From the U.S. Government Accountability Office, www.gao.gov

Transcript for: Artificial Intelligence in Health Care

Description: We discuss a GAO Technology Assessment exploring the future
of AI in drug development.

Related GAO Work: GAO-20-215SP: Artificial Intelligence in Health Care:
Benefits and Challenges of Machine Learning in Drug Development 

Released: January 2020

[ Background Music ]

[ Tim Persons: ] What we all desire is less time, less cost, more cures,
and just better things for society in general.

[ Matt Oldham: ] Welcome to GAO's Watchdog Report, your source for news
and information from the US Government Accountability Office. I'm Matt
Oldham. Only about one out of 10,000 chemical compounds makes it from
initial drug testing to FDA approval for marketing in the US. And
machine learning - a field of artificial intelligence - is currently
being used to help drug development. Tim Persons is GAO's Chief
Scientist and a managing director of the Science, Technology Assessment,
and Analytics team, and we're going to talk about a report assessing the
use of AI technologies in this field. Thanks for joining me, Tim.

[ Tim Persons: ] Thanks Matt, It's great to be here.

[ Matt Oldham: ] So, what are the benefits of using AI to develop new
drugs?

[ Tim Persons: ] Well, you cited the key statistic. When you think about
it, and kind of in a baseball analogy, as a batting average; one in
10,000 is not very many that can get through. So, what if with the
assistance of artificial intelligence, or machine learning, these are
just systems that can help bring more insight, reduce uncertainty. What
if you could kind of sweeten the pool, and maybe get it to one in 1,000,
one in 100? Then you really are changing the equation for how drugs are
discovered first, and then what is selected to then go into development
for what hope to become cures for dreaded diseases like Alzheimer's now,
and variants of cancer, or other things. And you have to bear in mind
that, from a market perspective, development of a drug can exceed $1
billion. It's not a cheap endeavor to be able to do that. And so, if you
can sweeten the pot, right? If you can increase your batting average, as
they say, and it could save money, and it really could reduce a lot of
the time toward the discovery and development of something. But just
imagine doing more of that, and what that would do for general good of
the patients and for society, as well as the reduction of costs to the
medical system.

[ Matt Oldham: ] So, there's the machine learning that's being used
today, and there's a potential benefit for artificial intelligence
tomorrow. Did you find any challenges getting from here to there?

[ Tim Persons: ] Yes. So, as with anything, there are research gaps in
any sort of emerging field. And so there are, of course, needed research
areas in the basic sciences, like the biology, chemistry, and machine
learning, but all of those things together. What's also a key challenge
is the shortage of high-quality data fit for this purpose. So we have a
lot of data, but we need the high quality data that are fit for the
purpose of machine learning to be able to have this effectiveness that's
desired to address this challenge. So, sharing the data is one of the
key things. That's an easy to say, but hard to do thing, especially when
you're talking about clinical trials, data, or things that touch a
human. You, of course, want to protect privacy and civil liberties of
individuals. At the same time in large populations that are anonymized,
you want to be able to use that information to help train or feed a
machine so you get the better outcomes and do that. And then, finally,
there's just a skills gap. We have a lot of very intelligent capable
folks working in this area. This field is growing, but you have to be an
expert in multiple fields in this area; in computer science, for the
machine learning part, and in biology and the Medical Sciences for the
medical part.

[ Background Music ]

[ Matt Oldham: ] So, it sounds like AI could help shrink the cost and
time it currently takes to bring new drugs to the market, but challenges
like research and skills gaps, or acquiring high quality data and data
sharing concerns could be limiting factors in this field. So Tim, what
can policymakers learn from this report? 

[ Tim Persons: ] I think that they can learn that there's a great deal
of justified excitement about the potential for AI for drug discovery.
We all feel the need for better, and more efficient, and lowered-cost
processes to discover and develop new drugs. I think understanding the
complexity of that space, the excitement of AI, what that can bring to
the policymaker in terms of the potential, and yet still the complex
challenges that in a multi-sectoral sense have to be coordinated and
work together. I think that's the bottom line for the policymakers

[ Matt Oldham: ] What's the bottom line of this tech assessment?

[ Tim Persons: ] I think there's tremendous potential on this. I think
it's a key way forward to find all of these drugs. These functionals
sweeten the pot of what goes into clinical trials, even supports
clinical trials itself along the way, and it augments that process,
hopefully. Again, what we all desire is less time, less cost to it, and
more desired outcomes, more cures, and just better things for society in
general.

[ Matt Oldham: ] Tim Persons was talking about a GAO Technology
Assessment of machine learning in drug development. Thank you for your
time, Tim. 

[ Tim Persons: ] Thanks, Matt.

[ Matt Oldham: ] And thank you for listening to the watchdog report. To
hear more podcasts, subscribe to us on Apple podcasts. And make sure you
leave a rating and review to let others know about the work we're doing.
For more from the Congressional watchdog, the US Government
Accountability Office, visit us at GAO.gov.