Tools4Patient Webinar March 2019

Transcript

Erika Smith: Good morning and good afternoon everyone, depending on your location. My name is Erika Smith. I’m the VP of Business Development at Tools4Patient, and I’d like to welcome you to a webinar entitled A Patient-Centric Approach to Characterizing the Individual Placebo Response.

Erika Smith: Questions can be submitted using the chat function on the lower right side of your screen, and please note that any questions we don’t get to today will be addressed by email within the next 24 hours. So now, I’d like to introduce Doctor Dominique Demolle, CEO of Tools4Patient, who will be presenting today’s webinar.

Dr. Demolle: Thank you very much, Erika. So, good morning. Good afternoon, everybody. The presentation of today is about the characterization of the placebo response. But, from the perspective of taking the maximum advantage of patient characterization, and patient-centric approach. We are going to cover briefly the placebo effect in clinical drug development, as we suppose that many of you are very familiar with this problem when conducting clinical trial.

Dr. Demolle: We will cover why we have used machine learning to predict placebo response. In fact, evidence we could have on the clinical trials, clinical drug development. And we will complete the session by summary, and take away messages before answering your questions.

Dr. Demolle: As we all know, unfortunately as we speak still today, about one-third of the failure happening in phase three may be still related to the placebo effect. In fact this phenomena happens in many different indication reviews. See that in the next few slides. And of course, the immediate associated, we could call “side effect”, is to delay the drug on the market and access to innovative drugs to the patients.

Dr. Demolle: So, as mentioned, this phenomena could happen in a variety of disease, and indications. Of course the most well known are CNS. I mean depression for more than two decades has been extremely well known for a huge placebo effect phenomena. But, this is observed also in respiratory disease, in immunology, in inflammatory disease. And day after day, since we have created the company, we are contacted by sponsor regarding many of the different indications, like the mythology of calmology for example.

Dr. Demolle: So, it’s a huge and complex phenomena. Now, why is it complex? It’s multifactorial. So, by nature, when you give a placebo drug to a patient, you may elicit a clinical improvement, so called the placebo response. This might be related to different type of things. The first thing is the patient is entering a trial. Per se, this impacts the way the patient is going to answer to the placebo.

Dr. Demolle: Of course, study bias may influence the error on your efficacy measurement, it’s also a source of noise. The natural evolution of the disease. The genetic, a few years ago, some genes have been associated to the placebo response. Regression to the mean, a huge effect on the placebo response. And finally what the academician called the true placebo effect, which is the one that is really, uniquely related to the patient.

Dr. Demolle: So, if you say placebo response, at the end it induce variance. And really this is this variance, it will be about during this presentation. This variance that could inflate your standard error, and of course jeopardize your true evaluation of the treatment effect. There might be, also, bias.

Dr. Demolle: For example, you may, in small side clinical studies have an imbalanced distribution of possible responders, non-responders in your study. That is also, unfortunately, a source of difficulties when comparing the different CD arms.

Dr. Demolle: So, could the uniqueness of the patient help to characterize this placebo phenomena, and really take advantage of this information for your CD analysis going to the patient? I mean, clearly since the time this placebo effect has started to be characterized, in pain for example, where we know it’s of two phenomena. It’s related to the opioid and the dopamine pathways.

Dr. Demolle: So there is clearly a pharmacological path we associated with clinical improvements, where pain decreases after placebo intake. It’s extremely evident to understand that it’s something unique to the patient. I mean, each brain is different from each other. And also, because the the trigger to induce the activation of the biased pathway, maybe, for example, related to the psychology of the patient.

Dr. Demolle: And if you take that into account, so the anatomical property of a patient plus things like the personality of the patient, it becomes very evident that it’s a key part of the puzzle. So that it’s not only about that, it’s also about the medical history, about the disease intensity, about the demography. A lot of work has been done in the literature to try to identify the source of the type of patient-specific feature, or characteristics that are related to the placebo effect.

