Better Dose Decisions and a Culture of Statistical Innovation

Marie-Karelle Riviere

Chief Scientific Officer at Saryga

Marie-Karelle is a statistical expert with strong technical competences, driving innovation for more than 10 years in both industry and academia. She provides methodological support to statisticians on complex innovative approaches for clinical trials, with a specialization in early phase dose-finding oncology. She has excellent programming skills and a demonstrated experience in developing R packages.

Anaïs Andrillon

Senior Statistical methodologist at Saryga

Transcript

00:00
You are listening to the Effective Statistician Podcast. The weekly podcast with Alexander Schacht and Benjamin Piske designed to help you reach your potential, great science and serve patients while having a great work-life balance.

00:21
In addition to our premium courses on the Effective Statistician Academy, we also have lots of free resources for you across all kind of different topics within that academy. Head over to theeffectivestatistician.com and find the academy and much more for you to become an effective statistician.

00:49
I’m producing this podcast in association with PSI, a community dedicated to leading and promoting user statistics within the healthcare industry with the benefit of patients. Join PSI today to further develop your statistical capabilities with access to the ever-growing video-on-demand content library, free registration for all PSI webinars, and much

01:13
head over to the PSI website, tap PSI Webblog to learn more about PSI activities and become a PSI member today.

01:30
Welcome to another episode of the Effective Statistician. Today, I’m super excited to have two statisticians from France on my podcast. But of course, we’ll speak English. My French is really terrible. I had a couple of years in French and that was a disaster. Anaïs and Marie Carell. And we’ll talk a little bit about early phase things, but also about how we can be more innovative.

01:59
within companies. And so that’s a very interesting topic that we are bringing together. Anaïs, maybe you can start and introduce yourself. Thank you, Alexander, for having us. It’s a real pleasure to be here. I am Anaïs Andréon. I am a Senior Statistical Methodologist at Sariga. I joined Sariga more than three years ago, right after finishing my PhD in Biostatistics. I completed my PhD at INCERN.

02:27
a French research institute and my research focused on dose finding clinical trials in oncology. And while doing my PhD, I was also working as a methodologist at APHP de Paris University Hospitals for about three years where I helped design and analyze clinical studies. And I also did a lot of teaching, which I really enjoyed. I started my career in a very academic environment, but at the end of my PhD, I faced

02:57
this dilemma that I would say most PhD students face. Should I stay in academia or should I move into the pharmaceutical industry? And that’s when I had this discussion with Gaëlle, the CEO of Sariga. She told me about her company and it actually matched exactly with what I was looking for because I wanted to continue that methodological research work. But at the same time, I also wanted to be involved in innovative clinical trials.

03:27
So Sariga felt to me the right balance. Overall, my background is mainly in early phase triumphs, but since joining Sariga, I have the opportunity to work across all phases on wide variety of projects, including adaptive triumph designs, both by Asian and Frequentist. Awesome. Marie-Carelle, please introduce yourself. Hi, I’m Marie-Carelle Rivière. I’m a Chief Scientific Officer at Sariga.

03:56
For those who don’t know, Sariga is a consulting company dedicated to statistical innovation and decision-making in healthcare. And our job is to support pharma and biotechs with advanced statistical methodology to optimize drug development and clinical trials. That’s also where we have close collaboration with Academia and we publish novel methods. I’m also recently a fellow at INRIA.

04:22
So the French National Institute for Research in Digital Science and Technology, in addition to my role in Sarega. So my background, I hold a PhD in biostatistics and especially in designs for early phase dose finding in oncology. My PhD, it was a collaboration between industry and academia, Cervier and INSEAM, which is the French National Institute of Health and Medical Research.

04:49
So I had the opportunity to work on my PhD on research project, but also on early phase clinical trials, because that was a nice experience. Then I chose to do a postdoc in pharmacometrics at Bishaw Hospital in Paris. And during approximately seven years, I joined the statistical methodology group at Sanofi.

05:13
It was a very nice experience where I was able to support all phases and all therapeutic areas and it was very nice. And today, since 2022, I joined Sariga where I lead the methodology team. So in fact, I’m managing the team. So we are currently seven plus me and Gaël and I’m running the project. And because I really love methodology.

