Gaining trust & driving change | Victoria L. Prince | Data Science Hangout

Hey there, welcome to the Paws at Data Science Hangout. I'm Libby Herron, and this is a recording of our weekly community call that happens every Thursday at 12pm US 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.

I would love to go ahead and get started by introducing our featured leader today, Victoria Prince, Senior Manager of Statistics at Takeda Pharmaceuticals. Victoria, welcome. I would love you to tell us a little bit about yourself, your role, and what you do for fun outside of work.

Thank you so much. I'm so happy to be here. I love hangouts. I try to join in whenever I can, and you guys are amazing. The whole Paws at team is phenomenal. I love working with you guys and hanging out as well.

I'm from Ukraine. I grew up in Ukraine during the 90s. It was my school years. Then I went to a polytech university there in the early 2000s. Then at some point, my sister moved to the US in the early 2000s. I started exploring ways to also follow that path. She is much older than me, and I wasn't really ready to enter the workforce in the US without much experience. I was thinking, well, let's try grad school. Why not? I just decided to apply to a whole bunch of places in New England. I was very, very lucky to get accepted into one of the very good schools here. I went to Harvard Stats here. There is a funny story with that, which I'm happy to share if people are interested. After coming here and pursuing that path, the Stats PhD, I decided to stay. I tried different things along my career path as trying a little bit of teaching, supporting research, doing consulting on the side. A year and a half ago, I decided to do a major transition. So far, I've only been in the higher ed environment, even though I explored multiple avenues there. I decided that's enough of that and transitioned to the pharma industry, where I'm just enjoying every minute of it as well.

Driving in this morning into work, I was thinking, how would I describe my approach to career, to life? Three words popped up. It's flexibility, curiosity, and dedication. I try to be flexible in a way that pushing myself outside the comfort zone, flexibility in that way. Then once I see something there, I'm trying to become curious about it. Then if I truly feel that I can add value to this or it's going to add value to me, I try to be very dedicated in pursuing that opportunity. I hope that will resonate with people on the call today.

Three words popped up. It's flexibility, curiosity, and dedication.

Transitioning from academia to pharma

I really would love to ask you a little bit about, just to dig in a little bit more about your transition from academia to the pharmaceutical industry. I know that that's a topic that a lot of us can relate to. What were some of the things that were really pushing you?

Pushing you out of academia? Well, so my industry, we're leaving something, but we're also moving towards something. Exactly. Another great question. My journey through academia wasn't a classic one because very early on, pretty much during my study, I realized that I'm probably not going to want to pursue the classic academic route, which is become an assistant professor, publish a whole bunch of things, get a tenure, et cetera. I didn't have that urge to be a theoretical. There are, of course, applied professors as well, but that whole environment didn't really draw me. Instead, I found paths for a data scientist, for a statistician, but in a much more applied setting.

One of the things was I did some institutional research, so basically crunching, actually, university data, which Harvard, being one of the most famous universities, and a lot of famous people are going through it, that eventually become famous. It does collect a lot of data about students, about faculty, about alums, et cetera. There's really a great potential to crunch that data and learn.

Then I transitioned to a place, again, within that university that allowed me to support research. All of that was a really great, fun experience, but what I realized, actually, after I transitioned to industry, that it was a lot of depth as far as exploration of theories, of domain knowledge in different areas, but at the end of the day, all of that felt theoretical.

One of the things that I just love being in, specifically pharma industry, that at the end of the day, you're asked, okay, so what is the sample size that we need for the study? These are actual human lives that you're basically calculating how many people, in your opinion, are going to have a chance to receive this treatment under investigation. That feels so much more impactful to me.

Now, at the same time, I felt that I'm at an advantage comparing to colleagues who went to industry right away, because when you start focusing on something from day one, you will become an incredible expert in that domain knowledge, but of course, there's only 24 hours in a day. It limits your maybe ability to explore, to widen the knowledge. I feel like by going through these multiple different experiences, I had a chance to widen, and that width now helps me see where I can add value here beyond the statistics, beyond what my colleagues are their domain expertise in specifically how to design clinical trials and all the wisdom and experience that comes with being in this for 10, 15 years.

