[House Hearing, 118 Congress]
[From the U.S. Government Publishing Office]
ARTIFICIAL INTELLIGENCE AT VA:
EXPLORING ITS CURRENT STATE
AND FUTURE POSSIBILITIES
=======================================================================
HEARING
BEFORE THE
SUBCOMMITTEE ON HEALTH
OF THE
COMMITTEE ON VETERANS' AFFAIRS
U.S. HOUSE OF REPRESENTATIVES
ONE HUNDRED EIGHTEENTH CONGRESS
SECOND SESSION
__________
THURSDAY, FEBRUARY 15, 2024
__________
Serial No. 118-52
__________
Printed for the use of the Committee on Veterans' Affairs
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
Available via http://govinfo.gov
__________
U.S. GOVERNMENT PUBLISHING OFFICE
55-186 PDF WASHINGTON : 2025
-----------------------------------------------------------------------------------
COMMITTEE ON VETERANS' AFFAIRS
MIKE BOST, Illinois, Chairman
AUMUA AMATA COLEMAN RADEWAGEN, MARK TAKANO, California, Ranking
American Samoa, Vice-Chairwoman Member
JACK BERGMAN, Michigan JULIA BROWNLEY, California
NANCY MACE, South Carolina MIKE LEVIN, California
MATTHEW M. ROSENDALE, SR., Montana CHRIS PAPPAS, New Hampshire
MARIANNETTE MILLER-MEEKS, Iowa FRANK J. MRVAN, Indiana
GREGORY F. MURPHY, North Carolina SHEILA CHERFILUS-MCCORMICK,
C. SCOTT FRANKLIN, Florida Florida
DERRICK VAN ORDEN, Wisconsin CHRISTOPHER R. DELUZIO,
MORGAN LUTTRELL, Texas Pennsylvania
JUAN CISCOMANI, Arizona MORGAN MCGARVEY, Kentucky
ELIJAH CRANE, Arizona DELIA C. RAMIREZ, Illinois
KEITH SELF, Texas GREG LANDSMAN, Ohio
JENNIFER A. KIGGANS, Virginia NIKKI BUDZINSKI, Illinois
Jon Clark, Staff Director
Matt Reel, Democratic Staff Director
SUBCOMMITTEE ON HEALTH
MARIANNETTE MILLER-MEEKS, Iowa, Chairwoman
AUMUA AMATA COLEMAN RADEWAGEN, JULIA BROWNLEY, California,
American Samoa Ranking Member
JACK BERGMAN, Michigan MIKE LEVIN, California
GREGORY F. MURPHY, North Carolina CHRISTOPHER R. DELUZIO,
DERRICK VAN ORDEN, Wisconsin Pennsylvania
MORGAN LUTTRELL, Texas GREG LANDSMAN, Ohio
JENNIFER A. KIGGANS, Virginia NIKKI BUDZINSKI, Illinois
Pursuant to clause 2(e)(4) of Rule XI of the Rules of the House, public
hearing records of the Committee on Veterans' Affairs are also
published in electronic form. The printed hearing record remains the
official version. Because electronic submissions are used to prepare
both printed and electronic versions of the hearing record, the process
of converting between various electronic formats may introduce
unintentional errors or omissions. Such occurrences are inherent in the
current publication process and should diminish as the process is
further refined.
C O N T E N T S
----------
THURSDAY, FEBRUARY 15, 2024
Page
OPENING STATEMENTS
The Honorable Mariannette Miller-Meeks, Chairwoman............... 1
The Honorable Julia Brownley, Ranking Member..................... 2
WITNESSES
Panel 1
Mr. Charles Worthington, Chief Technology Officer/Chief
Artificial Intelligence Officer, Office of Information and
Technology..................................................... 4
Accompanied by:
Dr. Gil Alterovitz, Ph.D., Director, VA National Artificial
Intelligence Institute, Veterans Health Administration,
Department of Veterans Affairs
Dr. Carolyn Clancy, M.D., Assistant Under Secretary for
Health, Office of Discovery, Education and Affiliate
Networks, Veterans Health Administration, Department of
Veterans Affairs
Panel 2
Mr. Prashant Natarajan, Author, Topics: Artificial Intelligence,
Machine Learning............................................... 16
Mr. Gary Velasquez, Chief Executive Officer, Cogitativo.......... 17
Mr. Charles Rockefeller, Co-Founder and Head of Partnerships,
CuraPatient.................................................... 19
Dr. David Newman-Toker, M.D., Ph.D., Director, Armstrong
Institute Center for Diagnostic Excellence, Johns Hopkins
University School of Medicine.................................. 21
APPENDIX
Prepared Statements Of Witnesses
Mr. Charles Worthington Prepared Statement....................... 31
Mr. Prashant Natarajan Prepared Statement........................ 33
Mr. Gary Velasquez Prepared Statement............................ 36
Mr. Charles Rockefeller Prepared Statement....................... 39
Dr. David Newman-Toker, M.D., Ph.D. Prepared Statement........... 43
Statements For The Record
Dr. Pratik Mukherjee, M.D., Ph.D................................. 53
Society for Human Resource Management, (SHRM).................... 55
North America Siemens Medical Solutions USA, Inc................. 60
Johnson & Johnson................................................ 66
ARTIFICIAL INTELLIGENCE AT VA: EXPLORING ITS CURRENT STATE
AND FUTURE POSSIBILITIES
----------
THURSDAY, FEBRUARY 15, 2024
U.S. House of Representatives,
Subcommittee on Health,
Committee on Veterans' Affairs,
Washington, D.C.