05:40
I organize the team so I can still have time to actively be involved in the science itself. That is pretty cool. And we will talk about this a little bit further. You also presented what we’re talking about here from a kind of content perspective at the PSI conference. If you want to look into going into slides around what you’re about to listen to, that will be in the show notes on the Effective Statistician homepage.

06:09
This all has to do with Project Optimus and there’s been a previous episode where we had a deep dive into Project Optimus, but maybe you can have a quick kind of overview what Project Optimus is and how your research actually relates to it. The Project Optimus is an FDA initiative. It was launched back in 2021 by the FDA and they published the guidance on it in 2024.

06:39
Basically, it’s about rethinking how doses are selected in early phase oncology trials. So this project arose from a shared understanding from different experts, including clinicians, pharmacologists, statisticians, and regulators. And this shared understanding is that doses and schedules of new therapies are often inadequately characterized by the time a drug reaches the market. And that’s why the traditional dose-finding paradigm should be revised.

07:09
So if we can explain it quickly, historically the paradigm was the more is better. So it comes from chemotherapy. That means you push the dose as high as patient can tolerate it. And this is what we call the maximum tolerated dose, the NCD. Because there were this assumption for chemotherapy that efficacy and toxicity both increase together as the dose increases.

07:34
But this logic breaks down for modern targeted therapies and immunotherapies because efficacy can plateau well before you hit the maximum tolerated dose. So the highest dose is not necessarily the best one. And we would like to recommend lower dose that provide the same efficacy but much less toxicity. And in the past for many approved cancer drugs, we push for a dose level that turn out to be too high.

08:02
And post-approval, they had to go back and modify the dosing regime and based on safety and tolerability results, which is the worst case for a company. So that’s the context in which project of things was born to really reform how we optimize the dose. But as you said, you had this previous episode, Alexander with Ayoun and Karji, where you covered it deeply. So maybe I can continue on what this recommendation means for industry instead of on the initiative again.

08:32
So in practice, Project Optimus is asking for industries to move beyond simply targeting the highest tolerable dose and instead find the dose that provides a trade-off between efficacy and toxicity. So a dose that optimizes the balance between benefit and risk. And this is what we call the optimal biology countries. And to achieve this, LDA recommends, we can say that they recommend two main things.

08:59
The first thing is that the dose decision should be based not only on toxicity, binary toxicity endpoints, but also on efficacy data, tolerability, patient reported outcome, PKPD data, activity data. So the idea is to have a more comprehensive approach integrating all these multiple endpoints to inform the dose decision.

09:23
And the second major recommendation from Project Optimus is to explore multiple doses early in the clinical development through the proper randomization phase. So once you have identified a few potential optimal doses after the dose escalation phase, they encourage what we call dose optimization phase, where you randomize patients across candidate doses to figure out which one is actually the best.

09:50
So it sounds simple in theory, but in practice, it’s a real challenge for industry because it means finding a balance between regulatory expectation and what’s feasible for ongoing trials. So this situation really requires reactivity and adaptivity for industry. And that’s exactly what in that context that we do our research at Sarega because we develop metals and tools to help companies in this process.

10:17
of those selection and those optimizations while still converging towards an optimist compliant development. If you talk about efficacy endpoints in oncology, what would that be? Because I would assume things like toxicity, can also in side effects, you can see pretty fast, but efficacy probably takes much longer to actually observe. Yeah, a lot of methods for that because indeed, yeah.

10:46
For efficacy, sometimes we observe it like not at the first cycle, but for example, objective response rate or for sure, PFS, we can observe it. Now, so there is methods for that that you can take into account that these are in fact time to events. When you decide to include a new patient in the trial, you don’t have yet all the complete observation of this endpoint. So there is methods for that. We will go more deeply in the method later we propose.

11:16
but you can do different steps in your clinical trials. So for the first step, when you include your first patient, you can base your decision only on toxicity and use the efficacy endpoint when they are available. So you can do in a step. So those escalation step is usually based on toxicity endpoint and the dose optimization phase is this phase when you use all the possible endpoint, but you already have included many patients. So you already…

11:45
have data available in your trial. And we used preliminary evidence of efficacy. if I can give two examples of what we used for some clients. So as Anaïs mentioned, it was ORR or tumor shrinkage that enabled to drive the decision. So let’s talk through your methods that you’re proposing. Maybe if I may. All this context, that’s also why there is a new special interest group.

