Are We Ready For AI In Clinical Trials?
December 14, 2018 | It seems that everyone wants to jump on the artificial intelligence bandwagon. But is an industry as carefully regulated as clinical trials ready for AI and machine learning? Is it possible to apply these algorithms without losing patients in the process?
Francis Kendall, Senior Director at Cytel, thinks it is. He’s not rushing in without reservations, though. Kendall sees some big challenges for AI and machine learning to overcome before they will be useful in clinical trials—both technical and practical. But he is confident that the world of artificial Intelligence and machine learning is going to change the clinical development paradigm.
On behalf of Clinical Informatics News, Marina Filshtinsky spoke with Kendall about how we can apply artificial intelligence to clinical trials: what are the challenges, where are the first applications, and how patients fit into a machine learning environment.
Editor’s note: Marina Filshtinsky, Executive Director of Conferences at Cambridge Healthtech Institute, is planning a track dedicated to Artificial Intelligence in Clinical Research at the upcoming Summit for Clinical Ops Executives, SCOPE, in Orlando, February 18-21. Kendall will be speaking on the program. Their conversation has been edited for length and clarity.
Clinical Informatics News: We are witnessing the rise of artificial intelligence and machine learning in many industries, as well as in all stages of drug discovery and development. What challenges stand in your way of harnessing the power of artificial intelligence in clinical trials?
Francis Kendall: Three things: availability of data, availability of skilled resource, and the opinion of pharmaceutical companies and the regulators.
I think the big one is the availability of data. To use artificial intelligence and machine learning, you need to have vast amounts of data to run the algorithms. There are some algorithms that you can use to reduce the size of the data, but to get the full power you need large data. So it will take pharmaceutical companies and healthcare data providers to work together to make that data more easily accessible.
The second area is to run these programs and develop the code for machine learning and artificial intelligence, you need the skilled labor to do that. And there's not a lot of Phramecutical experienced data scientists about. A lot of data scientists who know how to do machine learning are going to industries like finance or social media. So attracting these people, developing people in the industry is a challenge and a must if we are to adapt these tools successfully.
And then the final concern is that pharmaceutical companies are usually very conservative in their approaches when they're bringing drugs to submission or marketing. They of course really don't want to upset the regulators. So they're waiting for the regulators to give them clear guidance of where and when artificial intelligence and machine learning can be used. The regulators, on the other side, are saying, “Come and show us what you can do, we can have a good intellectual discussion, and maybe set some parameters going forward.” In a way the FDA’s 21st Century Cures act is a signal that the FDA are willing to embrace this new paradigm
What are the key areas of artificial intelligence applications in clinical trials?
In clinical trials, it probably goes into two clear areas: the production of evidence, and how can we improve clinical trial operations. Let's take that one first.
Clinical trial operations really means, “How can you improve the design of the clinical trial? How can you improve recruitment and selection?” There's a number of companies out there at the moment that provide tools and are starting to base it on large datasets and machine learning. Here we can start to find patient populations and identify if they are close to clinical development sites so you can actually start targeting where your patient populations to join the trials
Now we are starting to realize, if you've got these patient populations, you can start testing the protocol on these model patient populations. “If my inclusion/exclusion criteria are these, then this is my patient target population. If I change this, it's going to be a different population.” You’re going to have to think about redesigning your protocol to optimize your recruitment cohort.
The other application of machine learning is really about evidence—and this goes across the whole spectrum from research: looking at the molecule, how the molecule develops; looking at genomics; looking at clinical trial data and the patient’s own health data. Can we put together the genomics profile with the clinical data and start to understand patterns and response rates. Or we can start to look at overall, if we have a big enough patient cohort, to start seeing if there's patterns in the patient cohort. We can also now, with the availability of data, have patient data not only that's included in the clinical trial, but have it longitudinally before they enter the clinical trial, and then follow them afterwards. And again, put on top of that algorithms and artificial intelligence, which will start to predict some outcome, or populations which are going to be successful. This is one way to develop a precision medicine approach.
There's a real range of applications that machine learning and artificial intelligence can be applied to in the clinical trial paradigm.
You’ve already identified lack of data as an issue. Are the data science and data management disciplines ready for AI and machine learning?
I think that's a really good question, and I think it's a little bit deeper than saying is it ready or not? Everybody has to realize that large datasets—some people say big data or some people call it real world evidence—will have a different dynamic than clinical trial data. We try to make clinical trial data as robust as possible, so we check and double check. That's why you have strong clinical data management groups in the pharmaceutical companies. They're checking on all the data.
Now if you start to go into the big data paradigm and the real-world evidence paradigm, that's different, you're going to have to accept that there's going to be some more missing data points. You're going to have to accept that not all data is collected to the same standards. But we'll be able to use techniques in machine learning to actually reduce this missingness of lack of standards noise. Also with new forms of data from sensors, activity trackers and apps, we can collect streams of data. Now these streams of data will be useful, but again, we'll have to take out the noise in those data. I don't think the current clinical data scientists have yet developed enough skills yet to take advantage of data, but I think we've got to develop those skills and really explore the data to find out where it can benefit health outcomes
And I think we have to look outside the life science of industry, to see how health data is being used , for example look over at what's happening with health applications and apps and devices that the FDA are approving, that a patient can use, like the diabetes monitoring app, which has proven to be better than actually taking first line diabetic treatment, or the app that measures your ECG. These are starting to be used in practice, and therefore we have to look at how they're managing and analyzing the data, and maybe bring some of those skills into the life science industry.
Would you put blockchain in the same category as artificial intelligence and machine learning? What will its role be in clinical trials?
I think the jury is still out. If you'd asked me a year ago about this, I would have said unanimously this is the way to go forward. But I think the jury's out on where it can be used. The whole discussion of blockchain has really opened the debate about ownership of data. And I think that's a great thing that blockchain has done for patients and health data. Blockchain in its various forms is a security technique. There probably isn't a one-size-fits-all approach. I think there's got to be some more work on consolidating approaches before we can say whether it's got value. The big place that we want to go is that every patient can use a security mechanism to own their data, decide who has access to data and what their data is being used for.
Blockchain is one solution to that, but there are other approaches that can do that as well. Blockchain is probably going to be a technology that eventually is behind the scenes, and not something we'll talk about in the future.
With all of these new technologies, is there a space for patients?
I think there is. We've opened the debate about patients and their voice in clinical trials by saying, “Who owns the data? How should data be used?” I also think with the advent of sensor devices and apps, the patient has to be taken into account. For example, if the patient is using an app to monitor their diabetes or monitor their blood pressure, and they go into a clinical trial and they're not allowed to use these apps , I do not think that will be considered ethical. So I think Clinical Trials will have to be designed around patients collecting some of their own data.
I think patients, with the new technology, will have to have a greater voice. We're seeing patients already having a greater voice in the design of clinical trials and also be involved in the results produced by clinical trials. I see that technology such as AI, machine learning, and blockchain are enabling that discussion to go at a faster rate.