In the field of statistics, we need specific skills necessary for being successfull. Debarshi Dey, currently the head of the Statistics and Programming and Data Management department at Morphosys, has spent the past 13 years in the pharmaceutical industry. With a PhD in statistics from UC Riverside, Debarshi believes that to be a “killer statistician,” one must adopt the roles of a detective, lawyer, and storyteller. 

Debarshi and I discuss these key characteristics that make a statistician effective.

We also explore these key takeaways from our conversation:

  1. A statistician should act as a detective by asking questions to uncover what is really being asked of them and navigating uncertain spaces to reach conclusions.
  2. When acting as a lawyer, statisticians must keep in mind who their audience is in order to present information effectively.
  3. A statistician should act as a journalist that understands how evidence is used down the line, such as when presenting figures or tables to clients or stakeholders.

Being a killer statistician is not just about number crunching and data analysis. It is about adopting multiple roles to uncover insights, communicate effectively, and ensure real-world impact. Developing these skills requires a combination of methodological proficiency, curiosity, and awareness of the context and audience. By following Debarshi’s advice and curating role models, we as statisticians can become successful and impactful. Listen to this episode now and share this with your friends and colleagues who can learn from this! Resource: What makes a killer statistician?  

Never miss an episode!

Join thousends of your peers and subscribe to get our latest updates by email!

Get the shownotes of our podcast episodes plus tips and tricks to increase your impact at work to boost your career!

We won’t send you spam. Unsubscribe at any time. Powered by ConvertKit

Learn on demand

Click on the button to see our Teachble Inc. cources.

Load content

Debarshi Dey

Global Head of Biostatistics and Data Management

Debarshi is currently Global Head of Biostatistics and Data Management consisting of approximately 30 Biostatisticians, Statistical Programmers and Data Managers across two global sites in Munich, Germany and Boston, USA for a Bio-Pharmaceutical company engaged in bringing novel drugs in Oncology and Auto-immune disease to patients.

With more than 11 years of experience in drug development in the pharmaceutical industry, Debarshi has extensive and proven track-record of impactful data-driven leadership in all phases of drug development and successful regulatory interactions regarding New Drug Application (NDA) and Biologics License Application (BLA) with USFDA, EMA, and PMDA.

Debarshi was the Lead Statistician of a pivotal oncology study with a real-world component, which resulted in a successful FDA approval in July 2020. In the course of the path to approval, Debarshi provided strategic leadership, led multiple interactions with FDA and provided strong statistical input in regulatory aspects of the submission. Debarshi has designed Phase I-III clinical trials, with complex adaptive features, calculating optimal sample size taking advantage of advanced simulation methods.

In his current role, Debarshi has been instrumental in setting up a fully functional Statistics and Data Management Department from scratch , including a CFR 21 Part 11 validated SAS programming environment for performing in-house programming.


Killer Statistician

[00:00:00] Alexander: Welcome to another episode of The Effective Statistician. Today I’m talking with Debarshi and I’m super happy about the topic. We will talk about what makes a killer statistician. I stepped over his post on LinkedIn and really loved it. Another reason why LinkedIn is so great. Welcome on the show Debarshi. How are you doing today?

[00:00:27] Debarshi: Thank you Alexander. I’m doing great and thanks for liking my post and reaching out and I’m looking forward to having this conversation.

[00:00:34] Alexander: Before we dive into what makes a killer statistician let’s dive a little bit into your story. What have you done so far and what are you doing now in statistics and especially what intrigued you about, these aspects of a killer statistician?

[00:00:51] Debarshi: I’m Debarshi. I am currently heading the statistics and programming and data management department at Morph Forces. I’ve been in the pharmaceutical industry for about 13 years now. I started my career at Novartis. I worked there for five years. Then I worked at a medical device company called Lebanova for a couple of years. And then since then, since 2018, I’ve been with Morphosis. We built up the department in the last five years. We are around 25 people. Before that I did my PhD in statistics from university of California at Riverside.

And I grew up in, a very densely populated Indian city called Calcutta, now called Kolkata. So I’m from India and but yeah, I’ve been in, lived in three continents and I really like that about myself, my journey.

[00:01:41] Alexander: Okay, awesome. So in your post you talk about, A killer statistician and he speak about yeah, of course as a statistician, we need to be good on the methods part. Yeah. On whatever we work on, multiplicity and adjustment or kind of regression models, what can, whatever the job requires from a technical point of view. But you also talk about three different roles we need to also play. What are these three?