The subcommittee met, pursuant to notice, at 10:01 a.m., in
room 360, Cannon House Office Building, Hon. Mariannette
Miller-Meek [chairwoman of the subcommittee] presiding.
Present: Representatives Miller-Meek, Brownley, Deluzio,
and Budzinski.
Also present: Representative Rosendale.
OPENING STATEMENT OF MARIANNETTE MILLER-MEEKS, CHAIRWOMAN
Ms. Miller-Meeks. Good morning. This oversight hearing of
the Subcommittee on Health will now come to order.
Today marks our subcommittee's first hearing dedicated to
exploring the transformative potential of artificial
intelligence (AI) in healthcare, specifically in the VA. This
powerful technology is being used in healthcare systems
throughout the world.
As a physician and a 24-year Army veteran, I have witnessed
the evolution of healthcare in both military and civilian
worlds. While progress tends to be incremental, occasionally a
process or technology emerges that pushes our boundaries out
significantly. The integration of artificial intelligence or
augmented intelligence in healthcare offers this opportunity.
AI creates possibilities to improve diagnostic accuracy,
predict and mitigate patient risk, identify appropriate
interventions earlier, be a consultative resource for
providers, reduce the administrative burden, and save money.
AI, we are told, promises all.
While AI holds great promise, the reality is that it is a
new, developing technology, and we are still figuring out what
is possible and practical and ethical AI. A previous technology
modernization subcommittee hearing addressed the pitfalls of
AI, particularly in data privacy. Today's hearing will focus on
AI's potential. To tap into that potential, VA must first
develop a strategy to use AI, test applications, and, finally,
procure and implement successful AI strategies across the
organization.
As with data privacy, care must be taken when using AI for
clinical purposes. If the data AI learns from is incorrect or
biased, it can make incorrect predictions that results in over-
or underdiagnosis or mistreatment. These are not just concerns.
They have happened in real-life situations outside of the VA.
One promising AI technology for the diagnosis of sepsis, an
often fatal condition with rapid onset, generated alerts for 18
percent of all hospitalized patients, but completely missed 67
percent of the cases diagnosed. This kind of error compromises
not just efficiency, but patient safety. We will examine how VA
is developing use cases guided by various executive orders, and
how VA plans to implement successful AI use cases at scale
across the healthcare enterprise.
Of course, VA and VA healthcare do not exist in a vacuum.
The VA is not an island. AI efforts within the Federal
Government are proceeding in a parallel, while private industry
is significantly ahead of the public sector. Even within
Veterans Health Administration (VHA), this subcommittee has
heard that efforts to use AI are fragmented, with Veterans
Integrated Service Networks (VISN)s pursuing individual
projects that are sometimes duplicative of VHA's efforts. A
priority of ours is to ensure VA moves forward with a cohesive
strategy synchronized between the VA central office, VHA,
VISNs, and Veterans Affairs Medical Centers (VAMC)s.
It is also critical that we understand how VA will choose,
assess, and implement successful AI projects at scale for the
benefit of all veterans and in conjunction with private sector
entities that have already been developing and utilizing this
technology for some time.
We are joined by distinguished witnesses from the tech
industry, academia, and the VA. Their insight will enlighten
our discussion of the VA's use of AI and its potential to
augment VA healthcare.
I believe in the promise AI offers, and I look forward to
hearing from our witnesses about their efforts and vision for
the future of AI to provide what is best for our Nation's
veterans.
With that, I yield to Ranking Member Brownley for her
opening statement.
OPENING STATEMENT OF JULIA BROWNLEY, RANKING MEMBER
Ms. Brownley. Thank you, Madam Chair. All of us gathered
here today have no doubt heard something positive or negative
about artificial intelligence and how it will change the way we
live our lives in the coming years. For most of us, this is a
very new technology, and it will continue to evolve as we work
to better understand how it functions and how we can apply it.
It is also undeniable that this technology is already in use
across various sectors of government, including at VA and in
private companies. To ignore that fact and not support VA's
participation in AI research and implementation of this
technology would be to allow VA and our veterans to be left
behind.
Today, we will hear from our VA witnesses about how they
are approaching this technology, identifying ways to implement
it in veterans' healthcare, and taking steps to ensure AI's
benefits are amplified and its risk minimized. We will also
hear from the companies and individuals working in this field
about ways they see this technology can change how VA provides
care and their experiences in engaging with VA on this
technology so far.
VA is the largest healthcare provider in the country. Its
implementation of AI technology can be a model for other
healthcare systems, which makes it all the more important that
we ensure VA and other AI users establish best practices,
procedures, and guardrails early on in the implementation. AI
technology has the potential to revolutionize how veterans
receive care and ensure better health outcomes.
Providers using AI can potentially identify cancers more
easily, improve patient outcomes, and identify how well
treatments are working to manage chronic conditions. AI can
help providers review imaging scans, and focus their attention
on areas where the technology thinks there might be an issue.
This will also allow patients to get results, good or bad,
faster, and it can help predict disease progression and
potential complications, allowing doctors to more effectively
manage symptoms and apply preventive measures before the
patient's disease progresses further.
It also has the potential to lighten the burden of
administrative tasks for providers and allow them to provide
more engaged and personable care. It can help providers offer
more targeted outreach to veterans who need additional support,
and it can help track and predict risk factors that will allow
mental health providers to intervene sooner for at-risk
veterans.
However, as with any new technology, we must ensure that we
are approaching its use strategically and deliberatively.
Careful implementation will allow VA to establish entrust in
the technology and encourage veterans and providers to see AI
as a tool to solve problems rather than a murky technology with
potential risks. It will be important at this hearing and as
this committee continues to oversee VA's work in this space to
ensure that the patient experience is centered.