Dr. Demolle: Unfortunately, it is a lot of information. And if you would like to take all those different components feature of a patient into account, this would be too many, and you would not know which ones are really the most pertinent one. In particular, if you think about the psychological traits of personality, I mean you would have to submit your patient to a lot of quite different type of questionnaire. So, clearly not optimal.

Dr. Demolle: This is why, in fact, we felt about at the very beginning when we created the company, about the ideal approach. Ad what, as characteristics, it should have to be really usable in a clinical trial setting. So, okay, we have a tremendous amount of data describing a patient. But at the same time we could, we should take advantage of that for that to be easy of use, and really adding a minimal burden on the patients or the staff knowing that the clinical studies are already sufficiently complex enough.

Dr. Demolle: So metadata, an approach that has to be easy to implement, and at the end of the day we have to address the variance. So the idea we have, was could we from this set of different features come to as single covariate. And from now, I’m going to speak about the placebo covariate. Meaning, really, for a patient, is he or is she plus or minus on placebo responders.

Dr. Demolle: And this would be a composite covariate. Why? Because we will take into account all those different characteristics of a patient, and we use that to a single value which will be his or her placebo covariate. So, to achieve that goal, one of the only means was to use machine learning to predict the placebo response. And this is what we are going to see in the next few slides. So what we did when we started with the development of this technology about five years ago, we conducted our own clinical trials, and I’m going to show it on the data of those very early studies.

Dr. Demolle: So, we conducted our own clinical trials, and we have collected patient data. Including, evidently, some psychological data in the homemade psychological questionnaire. And based on the information collected, we have developed models, algorithms, to produce this famous placebo covariates.

Dr. Demolle: So, really thinking about the previous slides we have seen, is taking the characteristics of a patient to come to a single value, which is the placebo covariate, for that to be used in future analysis. Why? Also, to come to a single covariate. You may be aware that there is a guidance in Europe from the EMA that specify that only a few covariates should be included in the primary analysis.

Dr. Demolle: The reason for that are multiple. Certainly if you have many different covariates, it’s very difficult to use them individually, except if you have very large sample size. Because otherwise it’s, in fact, more negative than positive. And we will see that in the next slide. Indeed, if you have too many covariates, what may happen, and you should see it on the graph, is that in the x-axis, you see when the number of covariates increase on the y-axis, you see your study power decreasing. So it’s always the dilemma about having covariates to well characterize your patient at the baseline, without impacting negatively your study power.

Dr. Demolle: So, to move on to this placebo covariate, we have decided, and it was a strategic decision to have very stable models. So, we decided to use a regular linear regression. Such as ridge regression, linear support vector regression, Gaussian processes. So it’s also called machine learning, but it’s not really, it’s not like for example a random forest, or a neural network.

Dr. Demolle: And also, in term of feature selection, we have applied also a very usual practice like data driven selection via Support Vector Machine Recursive Feature Elimination, we have applied ranking from least to the most relevant features, and ultimately at the end, we have picked up the placebo covariate model with the best performance and the associated set of feature.

Dr. Demolle: So, let’s look now at the impact on the clinical drug development. The first thing we are going to cover now is the, the data that we have been able to generate when we start developing the technology in neuropathic pain. And then I will show a graph displaying some information about osteoarthritis as well.

Dr. Demolle: So, start with neuropathic pain. We conducted, in fact, two clinical trials to start with. One to develop the model, and the second one to validate the model. In other words, to see if we were able to predict in the second trial what would be a placebo responder or not, based on the model developed in the first proof of concept trial.

Dr. Demolle: So, let’s move on then. On this table, so a little bit of a complex table, but the important box to look at is the placebo responder, where you may see that about 30% of the patients have elicit a placebo response. Meaning, a decrease in the average pain score superior to 20%. And in practice, it was translated by an average pain score reduction of 2.5 on a scale from zero to 10. in the validation study, which was much more complex because it was a multicenter clinical trial, the etiology of the patient were multiple.