I'm rapidly learning from them about that. I'm still very much a novice in that, and I'm so grateful for them to be very patient with me, but at the same time, one of the things that I realized that when you transition to pharma, especially 10, 15 years ago, as a statistician, even if you were trained in open source, a lot of the times you will be funneled into more of a proprietary software route. I didn't have to do that. I actually had a chance to hone my skills in all kinds of different open source and other tools, and that really helps me right now to see a bigger picture in terms of how can we increase the access to tools in our country, I'm sorry, in our company, and what would be the advantage of that so I can like add value in that where, you know, bring from my other experiences.

Becoming the go-to R person

Yeah, that actually struck me. When you and I have spoken before, the talks that we have had about you becoming sort of a de facto person, the go-to person, almost like an internal IT person to help everybody get like updated versions of tooling, updated versions of packaging, stuff like that.

Noor, do you want to unmute and ask your question live?

Sure, I apologize in advance if I sound a little hoarse. Okay, so it's like a two-part question, so kind of cheating, but broadly, what would you say your day-to-day is like at Takeda? And then secondly, this is more of a broader question, what direction would you say, if you can say, is Takeda heading towards both scientific focus as well as technology focus? My knowledge of Takeda is primarily from the gastrointestinal side, because that's the side that I used to work in for my PhD, actually was on gastrointestinal, like IBD and so forth, so that's my familiarity with Takeda, but I don't know if they've expanded their portfolio and so forth, so I will stop.

Thank you, thank you, those are great questions. I definitely can speak to the first one, that one is easier. My day-to-day, I would say, I guess it changes depending on what is the immediate kind of deliverable that I have, so I recently transitioned, so I started in a group that wasn't as much kind of heavily involved in the day-to-day of clinical trials, it was more of a kind of like a consultant, and it was a good thing, because I was very new to the whole industry, I needed really to get up to speed, so it was more like a consulting unit that helped with like ad hoc analysis, something that is outside of the realm of like the usual day-to-day for clinical trials, but I soon realized that without that experience kind of in the trenches, so-called, I cannot really kind of like, you know, really grow and, you know, have a successful career in this industry, so that's where, about a month and a half ago, I transitioned, so my day-to-day changed pretty dramatically.

So now I think it's kind of a mix of some downtime, like kind of like, you know, thinking time, as far as like quiet time, I would say, figuring out, okay, what is the next methodological thing that I need to catch up on, like, for example, Bayesian adaptive designs, you know, something that I know kind of, I was, you know, fortunate enough to have the fundamental knowledge of Bayesian statistics from my training, but, you know, these specific things, I still need to kind of sit down and understand exactly what's going on there, so that's one, and also a bunch of meetings where we, you know, figure out what are the next steps, as far as, you know, the progression of the indication that we're trying to study and design, you know, a series of trials, and sort of learning, and also talking to different stakeholders, like, be that regulatory, clinical, medical, you know, clinical writing, all of these folks are incredible experts in their own, kind of, as I am in my little area of statistics, they're experts in their little area, and it's just so incredible to see how many people we need in a room, like a virtual room, to make sure that the studies we are designing are sound, are ethical, are, you know, the best that we could be, and the most efficient.

And the side gig, as Libby mentioned, that I have right now is kind of this, a little bit of an unofficial go-to person for questions about R, from basically very different parts of the company, and that was actually a deliberate effort over the course of a year and a half, I tried to make myself known as that person, because one of the things that I, kind of, realized through, kind of, trial and error, but then it turns out there is, like, a whole theory of organizational behavior behind it, is that if you want to, kind of, gain, sort of, not influence, but sort of, if you want people to start trusting you, going to you, is you need to try to understand what value you can add to their lives.

