Extrapolation to paediatrics

Interview with Andrew Thomson

What is extrapolation?
How can we use extrapolation in paediatrics?
What are the main challenges?

Paediatric research always comes with challenges, and understanding paediatric submission is vital. There’s always a lack of treatment in this area. In this episode, you’ll understand what you can do to get evidence through extrapolation for the children population.

Stay tuned while Andrew and I talk about the following:

• What is extrapolation?
• What are source and target populations?
• The guidance states: “Where possible, quantitative methods should be used for the collation of available data and the investigation of potential modifiers of the treatment effect”. How could such quantitative methods look like?
• What are the major challenges?

References:

• Innovative approaches in the development of paediatric medicines
• Adopted reflection paper use extrapolation development medicines paediatrics

Andrew Thomson

He is a statistician at the EMA Taskforce dedicated to Data, Analytics and Methodology, joining the Agency in 2014. He supports the methodological aspects of the assessments of Marketing Authorization Applications, as well as Scientific Advice, and methodological aspects of Paediatric Investigational Plans. Additionally, he is the EC lead for ICH E11A where he leads the statistical work stream.

Before the EMA, he worked at the UK regulator, the Medicines and Healthcare product Regulatory Agency. Here he worked initially as a statistical assessor in the Licensing Division, assessing Marketing Application Authorizations and providing Scientific Advice to companies. After rising to Senior Statistical Assessor, he moved to the Vigilance and Risk Management of Medicines Division, to be Head of Epidemiology. Here he managed a team of statisticians, epidemiologists and data analysts providing support to the assessment of post-licensing observational studies and meta-analyses. He also managed the team’s design, conduct and analysis of epidemiology studies, using the UK Clinical Practice Research Data link.

Listen to this episode and share this with your friends and colleagues!

Transcript:

Alexander: You’re listening to The Effective Statistician podcasts, a weekly podcast with Alexander Schacht and Benjamin Pisken designed to help you reach your potential, lead great sciences and serve patients without becoming overwhelmed by work. Today we are talking about pediatrics and how we can use extrapolation to help have data or evidence for a pediatric submission. And here we have Andrew Thomson from the regulatory side here. So stay tuned for this really, really good discussion with him and now some music.

Understanding about pediatric submission is really, really important. Maybe you have kids yourself, I do have kids and this always lacks treatment in that area. So I think as an industry, we need to become better in serving this population. Of course, there’s certain diseases that, well there’s really no kind of pediatric indication but very often there is. And so this episode will help you understand what you can do to get evidence through extrapolation for this population. 

I’m producing this podcast in association with PSI, a community dedicated to leading and promoting the use of statistics within the healthcare industry for the benefit of patients. Join PSI today, to further develop your statistical capabilities with access to the video on demand content library, free registration to all PSI webinars and much much more. The reduced rate is only 20 pounds for non high-income countries and 95 pounds for high-income countries. And of course, there’s the PSI conference in Gothenburg coming up. So, head over to psiweb.org to learn about the conference and all the other PSI activities and hopefully see you in Gothenburg. 

Welcome to another episode of The Effective Statistician. Today, I’m really, really happy to have Andrew Thompson here, who is really an expert in the topic that we are talking about? And not only that, he is also coming from the regulatory side of things, which I think is maybe the first time I have someone from the EMO, the FDA, on my podcast. So, I’m really honored about this. Hi Andrew, how are you doing? 

Andrew: Hi Alexander, I’m very well, thanks. 

Alexander: Very good. Before we dive into the topic, maybe you can speak a little bit about your career and why you actually went into both statistics in the first place? 

