采访嘉宾：

Brian Corrigan is Vice President and Global Head of Clinical Pharmacology at Pfizer. 

Brian received his B.Sc, Pharmacy from the University of Alberta, Canada (1989), and Ph.D in Pharmacokinetics from the University of Alberta (1996), and received the Distinguished Alumni Award for lifetime accomplishments in 2021. During his 22 years at Pfizer, Brian’s work has focused on application of Clinical Pharmacology and Pharmacometric approaches to facilitate decision making in drug development, most notably in the neurosciences and pain field, working on Neurontin and Lyrica and other .

作者：

Yuancheng Chen PhD (陈渊成) is a research associate of Phase I Unit, Huashan Hospital, Fudan University. He is a visiting scholar of Uppsala University. He serves as a committee in the Professional Committee of Pharmacometrics of China, Chinese Pharmacological Society. He is also a young committee member in the Professional Committee of chemotherapy pharmacology, Chinese Pharmacological Society, as well as a member of PK/PD study group in Professional Commitee of Clinical Evaluation of Drugs, Chinese Pharmaceutial Association. His research focuses on the clinical pharmacology, especially the pharmacokinetic/pharmacodynamic (PK/PD).

陈渊成, 博士, 复旦大学附属华山医院I期临床研究室助理研究员。乌普萨拉大学访问学者。兼任中国药理学会定量药理专业委员会委员、化疗药理专业委员会青年委员和中国药学会药物临床评价研究专业委员会PK/PD学组成员。研究方向为临床药理学，主要从事药动学/药效学研究。 

陈渊成: The topic of the session is “The application of real-world study and data science in the development of new drugs”. Can you brief introduce the research status of the real-world study and its challenge?

Brian: For example, the real world evidence guidance from the FDA that was issued last fall. Of a key point. But we've also seen that even within health authorities, they're starting to work together on reviews and provide common platforms for review. So for example, the project in August, is a project from the FDA where different health authorities can actually review a drug together. So you can see the feedback from each of them, for example, authorities, the sponsor can basically put the information up, so they all can review. There's also creation of cynical, a cumulus energy, which the platform that will allow one filing for use by various health authorities. If we use real world data, we can do it in the standardized way. So I think some of those regulatory changes have been very important. And then we're just saying that changing the overall ecosystem as well so. So we have partners now like that there are called data aggregate as the making that. Their business may be pulling together data records from across different desperate health systems. So that you can sort of make sense of it or bringing together electron health claim records for us to look at the information.

Of course, all of that is I bm. The backdrop that's really kind of driving. This is just the technological advances. We're seeing it's increased availability of variables. You never watch. I think I can give an etcg on my second. Look at oxygen saturation, for examples, that type of data is available more and more readily. We're using things that allow patients to be at home to do some of the measurements. We're moving to what we call a patient center, capital approach, where the patients can stay at home, tele health approaches and economic history and some of that. And then the other technologies, we're looking at micro micrometrics, patient centers and technologies were as opposed to maybe taking a venous blood sample. We might take a capillary blood sample, and then a quarter get back. So we make it very easy for the patients through to do this type of thing at home. Some changes are as well as the opportunities the field is growing and it's emerging or seeing more examples of the tubes.

陈渊成: In China, the real-world study just begins 1-2 years ago, right? My question is, are there many the real word studies performed in US?

Brian: Again, I think the FDA has actually been a leader in the regulatory science and thinking about how can we use of text and data approaches. For many years, we have something called post approval commitments that may involve looking at safety. It may involve, for example, looking at the populations that haven't been studied in the primary studies. Those types of scenarios are ones that have been primarily where the use of real-world data, real-world studies, design, and pragmatics study designs, etc., have been utilized. I think, if you've been following, some of the information is not just in the US though, and I think we're seeing the use of new ways of thinking about the use of data to inform drug programs.One of the examples that I'm going to use in my presentation is just the way the divisor is collaborated with the Israeli Industry of Health.

Functionally, what we did was we ran when the pandemic move, the vaccine was first available. We made the vaccine available in Israel, the entire country. You lies the device, a vaccine. We've been able to get data from an entire country looking at, for example, safety and efficacy. Really great example, as we were thinking about, do we need a booster shot? Of looking at the data from across the country, breaking down the information by the time that the people actually were vaccinated. To show that there's decreasing effectiveness as a function of time. As the delta variant is changing, we were able to the mystery of health was able to utilize that information to be able to make sense of, is this a rise in the delta variant? Or is this basically a winning of immune response from the vaccine? They did the work, they did that analysis. So that day it was a very unique partnership between twitter government Ministry of Health and a sponsor. Utilizing data really sort of unprecedented levels, things like adverse advanced minorities, for example, where we might study a thousand or 2,000 patients.