AI experts have generally acknowledged that AI will
necessitate changes to workforces across many sectors. Some of
these changes include applying AI to lessen provider burnout
and improve the diagnostic and patient care tools available to
providers. We must ensure that we are taking advantage of these
benefits to the highest extent possible.
However, when it comes to healthcare, removing or lessening
the human element that providers offer in healthcare could be
damaging for patient trust, comfort, and outcomes. Even as we
find productive ways for AI to be implemented, we must take
measures to ensure VA is continuing to robustly hire, retain,
and I will emphasize retain, and protect its clinical
workforce.
Additionally, we must ensure that as providers begin
utilizing AI technology more frequently, that VA can continue
to recruit and train a workforce that is able to use and
troubleshoot the technology. It is clear this is an exciting
and productive time to leverage this technology as we strive to
approach it with the same rigor and oversight we apply to all
our work on this committee.
I look forward to working with our partners from VA, the
private sector, and academia to ensure that we leverage its
benefits to the maximum extent possible for the betterment of
veterans care. I look forward to hearing from our witnesses
today.
With that, Madam Chair, I yield back.
Ms. Miller-Meeks. Thank you so much, Representative
Brownley. Not so rare bipartisan agreement here.
I would like now to introduce the witnesses for our first
panel. Mr. Charles Worthington, chief technology officer and
chief artificial intelligence officer at the Office of
Information and Technology, Department of Veterans Affair; Dr.
Gil Alterovitz, director of the VA's National Artificial
Intelligence Institute, Department of Veterans Affairs; and Dr.
Carolyn Clancy, assistant undersecretary for health at the
Office of Discovery, Education, and Affiliates Network,
Department of Veterans Affairs.
Mr. Worthington, you are now recognized for 5 minutes to
deliver your opening remarks.
STATEMENT OF CHARLES WORTHINGTON
Mr. Worthington. Good morning, Chairwoman Miller-Meeks,
Ranking Member Brownley, and distinguished members of the
subcommittee. Thank you for the opportunity to testify today on
the Department of Veterans Affairs' efforts in exploring
current and future possibilities of artificial intelligence.
My name is Charles Worthington, and I am the chief
technology officer and chief AI officer in the Office of
Information and Technology. I am lucky to be joined here today
by Dr. Carolyn Clancy, VHA's assistant Undersecretary for
Health, and Dr. Gil Alterovitz, the director of the National AI
Institute and VHA's chief AI officer.
VA is committed to protecting veterans' data while
responsibly harnessing the promise of AI to better serve
veterans. While AI can be a powerful tool, we must adopt it
with proper controls, oversight, and security. The Department
is taking a measured approach as we begin to scale AI solutions
to ensure that we are adopting these powerful tools safely and
aligned to VA's mission.
Adopted in July 2023, VA's trustworthy AI framework
outlines six principles to ensure that AI tools are purposeful,
effective and safe, secure and private, fair and equitable,
transparent and explainable, and accountable and monitored.
This framework was designed to align with previous AI executive
orders, Office of Management and Budget (OMB) memos, and other
Federal guidance, as well as VA specific regulation and policy.
Over the past several years, VA has created the
foundational guardrails it needs when considering AI tools have
a significant potential to improve veteran healthcare and
benefits. This foundational AI strategy has given VA a critical
head start on developing policies to govern our use of AI in
production. I believe that creating this clarity on our
expectations will be critical for our partners in the private
sector who are creating much of the AI technology VA and other
government agencies seek to use.
VA has long been a leader in healthcare research and at the
forefront of technology. We have led the way in various
innovations like the development of the first electronic
medical record, early adoption of telehealth, 3D printing, and
more. To support VA's adoption of AI in the healthcare setting,
VA established the National Artificial Intelligence Institute,
or the NAII. It is a collaborative effort among field-based AI
centers and was pioneered by Dr. Alterovitz and his colleagues
in VHA. This network brings together data scientists and
clinicians to enable AI research and development, explore the
application of AI in healthcare operations, and test AI quality
control systems.
As reported in VA's 2023 agency inventory of AI use cases,
VA has over 100 AI use cases tracked, with 40 of those in an
operational phase, with examples spanning speech recognition
for clinical dictation to computer vision for assisting with
endoscopies to customer feedback sentiment analysis modeling.
Most recently, VA launched the AI Tech Sprint, an annual
requirement of the Executive Order 14110. This sprint has two
tracks focusing on how VA can use AI to address provider
burnout by assisting with documenting clinical encounters and
with extracting information from paper medical records.
By investing in these projects, VA aims to learn how AI
technologies could assist VA clinical staff in delivering
better healthcare with less clerical work, enabling more
meaningful interactions between clinicians and veterans.
In closing, the Department believes that AI represents a
generational shift in how our computer systems will work and
what they will be capable of. If used well, AI has the
potential to empower VA employees to provide better healthcare,
faster benefits decisions, and more secure systems.
Similar to other major transitions, such as cloud computing
or the rise of smartphones, VA will need to invest in and adapt
our technical portfolio to take advantage of this shift. With
the strategies, policies, and programs already in place, the
Department will continue in its mission to protect the
integrity and privacy of the data entrusted to us by the
veterans we serve.
Madam Chair, Ranking Member, and members of the
subcommittee, thank you for the opportunity to testify before
you today and to discuss this important topic. My colleagues
and I are happy to respond to any questions you may have.
[The Prepared Statement Of Charles Worthington Appears In
The Appendix]
Ms. Miller-Meeks. Thank you, Mr. Worthington.