12:15
on early phase that was recently set up within the PSI. I’m colleague and Anaïs is also part of this group. And the objective is to bring people together and share practice and share some ideas on what is done, what we could do and so on. And it’s open to new members. So anyone who is listening to the podcast, it’s very welcome to join. Yeah. I wanted to give you this information. How do people join kind of?

12:42
contact you via LinkedIn or send you an email? They can send me an email. There is a webpage on the PSI website with our contact. Okay. Awesome. Very good. We’ll link to that in the show notes as well. And let’s mention these kinds of things at the end of the podcast as well. If you’re now listening to this and think, oh, I need to remember all of these things, we’ll have a couple of action items at the end so that you really get everything. New method.

13:11
We have a couple of different steps and you already highlighted kind of the dose selection and then dose optimization phase. Can you walk us a little bit more in a more detailed view through these kinds of different steps and methods? Yeah, our method is called UDES-P for utility based on dose, exposure, safety, pharmacodynamics, efficacy. So it’s in several steps. The first step is the POPP-PK model.

13:39
So we link the dose regimen, so it can be dose and schedule, which is interesting. It links into the exposure because we know that two patients on the same dose can end up with very different exposure. And that’s maybe very interesting to go through the PK part. Then once we have this pop PK model, we want to model each endpoint in function of exposure. So any endpoint that matters for the decision,

14:08
can be included in the approach. It’s not fixed. It’s very flexible. So we talk about safety. So it can be, as usual, those limiting toxicities. It can be grade 3, 4 TEAE or TAE leading to those modifications. We also talk about activity. So for example, we already integrated target engagement. And for efficacy, as we mentioned earlier, it can be tumor shrinkage, ORR, or…

14:38
any endpoint that matters in the decision. So we have this first part, those regimens to exposure, then exposure to any endpoint of interest. But then we need to go back from the exposure to the regimen because clinicians, they don’t prescribe an exposure, they prescribe a regimen. And so that’s why we combine the two. And because we are in this biogen framework,

15:06
At the end, we have the full distribution of every endpoint for each dose regimen. And so we have also uncertainty around our estimations. And finally, what we do is we have a utility function that brings the endpoint together into one measure for each dose regimen to guide the decision. This utility is constructed with the clinicians.

15:33
because they are ranking the endpoint by importance and they are giving some information to have a trade-off between the endpoint. For instance, we are asking them, it’s just an example, how much extra toxicity would you accept for 10 % more tumor shrinkage? Okay. So it’s like a preference survey. Exactly. Exactly. And with all these questions and preferences, we are eliciting the weight

16:03
of our utility function. We have a utility for each dose regimen and one we need to choose and we made some propositions with HMIS and we suggest to look at different criteria. to have the utility at each dose regimen and their credible interval, a rule could be to take the dose regimen with the highest utility, which is very simple.

16:31
But something we also explored because we wanted to take into account the uncertainty is the probability of each dosing regimen to have the highest utility. Okay. Yeah, because one is basically, if you think from other examples, there could be a network meta-analysis and you get a rank out of it. But of course you also get probabilities for being the best out of it. And that’s a very similar approach. Exactly. And we recommend to look at

17:00
all these criteria together to make the decision. So that’s the approach and the approach can be used at different times during the dose finding. So because evolution is quite slow, a few years ago we were still using the 3 plus 3. So first we thought applying our approach at the end of a dose escalation phase, for instance with a model-based approach. But the objective then is to

17:26
follow Optimus recommendations and have a dose optimization phase. So the idea would be to first have a dose escalation phase based on a model-based approach, then use our UDSP approach to have this probability of being the best dose, and then run a dose optimization phase. And there are several possibilities depending on what are the constraints. So we thought randomizing patients to different dosing regimens

17:56
according to their probability of being optimal. Let’s imagine, for instance, the company can only select two doses. We can do that with the two best doses identified by the method. The constraint with the method is that if you want to use the PK part, because it’s quite flexible, you can imagine not having this PK part and modeling directly the dose to the endpoint, but it was really of interest in the project we supported to have it.

