What will be the role of health economics in the future EU HTA?

Transcript

What will be the role of health economics in the future EU HTA?

[00:02:17] Alexander: Welcome to another episode of The Effective Statistician. Today, I’m talking with Katrin and Min-Hua. Welcome to the show.

[00:02:25] Katrin: Hello.

[00:02:27] Jen Min-Hua: Hi.

[00:02:27] Katrin: Great to be here.

[00:02:29] Alexander: Thanks so much. So let’s start with a short introduction. Min-Hua, do you wanna go first?

[00:02:34] Jen Min-Hua: Hi, I’m Min-Hua Jen, Director of Statistics from Eli Lilly, and also a group lead for our HTA team to support oncology portfolios across the globe.

[00:02:45] I have a PhD by joining epidemiology and statistics and graduate PhD from University of Bristol and have extensive experience applying statistics to clinical research, vitaminology and health economics and outcome research in academia as well as in the industry setting. And as I mentioned, I currently work for Eli Lilly as the group lead of our Oncology Statistics International Business Unit.

[00:03:15] So I supervise my team, the patient access medical affair worldwide for our oncology portfolio, and also advance Eli Lilly’s internal capability with technical and innovative contributions and try to build the best practice for Eli Lilly. I’m also a member of various special interest group, the PSI, ISPOR, and Fuse, and was a member for the Precision Medicine Oncology SIG at ISPOR before becoming the co-chair of the Oncology SIG and also the executive committee for the R Validation Hub project.

[00:03:54] Alexander: Cool. Very good. Quite a lot of work, and by the way, it’s awesome to work together with Min-Hua. We work together on the same team, so that’s great to have a former colleague here on the show. Let’s turn over to Katrin.

[00:04:07] Katrin: Yeah. Hi, I’m Katrin. I’m working for BMS. I am the Director of Statistics, and I’m leading the group of statisticians who are [00:04:15] responsible for market access and HEOR specific tasks.

[00:04:19] So we are supporting all the countries in the HTA submissions with the clinical analysis they need, and also all HEOR-related publications when they’re based on patient reported outcomes. Before that, I was leading the statistics group in the German affiliate, so I have a lot of knowledge on the GB-A and the German AMNOG system, which really helps me a lot because those are normally the most time challenging and request we get from the team from our stakeholders because they need so many analyses.

[00:04:50] I have a PhD in Data Science, and I’ve worked within statistics since 2007. First in a CRO, then in clinical development, then for Boehringer Ingelheim in the Medical Affairs Team, and now I’m with BMS since 2014.

[00:05:04] Yeah, and I’m happy to be here to talk about the European HTA and the implications especially when it comes to economic modeling. That’s really, really important and an interesting topic today.

[00:05:14] Alexander: Yeah. Awesome. And [00:05:15] this is the podcast episode number four in the series that the HTA Special Interest Group is working on.

[00:05:24] So, and if you haven’t listened to the other three episodes, then just scroll back in your podcast player and listen to them. We’ll all mark them so that it’s very easy to see. And just a short blurb. If you don’t know what the HTA SIG is doing, then head over to the PSI homepage where you’ll find an overview of all the special interest groups and also SIGs and there’s also the overview for the HTA SIG.

[00:05:50] And if you wanna join, there’s the information on how to join. Very good. So let’s talk about economic models and European HTA. So in terms of economic models, first maybe what actually is that?

[00:06:09] Jen Min-Hua: I think the purpose of the health economic modeling is to evaluate and support [00:06:15] decision makers in allocating their limited healthcare budget resources to achieve the best possible outcome for patients.

[00:06:25] And so deciding to invest one in one health intervention means you have to not invest in another. So decision makers will be willing to invest in a health intervention only if its value sees that of the alternative one. So in this case, something called cost effectiveness model

[00:06:46] So in this case, something called cost effectiveness model will draw a lot of information from various resources to help analyze which intervention represents the best value for the decision making. And then there’s various types of economic model evaluation. And the type of the evaluation differs in how they account for the cause and benefit of intervention. And what type of the model is the most appropriate depends on the capacity of capturing the impact of the intervention.

[00:07:19] And also depends on, for example, the variability of the data and the requirement in decision making. And mostly, you will see them being informed by large amounts of data processing into the model. So that will include maybe the cost, efficacy, safety, and quality of life.