[00:02:15] Debarshi: Yeah, so this is coming from my personal experience and what I have seen really, makes a difference. If you want to have that extra edge, you are right. You absolutely have to be very good in your statistics with your methodology. You have to have a deep understanding. But that is not enough in the pharmaceutical industry. The three roles which I found would really make you a killer statistician are you have to be a very good detective. You have to be a good lawyer, and you have to be a good journalist, if I can say, or a storyteller, right?

How do you present your case now? Why a detective is given a problem, a mystery to solve. And then he has to go after the truth, right? So there are a lot of unknowns. So he has to navigate through a lot of unknowns to reach his conclusion, right? What is out there? That’s true. Like when we are talking in the industry with the clinicians, they have all this fuzzy somewhat ideas that they want and here I think as a statistician, you have to dawn the role of a detective to understand, okay, what exactly is my clinician wanting? What is. He tells the clinical objective, but then what is the end point?

Many of them are already out there. The estimands framework is so popular and getting discussed. So much. So all of this, if you think is like navigating the unknown region, trying to find and trying to travel towards the truth. Especially in the frequent is paradigm. We know that the truth about the parameter is somewhere out there and when we collect a sample and try to estimate the truth. So if you think that kind of fits with the detective mentality the next Yes.

[00:03:58] Alexander: Yeah. Let’s stay with the detective for a moment. Yes. I can completely see that it’s not just about the truth, but it’s also understanding what really is a problem here.

[00:04:08] Debarshi: Yes.

[00:04:09] Alexander: I think about a couple of different scenarios, but one that directly comes to my mind was, I was given the objective to help with a new study. And we were given this criteria for this study should show. And these were so high. In terms of the evidence that’s, say it needed to show, and this security of the study needed to be that higher, that you actually didn’t need to study in the first place.

Yeah. So if you would know all of that, you wouldn’t need to study. Yeah. So yeah, people wanted to make it super, super safe. This was head to head study and people only wanted to run the study if it was sure to be shows that we have a great difference. But of course you can only be sure if you know already a lot. And this kind of, yeah, probability of success. Has upper limit that it can reach that is given by your prior knowledge. And here the problem was why were these where were these guidances coming from? Yeah. What, why was the study actually needed in the first place? Yeah, because it was not a study for regulatory purposes, so it wasn’t, an FDA or an email requirement or something like this, but where was it coming from? And that’s it needed the detective. Yeah. Talking to all kind of different people, understanding who said what was the kind of the motivations of the different people. Yeah. Where was the story coming from?

[00:05:54] Debarshi: Exactly. And Alexander, maybe unconsciously use the word. So there is one word which is so common in both the detective paradigm and the drug development we do today. And that is evidence. Call our world, evidence best medicine, right? Yeah. So to get a new drug in the market, you have to provide the right evidence. And this exact same word evidence is used by detectives, right?

What evidence can I gather to, to come to my conclusion, finding out the culprit, right? So you, there you go. How important to have this mentality, right? Yeah. The other thing is that when you as a detective, You ask a lot of questions. Because you are trying to see, you are trying to also see beyond what is spoken. Beyond what is obvious, what is the underlying texts. As a statistician, you also have to do that. You have to go beyond what is obvious. If, as you said, if it is obvious, then I don’t need a study. Correct. If you already knew what it is, right. The sample, you don’t need to have a the element of randomness is what keeps statisticians in job.

If there was no randomness, if there was no chance element, yeah. Then don’t, wouldn’t need a statistician. You would’ve just needed maybe an accountant just to add up some numbers. Correct. So I think this navigating in this uncertain space, understanding, questioning the assumptions, understanding to see what are the subtexts below what is spoken out. This is very important. So as a statistician, basically those who are listening to this podcast, I will say, reflect on what kind of questions you are asking to your stakeholders, what kind of information you’re trying to seek out. To your stakeholders. I always tell to my team that statisticians are not glorified calculators. We are not there to just calculate, we are there to seek out and give structure to a quantitative structure. To a qualitative scientific question.

[00:07:51] Alexander: Yeah. The other point is also it means you need to talk to lots of different people. Yeah. So have you ever seen a detective that only sits in front of his desk in any movie? No. They talk to lots of different people. They go to different places. Yeah. They see where did the crime happen? Yeah. Yes. And as a statistician, you also need to look into how did the data happen? Yeah. What, where is it collected? Yes. Why is it that you know, bad or good or, where are different features of the data coming from? And so you need to understand all these kind of different details and set forms as a bigger picture. So as a statistician, don’t just start your PC, talk to people.