And if the incidence is one of a thousand, you may not see it. When they're looking at it and they're looking at 2 or 3 million people, you can get much more effective estimates that are much better estimates of those types of rare events. We use that data better to capture inform the advisory committee of the FDA and have a successful resolution to the: do we need to be questioned whether we need boost or not?

陈渊成: The second question is, the data science is emerging rapidly in these years. So can you brief introduce the data science?

Brian: There are many definitions of data science. I'll just start with money. And I think there's some common themes in the definitions. And No.1 is typically defined as an interdisciplinary field. You have people who have statistical backgrounds, computer science backgrounds, and so on etc.

陈渊成: I searched your published papers, and I see one of the papers was entitled “Data standards for model-informed drug development: an ISoP initiative”, and you publish another paper in 2020 last year. The title for this comment is “Artificial Intelligence and Machine Learning: Will Clinical Pharmacologists Be Needed in the Next Decade? The John Henry Question”.

Brian: So you can stop.The paper was maybe a little bit slightly different perspective on AI and machine learning. So like you're saying there, if you think about it, the definition of the sciences is really sort of that interesting, bring scientific methods to scrap data and understanding from various sources that's not dissimilar to the definition of hormonal nutrition. That's what we do. We develop models. So what's different? Are we all gonna be data scientists next week? And I think that paper that john henry paper is suggesting that: No, you're not.

Now, there are things that will help. From the data science perspective, I think many of the things that we think about with machine learning, for example, our pattern recognition. So taking unstructured data being able to understand it. Same thing. Maybe it's genome wide data. We're looking for patterns of genes that might suggest how drug might work or in the target. But at the end of the day, you still need somebody who has that underlying scientific understanding of the field of clinical pharmacology. To be able to take that information, turn it into hypotheses that you convince her test. That's one part.

Second part is we take that information. And from AI data perspective, we then we take it using our understanding, put it into what I call expert systems. What are expert systems in pharmacometrics? System pharmacology, PBPK, right? We take all the learning to put together. Maybe it's a disease progression model. Where we have an understanding sort of longer two more nature of the disease.

Then the last part where I think artificial intelligence is not ever going to replace someone is right where you guys are sitting. I don't see you work at the website, you work in the hospitals, you work with doctors. You communicate your understanding of the drug, how it works through interactions. I don't see that changing, right? I never see that you will be replaced by a computer in your hospital. That's the good news. That part of artificial intelligence will always be artificial, it will never replace natural intelligence. So I had a little different perspective on that in general.

陈渊成: The third question is, which will be the breakthrough in the field of data science in the following 10 years? What is the challenge in the development of data science?

Brian: I will go back in the realm of chemical pharmacology. I think what I don't see. Let me say what I don't think will be the breakthroughs. But I think the actual technologies, right? Let's say, for example, we're talking about deep learning or neural networks or something like that. I don't see those as revolutionary that we continue to evolve and get better at them. I published my first neural network paper 26 years ago. It still hasn't become sort of the standard. It's one more tool that you can use. It's one more method for you to sort of utilize it. What's gonna revolutionize the field? Are the things that aren't so exciting data standards within your country, from another country, right?

From and cost insurance plan when you can bring data together effectively and easily. That is going to revolutionize how we use it. If there's a little work that needs to be done in the data preparation, then it makes it very easy for us to use it. The changes in the privilege toward background.

With more and more examples, I think it's becoming easier and easier for us to either use design real world studies or use real world evidence. And I look at that example from the Israeli Ministry of Health: Using different types of data in new ways with new departments. I think those types of things are going to be the things that really drive the change. I can see, for example, for us. You maybe have a director of interaction. I see that it will become very common for us to basically look at insurance, claims or hostile records, and as opposed to sort of trying to get an estimate of how often is someone else on. Maybe this is three a substrate. We could just ask the case. And we can estimate of 1 million patients that were with this disease. 16 % of them were on that drug.

So we get very accurate. We can use data in a new way we think of pharmacology. Also, I think for some of our populations that we typically only sort of address later in a normal adult program, we've studied pediatrics later. If you're working in a hospital that often we don't have dosing recommendations for pediatrics, I think there could be sort of some opportunities for us to think about.