Andrew: Yeah sure. So I started like many statisticians doing math at University and I found I wasn’t particularly good at either pure or applied math, but somehow I had a knack of being okay at statistics. And then, my final year, there was an advert screen in one of my stats lectures for an MSC place in statistics of applications in medicine, which would be funded. And I thought, that sounds like an interesting thing to do for a year and the funding certainly helped and never really looked back since. And I started off wanting to be an academic and I worked at Imperial College and then did a PhD at the London School of Hygiene. And then towards the end of my PhD, I wasn’t so sure that I wanted to be academic after all, when I looked at the alternative options out there for me, and at that time, it just so happened that the MHRA were recruiting for statistical assessments and I apply and I was very fortunate to offered a position and so I started work doing statistical assessment looking at dossiers across all therapeutic areas and it was an incredibly interesting job with a huge learning curve that never really stopped spiking up for too early. There was always something new happening and it was a very rewarding and interesting role and after a few years of doing that, then moved onward and upwards to be head of epidemiology at the MHRA. Looking more at the post licensing side which is very interesting, especially that we’re doing kind of starting rolling out of vaccine safety studies, the experts in my team were looking at that and these have become much more high-profile over the last two years as we say.

Alexander:  Yes, yes. I think the post marketing association work also from the regulatory side has much more kind of visibility nowadays than it has in the past, for sure yeah.

Andrew: Yeah, definitely. And I think that leads on next to the EMA, which is here may be very much integral in making sure that visibility is there. And so I’ve been at the EMA for about 7 years now. A lot of work I do is internal consultancy from a statistical point of view. And also, you have to find ways to go forward when there are perhaps differences of opinion between different member states. We still have to have one unified position, so finding the optimal scientific role route for that is one of my key roles. I sit in the task force, dedicated to data analytics and methodology, which has been going for a couple of years now, and it’s a real hub for quantitative methods of the FDA. I started getting involved in extrapolation actually about 6 years ago, I’d say when the reflection people were being drafted and it’s very much a kind of cross collaboration between clinicians and statisticians, and yeah from statisticians and modelers in particular, so to bring those three disciplines together and to get everyone at the same table and singing from the same hymn sheet, help stir that through and yeah. And so I’m now on ICH E11 as well where I, which is pediatric extrapolation. I’ll talk a bit more about that later. That’s out very soon and I leave the stats something on that. 

Alexander: Yep, very very good. I think this is a really, really important topic because I myself have three little kids and I know that for lots of lots of different compounds. You know, says very, very limited data for children and so, but still, you know, of course, children need to be treated. And I think there’s a lot of Physicians that you need to treat or pediatricians that need to treat patients irrespective of whether there’s an approved product or whatever. And really rather have them some really good solid data and quantitative assessment, then, you know, just maybe some anecdotal things and just personal opinion. 

Andrew: Yeah. I think that’s very on point, yeah, at a holistic level extrapolation is about bringing drugs on label, so that all prescribers have information on safe and effective doses in the pediatric population. So it’s not just those who have used off-label and those who are happy doing so but so every treating physician should have the accounts in treating. 

Alexander: Yeah, it’s really about making data available for good decision making and in the end that’s also one of my personal goals to improve that. So when we speak about extrapolation, what actually is that, because when I’m just thinking back of my stats parts and I think, okay, you have a maybe haven’t our aggression and kind of, most of the observations that we don’t know, 50 and 150 and then you try to extrapolate, what happens beyond 200 or something like this? 

Andrew: Yeah. 

Alexander: Is that the same thing? 

Andrew: Well not exactly no, I think, you know, maybe the principles are the same in that we want to try and bring the label down from adults to include younger age groups in it. But when we define it quite clearly as being, it’s based on information and one or more source populations such as adults, or children. That’s relevant to a target population. So for example, another pediatric population in a way that can be quantified and used as a basis for further development. And so effectively, that’s making use of information in one population to guide the design and analysis and interpretation, in another. And we’ve actually phrased a lot of our reflection paper, it’s age neuter. We don’t talk unnecessarily about pediatric extrapolation because it applies across development where you could have such populations. It just so happens that pediatrics is perhaps an obvious main use case because you have  this source population of adults, you have a target population of children and you know that you want to be able to do as rational a development approach as possible. So it’s the principles that apply across development but the key use case now is pediatric. 

Alexander: Okay, so it could also apply for example, all the population, it could apply for the kind of potentially patients that you know were previously excluded from studies for whatever reasons, these kinds of things. 