[00:08:44] Debarshi: Yes. No. I think you make an excellent point about where the data is coming from. What is the context, right? Because that makes so much of a difference, right? I will give you one current example. You know this when you said that it just popped in. So we are doing something called an anchor based analysis. I don’t know if anchor based analysis is when you have regression reported outcome endpoint. But you are not sure about what is the clinically meaningful treatment effect.

Yeah. Because the patient got outcome. It’s a, yeah. And so what you do is you try to anchor it to another more well accepted scale. And then if you know the, what is the treatment effect in that scale from that, you try to derive. Your patient reported outcome scale, what is the clinically meaningful, right?

So the other scale we have is something we are using patient global impression of change, PG. And this is something where patients are asked after 12 weeks, after 24 weeks, how are you feeling? Are you feeling better now? Are you feeling okay? Are you feeling worse compared to the beginning of treatment? And this question is asked, as I said, at different time points. Always the question is, how are you feeling compared to the beginning of treatment? Now you see this is a very, okay, as a, if I’m a statistician who’s not wearing a detective hat, I will take this data and I will run my model and get my estimate very good. If I’m wearing my detective hat, then this is how I will think. If I’m asking this question after 12 weeks, how are you feeling compared to start of treatment? The patient will say whatever, but if I’m asking this question after, say, 36 weeks, how are you feeling from start of treatment after 36 weeks?

You forget how you were feeling 36 weeks before compared to how you know that at 12 weeks. Because then you really, there’s a patient, the patient has gone through so much in the last 36 weeks. It’ll be really asking a lot if to, for the patient to remember exactly how he was feeling and then contextualize his score. So as you go over time, basically the point is you will have less and less confidence on your anchor compared to when you take the anchor at an earlier point in time. So this is where exactly Alexander, when you said that you have to understand where the data is coming from. How the data is getting collected, when the data is getting collected. You try to, you connect the dots and know what makes sense.

[00:11:10] Alexander: Yep. And you can also not only look back where the data is coming from, you can also look forward where the evidence going to. You’re based in Munich. One of my, most influential days I had during my, what, many years career. Was a day in Munich. I spent a day in Munich with sales representative at the time visiting psychiatrists. Because I was working in that area at the time, and that’s the first time I saw. How is the data that we produce goes through these kind of promotional channels and ends up in a discussion between a sales rep and a physician?

[00:11:56] Debarshi: Wow.

[00:11:57] Alexander: And one of the things I noticed was there was not a single table. They always had data visualizations, always a figure because of course they had very limited time. Yeah. Sometimes they had just had maybe two minutes. Yeah. To talk to the physician. And so you need to present your data very effectively. And nearly all of these data visualizations were not copy and paste from a manuscript or from the presentation or whatsoever. They were always adapted for the specific situations, the message, the audience, and so on. And there, I learned a lot about, okay, it’s not just enough to throw a table over the fence. You need to also enable all the people down the evidence pipeline to understand it. The strengths, the limitations and all these kind of different things. So can work effectively with it.

[00:13:02] Debarshi: Yes. And this, you are very right, Alexander, and this brings in I think the next, if we can go the lawyer part.

[00:13:10] Alexander: Yeah.

[00:13:10] Debarshi: Where you have to have, the very fact that, okay, you’ve present data visualization or whatever. You know what to present, whether visualization or a table. If you’re writing a dossier for FDA submission, then you of course have to have tables. You can’t. But I think the underlying thing is you have to know who is your customer. Who is, why am I doing this? For whom, many times I’ve seen statisticians are so excited about statistics, they forget. For whom they’re doing this. I think this is something we should always, when I’m presenting some data or something for my, for the CEO, right?

Sometimes the CEO reaches out and asks for something or my boss or for the board, whatever. I have to present it in a particular way. When I’m going for a face-to-face FD meeting, I have a type C meeting. I have to do it in a particular way. When I’m having a side meeting on the sidelines of ASH or ASCO with the key opinion leaders, I have to present it in another way. I think this is very important to always ask, who am I doing this for and how can I best communicate my message? It cannot be one size fits all situation.

[00:14:18] Alexander: Yeah. Completely agree. And the interesting thing about the lawyer is that it’s not just about the evidence, but also all the rules around it.

[00:14:34] Debarshi: Yes.

[00:14:35] Alexander: So you say and the rules are different for every court. Yeah, it’s different in US compared to Germany, or to UK, or to Japan. Yeah. You need to be kind. Some kind of international lawyer. Yeah. Because you need to know the rules in all these different places. You need to know how the F FDA works, how the email works, how the P M D A works. If you’re working in post commercialization, you need to know how the and G B A work or the hush in France or the ICE in UK or the C A E T H in Canada.

[00:15:18] Debarshi: Exactly.