All right, we have good estimates, good predictions of what the dose like to be. Maybe we don't need to do sort of the standard PK study, safety africa's study. Maybe they're pragmatic trial designs that will allow us to actually sort of. Those patients learn in a continuous matter. About whether the dose is right, allow those patients to have benefit from within the study, the benefit from the medicine itself. And get that information into the label, maybe years earlier, normally were more than 5 years from the time. We have a drug approved the time that we have a pediatric dosing recommendation in the label. I'd like to see that being 6 months.

陈渊成: Thank you. The forth question: I searched your published papers. Your team have done a lot of pharmacometric work in the field of nervous system. For example, at least about six papers (see the word document). Can you brief talk about them, and talk about your pharmacometric work within the field of nervous system?

Brian: Yeah, I spent many years working in our sciences field and our generation field, and it's a field where we have a long ways to go. We don't have good treatments for things like all san francisco, Parkinson's. And what we were trying to do is really sort of help with the design of trials that look at disease, modifying agents. Part of that is understanding how the disease progresses. And really, so a lot of our work is focused on sort of the understanding of the factors that influence how patient progression might be, so that we could design trials that could answer questions about whether this drug is modifying the effect.

We did that for a lot of work in Alzheimer's and Parkinson. Along the way, again, using outside sources, many, much that workers 10 or 50 years ago, some of those outside sources were that we used in some of those people. Theaters were simply the literature, right? We could look at control arms from studies that have been done for the past 10 or 15 years, and then sort of understand what the normal progression of the disease might look like, whether disease severity impacted that, whether things like different genetic status is might impact the disease progression.

We also later on sort of worked with a trasorsia, a consortium called a critical path institute to sort of cruel data, patient level data from across control arms. To bring that into our models, so that we can develop what we call disease drug trial models. Looked at basically get different drug effects, different disease, severities, and different trial aspects drop out. For example, how is it bring that information together? We develop models that could actually use patient level data as well as that sort of meta data from the literature. And then the last piece, I think, again, coming back to the general theme of the regulatory science, helping people, helping us to drive innovation. We developed the model, but we wanted people to use it. So we worked through the consortium with the FBA that developed a pathway approval pathway called the fit for purpose pathway that allows people who develop drug developer tools like this to actually have them reviewed and validate it is good for purpose, so that other people can use the tools in the same manner.

So that's not just the one off and where they can take the models and continue to evolve them, add new data, add new understanding, comes along to add to the models. So that was a way for companies to de risk by having the approval of a model by the FDA before the instance.

陈渊成: Thank you. The fifth question: in data science, artificial intelligence (AI) is an important aspect. So how we combine the AI and the traditional pharmacometric models to solve some practical problems in the new drug development? For example, develop new drugs in the nervous system area.

Brian: It's a good question. I think, as I said, I see some of the machine learning approaches ---- a break AI. AI is a broad term. So I can break it down at the, various components. I think one good use of some of the emergent of some of the data technologies. Some of the algorithms is around understanding unstructured data and large data sets that are structured. As scientists, we want to understand what may be driving, for example, all standards. We can use information, for example, from a do us studies things like that, look for patterns, for example, in the genes to suggest or maybe there's an atlantic or component, maybe there is various components that we didn't understand before.

The use of that helps us to drive what I would call sort of or develop hypothesis. You can say maybe there's an inflammatory component we should be looking at. We should be targeted these types of medicines for those indications. So can help us sort of pick the different targets that we might look at. So that's one part. In AI space. I also talk about process, body process, automation, process automation. So I think our day is you both are there at 7:30 at night. It's been a long day for you. There's a lot that we can do to automate some of the work that we do. With respect to data flow, things like that. Those elements of robotic process, automation will always that value and and in improve our ability to do the model modeling itself, whether or not the model is using AI approaches or are more classic sorts of approaches that we've used in the past.

I don't think is it's another tool in our tool box. So I think we have a question, and depending on the question, we use different tools to answer, I think it can be a tool, but it really facilitates a lot of the work that we do. Yeah, and I think those are probably the other important ones is I do believe that your domain expertise, right? And it is what is important is that it's sort of that magic ingredient in between allowing you to get from real world data into development and incorporation of that information into expert systems, like system pharmacology or PBPK or things like that, where we can actually make some use of and understand how does this title, the underlying science and really advanced the discipline of pharmacology.