We will now proceed to questioning. As is my practice, I
will defer my questions to the end.
I now recognize Ranking Member Brownley for any questions
she may have.
Ms. Brownley. Thank you, Madam Chair.
My first question is to you, Mr. Worthington. Thank you for
being here.
You know, when it comes to technology and this committee's
oversight of that and all of the initiatives and programs that
the VA has, technology has been very helpful on one hand and
sometimes has stood in the way of meeting the goals that we
have set out to do. As a consequence, you know, I am always a
believer that the VA should be leading the way, as you
mentioned, you know, some ways in which we have led the way.
That was a decade or two decades or three decades ago.
I think the research is current, do not get me wrong. In
terms of looking forward into the future, obviously, AI is
going to be very, very important.
I am asking the question to you is, with regard to AI and
its use in the VA today, right now, where do we stand compared
to private health care, teaching hospitals, and the like?
Mr. Worthington. Thank you very much for the question and
it is a good question.
I think that we are doing our best with technology when we
are using it to solve problems that are the most important
problems for the agency, and AI is no different. I think that
VA, in my opinion, we are right in the middle of the pack, I
would say, at adopting these things. I think that a lot of the
health industry, and I would love for Dr. Clancy to chime in as
well, is at the early stages of adopting these new paradigms.
Obviously, many systems went all in on electronic medical
records (EMR)s, which is sort of the basis for a lot of what
can happen now that we have digitized a lot of the healthcare
data. I think we are at the early innings of applying these new
technologies to that data to deliver better healthcare.
I think VA does have a number of these tools that are in
operation now, but I think we also want to take a measured
approach to make sure we fully understand how to monitor the
safety of these tools as we deploy them more broadly.
Dr. Clancy, anything you would add?
Dr. Clancy. Yes, I would say we are the middle of the pack
or possibly even further up than that. The measured approach
that Mr. Worthington described is one that no system yet has
put out in public or has figured out how to take all these
steps in a very, very careful way, you know, to balance
benefits while being very, very attentive to risks and so
forth. The chair gave an example of one that perhaps suffered
from an excess of enthusiasm, which was not to patient benefit.
I think there is a fair amount of caution all around. I
would expect by virtue of our size that in many ways we may
actually be in the lead, which would be a good place to be.
Ms. Brownley. That would be a good place to be. Dr.
Alterovitz?
Dr. Alterovitz. Yes, Alterovitz.
Ms. Brownley. I apologize. You are new to the VA, have been
in the private sector now for a while, I think, at Harvard and
other teaching areas. What is your opinion on this?
Dr. Alterovitz. Thank you for the question, Congressman.
You know, I think it is hard to define it as that there is
uniform progress. What I think we see is that in some areas,
for example, devices----
Ms. Brownley. In some areas what?
Dr. Alterovitz. Some areas, such as devices, medical
devices, we are well ahead. Medical devices, we work through
the biomedical engineering within VHA. Then in other areas that
may require more complicated integrations with different
systems and involve collaborations that need essentially
collaborations across the Department between different parts of
the organization. Those are the ones that we are working
toward, you know, finding ways to do that efficiently at this
time.
The other area that we have been definitely ahead of is on
this aspect of trustworthy AI. A lot of the work that we have
done ended up being in or supporting work that we have seen in
executive orders, some legislation and so forth from the VA. I
think that is partly because we do have that a very special
mission with the veterans. We are especially looking at those
aspects well ahead of time. Thank you.
Ms. Brownley. Very good. Thank you.
Mr. Worthington or Dr. Clancy, either one, you know, so
what is your Department's plan to take the projects from the
tech sprints and pilot phase and implement them as tools across
the VA?
Mr. Worthington. I think that is an excellent question
because I think we are all focused on how we are actually going
to use this to help veterans. We are very focused on not just
the outcome of the tech sprints, but some of the other steps
that we need to take to make it possible for VA to adopt these
at scale. Things around the contracting approaches, the
underlying technical infrastructure to support the hosting of
these tools or the purchase of them if they are hosted by a
third party, as well as the workforce.
You know, there is work that we are going to have to do
both on the AI practitioner side to make sure we have a
workforce that understands how to manage these tools, but also
on the user side. I think there is a lot of training we are
going to need to do with our staff about how to effectively and
safely use these tools. We are starting to make investments in
all of those areas now so that we are ready to receive
promising insights from things like those tech sprints.
Ms. Brownley. Thank you. I yield back, Madam Chair.
Ms. Miller-Meeks. Thank you, Representative Brownley.
The chair now recognizes Dr. Murphy for 5 minutes.
Mr. Murphy. Thank you, Mr. Chairman, and thank you all for
coming today. This is kind of gold rush material, I think, that
we are coming literally on the vanguard of all of this.
You mentioned medical records. I remember kicking and
screaming about 18 years ago when we would literally spend
about an hour just trying to put in an order set. We have come
a long way since then. We are still just literally on the
vanguard of this. We are going to have to go a long way before
this is really streamlined and integral to patient flow.
Just a couple of questions. Dr. Alterovitz, are we still
using Cerner at the VA?
Dr. Alterovitz. I am going to----
Mr. Murphy. All right. Maybe Mr. Worthington.
Mr. Worthington. Yes.
Mr. Murphy. Sorry about that.
Mr. Worthington. The Electronic Health Medical Record
Modernization (EHRM) Project, which is to migrate our Veterans
Health Information Systems and Technology Architecture (VistA)
instances to use the Department of Defense's (DOD) Oracle
health product, which was previously known as Cerner, yes, that
project is underway, and I believe there are maybe four or five
sites that have currently migrated.