[00:07:39] And so for that, I think the most we know now is from nine submissions that request for those economic evaluations. And you could see those models were mostly developed in the Microsoft Excel, but you can also see them in some other platforms such as R. And in terms of how they evaluate is really to look at the proportion, at least in oncology, to look at the proportion of patients going through different health states.

[00:08:10] And then you model them by assigning the utility and the course through time of the proportion of patients that go through those health states. And EMA comes up with something called summarizing them in a statistical ISA, which is then for the incremental course effectiveness ratio, and in summarizing all those costs and then the benefit from the treatment.

[00:08:34] And then inform the payer to see how much it might cost them to have a patient to have a quality of life, good quality of life for a year. And so this is kind of in a nutshell of what is the economic model.

[00:08:49] Alexander: So that was quite a lot. Let’s depict it a little bit. The first thing is, as a country, you have only a certain pool of money you can put into health. So it might go a little bit up and down.

[00:09:01] But overall, it’s more or less fixed. So as a government decision maker, you want to get the best hold for this money that you invest. And now whenever there’s a new intervention coming up, you wanna see [00:09:15] does it make sense to invest in this and will that overall give me basically more [00:09:22] health for that money?

[00:09:24] And now you just scan for other treatments out there that usually you compare against these other treatments. Now the way you wanna compare it is, of course, the problem is every disease is different. You have oncology, you have diabetes, you have dermatology, you have all kinds of different things.

[00:09:45] And if you wanna compare across all of these different diseases and therapeutic areas, you need to have some kind of common standard. And one is the symptoms are so different and so what is usually used is some kind of utility. And here, the EQ-5D is really used mostly to assess the health of the patients.

[00:10:02] And here, the EQ-5D is really used mostly to assess the health of the patients. Which costs are all taken into account? Is it just the costs of the treatment?

[00:10:16] Jen Min-Hua: It’s more than the cost of treatment. It could be the cost of, let’s say, if a patient has an adverse event, it would be hospitalization. And those will be part of the cost that will be incurred into the model as well.

[00:10:31] And it could be other things. It depends on what type of disease and hence the cost can be related to what might incur when the patient is not just having the treatment but things happen altogether as well.

[00:10:47] Alexander: So all these other costs mingled into the overall model and depending on the disease. Yeah, sometimes, you know, the treatment cost is really all what matters.

[00:10:57] And some other instances, maybe hospitalizations, surgery, other kinds of things can be much higher cost drivers. The probability of these, all these kinda built into models. So you mentioned Excel is used. Now that is, when I [00:11:15] first saw that, I thought, really, you used Excel for this kind of thing?

[00:11:20] Why is that the case?

[00:11:22] Jen Min-Hua: I think it seems to be commonly provided when people want to build an economic model. And when I heard about the reason why we use Excel is that maybe for payers, it might be an easier interface for them to look through what’s being implemented in the model. Then if you are using, let’s say, another platform could be basically you have the and also there’s the computer power issue as well.

[00:11:54] Alexander: Katrin, when I think about such a model in Excel, what does that look like from a kind of look and feel? Can you give a description for the listener, what that looks like?

[00:12:05] Katrin: So if you think of, for example, again for an oncology example, so you have your different health states. So, for example, if you think of an adjuvant trial, you have the patients [00:12:15] who are really free of disease.

[00:12:17] So you have one column in your Excel for the patients who are free of disease and you use the utilities and the data from your trial in this column. Then you have a second column for those who have a recurrence of progress, and then you have those who die, who really have no quality of life at all.

[00:12:33] So the utility, zero for them. And then you have three different columns and you have transition probabilities for the different time points. How, what is the probability that a patient changes its health state? And this is how this is done. So I’m not doing that myself, I have to say. So we are really responsible for giving the analysis that is needed for the modeling.

[00:12:55] But that is how the colleagues told me what it looks like, and I also had the same feeling. So using Excel sounds a bit weird for a statistician, but it’s because players are used to that too.

[00:13:05] Alexander: And also if you have an Excel, you have basically the tool and the data in one place so people can put alternative costs into it and [00:13:15] basically play with it.

[00:13:16] If you do wanna adjust models for different countries, you can put country-specific costs into this and stuff like this.