And that can only, in order to do that, it helps to understand your strength, like, truly, you know, realize what is the value that you can add, that it seems like, you know, it's kind of, like, lacking in the whole, you know, the place where you are, and so it was a very deliberate effort on my end to show my expertise, to be very, very generous with it, to, you know, basically, unless I'm in complete time crunch, to not, try not to say no if people are asking for help, maybe ask to delay, but sort of, offering that help and expertise, and then, you know, it kind of, like, sort of, from there, you know, the word spreads, and kind of, people start, and then once you, you know, people are, you know, open to, like, for your help, you know, it becomes mutual, and then you're, like, if you gain, sort of, their trust, and at the same time, if there's something that you need their support from, they're definitely much more willing to help.

If you want people to start trusting you, going to you, is you need to try to understand what value you can add to their lives.

Takeda's pipeline and scientific direction

Your second question about Takeda's direction in general, I can speak only in general terms that there are definitely at least half a dozen very promising compounds that, that are in, in a pipeline that we're expecting readouts, maybe this year, early next year, and they are showing some great results already. I am specifically working, actually, on DERM indications, so we're trying to explore and enter that DERM, or let's not enter, but actually be a sort of more, have more solid representation in the DERM space. GI definitely is a huge, as you said, it's still big, and it's going to become even bigger once the, you know, the, the drugs that are currently under development are, you know, going to show some readouts as well. There is plasma-derived therapies, right, there's a little bit of vaccine direction as well, and neuro, of course, neuroscience, so I think there is a drug related to narcolepsy, and actually there's some really cool technologies, so we have a whole subgroup of our larger statistical group that is quantitative sciences, basically, so they are also not necessarily focused on the trial work as much as they are focused on developing kind of sophisticated models for kind of predicting, using, say, biomarker data or other patient data on predicting the fact that they have a disease, for example, so that the diagnosis actually can be much, so that discovery of patients to enroll will be kind of more streamlined, so there's definitely a technology, technological kind of development going on as well.

Building foundational statistical knowledge

We have some questions in the Slido that kind of ask along these lines, so so many questions in Slido. Let's go ahead and get started there, and one of them, the very top one that has a couple of replies on it. It's an anonymous question, and it says, I'm an experienced data scientist, although I only have very basic level of statistical knowledge. I'm intimidated by working with those who are statistical experts like modelers. Is there any advice you can give me?

That's a great question. Something that I wanted to, so I have a former colleague of mine who I actually convinced to go to grad school for, she hasn't, of course not all of us have that opportunity, but if she was kind of curious about grad school, I was, I highly recommended her to kind of pursue that time, you know, maybe a couple of years to kind of dedicate to going back to the roots of, you know, like the educational roots and really emerging yourself in the kind of the fundamentals of the science, because I found that looking back, that's basically the the most important thing that I kind of took away from my experience and, going through the PhD in stats, is that I had a dedicated time and space and, you know, like brain kind of like space, I would say, dedicated to really understanding the, you know, deepest core, the core of the fundamental, because there's only very few fundamental principles.

I'm sure in any science, but in statistics, I would say, of course, there's, you know, a lot that builds on top, but understand, but, you know, once you learn the different approaches, it, and if you have great teachers, of course, that really helps, you realize that, yes, the fundamentally, you know, that there's variance, bias, variance, trade-off, there's causal inference, right, so and sort of getting a deeper understanding of those, starting from fundamentals, that would be, you know, that would be the advice that I would give someone who is, you know, fresh out of college. If people do have an opportunity to go back to school and at least like do a couple of, you know, like evening classes, but actually like classes with homeworks, like for real, you know, doing the reading, and it just, because that, that going back and taking the time to take those classes, I think, will pay off tenfold, hundredfold.

As far as working with people who have that deep knowledge, I would say, try to find colleagues who are willing to teach you, so I'm all about spreading and sharing the knowledge, and it's not as common in the industry, actually, this is another thing that I really appreciate it, and I really miss about the higher ed, even in the setting of kind of a non-academic setting within the higher ed. Being in that environment, it's very much inspiring to continue, for continued learning, for professional development, people are willing to, you know, give workshops, and you can always just kind of go to the other side and like attend a workshop. I used to, when I was a staff member, I would still go to those causal inference workshops that I would go to as a student, because they would still continue, and they are completely open to sharing.