18:25
If you want to have this PK part, of course you need to have a few cohorts included and PK sample available, which may take some time. And so you can apply the method only after a few cohorts were included and the pop PK model was built by the PK team. I see that. And I guess if you have, let’s say one pretty good regimen and then the second and the third are very close and then there’s…

18:54
bigger gap to the fourth and probably would go for three to better understand everything from a dose selection point of view. Exactly. If I can add a few things on top of that, maybe, and it’s maybe on the flexibility of the method as Marie-Claire already mentioned, but there is other things that make the makeover flexible, is that the end point of interest that drives the decision are not fixed by the method.

19:22
So you can use whatever endpoint makes sense for your study. So the idea is really to capitalize on all available data. Sometimes we include PFS, ORR, PD marker. It really depends on the compound of interest for your patient population. And secondly, what makes also the method flexible is that you can use different exposure metrics for each endpoint.

19:47
Again, it depends on the pharmacology of the compounds. For example, peak concentration might be more relevant to explain toxicity, but cumulative exposure might be better to explain efficacy. It’s an example. So this has to be discussed with the pharmacology teams. And that’s why also it’s important to work closely with them. And the third thing I would like to add is that the UD-SP method also works on, as Marie-Carré said, optimizing regimen schedules.

20:17
So it’s not just doses because two different regimen can have the exact same dose level, but a completely different rhythm of administration. And we end up with a different toxicity and efficacy profile. So with the UDSP design, you can compare those directly because we take into account the exposure in the modeling and not just the dose model. A really last interesting thing if I can add, which is related to exposure, is that we can…

20:47
include covariates in the population PK model. So we can see why a given patient ends up over or under exposed, which in fact opens up a real path toward more individualized dosing. Because in the example Marie-Carelle presented, when we applied the method for the clients, the exposure was different, for example, between the male and female, which led us to recommend potentially different

21:15
dosing regimen for men and women. So to conclude more globally on the method, if we step back from the first question, you ask about the Optimus project. So what does Project Optimus ask for? They first specify that the dose should be based on the benefit risk, not on toxicity alone, but built on different sources of evidence, safety, tolerability, efficacy, and PKPD.

21:42
And they asked for several doses to be compared early in the development through randomization phase. And this is exactly what UDES-P, for what UDES-P was developed, because it’s a quantitative tool, a transparent tool that helps teams to justify a dose. So it’s flexible. You can apply it whenever you are in the process, whether at the end of a dose escalation or for a full optimization phase, if you are ready for it.

22:10
Pretty impressive. Thanks so much. Very well thought through approach with a couple of nice applications of decision theory, especially if I think about the utilization or the utility based approach, which of course there’s probably lots of ways how you can get these utilities. It’s another research subject actually. Yeah. But if in the end you need to make some kind of trade-offs anyway, yeah.

22:37
By having that questionnaire approach, at least you make it transparent. How you make trade-offs. That I think is very often the biggest challenge. The trade-off happens by way of discussing things in a room and that’s very difficult to document. Thanks so much for that introduction and all that expansion on your presentation at the PSI conference and I very much hope your corresponding paper.

23:07
also gets out there and published quite soon. And what I see again and again is from you and your colleagues, quite a lot of presentations. If I just think about the PSI conference, there were several presentations and posters and things like this. As Marc Carrel, you just mentioned, you’re not a big team. You’re not a hundred or 150 or even more.

23:36
statisticians, you’re a very small team with less than 10 people. How can you drive so much methodological research forward while being a CRO? Of course, you also need to work on clients to basically pay your salaries. Yeah. And that is my kind of also working in CROs or have been working together with CROs for quite some time. That’s really my biggest question.

24:05
That’s an interesting question and that’s also interesting that you are presenting it as a tension between methodology and client’s work. Because in fact, that’s exactly why Sariga was built for. So Gaëlle, when she created Sariga approximately four years ago, it was really she described it as a company that supports statistical innovation and that is doing statistical methodology.

24:33
We don’t have a methodological identity by accident. It’s really on purpose and we are feeding it. first on the team, everyone holds a PhD. We all spend some time in academia and we are still continuing to have a collaboration with academia. We are encouraged to publish and we are going to several conferences each year. So

24:59
Why it’s working is because the main part of our work is to do consulting for clients. But we are supporting clients on complex issues. They come to us when they have specific issues that are not standard and they want our support. And that’s a very good example because the Udacity method we just presented is in fact a collaboration with a Switzerland biotech, BabyOFarm.