Mr. Murphy. All right. A couple, maybe it was months ago,
we had a hearing on Cerner, and then one of the gentlemen
mentioned it would be probably 5 years until it was fully
functional and all these other things. Here we are trying to
walk and chew gum at the same time. We are trying to get our,
you know, providers to really even learn the system, much less
now try to integrate artificial intelligence. This is really
going to be difficult and very, very challenging.
We had a witness a few months or a month or so ago who said
that the efficiency now for clinicians was 60 percent compared
to academic medicine, which is normally about 60 percent
compared to the community. This is really, I think, going to be
very disruptive in the learning process to clinical flows.
Can you expand a little bit what you meant with the DOD?
Are we now having a little bit better communication between our
two healthcare systems, DOD and the VA?
Mr. Worthington. Yes, the goal of that project is to
actually have both systems use one medical record system. That
is underway now.
I think you are raising a really critical point, which is
that many of these AI solutions, to be truly effective, need to
be carefully integrated into the existing workflows so that
they actually reduce burden and reduce the number of clicks and
not add yet another thing that the providers need to check or
open.
Mr. Murphy. I will tell you, I still have my very, very, I
think, well-founded concerns about Cerner being able to handle
this. It is just--it was a system made for smaller hospitals
and here you talk about the biggest healthcare system in the
country, I worry about their ability to, one, even deliver a
regular product, much less an AI product.
You know, one of the best things I thought about residency
is the fact that it was kind of like a buffet line. You had 5,
6, 7, 8, 9, 10 attendings and while you had to rotate with each
one, you took a little bit about what they learned, a little
bit about they learned. If you ask the same question to 10
attendings, oftentimes you get 10 different answers. This is
where the problem with bias is going to come in.
We learned that bias, especially the public, learned about
medical bias during the pandemic. We had one rule, one person
making the comments, one person doing this. This is going to be
a tremendous issue for us.
I am a urologist. I just recently looked at the American
Urological Association's (AUA), one of their ``guidelines.''
Remember what guidelines were? They were saying, hey, think
about this. Now I am hearing--I am seeing the clinician should,
should, should. This is--I think it is very problematic when
this happens.
When we are rolling out AI products and it is saying
should. Yes, where there is going to be a massive liability
concern, in my opinion, because what if you are staring in
front of patient and the AI generator says should, and you are
thinking, I do not think so? Then, God forbid, if something
else were to happen, who is liable? This is a major, major
concern.
Dr. Clancy, you want to speak to that?
Dr. Clancy. Yes. I am sure, Dr. Murphy, that you have heard
that many physicians prefer to use the term ``augmented
intelligence'' as opposed to artificial intelligence.
Mr. Murphy. Right.
Dr. Clancy. In other words, the human in the loop is quite
important. By way of example, right now in research, we have
teams working on developing artificial intelligence, predictive
rules, to try to identify which veterans are likely to do well
after an initial definitive treatment for prostate cancer and
which are likely to have far more aggressive disease and need
much more frequent monitoring and so forth. There is no plan
to--and we do not know enough to actually even get anywhere
close to should, but it is an incredible opportunity.
Mr. Murphy. Yes, I saw that comment in the guideline, and I
am like, I was dumbfounded. We cannot say that in medicine. We
cannot say should, have to, and all these other things. That
takes away absolute clinical aspect.
You know, I could ask you questions for 4 days because this
is such a target-rich environment. One of the things, and I
will just end this, you know, the medical records writing,
these are the bane of our existence.
I spoke with the head of another company I will not say
here, and their thought was, you could walk into the room, it
would have a microphone. You are just talking with a patient.
It is assimilating what you are saying, what the patient is
responding to. Then you just tell it, you know, I am going to
order this. Bam, bam. You walk out of the room and the notes
are done, the orders are done, the paperwork is done. That
would be a quantum leap, quantum leap, to addressing physician
provider burnout.
Dr. Clancy. That is exactly what we are testing, sir, in
this tech sprint that Mr. Worthington referred to.
Mr. Murphy. Yep.
Dr. Clancy. Having seen one of these tools demoed live, it
was quite amazing.
Mr. Murphy. Yes.
Dr. Clancy. We are going to be testing all of this in our
simulation center in Orlando so that people can figure out what
the workflows are. We have actually looked at one company's
product because at that point in time, June of last year, that
was the only one they thought was ready for prime time. I have
to say the teams were wildly excited. Like, when can we start?
Mr. Murphy. Yes, that is a big time. I have exceeded my
time. Just remember, AI is not going to take over my scalpel,
so. All right.
Ms. Miller-Meeks. Maybe. Thank you, Dr. Murphy.
The chair now recognizes Representative Budzinski for 5
minutes.
Ms. Budzinski. Thank you, Madam Chair, and thank you,
Ranking Member, for holding this important hearing today. I
want to thank the witnesses as well from the VA for
participating. Really appreciate that.
As we have heard this morning, there is so much potential,
and I believe that in AI, to better serve our veterans and
especially the veterans that I am honored to represent in
central and southern Illinois, which are predominantly rural
veterans. A part of the nature of artificial intelligence is
that it is constantly changing, which can lead to challenges
when trying to implement or scale up the technology.
My first question is really for the entire panel, and if it
is okay, we will start, though, with Mr. Worthington. What
steps is the VA taking to monitor and keep up with the emerging
research on artificial intelligence?
Mr. Worthington. Thank you for the question, Congresswoman.