[00:13:24] Katrin: Yeah, and this is how it’s done also in most of the companies. So you have one global model you start with, and then once you have the CHMP opinion, you adapt it to the different countries and then you can do that easily in Excel to include the different country utilities for the healthy people and the different costs of the different healthcare systems.

[00:13:41] So it’s very convenient for that.

[00:13:43] Alexander: Yeah. Very good. Now you just mentioned we as statisticians work really hard on some of the analysis that goes into these models. And the overall kind of framework for that is based on an acronym called PICO. So what is PICO and how does that fit into this economic model?

[00:14:07] Katrin: So maybe I can take that. So PICO is an abbreviation. So the P is for population. So the population of interests, so different [00:14:15] HTA bodies might want to have a look at different populations. Most of the time it’s dependent on the label, but also depending on the standard of care in that country.

[00:14:25] Then the I is the intervention. So the drug you want to compare versus your comparator. Then you have the comparator. This is the C, and this is normally based on the standard of care. It’s defined by the HTA bodies. It’s based on guidelines. It’s of course based on regulatory approval and also what’s affordable in the country.

[00:14:44] And then you have the outcome. So your variable of interest. And so what you want to see, so for example, you want to know the quality of life of the patients at week 52 in a trial or the overall survival in an oncology trial. So this is the outcome.

[00:14:59] Jen Min-Hua: Yeah. And also the PICO framework has been well developed and been accepted by the Cochranes, and so it’s increased the credibility if you follow the framework well, and so that you could [00:15:15] summarize the kind of the relative efficacy across treatment, via this systematic literature review, which use this framework to look at the literature review search strategy.

[00:15:30] And to ensure that more comprehensive than less of the bias in terms of the search and the focus of the study that should be included. So we are more well defined.

[00:15:44] Alexander: Yeah. This is also to say evidence-based medicine 101. If you want to learn more about this, yeah, for sure. Check out the Cochrane homepage. Cochrane. The Cochrane organization is the muscle of evidence-based medicine and evidence-based decision making in medicine.

[00:16:04] And the Cochrane handbook is quite fundamental for many of these HTA guidelines. There’s surely a lot of overlap between that and, for example, the NICE work, [00:16:16] also in terms of people that, also these different things there’s a lot of overlap. But that’s maybe another topic.

[00:16:22] And Min-Hua mentioned she’s coming from the University of Bristol. That’s for sure, also one of the universities that is very heavily involved. When I hear population intervention comparator outcome, I think that sounds a little bit like what we nowadays call estimand.

[00:16:41] So how does estimand and PICO go together?

[00:16:46] Katrin: Yeah. So the PICO and the estimand are, of course, directly related. So the population definition is given in both concepts, and the outcome of interest describes the variability to be analyzed in the estimand concept as well as the analysis strategy.

[00:17:02] It is different in the estimand framework as that on the estimand framework highlights the importance of the intercurrent events. That’s not given in the PICO, so it ensures that the summary measures are used, which are used to compare the treatments. [00:17:15] Conditions are made really explicit, so the PICO is less explicit. And this is really an issue because the estimand defines the trials do not necessarily match the estimand needed for HTA.

[00:17:27] So for example, when you think of the German system, which I know very well, and especially of oncology, the German GBA always wants to see the treatment policy estimand. They’re asking for that for all of the endpoints, also for quality of life and safety. And the clinical trials are normally set up so that you can analyze a vital treatment estimand only because the data is collected on treatment.

[00:17:49] This is what you want to see. You want to know what happens, why the patient is on the investigational drug. So this is not harmonized at all. And this will be one of the challenges also for the European assessment. How does the estimand and the PICO for that dossier then match, and how can that be solved?

[00:18:06] Alexander: Yeah. So what is your guidance and should the kind of, does a PICO basically needs to be more clearly defined or is [00:18:14] that all the other estimand parts are in it already defined in the systematic literature review so that when you see an outcome, oh, that is on treatment or that is treatment policy, and what kind of intercurrent events were taken into account and all this kind of different things?

[00:18:35] Katrin: So I think that would make the life for the statistician running the analysis and especially when you think of indirect comparisons where you have to rely on published data would make our life much more easy if you would really know, okay, what was analyzed exactly how it’s been into current events treated because this is sometimes not so clear.