25:26
And so that really shows that we can do innovation. can create new methods based on client’s work. Usually, this client’s work, usually it relies on complex methods that may already exist. And so we are looking in the literature to pick the right method and to apply it depending on the needs of the clients. But that’s already methodology because the method, maybe it was already published, but it’s not standard.

25:56
It can be unusual and not trivial. We need to implement it and it’s still innovation. And sometimes we end up with some questions and problems that don’t have an existing method published in the literature. And that’s also where we are doing innovation and we create our own approaches to answer the problem we are dealing with. And that’s something, for instance, we are doing with Servier.

26:26
on biomarkers methods. And as I said, that’s also what we did, for instance, with Debye Farm on UDSP. So every day we are working on methodological methods for clients. It can be simple or more complex. And that’s the main part of our work. But it’s true that we have also part of our work that are not built to anyone. And we spend this time for research. For problem projects, think…

26:55
that matter and that may be of interest at some point. For this project, we can do some collaboration with Academia or we can create working groups. So for example, we did a scoping review of early phase designs. So we review more than 700 papers of designs. And for that, we created a working group with people from Academia and from industry. And it was really

27:23
few research not built to anyone, but we really think that matters. And we are also participating in lot of working groups. Yeah. So basically you attract especially work that needs not just the standards implementation, but where there’s something complex in it. There’s something that needs a little bit more brain power than the standard study. Exactly. And I think that is what is motivating the team.

27:51
My team is really enjoying having complex approaches to set up. Usually our clients don’t come to us for standard analysis. They go to discuss with standard CRO and we’re here really to support complex issues. How do you select your people that join your company? So we have a process with several rounds of interview. So first, normally the first thing, as I said, is to hold a PhD.

28:21
And then we have several rounds where we are evaluating the technical skills of people, the adaptability and so on. So it’s a global decision with different steps. Okay. So do you look specifically for people early in their career or can anyone join that is also, I don’t know, 20 years in the industry? Oh yes, it’s quite large depending on the need and what people already have in the team.

28:50
So Anaïs, for instance, she joined just after her PhD. Now it’s three years that she’s with us, but it was right after her PhD. But we have, for instance, another person that worked at EMA before joining us and have a lot of experience. So we are open to all kinds of profiles. Anaïs, what made you join Saria? I think it was the balance between methodology and clinical trials applied to industry.

29:18
Because after my PhD, I really wanted to continue on that research work, but I was not sure to have a full academia career. And Yagel told me about Saneriga, she explained me, and I also wanted to join the story of Saneriga because the company was new and I was just finishing my PhD and I was really motivated to join a small company, really a dynamic company.

29:46
What do you suggest to other employees who also want to drive innovation forward in their companies? And they may not have the benefit of working like in your setup where it’s small, focused on these more complex projects. What can you do so that you can drive things forward even up to publication? That’s a good question. Indeed, it’s not straightforward. I think it has to be first person initiative.

30:14
We have to be curious, we have to be proactive. But the main skill is not knowing directly how to write a paper. It’s knowing how to spot things. And what I mean by that is being able to notice when there is room for innovation. So as Marie-Claire mentioned, this is when the standard method doesn’t quite work or has real limitation and you could go further than just applying the basic approach. So the first step is…

30:44
Really about learning to recognize where the potential methodological problems are and where the real opportunities are. And after that, the next step is to read the literature on this subject. Because when you are designing a trial, you have to take time to read what has already been published on the topic. And just doing that well is already a good contribution itself. So you start by proposing published methods that answer your question.

31:14
that’s already valuable, it leads to your credibility. And from there, with experience, you start proposing your own ideas and naturally you start developing more innovative things. But in practice, there is a few steps before publication. We said a conference abstract or a poster is already a good first step because it will give you that first push toward publication.

31:39
Co-authoring with academic groups or co-supervising master or PhD student is maybe the most realistic path when you are coming from industry. Working group as the PSI 6, for instance, are also a good way. So honestly, you don’t have to start with the first author in biometrics. There are plenty of steps before that.