Dr. Demolle: We have, in fact, similar pattern. We have 30% reduction in placebo, a placebo risk, 30% placebo responders with an average pain score reduction above 20%, and also a a reduction about 2.5 on the visual scale. So, those results were very similar to any type of placebo response that you could observe in a neuropathic pain study conducted by a sponsor.

Dr. Demolle: So the, the slide that we are going to, the graph that we are showing now is about during the duration of the trial. It was about four weeks. We are going to show the decrease in the average pain score. So, the change from baseline. Of course if you are, you see the bottom of the graph you are, plus you see the placebo responder because they have a decrease in the average pain score. Of course the up side of the graph will be the placebo non-responder, as there is no decrease in the average pain score, and even a slight impairment in the anesthetic status.

Dr. Demolle: So, let’s see first, the average pain decrease for the population. We see in green, the pain intensity decreasing week after week. In red, now we will see the pain decrease in the patients that we have in fact predict would be placebo responder. And in blue we will see the pain intensity in the patient that we have predicted would not be placebo responders.

Dr. Demolle: And the most important thing is the statistical significance of those prediction. And as you may see it is very, I believe, significant. So, to summarize these few graphs, the advantage of the technology was that we were able to discriminate between the placebo responders and non-responders.

Dr. Demolle: Evidently the advantage is also to have a continuum, a score that we don’t display here, but in practice we go from high responders to low responders as a continuum. Then the next slide, now, is going to present the advantage. Because you may say, “This is fine, I know the placebo responders, non-responders, but what does it bring to my, to my study and my analysis?”

Dr. Demolle: So, in what you have the variance, the proportion of the variance explain when it is using the composite covariate, and the placebo composite covariate, compare to a typical baseline covariate, like the average pain score, the age, and the gender in all case. And you see that in two neuropathic pain studies, and in osteoarthritis.

Dr. Demolle: And what you notice immediately is that when using a Placebell, you have a proportion of variance explained that vary from 25 to 40%. Of course, in comparison, you see that without the psychological traits of personality information, and also the advantage of the modeling, you are able to explain less variance than when taking into account this information, and it’s about let’s say, 12 to 15 percent. So it’s a significant difference in term of adding information. Of course, also if you explain the variance, what does it mean? It means that you may increase your study power, and let’s say you explain about 35% percent of the variance, you may increase your study power by 10%.

Dr. Demolle: Importantly, if you keep your protocol exactly as designed, you don’t change, your design, you don’t change your sample size. Using this type of approach, having this value brought by the covariate will in fact artificially generate additional patients. So it’s like your sample size is 30 to 40% higher. Which is also a very significant advantage. So you may use this information in different venues indeed. And last, but not least, using a covariate of this type may help you to reduce your risk of type two error, and help you with a a better quality of the decision making process.

Dr. Demolle: Of course, many of you listening to this presentation may ask himself or herself, “Okay, this is very nice. But what is going to happen when I’m going to submit my dossier to the agency.” Very soon when we start with this technology development. We have been in touch with the FDA.

Dr. Demolle: And as you may have already understood, we are not speaking about the biomarker, we are not speaking about the diagnostic, we are not speaking about only the psychological questionnaire. We are speaking about a mixture of a psychological questionnaire, and a statistical approach. So we were invited at a critical path innovation meeting. And at that meeting, which was a very nice scientific discussion, but the outcome was there is no guidance for you to submit your technology.

Dr. Demolle: So the recommendation was to submit the covariate, the Placebell covariate, like any covariate at the time of the IND discussion and IND submission package. A charge for us, through the patient, to provide the validation package of the covariate, and really support the sponsor in this approach. And of course all the documentation related to the development of the technology as a Placebell validation dossier would be reviewed through a third party process.

Dr. Demolle: The EMA, it’s a little bit different with EMA because there was first a specific guidance related to the use of covariate in statistical analysis. So, evidently, when developing this technology, we have taken advantage of this existing guidance. And also we met with the EMA at several locations.

