, or maybe Isabella can find it. And then we had a couple of people drop some of the other things that Allie mentioned there, like data.wa.gov from Andy Cooper. Thank you. Jason had added data.gov, overarching data.gov, which is an amazing resource. And the data portal on data.gov is where I go for a lot of stuff because there are state organizations, municipalities, there's all kinds of stuff in the data portal. And then everybody hop in the chat and just take a look at some of the answers that people are giving, connect with people, chat with them more, ask them questions.

Dylan says, I'm very interested in hearing these answers because I'm a data science professor and I want to help my students. We all want to help students. We want everybody to learn good things and have good takeaways.

AI in public health and resource allocation

Prajesh had said for PII redaction that they have cloud plugins as well. And there was a question around AI. So I'd like to transition, let's use Prajesh's comment to transition to a question that might be posed to both Ali and Paul. Paul, I put it for Paul in the chat, but I don't know if Paul's going to have a bunch of.

Oh, no, Paul did respond. OK, so the question was, how do you think about where you invest limited resources in this time of manufactured scarcity and public health? I have concerns about investing in these marketed as AI tools when we could instead be using financial resources to rehire and retain public health professionals.

And so Paul had said, like, this requires careful assessment and having a really grounded understanding in data science that can give benchmarking comparisons to see how things really perform. And so far, the best value lies in the small admin and the bureaucratic type tasks. So, Ali, where does AI succeed in your eyes and where is it maybe falling short as far as public health?

Yeah, I think advantage of AI is often like a scaling question of places where I've seen it do interesting work is around like digitizing records where it's like thousands and thousands of records in this work that just wouldn't happen because it's just like too much manual work that is not feasible and there's just not the resources to do that. So I think there's interesting opportunities there. I think I would always prioritize people over AI, especially at a time of limited resources. Yeah, that would be always my preference.

I think I would always prioritize people over AI, especially at a time of limited resources.

But I think, yeah, for that, like scaling, and I think it's often the work that wouldn't happen. I think there's some work where it's like that was the goal with our pilot around the overdose reporting is that this is really time intensive manual work that has to happen and is keeping people from other work. And I'd love to be able to replace more of that because it's like, yeah, just to do these kind of standardized reporting things is, yeah, it's like high priority, it has to happen, but it's very time intensive and it keeps people from being able to look at things deeper and do the further work that we would like to be doing. So I think it's always like it's ideally a partnership. I think you really need that human review and knowledge. So I think it really, really is a partnership, but I would like to be able to take more of that that work away from people. Fantastic stuff that's getting in the way. And that is not maybe bringing anybody a ton of value, personal value to do it themselves.

Breaking into public health data science

Okay, we had an anonymous question. And I know that Ali, you said that you were coming from cancer research, right? Not specifically public health as a background. This person is asking for someone who's interested in starting a career in public health with a background in science. With a background in statistics, data science and AI, how would you recommend breaking into the field? Specifically, how would you recommend gaining that domain knowledge to be qualified to break into the field? Is there anything specific that you did to kind of change your trajectory?

Yeah, I think it was definitely a bit of a different time when I broke into public health. It was, yeah, it was during COVID and there was just really, really that need for additional data science support and to build out systems that could scale and just how do you deal with more data and topics around like moving to the cloud and things like that. I think those are still like all current problems that we face and are still relevant. But it was like some of those were kind of newer problems. And so there was definitely more room to break in from a computer science background. And yeah, computational biology, similar skills. That says it that way. So I think you can always take your skills and they will be a value.

I think I always was just like personally interested in public health and I was really interested in infectious disease. So I was always just reading a lot about that. And it's like good to get, you can do little projects of getting started and trying out working with public health data. It's great way to, you know, get your feet wet of like, what does it mean? What does it look like to work with public health data? Yeah, get excited about it. Go read publications and papers. I feel like if you get the chance to talk to somebody, your passion for something and your interest in something will come through in your conversations with them because you won't be the one asking really, really high level questions. You'll be asking more nuanced questions, right? You'll be asking better questions and it will show through.

In our episode with Alex Gold that we had a couple of years ago now, maybe a year ago, he focused on like his hiring decisions being talking with people who are asking really good questions and what a good question looks like. And so I feel like if you want to get to the good questions, you have to go do the research and be really interested in the domain. That was a fantastic anonymous question. And I have another one that I think is so fun that I just saw pop up. Put it in the chat here.

Top R packages for Epi projects

It says, what are the top five R packages besides the tidyverse that you use regularly for Epi projects? I'll put you on the spot. We'll let you know. Okay, top five. Okay. So I've been doing a lot of nowcasting. I'm a big fan of the Epi nowcast and baseline nowcast packages. They're really well implemented and good ways to get started with nowcasting and have various levels of like the most basic models to like you can do a lot of complexity in your models in terms of like how fine grained of like different patterns you're looking for and incorporating like day-to-week effects and all these different reporting kind of patterns that happen. So I like those.

Record linkage. I guess I've done a different, not always are, but fuzzy join is always a good way. And I will put a plug in for one of my colleagues made a Rust implementation of fuzzy join called fuzzy join. That's cool. It's really fast. So that's an interesting one to work with.

I'll share that one. I've enjoyed recently DizNetwork. I was like, yeah, working with some like tree data and I was looking at like employee data like in structures. And so it's a great, it was so nice for interactive, like tree and graph visualizations. Yeah, I really enjoyed that. And it made it a lot easier to look at the data that I was working with.