32:02
And the last thing I would like to add is, and what I understood from a discussion I had with colleagues at conferences is not just about personal motivation. If you are fully loaded with operational work, you will have very little time for innovation and even less for writing and publishing. So even if you are part of a working group, if you don’t have time, it’s really hard to make room for it on top of everything else.

32:30
So the company has really to support it, not just allow it, but make it possible. Otherwise people end up having to use their own personal time to do it and it’s not sustainable. And I know people from who are literally in methodology teams, but they just don’t have a publication objective. And that’s very common in industry. Publishing on your own personal time is not realistic.

32:58
So it really has to be an objective by the company itself. And we have this chance at SORRIGA that it’s exactly the case because publishing is part of our personal objectives. I love what you’re saying for a couple of reasons. If you don’t have time to do it in your work, it’s over time, it just burns you out. Yeah. That’s, as you say, it’s not feasible. think it’s actually dangerous to do that over a longer period of time.

33:27
It also needs to be so important and so relevant that your company really supports it. If the method is important to actually be applied in a trial or any other kind of business situation, then it must be published. Because if it’s not published, you can’t cite it, reference it, especially when it comes for also for regulatory work, you need to have it published. And I know the first paper is always the hardest, but it’s

33:56
gets better over time. If you have submitted a couple of papers and then it gets easier. Yeah, it’s an exercise. You learn to do it. Marie-Claire, anything that you would like to add to that point? I completely agree because I was part of it. After my postdoc, I joined the statistical methodology group at Sanofi. And so I had in this group the opportunity to work on different methodological problems. And it was a great experience where I learned a lot and I did.

34:26
very interesting work on various complex methods, but clearly publishing was not an objective. And so I completely agree with Anaïs. We really need support from the company to publish it. And at some point I tried to say, okay, I will write a paper when I will have time and so on. And at the end, I think I never managed to have a paper finished. I have some…

34:54
parts of papers that I started, but it takes too much time. And when you have a personal life and I became a mom during this period at Sanofi, I didn’t have time. I wanted to spend some time with my kids and it was too much. So yes, I agree. need support from the company if we want to do that. Yeah. I think the environment is really important. And obviously they set the overall company objective in such a way that there is time.

35:23
And that also it’s within brand, you know, because whenever you publish something that helps with your brand, you become more visible, you become more trustworthy. So it makes it easier to attract new clients and especially attract the projects that you want to have. Exactly. And that’s why it’s working perfectly because we are encouraged to publish for this reason. It’s not useless because it’s also a reward for the company. So yes, I agree.

35:53
Yeah. And of course it’s great for the co-authors from industry companies that you work with. They also get a publication and who doesn’t like an additional citation on their CV. It’s really nice. Yeah. So I think that makes a couple of important suggestions for the listener. Yeah. Make sure that what you work on is really relevant, not just for you, but for your company. Create or choose an area where publication is also important.

36:23
Yeah. Make sure that also is in your objectives and that you have really time for this and don’t underestimate it. It always takes longer. And the first submission to a journal is not the end of the journey. There’s lots of more to come. And if you have a couple of people you can do this together with, it’s really important. But in the end, have in mind that it goes through various steps like

36:52
Anaïs mentioned that the first step is usually something internal, then it’s something presented at a conference, poster or presentation, and then it is later a paper. And then you can still come on my podcast and speak about it. there’s even more things you can do. But the podcast episode is super easy. It’s just one hour of work. Thanks so much, Anaïs, Marie-Carre. For that, we talked about lots of topics. want to highlight Project Optimus.

37:20
Again, and the link to that in the show note. Also want to highlight the PSIF by working group that we talked about. So have a look into this. We’ll also put a link to the slides to the method that we discussed. And of course we have put links to Sariga and these wonderful guests that I have today on the podcast. Make sure to connect. If you’re working in this area and you need an additional

37:48
zero, especially ones that does more the heavy lifting, the difficult stuff. There are some people that love doing this. Thank you very much. It was a nice moment with you. Thank you so much, Alexandre. Thanks a lot.

38:05
This show was created in association with Kiosign. Thanks to Rain for help with the show in the background and thank you for listening. Reach your potential, leap right, silence and serve patients. Just be an effective statistician.