Andrew: Yeah exactly. And you know there are indications where we have used extrapolation for that because as you say patients have been excluded male breast cancers, an obvious example. So yeah, it’s something that can work. I think perhaps the difference is that pedias is often after adults and I think that’s often as it should be in terms of development. We all want safe and effective medicines for children absolutely but if we have no real evidence that it’s safe and effective, in adults that I do think people start getting nervous about whether we should be starting straight away in the pediatric population. So, it is kind of inevitable. You have this order to develop in many situations. 

Alexander: Yeah, absolutely. I think it’s the same what we do, what we have seen with covid and Pandemics. At first, it was adults and you went into more and more, younger patient populations. So, in the document, and I’m referencing systems in the show notes, it says a lot about talking about the source and target population. So can you speak a little bit about what exactly it is? 

Andrew: Yeah, so the source population is, is what you are extrapolating from and the target is who you are extrapolating to and we would normally think so therefore, but if we’re talking about pediatric extrapolation, that your target population is some sort of pediatric population. 

Alexander: That it means that we have no data whatsoever for the target population? 

Andrew: No, it doesn’t. It doesn’t actually. And I think the purpose of this is to work out what data we need based on what we have already. So your sources might be adults so you could extrapolate from adults to children but children are not homogeneous not just small adults. They have different organ maturation of various stages. And so the extrapolation from those different age groups can be different. So you could, for example, extrapolate from adults, to adolescents. And then you could extrapolate from adolescence to 6 to 12 year olds and will just populate from six and 12 down to 4 and 6, and 2 and 4. In each of those situations, you would have some information in a relevant pediatric population of some sort, which you can then use to inform what information you would need in those different age ranges that you’re looking into. 

Alexander: Okay. So is it all kind of safety efficacy and pharmacology data or do you also use other data sources? 

Andrew: Well, in a reflection paper talk about the extrapolation concept and the plan. So we’ve stripped split extrapolation into two main steps on a more technical level, we have the extrapolation concept and the plan. And the concept is kind of held up by three pillars of kinetics, and disease, and clinical pharmacology. So the drug, the disease, and the drug disease. of the And we have a table in our reflection paper. And I think it’s not a box ticking exercise. It’s a box filling exercise. 

Alexander: Okay. 

Andrew: And we need to identify where the gaps are in the knowledge. So for example, we might know a lot about clinical pharmacology and adults which we probably should do by the time that we embark on pediatric extrapolation. We should know about the natural course of disease in adults as well that may be different in children, but we should find the evidence to do that. So, for example, real-world evidence data sources, may be to tell you something beyond what clinical trials show to show that actually the course of disease or the time taken to subsequent therapy or whatever is the information you’re looking for may be available in real-world data sources as well and of course with the preclinical models as well for the adme data. So there’s all sorts of possible data that could be brought to the table to fill in these gaps and the knowledge and to me, that’s why it’s quite an interesting area to be involved in because you can work with many different data sources and types and experts and field 

Alexander: That’s cool. Yeah, I completely see that it’s kind of pulling data from all the different sorts, clinical, epidemiologically, real-world evidence data, pharmacokinetic and pk data and combining them all into one kind of model, one kind of concept or framework. It’s quite interesting. 

Andrew: Yeah, it’s quite hard as well. I don’t think that should be underestimated. I think what we’re trying to do is to say how much can we trust the data in the adult population as being relevant for the pediatric population We’re interested in? So the adult data is what it is now and the studies have been done and we have that information but then the work on the clin pharm and the disease similarity and the adme data all adds together to say and because of all of this additional knowledge, this is how much we can trust the adult data. I think that it’s a knowledge synthesis process and also a large degree of clinical judgment as well. 

Alexander: Okay. Okay, so, in the guidance of states where possible quantitative methods should be used for the collation of available data. The investigation of potential modifiers also has a treatment effect. And I think in terms of modifiers the treatment effect is really about the differences between the target and the solve population but we are mostly interested in not other modifiers. 