[00:15:19] Alexander: Yeah.

[00:15:20] Debarshi: Exactly. Yes. And you have to present it in their language. All courts have a very specific jargon or, language vocabulary. So you have to speak in that vocabulary. The other thing I find very interesting about lawyers, I don’t know, I had in, in my life had to deal with lawyers and they always go to, state something called precedence. What was done in a similar case before. That really makes your case extremely strong. 99% of the time we can be sure that if we are facing with a situation someone else has faced that situation before. And you as a lawyer becomes extremely powerful if you can show that in a similar situation, a different, in a different case, this was the outcome. And then you argue coming from that position that really makes your case. Very strong and you will also know what exactly how to navigate, how to proceed in that case.

[00:16:18] Alexander: Yeah. Or if the other case says, wasn’t going in your favor. Yeah. How is your case now different than the other case? Yeah.

[00:16:27] Debarshi: Exactly.

[00:16:28] Alexander: Why is, should the judge. Or the regulators or the payers, now decide different for this one. Yeah. If your drug is second on the market or third on the market. Yeah. Okay. Who’s it different to the others? Yeah. Where’s the differentiation coming from? That is really important.

[00:16:48] Debarshi: Exactly. And as I said, the word evidence linked detective with the statisticians. I think this concept of false positive and false negative links, lawyers with statisticians, because when you are in a court, you can think of your false positive rate, type one error as someone who’s not guilty is pronounced as guilty. The court would like to, because you would the presumption is that the person is innocent until proven guilty. So the presumption is, That the person is not guilty. Yeah. In the same way the health agency says the drug is not working until you can prove that the drug is really working. Correct?

[00:17:26] Alexander: Nice. Yeah.

[00:17:28] Debarshi: It is. It is a very similar framework.

[00:17:31] Alexander: Love it. Love it. Yeah, absolutely. And then there’s the third part the journalist. Or storyteller. I guess you are not talking about stories in a way the stories in some kind of movies or something like this but tell me a little bit, what do you understand by stories here?

[00:17:52] Debarshi: Yes, by story I mean that you should have a plot when you just say, okay, my hazard ratio is point A two. For example, I’ll give you, give an example. We, in my team last year for a particular study, the top line data was read out. And, the statistician and the clinical team, they put together a few things and they said, okay, the response rate is 32%, right? They had few slides and some other things.

Now, when I looked at the slide, I didn’t know what it means. Is it good? Is it bad? Is it something that I should be disappointed with? Excited? What is the story? What is the context? What am what are the what is the standard of, what is the treatment effect for what is out there in the market? So you have to understand that whenever you are presenting data, you have to contextualize it. Whenever you’re presenting data, this is linked to what I said a little while, be before about having your customer your stakeholder in mind. What am I trying this person my customer to take away from the take away message?

What am I trying to convey? So many times I see that statisticians, of course, who are not that experienced, in their five, six years of experience, they present a lot of data numbers in their slides without telling, okay, what is the takeaway message from them? So what I meant by a journalist, by a storyteller is that you have to have a plot. You have to tell where it, why it is coming from, where it is coming, what is the interpretation, how should this be contextualized with what is out there so that there is a definite understanding of whatever conclusions we have drawn from the data.

[00:19:30] Alexander: Yep. Completely agree. If the journalist would say unemployment rate is 7%. Is that good? Is that bad? Yeah. Is it an improvement whatsoever? Yeah.

[00:19:40] Debarshi: Exactly.

[00:19:41] Alexander: There will always be okay, it’s an improvement. It comes from these kind of different factors. Is this driven by that person, this, external organization whatsoever? What’s the inflation? Has this kind of influenced all these kind of different things? And as a good statistician, you work with your cross-functional team to come up with a similar story. Yeah.

[00:20:07] Debarshi: Exactly.

[00:20:08] Alexander: And you can use all these different elements of stories. Yeah. Who’s the hero? Who’s the villain? Who’s the guide? Yeah. Here. Where’s the challenge that the hero is facing? Yeah. What is the inner challenge? What is that alter challenge? Do you have similar things in your data presentations as well?

[00:20:32] Debarshi: Exactly. No. Absolutely. And this is something, you cultivate, right? As or as a story, right? A story. There is no right story and wrong story. There is only a story which is interesting and not interesting.

[00:20:44] Alexander: Yes.

[00:20:45] Debarshi: You can’t say this story is right. Similarly, when you’re presenting data, when you’re putting forward this slide, there is no right and wrong. How interesting can you make it? Will the other person go remembering what you have said? So you have to always put yourself in the shoes of your client, of your customer, whoever it is. You can have clients within the organization, outside the organization and always present it in that way.