We have a really robust partnership with our colleagues in
VHA's innovation group, as well as the National AI Institute,
which is, I would say, constantly looking at the emerging
research on this technology and even doing some of its own
research. My part of the VA kind of steps in once things are
getting past that research phase and into something we want to
start testing with real veteran data or real clinical use
cases.
Then as we find those examples that are most impactful,
then we bring them into operations in a way that is somewhat
similar to how we would operate other IT systems. We are
following those same security and privacy policies that would
govern our use of veteran data in other cases as well.
I will defer to the other panelists if they want to talk to
how we are keeping up.
Dr. Clancy. A couple of other efforts. First, a lot of our
currently funded research from the Office of Research and
Development does not have AI in the title. By way of the topic
that is being focused on whether that is cancer research or
other problems, they are testing strategies to try to predict
who is likely to do the worst.
We saw a lot of this as well during the acute phases of the
pandemic. I am trying to get past saying we are done because we
are not. We were able to predict, for example, which patients
hospitalized with COVID were most likely to die within the next
several months because those would not be the people you would
want to be discharging first. You would want to be attentive to
detail and so forth.
We also have a team keeping up with the published
literature and things presented at meetings and so forth. There
is so much we need to know about the safe and effective part
that Mr. Worthington referenced that we are very, very excited
about it and do not want to leave any stone unturned.
Dr. Alterovitz. I just wanted to say a quote that I heard
from a former VA person that really research is kind of needed
in a couple of places. There is kind of need for research to
ensure that operations are really based on science. Right? Then
the reverse is also true in some sense. For research to be
successfully translated, right, into operations, you have to
push forward on that.
Connecting research and operations is a very important kind
of mutually symbiotic type of thing, where they work together
to create the best product on the operations side-the best
research that can actually be useful and leveraged. Interacting
from the beginning is something that we do at the VA to really
make sure that all the work that we do can be useful for the
veterans.
Ms. Budzinski. Can I just ask, have you found in this
research and this collaboration and partnerships you have any
ability specifically for AI to address some of the gaps in VA
care for rural veterans in particular? Have you had any
specific takeaways from the research thus far, I guess?
Mr. Worthington. I think the VA has a number of programs to
try to address that gap, including our telehealth program.
Overall, I think anything that can make our system more
efficient at identifying which patients are most in need of
specialty services, for example, could assist with things like
our telehealth program in getting those right--exact right
experiences to the patients that need them.
Beyond that, I do not know that there is specific AI uses
in the rural space, but it is a very interesting question.
Dr. Clancy. Well, I will simply say that we have a very
substantial initiative and investment in precision oncology.
This focuses on lung cancer and prostate cancer. From the
beginning, launching this 4 or 5 years ago, you know, the
overarching motto was leave no veteran behind. We are now up to
about 75 tele-oncology clinics and also working through the
extent to which we can engage those veterans in research
without making them come a phenomenal distance to the research
intensive institution.
There is a lot of work going on there, and I know that
cancer is a very, very big issue for rural communities. I mean,
a big fear.
Ms. Budzinski. Yes. Thank you. Thank you very much.
I am out of time, so I will yield back. Thank you.
Ms. Miller-Meeks. Thank you, Representative Budzinski.
The chair now recognizes Representative Rosendale, who is
the chair of the Subcommittee on Technology Modernization.
Representative Rosendale, you have 5 minutes.
Mr. Rosendale. Thank you very much, Chairwoman Miller-
Meeks, for holding this hearing and allowing me to participate
today. I appreciate the witnesses for being here. Good to see
you folks again.
I chaired a hearing last month in the Technology
Modernization Subcommittee titled, ``The future of data privacy
and artificial intelligence at the VA.'' This is an important
topic and something the VA must get right. I am grateful that
the committee is giving artificial intelligence the necessary
attention that it needs.
Mr. Worthington and Dr. Alterovitz, during last month's
hearing, I asked you whether you think the VA has a
responsibility to notify veterans when their health or personal
information is fed into an AI model or whether analysis that
affects them was done by AI rather than a person. Everybody
seemed very agreeable and supportive of that, that we actually
had this disclosure and that question was posed to them. When
are you going to put that notification and informed consent
procedure in place?
Mr. Worthington. Thank you for the question. We are working
with our VHA ethics group right now to better understand what
the approach should be on this topic. Obviously, this is kind
of an emerging topic, as you stated in the prior hearing. I do
not believe we have a specific time that we are aiming for to
implement this, but we are very aware of this issue, and I
think it is one that is spoken to in the Executive Order as
well.
Our thinking right now is that the use case inventory is
the basis for which we would want to make those disclosures.
Obviously, the use case inventory is a pretty technical
document, so we are going to need to do work to make that
understandable to veteran patients so that they can understand
how the VA is using AI and how their data might be put into
those models.
Mr. Rosendale. That is fine and good. Okay. I know you are
working on this. The problem that I see is that you are
literally putting the cart before the horse. You are utilizing,
okay, you are utilizing AI and you are not disclosing it to the
veterans. You are not giving them a choice. That is dangerous.
It truly is. It is dangerous and it is dishonest. There is no
really industries that are allowed to be utilizing different
types of techniques and tools, okay, without the consumer being
notified of what those techniques and tools are and how it may
impact them.
I will reiterate, this needs to be a high priority. You are
utilizing AI at whatever degree, at whatever level, and the
veterans need to be aware of that, and they need to have that
consent and to continue to utilize it is not right. Do you have
information that would show that the analysis of any type of
testing whatsoever can be done more accurately by AI, rather
than a doctor's bare eye, shall we say?
Dr. Clancy. We do not have that information, and I think it
is going to be hugely important. Women recently have been
offered the opportunity to spend another $40 to get an AI-
enabled mammogram reading. You know, to a person, most of the
physicians interviewed for this article said, I have no idea if
this is worth the money. Some people coughed up $40 and others
did not.