Yeah, I think, yeah, I'm going to add fuzzy join, which is from John Downs. He gave a presentation about it at Cascadia Conf this year, which I freaking loved. And I am trying to get John to come on the data science lab to talk about the challenges of fuzzy joins, how fuzzy join as a package solves that problem. John, if you're listening, Ali, go tell John, reply to Libby's email. Come on the data science lab with Isabella and Libby. Yes, waka waka. Daniel, I think you were there watching that talk with me. It was so good. We had such a fun time.

Career advice

Can we get in one more question in our last five minutes? Because I would love to ask you for a piece of career advice. Maybe we can even get into.

Yeah, I think one thing I've done really useful in my career is just like making connections in your organization to people who work in a very different area to you. I've always found it really helpful for like learning about the organization and things about the data that you're working with. And they'll often have knowledge about a data set that I didn't even know existed or how something was done in the past. And that's always really useful.

It pays to be in person. I feel like much easier to do in person. Bring food in was always a way to just draw people and you could meet a lot of people. But it's definitely harder. I work remote now based in Seattle, but we're working remotely for Washington. And so figuring out how to do that. I think like just scheduling kind of those like more informal coffee chats, virtual coffee chats has been really helpful.

When there's a chance to be in person, I like go in every once in a while and I'll tell everyone and try and get them to come. And then finding those people in your organization who seem to know everyone and taking advantage of that because they will be able to introduce you to people that are very helpful.

These are fantastic suggestions. I love that. Go find somebody. Go find their Rachel at an organization who's been there for a million years and knows all the people and is going to be fully open to just like introducing you to everybody. And now I get to do that too. New people will message me and I'll be like, have you met this person? Have you met this person? Have you met this person? I will be more than happy to invite you to talk to everybody. But yes, when you can go have coffee chats, do all that stuff.

And then finding those people in your organization who seem to know everyone and taking advantage of that because they will be able to introduce you to people that are very helpful.

SAS and the data stack at DOH

I had a question that came in that I thought was so funny in Slido because it says, this is being asked as I type, but I don't think we answered it. It was anonymous. And it says, I know some other data folks at DOH and seem to recall SAS being a major tool. Are other tools coming into the mix? This is being asked as I typed, but we didn't get to specifically address SAS. Are you a SAS user? Are there SAS users at DOH? Is it still in the mix?

Yeah, definitely. I'm not a SAS user. There are certainly still some SAS users. It's definitely less, I think, historically in public health, it was much more of a true used language. Less true kind of when I joined. I think the cost of a SAS license is always a good argument for moving away from SAS. But yeah, so I have helped a little bit with sometimes people migrating away from SAS. But it definitely does have a historic use. And there are some things that just will continue.

I wanted to see if there were any that were short enough that we could get them in the three minutes that we have left. I'm going through and trying to mark some as answered that I think that we answered because we covered like Ojo had asked something about data quality issues for duplications. We covered that. Nor had asked what the data stack looks like and whether or not you're more data engineering or like pathogen surveillance or a mix. I think we covered most of that. But do you want to cover any more details on that? You're split between like data engineering that you do versus analysis or like epigenetics or epidemiology.

Yeah. Yeah, I think I kind of fall in between in the middle. And my team in general does a mix of both. We're not like pathogen specific. Like we kind of are more like central data science resource. We're working on projects across anything that the agency covers. Yeah, I've done like respiratory and pertussis and other vaccine-preventable diseases and like zoonotic disease and then the drug overdose work. So, yeah, I kind of get to have a little bit of work in everything and just find wherever help is needed. To be able to move on to those projects.

So, yeah, it's a little more on the data science side in Linux, but I don't think I'm in my start at DOH was definitely kind of being more on that data engineering. Our stack may look like so many different things. There's so much work that is still just happening on network drives and things being shared that way. And then some things that are moving more into the cloud. And we have Unity Catalog and Databricks. And then there's a little closet cloud environment. It sounds like a big mix. I think a lot of people can relate to that. Yeah, some of the stuff we have in Unity Catalog, Unity Catalog is Databricks for anybody who's not aware. I'm not a Databricks user either. So I am. I just have Databricks imposter syndrome because I never know fully as a user.

Closing remarks

This has been so much information. I hope everybody learned a lot today about NowCasting, about public health in general. Allie, if anybody wants to connect with you, is it OK if I put your LinkedIn in the chat? Yes. OK. Definitely. I'll put my email below. I admit I'm bad on LinkedIn. So I'll put my email. I am also trying to break up with LinkedIn. I took it off my phone. Anybody who sent me a message lately and it's late to reply, that's because I'm trying to have good hygiene.

But yes, thank you. Allie's contact info is in the chat. Thank you, everybody, for showing up and hanging out with us today. We love hanging out with you. We have so many more amazing people just booked up through the rest of the year. Keep coming back and hanging out with us, whether you get a reminder email a day before or two hours before. Put something on your calendar for Thursday at 12 p.m. Eastern. And then also, if you would like to join us at the Data Science Lab, Isabella and I would love to hang out with you. We have all of our chat on Discord. We are there same time, same place. But on Tuesdays, you can go to pos.at slash dslab to register for that one. Isabella just put that in the chat.

We are rounding off the end of Testober. So we are doing testing for Shiny in Python with Barrett Schlerke, who is on the Shiny team here at Posit. Barrett's always a good time. He's open to so many interesting questions. Allie, thank you so much for joining us today. This was a lot of fun. I hope you had fun, too. Thanks so much for having me. Yes, and thank you for being a wonderful speaker at Virtual Day at PositConf this year. Absolutely loved working with you.

Everybody, I hope you had a good time. I will see you on Tuesday or I'll see you next week on Thursday. The chat file will be in the Discord because I just saw that Isabella saved it. But you can save it, too, if you want. There's three dots in the top right-hand corner of your chat pane where you can save the chat. Okay, I'll see you on Discord, everybody. Bye! Bye!