[00:21:10] Alexander: Yeah. And it needs to then drive a decision.

[00:21:14] Debarshi: Yes.

[00:21:14] Alexander: Yeah. A journalist always writes in some shape or form that he can enable the reader to make a decision, change his mind whatsoever. Yeah. Think about what is your goals that you want to achieve. Yeah. There’s also these kind of different parts of journalism that you can think about. Yeah, of course. There’s the, there’s news reporters. It, just talks about the news. But there’s also the person said comments on it. Yeah. So I think too often we only stay in the news reporter area and we don’t go into the commentary. Yeah. Okay. And that is where a lot of value comes in. Yeah. Where you know, your C E O asks you, Debarshi! What do you think are our next steps here? What should we do? Then you need to step from the news reporter into the commentator and give your own perception on things. Yeah. And that is quite valuable.

A journalist that comments on things,. We know they have work in this area there may be an expert on military or security or environmental health or any other area. They have a lot of knowledge in this area. So their comments help us make a decision. Yeah. They trust them. And in the same way, your CEO or your head of armor and your, any other partner Yeah. Will trust you in your interpretation of the things and what should be done now.

[00:22:58] Debarshi: Exactly. Yeah. You’re very right. I also think another what statistician in the pharma industry, we are actually at the intersection of science and business.

[00:23:09] Alexander: Yeah.

[00:23:09] Debarshi: This is, if you think of, we are statisticians, we know when diagram with two circles, science and business, and then there is a intersection that is where we operate, correct? And when we present our data in a storytelling. We are so much more powerful because then we can really influence in both of these things. If you think about the good scientists, when they present their findings in a, as a script that is so much more powerful than when they just present a law or a, a Newtons third law, second, whatever. Yeah.

[00:23:42] Alexander: Completely agree. Love it. So we have talked through all these different aspects of what you can do to become a killer statistician. Of course, you need to be great on also methodological skills and there’s a lots of training out there. And also invest in your skills, becoming a detective, becoming a lawyer. And that doesn’t mean you need to read law books, learn from lawyers and learn from journalists in terms of how you write a headline, how you present your data. There’s a lot of opportunity for you to make a bigger change for patients if you develop skills in these areas. Debashi, for yourself, what do you do personally to develop these skills?

[00:24:39] Debarshi: I listen to podcasts like this to start with, right? I think the more I interact with smart and accomplished statisticians, I think there’s so much I learn from them. That is something I will definitely encourage that, have some role models. I’m sure you’ve come across. We all have come across some, very good mentors, some very good bosses that we have worked with. That is something I do then I do a lot of reading, there are a lot of stats review from FDA and IMA on the websites. You see what, how they’re looking at the data, how they’re interpreting, what are they looking, what are the and then try to. When you are doing your stuff, try to match that because at the end of the day, we have a lot of customers, but the mean customer are the regulators, right?

So at the end of the day, we have to convince them that, that with our data. So that is something I do. What else do I do? I think I just I’m a very curious person, right? So whenever,

[00:25:37] Alexander: curiosity is great. Yeah.

[00:25:39] Debarshi: Yeah. Curiosity is something that, and I explore, right? Reading, listening to people, listening to podcasts, watching good content on YouTube. These are the different ways I keep myself updated and learn how to think in different ways.

[00:25:54] Alexander: Awesome. Yeah, so there’s lots of further episodes you can listen to here on the show about leadership. Just scroll through. So now up to 300th episodes that we have and you’ll find a lot about, Ciscos and these different perspectives. Follow Debashi and myself on LinkedIn where you will also see a lot of news on these kind of things and yeah, become an effective and killer statistician. Thanks so much, Debashi for this great discussion today.

[00:26:27] Debarshi: Thank you, Alexander, for having me.

Join The Effective Statistician LinkedIn group

I want to help the community of statisticians, data scientists, programmers and other quantitative scientists to be more influential, innovative, and effective. I believe that as a community we can help our research, our regulatory and payer systems, and ultimately physicians and patients take better decisions based on better evidence.

I work to achieve a future in which everyone can access the right evidence in the right format at the right time to make sound decisions.

When my kids are sick, I want to have good evidence to discuss with the physician about the different therapy choices.

When my mother is sick, I want her to understand the evidence and being able to understand it.

When I get sick, I want to find evidence that I can trust and that helps me to have meaningful discussions with my healthcare professionals.

I want to live in a world, where the media reports correctly about medical evidence and in which society distinguishes between fake evidence and real evidence.

Let’s work together to achieve this.