So I guess, and that's what's kind of honed this idea in me that we need to, like, we need to be generous with our sharing of knowledge, especially among colleagues. I mean, I know there's all proprietary questions, if like sharing knowledge across companies, even though, like, I so appreciate POSIT, you know, supporting the open source community and the development of these packages and kind of dissemination of those, you know, completely free, that's incredible, and that's kind of the spirit that I'm trying, I'm really hoping to support within any company where I work, so I guess, yeah, finding colleagues who are willing to mentor and to, you know, to sit down and explain, that's another, maybe, path forward.

Statistical techniques in pharma

Could you just talk a little bit about some of the statistical techniques that you get to use in industry? Great question. I don't, I never really kind of thought of it sort of as special techniques, because what was something that I learned back in my time in the business school, being sort of supporting research there, is that usually the professors that go, that end up in business school, a lot of the times they come from very, very different areas, thinking it could be finance, it could be econ, it could be accounting, it could be org behavior, and a lot of the times, so they had very different trainings, including they use different tools, and if they're in econ or kinematics, they use different, maybe statistical, or the names of the statistical techniques, but again, fundamentally, those are all kind of similar concepts.

So basically, I would, the approaches, the general kind of overarching directions that I would split the techniques into are either it's predictive, or it's inferential, right, so either you are trying to find some kind of causal relationship, or some kind of an effect that would be more of an inferential, or what you're purely interested in is that, okay, we have these parameters, they're predicting this response, what if we gain new parameters, what is going to be the response, you don't really care about how those parameters are actually predicting that, it's just, it's kind of like more like this kind of deep learning, the machine learning, the black box approach, and so I think I find myself, you know, throughout my career, working with both, and it's kind of a nice, it's a nice skill to understand, okay, which one is more relevant, which approach works better for this particular purpose.

As far as specifically a pharma statistician, I think the biggest, one of the biggest so far that I'm seeing, biggest area of statistical knowledge that is very important is the ability to power, so-called power the trials, in other words, understanding of the methods that are involved in calculating what is the sample size that we need to collect to detect a certain effect, and there's this Bayesian borrowing, right, there's this adaptive design, so all of that is, that's where they are looking at us as statisticians to really provide guidance. By they, I mean, you know, clinicians, everyone who is involved in the development of the trial. So I would definitely pay attention to those, because that's what I'm trying to, like, you know, practically catch up on as far as the knowledge on my end.

Sample size and powering studies

So Morgana, if you want to request an unmute and ask your question live, that would be great.

Thank you. So my question was, could you please talk more about your considerations for sample size? I'm familiar with some techniques that are used for survey sample size determination, like Slobin's formula, but I'm, I always get a little confused when determining other survey sizes and end up needing to do a lot of research. So I'm curious what kind of things you think about and what techniques you're using.

Right. No, this is a great question. I try to kind of go back to the fundamentals. So what does it mean to power a study or calculate a sample size? Well, there's like several kind of moving parts to it, so there's so-called type one error and type two error. So type one error is the chance of detecting something where it's not there, and the type two error is not detecting something that is there. So by calculating the sample size, we're basically trying to get the number that would be needed to kind of account for both of these. Usually the sample size determination is pegged to some kind of statistical technique that will be used to ultimately find that effect. So usually you kind of need to agree on, okay, once we collect the data, what is the statistical technique that we're going to be using to find that difference, that effect size, be that like t-test or chi-square test or ANOVA, whatever.