I did want to get back to your very important question
about ethics, though. I am just quoting from my colleague. We
are developing processes and standards right now. The first
step, we thought, was a very broad ethical framework about
protecting the privacy of veteran data, and that cuts across AI
and everything else. We will be happy to follow up with you as
we progress through that. Our lead ethicist in VHA is really
terrific.
Mr. Rosendale. Again, I appreciate that, and I do believe
that you are working on that. The problem is that you are
already utilizing AI, and the veterans, they do not receive
informed consent.
Mr. Worthington, during the last hearing, I asked you
whether anyone in the VA ever rejected an AI use, and if so,
why? You took the question back. If they are not getting
consent, if they are not getting disclosure, then it is
probably not likely that they are. Have you already had any
veterans rejecting the use of AI, even without this consent?
Mr. Worthington. I am not aware of specific examples of
that. I think that, you know, in many of these cases, the AI we
have in operations today is tied to, like, an Food and Drug
Administration (FDA)-approved medical device. For example, we
have a product called Clear Read, which is a tool that assists
with radiology scans, chest Computed Tomography (CT)s. These
features are being added to existing products incrementally and
in many cases being adopted.
I think there is this new interest of the AI technology
with a broad definition of what would constitute AI. I think,
as Dr. Clancy mentioned, this is a topic that I think our
ethicists are going to have to kind of understand. What new
requirements should we create versus what can reuse our
existing guidance on standards of care and other sort of
disclosures? How much will that cover the bases?
Mr. Rosendale. Thank you. Madam Chair, I see I am out of
time. I yield back.
Ms. Miller-Meeks. Thank you very much. Representative
Brownley had another follow-up question, so I will yield to
her.
Ms. Brownley. Thank you, Madam Chair. I appreciate it. I
just wanted to ask one last question.
I do not know offhand, but I think the majority of medical
centers are associated with medical schools and teaching
hospitals. I am wondering, are there partnerships out there
working, and is that happening really across the board with
medical centers and medical schools?
Dr. Clancy. Absolutely. We are affiliated with literally
every single medical school in the country and many, many other
programs associated with other health disciplines, which is
just an awesome asset to have in the research space. Many of
our docs, about 60 percent, and it is a higher number of those
who are active researchers, actually have dual appointments
with an academic affiliate. Yes, there is a lot of
collaboration going on, and we look forward to more of that
here. Yes.
Ms. Miller-Meeks. Thank you. The chair now recognizes
Representative Rosendale for an additional minute.
Mr. Rosendale. Thank you very much, Madam Chair. I do
appreciate. I just have one more quick question.
Mr. Worthington, the VA holds an unparalleled wealth of
veterans' data. Far too many companies are already interested
in monetizing this information. We have seen them actually
buying it to skirt around the privacy laws, okay, especially
when it is in regards to government agencies doing so. It seems
to be getting even more tempting. AI companies compete by
consuming the most data to train their models, and some of them
already have covered nearly everything on the public internet.
How are you going to protect veterans' health data as it
becomes a more and more lucrative prize for these companies to
get their hands on?
Mr. Worthington. Yes, it is an excellent question,
Congressman, and we believe very strongly that protecting
veterans' data is pretty much job one, especially in Office of
Information and Technology. I think we are lucky that we have a
lot of existing policies around how veterans' data can be used
and how it cannot be used. We would expect that those would all
continue, even in this AI use case. It will be really important
that all of our vendors understand, which is the case today,
but that----
Mr. Rosendale. Are you putting language in place in any of
the agreements with your vendors to make sure that that
information is protected and not monetized?
Mr. Worthington. Yes. I believe that our existing contract
vendor relationships already have clauses that say they can
only use this data for very specific reasons, if indeed they
even have access to it. Oftentimes, this data is stored on a VA
system. It is not given to a vendor to have in their system at
all.
I do think that because, as you mentioned, the value of
this data is uniquely increasing in the age of AI, I think it
is something we want to look at to make sure that we are very,
very clear that the data cannot be used for any purposes other
than what is in the contract.
Mr. Rosendale. Thank you very much. Madam Chair, thank you
very much. I yield back.
Ms. Miller-Meeks. Thank you. The chair now recognizes
herself for 5 minutes.
I appreciate the great questions by our members, both on
the 1 million in prizes for the conclusion of the tech sprints,
and I appreciate your answer to that. A follow-up question to
that is whether there will be barriers to implementation for
these technologies. Dr. Worthington?
Mr. Worthington. Mr. Worthington. I wish I was a doctor
sometimes. Yes, I think that contracting for technology is
obviously a pretty complex topic, and there is a lot of rules
around how that works in the government, and that is one of the
things that, you know, we work on being good at. There are
things like Federal Risk and Authorization Management Program
(FedRAMP) for cloud services, which is a policy designed to
ensure that cloud providers have some of those data privacy
protections that we just discussed.
One of the challenges we see in the health space in
particular is that while VA is a big healthcare provider, in
the scheme of the American healthcare industry, we are
relatively small. Oftentimes many of the healthcare tool
providers, they are not really familiar with FedRAMP as a
compliance regime that they would be focused on. Now, they have
a number of other compliance regimes from the health industry
that they focus on, but FedRAMP is not often high on that list.
That is one example of some of the challenges we sometimes have
at doing acquisitions of enterprise tools in this space.