Andrew: Yeah, I think that is the correct interpretation. I think in terms of quantitative methods what was the first thing that springs to mind is meta-analysis, it’s a standard part the tools but we also have you know, that metric colleagues, they’re very good at dose exposure modeling, and exposure response modeling, and all of those will be potentially different in adults and children. And think what we would like to do is to link that together with better predictions for efficacy because we can accept that there may well be different efficacy in children compared to adults and that’s okay because as long as we know what that is and we’ve quantified it and we’ve got data to robustly confirm that and we don’t necessarily need to demonstrate have a standalone demonstration of the efficacy as it is perhaps maybe we need it to confirm that do is modeling is sufficiently robust.

I think it’s important here to recognize that we don’t have equipoise per se because we have the adult data where something has been demonstrated, but we’re not operating in a vacuum and the requirements for what’s required for Pediatric development will have to take that information into account and so if we’re not demonstrating efficacy, but instead of confirming predicted efficacy as is we expected, then that will require a different amount of evidence to demonstrate that. And I think from a practical point of view that or theoretically a bit of both really always needs to be less because otherwise, we’re not doing extrapolation. If we do this whole exercise and do lots of modeling then conclude that we need more data than we would have done doing standalone efficacy then then something’s gone wrong.

Another thing is that I think it’s important to note that we don’t always need data to verify predictions either, the purpose isn’t just to kind confirm the model for the sake of it. Now there’s uncertainty there, we don’t know we need the data to be convinced but there are situations certain anti-infective what you’re trying to do is you’re trying to get a sufficient amount into the bloodstream to kill the bug. It’s not about generating a receptor driven response for example. So as long as we know what pk is and we have a safe space that we can use in Pediatrics then we don’t need any additional efficacy data depending on the situation. 

Alexander: So, that’s exactly where kind of the disease comes into play. The better we understand disease, the better we can be sure that certain kinds of relationships hold true. 

Andrew: Yeah, very much. So when I think, you know, there’s been huge advances in many areas where we’ve got a really good understanding of the disease. And the day on the understanding is improved and that’s driven drug development, kind of immunological challenges in children for example and ulcerative colitis and pediatric trains and things like that you know, we have a wealth of products compared to what we had 20 years ago and huge more understanding of the immune responses of the disease and the drugs used to target.

Alexander: If I’m thinking about data from the source population, do I need to kind of treat different parts of the data with different kinds of weight so to say, or how much it helps me, just if I’m thinking about let’s say I want to extract to kind of teenagers and I have study, you know, above 18 years old, is the data kind of said I collect between 18 and 25 is that potentially more relevance? And let’s say that they don’t also be over 50? 

Andrew: Yes, it is. And I think that the challenge that we have is that we companies can’t always precisely control the age distribution of people who go into their own development programs. But I think we all when looking at say extrapolating from adults to adolescents would think it is more relevant. I think again we’re statisticians would need to get involved with the appropriately designed subgroup analyses and interpreting them because ultimately if you don’t really see a difference in the age subgroups that perhaps lens not wait that extrapolation is more possible or as better option but that you could use the entire adult data set and be confident using the adult data set and not just the 18-24 year olds but think we do start at the default position that some data more relevant than others here. 

Alexander: Yeah. I see that and so that’s actually quite interesting also, if you kind of plan for going into more younger populations having potentially more patients on the lower end of the spectrum, could help you to gather more data in that regard. 

Andrew: Yeah, definitely. And I think one of the challenges that exists I think is encouraging companies to design adult development, with an eye on the inevitable pediatric development that will come afterwards, we make companies agree pediatric development plans, after adult clinical pharmacology studies often paraphrase to at the end of phase one but sometimes a little more nuanced than that. But we would like to be able to make sure that the trials would be relevant so we can utilize as much and yes, that’s about age distribution. Sometimes it’s about things like collecting relevant, important biomarkers or something similar which companies might be thinking about doing anyway but if we have a conversation, we’re saying this could be very crucial to facilitating an expedited pediatric extrapolation plan. Hopefully, we can encourage people to think about the design of adult studies with more than just one eye on Pediatrics. 