So then stepping back from that, all of it can be done as a simulation. So you can envision your entire trial. And I find that's the most intuitive for me to understand what's going on behind the scenes is you just start basically simulating your trial. So these are my control, this is my treatment, this is what we observe, this is the test that I apply. Okay, am I rejecting or not? So in other words, you can soon realize that you don't really need the fancy packages that do it for you. There's much more value in understanding what, in sort of tracing those steps yourself. So I would actually, and with the accessibility of these LLMs or the larger language models that help us with coding right now, I feel like there's really, I would just kind of start from there, just start simulating. And you don't really need mathematics if you do that.

Getting into pharma and entry-level roles

We've got about eight minutes left, and one of them is from Will. Will says that he graduated with his master's in statistical practice at the start of 2023, and has had one internship, but no full-time jobs. So any advice for unemployed people in this area?

So I think going back to those three things that I mentioned earlier that helped me throughout my career is flexibility, curiosity, and dedication. So I think maybe see, just do maybe some like informational interviews, reach out to people that you think you'll like what they're doing, and ask them how they got there. Maybe they will have some kind of internship, try another internship. So maybe just kind of widen that horizon where you're thinking of applying yourself.

Something that helped me throughout my career, actually, I probably had dozens and dozens of little consulting projects on the side. Even just starting from grad school, it was just going to be maybe like for a pharma company, I would, you know, join their HR department for a little bit and help out to crunch some numbers, right? Or I had the craziest gig that I had was with, which was, I think it was People magazine, out of all places, right? So what would a statistician do there? But they actually approached me and asked me to analyze the data on fashion choices and the chance of winning an Oscar. So basically whatever the celebrities wore that day and whether they won an Oscar or not. I mean, obviously it was just kind of like a little bit of a silly sort of kind of like approach, I guess, or sort of like a fun way of doing it. But yeah, I was like, oh, somebody who wore red lipstick was twice as likely to win an Oscar that day. So, you know, but again, like this was a really cool just exposure to, you know, how People magazine operates.

Yeah, but also big question here. How did People magazine find you? Because if you weren't working out loud before that, and if you weren't making yourself visible, they never could have reached out to you, right? Exactly. But yeah, so be visible. Yeah. I mean, that's another thing, actually, that I was trying to.

Something that I kind of, in retrospect, what helped me a lot within Takeda is to kind of put my name out there, not only in the sort of volunteering or, you know, offering help, but actually, you know, so we are organizing these POSIT or R workshops throughout the year. And I just kind of send out invites to hundreds of my colleagues, and they see my name. And then, you know, when I kind of like run into them in the car, it's like, oh, Victoria Prince. Oh, I've heard, I've seen your name somewhere. It's like, yeah, probably those emails, the invitation to some R workshops. So, you know, just make yourself visible because, you know, because you're doing great work. So it's important for people to kind of realize that and then know that you're a go-to person.

R adoption and removing roadblocks

Can you talk more about the R adoption initiatives that you have helped drive since you transitioned to industry? Oh my God. Maybe advice on like how we can also be a force for that sort of like updating technology and stuff and, and, or moving to open source. Yeah. I mean, it's, to me, it's still kind of a learning experience because I'm trying to understand how to make a bigger impact. But yeah, I mean, it all started when, so when I joined Takeda, one of the, coming from a place where I basically like had like all the tools that I could possibly use and like, you know, on my laptop where, you know, state SAS, SPSS, like whatever professors wanted us to kind of use it, it would just, we just did it. So I, it's kind of, it was important to me to be like tool agnostic and R definitely is my go-to tool for anything statistical. So coming here and realizing there's really not, there's kind of like a little bit of a vacuum as for the people that are using R, but there's no kind of, you know, more like a general understanding was the kind of best practice to do it.

I started kind of like on a journey. Okay. So trying to understand who is in charge of this system versus that system, what's the best way to use it. And so, and what a year later, once I collected all of those bits and pieces, what I did is, and actually I was grateful that people here recognized sort of my kind of grassroots efforts and gave me a sort of like this unofficial kind of, they call it like R resource hub leader role that, so now it's like, makes it a little bit more official that I can sort of, it kind of opens doors a little bit more, I would say. But what I ended up doing is created a little SharePoint website where I kind of almost downloaded everything that I held in my brain and sort of organized so that other people could now, you know, make use of that. So it's kind of deliberate effort to collect and get to the bottom of things to collect information and then making sure you share it with others. And as you share, you realize that people actually share back and what they know, and it sort of enriches what you know.