陈渊成: Thank you. The sixth question is, for the drugs in nervous system area, can the real world study be performed? Can you talk about it?

Brian: I think there's a lot of potential for real world studies to be used within that in the realm of exercise. And even I would say what we're seeing is that we can use. We think about real world trials, right? There's a number of different trial types that we can have sort of those pragmatic trial designs. I think, for example, one of the things that we have sort of always have that challenge, and that makes the drug development process long as a role in patients.

So in many of those designs, especially after for thinking about answering questions that randomize control trial design is very effective. But then afterwards? If we have put somebody at, an open label extension, for example, the patient may not want to stay because they may or may not actually be receiving drugs. So the use of brilliant data from the outside hand provides the patient benefit. So I think the utilization of, for example, natural history data model models that understand some of the information beijing approaches. So that allow us to randomize more than 1 to 1 with a controller, will be readily acceptable and should be readily acceptable to health authorities, provide more benefit to patients so that they want to participate in clinical trials. And I think some of those types of hybrid signs are probably on the are being used now. We also have examples of where maybe not so much as many in neuroscience as yet, but rare indications where we're able to stand the label or different populations.

One of the examples I'll talk about is breast cancer. We have a drug where we have approval for breast cancer in women. Male breast cancer is much more rare. And so we were able to use real world evidence to show the use of our drug and having the impact that it was having on male breast cancer patients and make the case effectively. That we should extend our label to include men for breast cancer. So I think there's a lot of examples up in out there.

Now, I think a number of companies have examples of sort of the use of those pragmatic or external comparative studies to demonstrate effectiveness. If many of them are coming in the ontology space for where to see so than the neuroscience.

陈渊成: The next question is, whether the real world studies in US or North America make up the unmet needs of traditional clinical trials?

Brian: I think it's a good question. And what I would say is they believe that is the case. I don't think that the standard randomized control trial can answer every question effectively. I mean, it does a very good job of providing confirmatory type answers. Does the drug work or no? And that dose. But there are things that you can't necessarily answer within practically within that study. So you can't answer questions about rare adverse events, right? That's a very easy place for us to use real world trial designs. What's the incidence of mile card? I guess, with a vaccine, for example, is something like the Israeli Ministry of Health will give us so much better answer that type of data.

There's many other types of questions. For example, what is the actual effectiveness of the drug in a real senate? So we do this control trial, and then we put it out into the general part population, and we get different treatment effects. And there's a term for that's called the effectiveness gap. So the real world trump doesn't necessarily tell us exactly how the brother is going to work in your hospital. And so a real world design that has less inclusion criteria. That maybe doing a controller that will give us actually information about drug effectiveness and safety that might actually better reflect what actually happens with the drug. So I think there's a lot of areas where we can use other types of trial signs that as a process of sort of randomized controlled trial design, we used to answer different types of questions.

陈渊成: The last question: can you brief introduce the important research team in the field of data science? I searched the internet yesterday, for example, the Manchester University, Duke University and Chinese Hongkong University is recruiting the students in the specialty of data science. How to culture the qualified personnel who work in the field of data science. Can you briefly talk about it? Thank you.

Brian: Yeah, I don't know. This is the most difficult question for me, because I'm not familiar with some of those groups. I think the science. We are for more data scientists. Let's talk about data scientists. Help the health field health care. Uncovered college is just one customer. So there are probably some of the biggest data science fields such as banking, finance, and things like that. And many of these are groups you're talking about are doing work there as well. I think, again, coming back to it, each of these centers.

And the data science is actually, if you're looking at some of the components, are the computer science, that perspective components are the same. That you understand that, the understanding of math and statistics is the same right across all of those institutions and all of them, virtually every university does a good job of teaching those skills. Very good. But whatever, however you're applying it, and that's where those fields differ. And I think effective universities. Again, coming back to the definition of this of data science, they're interdisciplinary. The most different groups, the computer science group can talk to the faculty of pharmacy or the faculty of medicine and they work well together. That's where you probably have the very best training and very best ideas come from out of those programs, but that can happen anywhere. It can happen. And it is happening. We were, I don't think there's anybody that really sort of owns this field.

陈渊成: Okay, that's fine. You provide many information on the data science, and the real word study.

分会信息

分会主题12: 真实世界研究与数据科学新药开发中的应用: 中国博鳌与美国实践与策略

分会时间：12月6日上午 上海分会场·远程会议中心

分会主席：刘晓曦 博士 马广立 博士

刘晓曦 博士