Alexander: Yeah, that’s really good. I think I see it also from a HDA  perspective, there’s much more that we can do with a good phase free package, than just getting regulatory approval for the adults which of course you know it’s also quite a hurdle to jump over anyway but yeah that’s really good. What first challenges do you see? You know what kind of your biggest concerns that you have? 

Andrew: Well I think they kind of fall into many different areas and there’s the kind of the logistical challenges of extrapolation, in the timings of this, as I said, you know, you’re supposed to agree your pediatric development plan at the end of, including the sample size, the end of phase 1, and adults. And  the timing of the interactions with the regulatory authorities is difficult because of what we’ve said in the reflection paper. We’d like to be as iterative as possible because we recognize and accept that new data comes out through adult development which will influence pediatric development. But also what we would like to do is to minimize the amount of modifications companies have to make in their pediatric plans, as well. We’d like to be able to say yes, let’s everything go smoothly and if it all happens as you hope it will, then carry on and if not come back and talk to us later but finding words to do that to fit with our legislation just involves a lot of upfront planning and it is hard that we would like to agree as much as possible as early as possible. I also think that pediatric development is global and it almost inevitably is and that means that companies will have to agree across agencies globally. You have to get agreement between different EU committees such as the Pediatric committee and CHMPR licensure but also between EU and FDA and indeed beyond the ICH E11 that hopefully, we’ll go some way to ensuring that’s going to be more straightforward in the future for companies.

Alexander: So if you just kind of pre-specification thing or kind of pre agreeing to things before we collected data I think one of the paradigms that we have in drug development is to kind of avoid fishing for significance to these things. I think that is particularly, you know, difficult for these models because you pull data from different sides and you have the different studies coming out at different time points. Is that kind of the background for the pre specification and kind of iterative approach?

Andrew: Well, I think if we agree, a pediatric sample size at the end of phase 1 and adults. And then we have Phase 2 and phase 3 data which don’t necessarily line up exactly as the company thought they would, in either direction, both impressively or less impressively then, I think that information will need to be taken into account into the Pediatric development. And obviously, if you have a drug, that turns out to be an absolute wonder drug that has a huge treatment effect and there’s good evidence that the mechanism of action is going to be similar at least, in some pediatric subgroups. Then we will be more comfortable saying it’s reasonable to conclude that the benefit risk is positive in children. Whereas, if you’ve got phase 3 package, where you didn’t get as big treatment effects you hoped for and you just squeaked over the line with P equals naught point 1, naught point 9 or whatever it is. Then if it’s inevitable that the situation of how comfortable we would be extrapolating. Thus, how much data or what type of data is going to change? I also think that the other main challenges are the design and the analysis of these studies. 

Alexander: Yeah. 

Andrew: And the design is quite high level, whether you would require a randomized control trial, but a different level of evidence, or if you require a single arm trial, or if you would require something that could rely on some kinetic data because you’ve got sufficient. 

Alexander: Yeah, I think that’s a very good point. In the end, it’s about how comfortable you are with the decision making and the overall robustness of the data is important. 

Andrew: And you know that we do talk about extrapolation being a spectrum and I think we’ve moved away from decision trees or a way of saying, you know, if this then that. But nevertheless, there are a finite number of study designs available to developers and so working out given all of this information all the uncertainties which are the trial designs that we want would be part of the one of the main challenges. I think another one which needs to be mentioned, is about Bayesian and kind of frequentist designs. And I think it’s a very potentially, a very rich area for the use of Bayesian statistics because if you have some sort of prior in the adult data because you do have prior information and when people hear the words prior information it’s only natural I think that people think are that means I should be doing Bayesian statistics. 

Alexander: Yeah. 

Andrew: And in the FDA have been driven by white 4246, where their legislation says, they have to investigate Bayesian methods. And it’s an obvious place to start really because of this prior information. I think we’re still not entirely sure of the utility of all Bayesian methods and that they’re not all created equally.