[00:18:54] And I think also for statisticians still working on clinical trials, having an understanding of the different concepts of estimands needed for the different parts of the drug development in the clinical trial and then later in HTA would also help to be better prepared for those analysis and to better understand those concepts.

[00:19:10] Yeah. But I really would recommend that this would be more precise also with the HTA scope.

[00:19:15] Jen Min-Hua: I think for HTA, because typically, it evaluates the effectiveness of a policy rather than a treatment, as we all know. So it is really nearly impossible to replicate feasible policies for multiple payer, right, just by using, I think, oncology tend to have just one pivotal clinical trial to support each indication, and also I think, for example, the standard of care would like to be different across country, probably due to their availability and also related to their previous submission as well.

[00:19:53] And also the treatment evolution also contributes to this complexity as well. [00:19:59] So for example, by the time when the study is for data log and look at the result, the timing of having that comparator might be changed compared to when the trial has been designed at the time.

[00:20:15] And there’s also another issue as well. For example, the physician medicine era we are now facing, so you would use the biomarker and other tests to select the treatment that is most likely to help the patient, right? But with this evolving landscape, we are likely to face situations where we might need to compare the treatment of the differences in population as well as the study design across different treatments.

[00:20:43] So yeah, so all those in so much complexity when you are trying to summarize the relative efficacy across treatment and compare that to fit an economic model and suitable for each country to address their peers scoping requirement is just very complex.

[00:21:02] Alexander: Yeah, we are already going into the topics of the network meta-analysis and into our comparisons because this is really what feeds into the economic model.

[00:21:13] If you wanna just say that you have the risk, you have the transition probabilities, all these kinds of different things. But of course, you need to also have the treatment effects. And in order to get treatment effects across many different compounds, more or less the only way to do that is by doing network meta-analysis and indirect comparisons.

[00:21:32] And as you just mentioned, there’s so many problems that come with it. This population, these pretty much everything that is in the estimand statement, all right? And that makes it very often very demanding. But at least you need to be transparent about it. You need to kind of check for the different assumptions. You need to do sensitivity analysis across that.

[00:21:55] Katrin: Exactly. And knowledge about the estimate framework then used for the different trials would help to be able to do that assessment. And of course, Min-Hua said it will not be possible to have the estimand yet. The HTA bodies want to see in each and every trial, but at least to have the knowledge, what are you comparing [00:22:12] with would really help.

[00:22:13] Jen Min-Hua: Probably also address those kinds of the certainty of the evidence, right? Different definitions across trials, different population. All those all need to be considered and documented and be transparent for sure.

[00:22:26] Alexander: Yeah. So just from this kind of comments, you can probably get an understanding that is everything but a straightforward analysis, where you, oh yes, this data set, you run an NMA over. Your job is done. Not really.

[00:22:44] Jen Min-Hua: That’s true. And it would be much more complicated having now the European HTA in mind and to try there to really fulfill all the different country’s needs, the different payers’ needs within one dossier, having a bunch of PICOs, and a bunch of different meta-analyses.

[00:23:01] Being available then also to everybody. And those have to be fitted into the economic models afterwards. Somehow you might have deviated with slightly different subgroup definitions [00:23:10] subpopulations. So there’s a lot of challenges there to come and that will be really interesting how that was and how that will be fitted together with respect to the timelines we have there.

[00:23:30] Alexander: Yeah. Yeah. I can only imagine if you already have lots of sensitivity analysis on the NMA and then that goes into the models, the cost effectiveness model that adds further complexity, you cannot, very easily end up with a lot of factors to consider.

[00:23:38] Jen Min-Hua: And to that, I think as a statistician we, it’s probably not uncommon that we will face this kind of challenge. And then when the trial has been done, it has been published, so there’s nothing else you can do in terms of what you can get from those reported outcomes from the publication, right?

[00:23:56] So as a statistician, sometimes we’re trying to think out of box and trying to address those shortcoming when you’re trying to do an NMA by taking some innovative approaches, like I can give an example [00:24:10] that Lilly has done this submission for the second line non-small cell lung cancer. And so for that indication, because it is for the older non-small cell lung cancer patients, so it’s a broader indication.

[00:24:24] But during the scoping, the comparator being assigned, for example, the nintedanib plus docetaxel, these are for nonsquamous. And then you have another comparator assigned to you saying it’s imatinib that is EGFR positive and so on. And also like other immuno-oncology, which has, using the PDR-1 expression.

