If you are Health Data Scientist Manager or Director, let me know if you are interested in sharing the story of your career journey. You can email me at ahobby@healthdatasciencenewsletter.com
Summary
Andrea Hobby interviewed Austin Barrington who is a Health Data Scientist and Actuary and previously worked as a Senior Director of Health Data Science at Blue Cross and Blue Shield.
Tell me about how you got into data science.
I became a teacher actually while in school. I had a good feel for programming and all things technical. And then, I became a teacher, and there was a need. I was a middle school teacher. And there was a need for data analysis because we had all these standardized tests. And we had all these what they call the essential knowledge and skills, so they were skills that were numbered off. A student will convert fractions to decimals, which is a skill. And so, each question on the standardized test had one of these skills attached to it. We get these results, and then we must figure out what to do with them. So, I transitioned from my role in my first year of teaching to data analyst and teacher. So I would take all the standardized test results and plug them in and associate them with the district's demographic data to say, for example, boys between sixth and seventh grade who have English as a second language are struggling with these three skills. Or kids who are new to the district or new to this class are struggling with these specific skills. Or for instance, this teacher is especially good at teaching XYZ to Hispanic kids. And so we would come up with these associations and say, Okay, well, what is it that you're doing differently? And so that was where my data science roots started. I wasn't building predictive models or anything. Still, I got interested in the data analysis and what we can do with this instead of just saying, cool, you're a terrible teacher, or fantastic, you're a good teacher, or like you're a student who's like, not learning very well. So we started pairing up teachers who were good at teaching subjects to kids who could be doing better at those skills. So that was when I was a teacher. And then, I moved into health insurance from there and realized there were many opportunities to apply predictive modeling. So even though I've taken actuarial exams, my first data science project was coming up or assigning the value to a specific intervention. And saying, Okay, if we didn't have this intervention, would we have saved money or lost money? And so, I built a regression discontinuity model to do that. And that was the start of my data science career.
What was how was your process to get ready and prepared for data science interviews?
Good question. So I was too early to the data science area, where this would sound terrible, but most of the people who interviewed me had yet to learn what they were asking. And so yeah, exactly. I would prepare as I would. I read things I would like to go on Glassdoor and see what they're asking me. I would go on Towards Data Science and read a couple of articles to have some things fresh on my mind. I go there because there's usually how do you stay proficient type of question in interviews. Also, I do a lot of reading. I read a lot of magazines and I read a lot of articles, but mostly I like to practice myself.
What do you like to read to stay up to date and anything that you would recommend doing for an up-and-coming health data scientist?
This was very specific to health, but the actuaries in general, but let's see, here we go. So the Society of Actuaries puts out in this Predictive Analytics and Futurism newsletter. So it's got, I don't know, eight, eight or so different articles each month. And so it's fun to see applications that are more of my thing now in a consulting role, where I do not have to stay up to date on the latest methodology because, honestly, most people want the most straightforward methods. I'm staying up to date on use cases and trying to see where we can expand our reach.
When you were more focused on staying up to date in with new data science methods, how many hours a week were you working? How much time did you focus on keeping your skills up to date outside of work?
Yeah, so I had a subscription to Coursera, so I would use all the Coursera courses and stay up to date on them, like the technical skills piece. And then internally, at my last job, we had like bi-weekly skills challenges, and so that was another way we did it is to say like, okay, so you each had to rotate through and come up with a problem and say, Okay, well, if you needed to do such and such, you know, how would you do it? And then we would like to greet each other, essentially. So that was a big way. And then SAS was, you know, now I don't really use, but so yeah, I mean, I still consider myself like an expert in SAS, but I try not to start new projects in SAS, kind of my SAS Projects are in maintenance. So Python or R is what we typically use. Here Python is more recent. And, when it was SAS, I went to several other SAS conferences and read Bart Baesens's materials. If I remember correctly, I learned how to use Enterprise Miner from Bart.
What is your favorite I guess data science product that you've worked on in our health data science because everything's healthcare-focused?
Oh, sure. My favorite. I am trying to remember what it was called. But I can describe it. So we would take the whole population and cluster them into 12 to 15, Depending on the version of this project, or the 15 clinical, clinically homogenous groups. Okay, your claims costs are driven by your maternity, or your cancer causes your claims costs, etc. And then once we had folks divided up into those clusters, we would then train a model to say what was going to happen to them next to what was the likely pathway of claims that was going to occur for individual people. So for maternity, we would see a lot of you're going to have a such and such lab, you're going to have a home birth delivery at home. Then you're going to have this emergency, visit an ambulance, and then an emergency visit. So you're seeing specific claim types, but it was cool because we were able to implement it and give the nurse advocates at Blue Cross the actual tools to say, hey, we think there's a high likelihood that, like, your pregnancy is too risky to try a home birth. So they tried to intervene early before those events. So that was my favorite, most impactful in my mind, like actually having accurate information instead of here's a risk score. You know, we think you're a high-risk pregnancy, and here's a number, and nurses are just like, cool a number, but showing the clinical events that we predicted what happened next was meaningful.
What advice would you give for up-and-coming health data scientists?
Don't underestimate the importance of domain knowledge. You know, there's going to be this battle between actuaries and data scientists everywhere but insurance everyone insurance but in health. There's so much specific health data that often gets ignored, especially when data science departments are, like, housed under the IT function, because then it's like, okay, like, everyone specs out all of our work for us, right, like we have an entire scope of work. So we follow that, you know, so if it's, I've hired some data scientists with three years of health experience, and they still don't know the difference between a PPO and an HMO. It's an essential thing for you to know in this business. And so yeah, don't know that, you know, focus on that just as much as learning new methodologies and staying sharp on your software engineering skills. And then be open to marginally data science-driven things unless you want to be a machine learning engineer. You know, like put your aspirations in check. You know you don't often see a CEO that's a data scientist. And so if you want to be a CEO, you got to think outside the box and be like, Okay, well, maybe programming needs to be less and less of my focus if, depending on what level I'm like aspiring to be at. But yeah, domain knowledge and number one. Yes.
Anything else you want to share?
Having a technical portfolio is excellent. But, also having a portfolio that a hiring manager or hiring managers manager would want to look at, too, would be ideal. You want to show the context behind the project. Do not just show me your code in GitHub when I click on GitHub and there's no readme. I want to know why did you send me to your GitHub. I'm not going to read your code. I want to read the project you wrote the code for. And then I understand you probably did the project, alright, but I want to know, are you going to be able to contribute in a meaningful way with the code that you can write? And are you able to explain it to me?
Resources:
https://www.soa.org/sections/pred-analytics-futurism/
https://www.sas.com/en_id/training/campaigns/live-web/bart-baesens.html