Dr. Demolle: We were invited at an innovation task force to present the technology. As an outcome we where we recommended to move forward to a qualification advice procedure. This is ongoing. We have already nice positive feedback. Mainly the fact that, yes, a composite covariate baseline, a covariate of this type, maybe useful for predicting a subject placebo response.

Dr. Demolle: Also that the questionnaire was a really innovative approach in this instance. And finally that, indeed, to develop the technology in different disease indication, and/or set population. The agency was feeling comfortable with an approach of tailoring first to the technology, the algorithm, and productivity for a further implementation in the following trial.

Dr. Demolle: So last but not least, we have spoken a lot about this covariate as a baseline covariate. So, regular wounds apply. It must be defined before conducting the analysis. So the Placebell covariate, evidently must be available before the inclusion of the first patient in the trial. And of course, included in the statistical analysis done.

Dr. Demolle: And as already mentioned, we provide the sponsor with the report concerning all the individual covariate that has been calculated for the patients who have participated in the trial. And in parallel, we prepare a full validation dossier that could be submitted independently to the agency, to protect our IP.

Dr. Demolle: So, a few take away message to conclude this presentation. First, maybe about the benefits. As an overall, it’s like a a win-win, a benefit for all the stake holders of these processes of drug development, being a patient, pharmaceutical company, or biotech company, or the regulatory agency. In the sense that for the patient the advantage is inevitably the acceleration of access to new therapy. Maybe, also reducing the unnecessary exposure to drugs, and having, really, the patient at at the center. I mean, he or she is really taken into account, because all these characteristics are used.

Dr. Demolle: For the industry, the improved efficacy evaluation of a more accurate evaluation of the real treatment effect. A lower risk of a inconclusive trial, and better go, no-go decision. And certainly at no risk, and we will see in the next few slide, why.

Dr. Demolle: For the agency, it’s moving in the direction of really improving clinical development. There are several exercises conducted now about putting efforts to reduce the clinical research cycle time. This is the type of approach that could really help. And, of course, for the agency, it’s a technology that at the end, it looks familiar because it is a covariate. So, from an agency perspective it’s no waste because the analysis could be conducted with and without.

Dr. Demolle: Remember the key performance indicator that we have mentioned at the beginning. We say it has to be easy, it has to be quick. So let’s see how looks like for the patient, it’s a self administered questionnaire. It’s definitely much shorter than having to complete multiple personality questionnaires to end up at the same result.

Dr. Demolle: For the sponsor, easy. The questionnaire, only once during the whole duration of the trial. And data transfer, yes. We need, evidently, other data like the baseline information, disease information, et cetera. But nothing really beyond the standard practice we are used to.

Dr. Demolle: No extra burden for the the patient or the the site, in the sense that it’s 35 minutes would be taken to complete the questionnaire. Again, only once for the full duration of the study, and evidently it could be available on an electronic fashion as well.

Dr. Demolle: So the advantage of the Covariate solution, now there is no need to exclude patients. You may enroll in your clinical trial placebo responders, because you will have information about the placebo responders or nonresponders provided by the covariate. It’s a no risk. Why? The study, as it is designed, will remain. No change in the design, new change in the sample size. So just adding a questionnaire could provide a lot more information by the value of the covariate.

Dr. Demolle: So, it’s really powerful, as it may help with your study power, and also it allows adjustment for multiple aspect of the placebo response ,and probably would be a matter for our next webinar. It’s very flexible, because it’s easy to integrate. And also, it could be adapted. So, from one indication to another, I mean, starting with a first study in your population, or your indication where we calibrate to the different disease end point, or specificities of the population. And when done, it’s ready to move on for the clinical trial in your drug development.

Dr. Demolle: So, I think that I will stop here now. So, thank you so much for having listened to the presentation, and I’m going to turn to you, Erika, to see if some questions have been posted when I was speaking.