[00:24:47] So all those different biomarkers just make it not possible to summarize all your network of all the studies because they have different populations. But then one could say, oh, you can just go into this population doing that comparison.

[00:25:02] However, then you would come to risk that you might not be able to address all the comparators that’s [00:25:10] assigned to you by the payer. And so, we have taken some approaches by hybrid methodologies to address this issue. So this innovative approach is the hybrid two method. So it’s a network meta-analysis of approach with the fractional polynomial and also has a hierarchical exchangeable model.

[00:25:33] So that means that you are able to perform different interactions between the treatment and the population that address this issue. And also, I think at the time, this was not in the guidance. And of course, we’ve been asked by the NICE ERG team many questions, and you have a back and forth to address why we think this is an appropriate approach.

[00:25:56] And to capture the right treatment efficacy relatively compare our compound with other comparators. And so to cut the story short the end is that NICE has accepted [00:16:10] these approaches and then also has grown as the end of life criteria. And we have this approach published in BMC Cancer 2019. And now we have seen a webinar last year that ERG and NICE mentioned about the potential of that approach to be in their future guideline as well.

[00:26:31] So I think that what I’m trying to say is that the innovation could also come in to address some of those shortcomings, but of course, it’s not penicillin that can address all the issues.

[00:26:42] Alexander: Yeah. That’s a really, really nice story. And it shows that by having a statistician that is able to understand, see the problems, and come up with an innovative solution, you can overcome really good hurdles.

[00:26:59] And I’m pretty sure you stumbled over, oh, that we have never done this that way. But if you have really kind of good relationships, good standing in your [00:27:09] company, then you can go with these kinds of innovative solutions and really make a difference to the company, but even more to the patients.

[00:27:18] So that’s a really, really nice story. There’s one thing that, if you are coming from the clinical trial field, that’s always really important, is pre-specification. Now, what role does pre-specification place in this process?

[00:27:36] Katrin: So maybe I can start with the answer on that. If you look at the current guidelines, pre-specification really plays a huge role there.

[00:27:45] So the current guidelines the EUnetHTA is currently establishing for the analysis for the joint clinical assessment. It’s really written down that those which are pre-specified have the highest statistical rigor with the most certainty with respect to outcome.

[00:28:01] And all others which are not pre-specified should be clearly labeled as post hoc, as uncertain, as less validity.

[00:28:10] However, there’s an ongoing discussion on what does pre-specification mean? So pre-specification can mean a lot of things. So of course, you have your SAP for the trial and you have pre-specified what needs to be done for the regulatory approval and for the CSR. But you also have the guidelines in which the analysis to be performed are pre-specified.

[00:28:29] You have the rules of procedures of the HTA bodies. You might have a different document for the HTA pre-specifying analysis, and you also have best practice from other submissions. So there’s an ongoing discussion on what pre-specification means, and I think we as statisticians have to be a bit careful with those wordings pre-specified and post hoc because it has much more different colors.

[00:28:53] Also, when it comes to statistical rigor, how valid is a result or not. So that’s a very good question.

[00:28:59] Alexander: Yeah, completely agree. Actually, when you scroll back in this podcast player, probably by something like three years, you’ll find an [00:29:09] episode that is called something like 50 Shades of Pre-specification because there’s everything in it.

[00:29:18] And also how detailed is it? Is it just a subgroup that is specified? Is this subgroup plus the endpoint? Is it subgroup plus endpoint and population? And exactly how the analysis will be done and all these other things.

[00:29:34] Katrin: Yeah, and I think it’s really important that you as a statistician really know about those different shades of pre-specification.

[00:29:42] And what we really have to do is really avoid data dredging, avoid cherry picking, and avoid data-driven analysis. However, as an HTA statistician, you have to be a bit more pragmatic when it comes to pre-specified analysis, yes or no? Because it’s not meant only the multiplicity and testing hierarchy in the trial.

[00:30:00] And sometimes you pre-specify, let’s say, an analysis flow. You define, okay, what do I do if I get these results and pre-specify this instead of really using the exact method. But we have to be really careful to not take data-driven decisions. And I think that really is an important role of a statistician to find that balance between pragmatism and statistical rigor in the HTA field.