Alexander: As easy as the one thing where you get the information is of course from the adult populations. The other thing where you get information especially about the kind of standard of care or placebo treatment is, of course, we’re both evidence data. That is another piece of the puzzle that you can fit in and I just recently recorded a podcast about how you can use real-world evidence data, to kind of, help with your comparison data for clinical trials. Isn’t that going in a similar direction? 

Andrew: Yeah, it is and I think we tend to use the phrase augmented. I think this is probably the buzzword at the moment that we would probably use, that augmenting the control arm potentially using other randomized controlled trials, control arms is one way, but sometimes, they don’t exist or you don’t have anything that you think there’s actually relevant. And so, you might be able to use that. I mean, if you want to use a successful randomized control trial arm of the map active and show superiority against that I think we might be quite happy with that. There will be situations where real-world evidence is suitable for that. It will depend on the design and the disease and the indication  and all of the things that go into these thinking and adults apply in this situation as well, but it’s clear that the door is not closed here and there are options available but there are differences of opinion between some regulatory statisticians about whether would be better to borrow the treatment effect from adults or whether you would prefer to just augment a control arm. I don’t think there is a unified position as to which of those two approaches would be preferable and there are strong views on both sides of the argument on that. 

Alexander: Yeah, I can foresee that, it is not a closed discussion that could kind of go on for quite some time. Maybe we can kind of have one time come up with a kind of unified approach there. It takes both sides into account. 

Andrew: Well, it’s a possibility, I think that’s the good thing about well-thought-out Bayesian designs, for example, if you’re not making multiple changes and you’re not having multiple interim analyses, and actually, it’s quite the clean Bayesian design so to speak, but you just wish to interpret the data in light of the prior, then you can run a pediatric program successfully and analyze it in a frequentist fashion and analyze it in a Bayesian fashion and come to conclusions based on that. And indeed, if you read the assessment report from the FDA for Bolivia app for the purpose that they chose a Bayesian method as one way of characterizing the efficacy and then the EU, we chose more for quantifying to interpret the data. But the key point was it, because the trial had been sufficiently well designed in the first place both interpretations are able to happen and it’s got little bit on the post hoc side and it’s not necessarily the post child for example, one might use, but it is a good example of different regulatory authorities applying different frameworks for analysis and still coming to similar conclusions. 

Alexander: Yeah, I think that’s actually quite nice if you have different analysis approaches and vastly different conclusions, that always has been concerned anyway, so that’s a good point. Any other kind of big challenge that you kind of beyond the concerns that we talked about in terms of analysis and logistics. 

Andrew: Yeah, I think again, the design measuring what you need in a pediatric trial and other trials is something that needs to be rethought out. Another really big one is communication and that’s the communication between companies and regulators and also between other stakeholders as well. And we need to be clearer regarding the acceptability of certain designs in particular, with Bayesian wants the operating characteristics. And the type error and the type both conditional unconditionally. What the important metrics are for success for regulators because if we don’t tell you in advance what our metric is, the successes and it’s quite difficult for a company to go away and in particular to the power and the size of their study. 

Alexander: Yeah. 

Andrew: Yeah. So we do want to know properly what these operating characteristics are and their companies say it will be this and we can accept that. I think some of the more technical Bayesian designs such as dynamic borrowing where we don’t know in advance how much borrowing is going to take place because it depends how well the data matches the prior. Things like that certainly, we understand the models which are not entirely sure that they routinely robustly provide all the outputs we want if all the borrowings left models. So we need to perhaps better communicate between us, between industry and regulators as to what our standards and expectations are. And then once these have been successful to kind of tell the outside world that because some people will inevitably look at a trial and say all that had a p-value of naught point 1. So therefore, it’s a failed study and that’s not how we would interpret it in this situation and I’m trying to explain to the outside world but from our point of view actually given the experimental circumstances and everything else, we know actually that’s a success. 