So it wasn't kind of like, to me, it wasn't a challenge of like getting people to start using R, it's more like removing the obstacles for them.

It wasn't kind of like, to me, it wasn't a challenge of like getting people to start using R, it's more like removing the obstacles for them.

Yes, remove roadblocks. There's a concept that sort of, I think it's called force field analysis, where you're like, you have this goal of what you're moving towards, or what you would like to move towards, and part of the analysis is like, how can we push towards that, but also how can we remove the roadblocks so that natural momentum towards that is not impeded?

Statistics, data science, and domain knowledge

Arsene has asked a question, how do you see the relationship between statistics and data science? So more specifically, do you see clear lines between stats, ML, AI and other domains or subdomains?

Relationships between them? That's a great question. There's a whole bunch of these, there are different versions of these Venn diagrams, right? So there's usually overlap, there's math and stats, there's computer science, and then there's knowledge domain, and all of that overlap equals data science, right? So I guess I would probably subscribe to that.

Knowing the methods that you're applying, like to the best of your ability, try to get to the bottom of exactly what's going on behind the scene. To me, I found it extremely helpful so that I can right away see if that approach is not going to work, or if that approach is applied, maybe not necessarily in the best way, and how to improve upon what we've already applying. And I know these are abstract things, but the bottom line is to spend as much time as you can, as feasible, to understand the methodology. Then be tool agnostic, so computer science part, right? So try to understand kind of like how the, you know, these are just tools, basically. They help us apply that science, right? Apply that statistics. And ideally, we want to use the best tool that would be helpful in this situation. So in other words, like I wouldn't focus like, oh, well, I know R, then that's the only thing that I'm going to use. Sometimes Python actually is better, sometimes. SAS is great. So all of that, you know, try to be as, like, to develop breadth in the tools.

And then finally, like something that I'm trying to catch up on right now is, of course, the domain knowledge, the realization. What are the, because of course, math and, you know, coding is great, but ultimately you want to add value. You want to tie it all back to real life, and that's where the domain knowledge comes, and that's where so wherever you apply that, those skills, it's important to understand that the stakeholders, what are the, what are the things we need to consider from real life, you know, beyond the methodology itself. So I always, when I approach a new project, I always just sit down with the, with the clients, with the stakeholders, and ask, okay, what is the, what is the question? Give me in plain language, the question that we're trying to answer, or what, how are we going to use this before kind of throwing the machine learning, like the machine learning kind of model on it? Is it really necessary? What is the ultimate goal? So try to start with that.

Career advice and final thoughts

I just wanted to sneak in a question about career advice. Do you have any career advice broadly for anybody that's really helped you over time?

So career advice, I would say flexible, curious, dedicated. Right. So there is, I feel like it's still a great time to be a statistician, to be a data scientist, because data are everywhere. Now, of course, there's these deep learning models. There's LLMs that fear may take over. But I think we need to make sure we keep in mind what added value we can offer. So understand your skill set. If there's something, there's a gap, if you try to fill in. But I would say just kind of be flexible, just really explore, widen that reach. There's some very unexpected places where a statistical and data science skill and knowledge can be applicable.

Awesome. Thank you so much, Victoria. We probably don't have time to fit anything else in. Everyone's jumping for their next call. Thank you so much, everybody, for hanging out with us. We still have so many questions that were asked in Slido. I'm going to try to gather some of them, send them to Victoria so she can see them. And also maybe we can open up that discussion on LinkedIn, and everybody can chime in and help answer from the community. So Victoria, thanks so much for hanging out with us today. I hope you had a good time. Thank you. It's been a great pleasure. Thank you so much. Wonderful questions. And Libby and the team, thank you so much for doing all the behind-the-scenes.