[00:30:24] Jen Min-Hua: Also to Alexander, you mentioned that this pre-specification could be referring to outcome, could be referring to subgroups or any other elements. And I think we all know that patient access to treatment is as important as to receive a positive licensing approval.

[00:30:42] So I think it’s also important at the process of kind of more trial design or even earlier than that, you have a strategy framing statistician from the HTA should have also some voice there to bring in the need from the HTA, different payer, but also what are the other elements like what are the right instrument, what can be accept by this payer, but not [00:31:09] accept by other payer.

[00:31:10] And you need to really balance out the pros and cons before you go ahead with the final study design. So I think that’s also important for the HTA statistician to really have early involvement during the process.

[00:31:28] Alexander: Yeah, absolutely. I think that is one of the big things I learned in my career.

[00:31:34] It’s like always if you put a statistician too early on the project, there’s not a lot you can rescue. And you may have a statistician on it, but the statistician that is mostly experienced in the regulatory frame and now supports FDA and EMA and MHRA and this kind of thing.

[00:31:58] But they may not be aware of all the problems, the challenges that come with HTA, the things that you need to have in your study to be able to have, to even make indirect comparisons. I’ve seen that where, oh, we wanna use this new endpoint in all our studies and then it’s a bad idea because now you can’t do any indirect comparisons anymore.

[00:32:23] And there’s something really nice and innovative in it, but you can’t demonstrate how you’re really better than the others because you’re the only one who would use that endpoint. So see, having this kind of discussion early on in terms of the design of your phase three studies is absolutely vital. And when it comes to oncology, probably even phase two studies because if you’re lucky, these are the ones that go forward into the HTA dossiers.

[00:32:51] Katrin: That’s true. And sometimes it’s singled out on trial. And then you have to be much more innovative and.

[00:32:57] Jen Min-Hua: All the evidence gap there, right? Yeah.

[00:32:59] Katrin: Yeah. Two propensity score methods are different things. Yes, we got evidence to do the comparison. So it’s getting complicated and complicated then. Yeah. But it’s true. And that’s really something I would recommend for statisticians.

[00:33:11] So I’m also curious to learn about it. Well connected, also cross-functional. So not only the development statistician and the HTA statistician, but also with the commercial colleagues, with the market access colleagues to better understand the background and to be able to make the right decisions.

[00:33:26] Alexander: Yeah. There’s usually someone also that is assigned from a market access perspective to these compounds early in the process.

[00:33:37] And that can help at least get input from the major markets. You’ll never be able to get input from every small country. But if you can at least get from Germany, England, Canada, just what they wanna see, that would be really, really helpful. Thanks so much. That was an outstanding discussion about economic models, the European HTA process.

[00:34:05] And it gives you a little bit of an insight into what the HTA Special Interest Group is doing. And there’s a lot of work going on currently around European HTA processing, kind of harmonization of these. That will be a very hot topic for a couple of years, I’m pretty sure about the future of Brexit. That’s kind of the nicest little bit and seeing the English side is a little bit outside of it.

[00:34:33] So there’s a lot of kind of core development. And then there’s other countries like Japan, US, Canada, South Korea, Australia. All of them have similar kinds of processes. So it’s not just about the EU. Katrin, Min-Hua, if you’re gonna give one kind of key takeaway for all the listeners, what’s that be?

[00:34:59] Who wants to go first?

[00:35:00] Katrin: Yeah, maybe I’ll go first. And so really my key takeaway is really be curious, look right and left. Not only focus on your [00:35:09] daily business, but really have a look at the whole development chain of the drug. Understand what’s behind so that you can take the right decisions and also be involved in the discussions, which are so important.

[00:35:21] Jen Min-Hua: I think the recommendation from me would be, as a statistician, it is always very important for us to have a broader picture rather than just looking at the data, analyzing it. Cause I think for the HTA the elements are broader than just analyzing the data, getting approval in terms of licensing, being involved with other elements.

[00:35:43] And so be open-minded, be connected with your other business partner and know the concept in a broader sense. And that would help the statistician to also provide the very relevant and also, to address the relevant questions from the payer.

[00:36:02] Alexander: Thanks so much. And yeah, again, check out the HTA Special Interest Group [00:36:09] homepage on psiweb.org.