Alexander: Yeah, I think that says it’s a good point that it becomes much more challenging to kind of communicate well all the evidence together and I think that is also potentially we need to communicate these models also in a different way rather than you know just taking a number like the P value you just mentioned more describe kind of the distribution of things. More describe kind of the relationship of the for example, in terms of the treatment effect modifiers. Potentially that there’s no effects there and to show kind of all the data is really robust. Would you think there would be potentially some more visual tools to show these kinds of things rather than tables where I’m just thinking about a kind of predictive analytics and Bayesian analysis, lots of this is really really much more kind of about distributions rather than estimates.

Andrew: Yeah, very much so. And I think one of the challenges we face with some Bayesian models, for example, is that the unconditional type 1 error can be computationally calculated, and that we can accept that and then, that can be plotted against various underlying parameters, and it’s not constant. And actually understanding where the type 1 error is raised as well as how much it’s raised by using these Bayesian methods is inherently a visual display. And I think there was a paper from the DIA base group quite a few years ago now and I could feel it was on the lupos case and that that one’s really good for giving examples of how you might display data like that. Other visualizations which I’m particularly fond of are thinking about what success looks like. And as an example, we will say a company will make a proposal for some analysis plan and say, okay? Well if we were to cease, let’s say 40 subjects per arm, randomize 1 to 1 and if we were to see 30 subjects successful on control, how many would we need to see successful, maybe 35 the 40 or whatever the numbers are. But then looking at different cut points, if you want to see what it was, 10 subjects were successful in control. How many? Really understand when would a company consider this trial a success and then we can say well actually that then translates to a certain treatment effect and we would say, well yes that looks like an important treatment effect we would be interested in and if you can go through that process then you can you can agree a program and sample size and success criteria quite quickly. And I do think that better visualizations for that certainly help focus the mind in terms of the relationship between the size of the treatment effect and the sample size you’ve got to hand. 

Alexander: Yeah. That’s a very very good point. You mentioned this lupos case study, I guess that is publicly available, isn’t it? 

Andrew: There’s always a wealth of information on regulatory websites. You can look at the EU, the European public assessment report and their FDA information is available on the website as well. And indeed, I think the regulatory websites are a very good resource for that kind of thing because sometimes it’s difficult to communicate with the outside world. It’s not necessarily something that people would write up in a paper for the outside world, actually, some of the most interesting extrapolation case studies I’ve seen are ones that have been presented at conferences. And one of the things that’s happened for the pandemic as I’ll be able to at least online attend a lot more of these meetings and have been previously when they might have been face-to-face. But there’s something that’s so great about the face-to-face approach that I’m hoping is going to be returning very soon. Well, but sadly, I’ve been exposed to a much wider array of things like case studies. You can see the perhaps the for that people webinars and so on and so forth. 

Alexander: Okay, let’s put a couple of these into the online appendix show notes and then as a listener you can have a look there, just head over to theeffectivestatistician.com and find these nice episodes with Andrew. Thanks so much Andres, that was an awesome discussion. We started really with kind of what actually is extrapolation and, you know, if we really want to extrapolate from source to the target population, whereas a lot of discussion about what type of evidence we need and where this can all come from, and that it comes from disease understanding, it comes from efficacy and safety data, it comes from PK data, it can also come from real world evidence data, and other data sources that help us to have a good understanding of the overall data to make a good recommendation for physicians, for patients. And we also talked about, you know, challenges like communication, logistics, having good kind of iterative planning on terms, all the data that comes in and whether that kind of changes our plans or how we can you know make sure that we have all our plans well thought through and if there’s some change in the data that we didn’t accept, expect that we can make amendments, make things faster or slower depending on the data. Are there any final thoughts that you would like to leave the listener with? 

Andrew: Yeah, I just want to mention ICH E11 which should be out very soon. Hopefully by the time some people download it may already be out so that’ll be out for consultation so please do comment that. And it helped us improve the document and also, finally, thank you for inviting me. It’s been very nice to talk to you. 

Alexander: Thanks so much. Did you enjoy this episode as much as I really enjoyed the discussion with Andrew, then let me know about it. Connect with me on LinkedIn and tell me what you think. This show was created in association with PSI, thanks to Reine who helps the show in the background and thank you for  listening. Reach your potential, lead great science and serve patients. Just be an effective statistician.