Ms. Miller-Meeks. In follow up to Dr. Murphy, it was an
excellent question. I had the same question on my mind. Without
the full implementation and the hiccups that the VHA has had in
implementing its electronic health record (EHR), that certainly
is going to impact and I think delay your implementation of
appropriate AI into the VA. Perhaps I will ask that question a
little bit more on the second panel.
Suicide prevention is a top priority for me, for this
committee, and for this subcommittee and the larger committee.
What is the VA doing in regard to using AI to better prevent or
predict veteran suicide? This may be related to my comment
about EHRs, and how are we ensuring these tools are the best
ones available on the market? Mr. Worthington?
Mr. Worthington. Yes, great question. I will just maybe
point out two examples of our current use of AI in operations.
One is we have a model called Recovery Engagement and
Coordination for Health Veterans Enhanced Treatment (Reach
Vet), which is designed to predict the veterans that are most
at risk for suicide as an outcome. That information then can be
used to inform the way that the doctors follow up with them or
the treatments that they prescribe when they are seeing them.
That model is in operations now.
To provide another example, we have a natural language
processing (NLP) model that is looking at comments that are
coming in through our customer experience listening. Most of
those things are like, you know, I went to the VA and the
parking was slow or whatever. Occasionally those comments will
indicate that this veteran might be at risk or need, you know,
help. Maybe they are indicating that they are having
homelessness problems. This NLP model can flag comments that
might be particularly concerning for follow up by a
professional that can read the comment themselves and decide if
some other action is warranted.
Those are just a couple of examples of how we are trying to
use these tools to help the VA with that mission.
Ms. Miller-Meeks. If I can, one follow-up to that, given
how important this issue is, if there is a flag, are you
working with the clinical side to make sure that that is
addressed in immediate fashion? We have veterans who have
committed suicide in the parking lot of a VA hospital because
they were denied care or thought not to be suicidal.
Dr. Clancy. Yes. Reach Vet that Mr. Worthington referenced
is focused on veterans who are enrolled in our system, and we
have seen a decrease in suicide attempts and a subsequent
decrease in all cause mortality. Hard to pinpoint that and say
which is associated with suicide or not.
We are also working with external contractors to use
various types of AI, working with veterans who are not enrolled
in our system. When we give the numbers about veteran suicides
and think about what our responsibility is, it is all veterans,
not just those who are enrolled in the Veterans Health
Administration.
Ms. Miller-Meeks. Thank you. Mr. Worthington, even though
the VA has a published and my opening remarks called augmented
intelligence AI strategy, it is difficult to find guidance on
how that strategy is implemented and how VA is faring against
each of their four stated objectives. Is the VA going to
publish key performance indicators (KPI)s so that us here in
Congress and the public can actively see the progress VA is
making in the AI space?
Mr. Worthington. Yes, I think that is an excellent
suggestion. We will be updating the AI strategy in the coming
year, and I think having KPIs would be a great idea.
Ms. Miller-Meeks. Thank you very much. There are no other
representatives here at this time, so on behalf of the
subcommittee, I want to thank you for your testimony and for
joining us today. You are now excused, and we will wait for a
moment as the second panel comes to the witness table.
[Recess]
Ms. Miller-Meeks. Welcome, everyone. That is my signal. I
am going to thank you all for participating in today's hearing.
Our witnesses on our second panel: Mr. Prashant, I said it
in my brain, Natarajan, an author on the topics of artificial
intelligence in healthcare and machine learning (ML); Mr. Gary
Velasquez, chief executive officer and president of Cogitativo;
Mr. Charles Rockefeller, co-founder and head of partnerships at
CuraPatient; Dr. David Newman-Toker, director of the Armstrong
Institute Center for Diagnostic Excellence at Johns Hopkins
University School of Medicine.
Dr. Natarajan will deliver his opening statement. You have
5 minutes.
STATEMENT OF PRASHANT NATARAJAN
Mr. Natarajan. Chairwoman Miller-Meeks, Ranking Member
Brownley, and members of the VA Health Subcommittee, my name is
Prashant Natarajan, and I have problems pronouncing my own last
name half the time. I am an author of four books on health
data, AI, and cancer. I have more than 20 years of experience
in building electronic health records, including Sono; also
medical imaging and building AI systems at scale. I have
brought about 100 AI applications and use cases to life with my
customers and my teams.
For the last 8 years, I have been volunteering as industry
advisor on data science and AI at the San Francisco VAMC and
University of California San Francisco (UCSF), where we have
been developing expert solutions for detecting traumatic brain
injuries, specifically using head CT. In my daytime, I work as
vice president of health and life sciences at H2O.ai, which is
a leading open source generative AI company.
It is my privilege to join you for this important hearing
on AI in the VA. It is a cause that is close to all of our
hearts and is happening at a pivotal time for our veterans, the
clinicians who serve them, and our industry as a whole. AI is
already bringing value to health systems, pharmaceutical
companies, and various organizations in the public sector. It
is happening right now.
Generative AI provides a lot of new options. It does that
and more by augmenting and amplifying the human experience.
Generative AI humanizes and empowers by democratizing access to
actionable insights using plain English, plain Spanish, or any
other language of your choice. Any patient can rapidly develop
a personal health AI where each individual creates the AI that
they need, in addition to what is created for them by others.
Veterans can now use Generative AI to better manage their
health and life. Similarly, clinicians can leverage GenAI to
address burnout, reduce human errors, and find the needle in
the haystack of expanding side effect knowledge.
Allow me to illustrate with an example of 11-year-old who
is using Generative AI to turn her baking hobby into
collaborative solutions and new value for her classmates. She
does this by creating new AI business applications and agents,
and she did not even know what these meant a year ago. She used
Generative AI to ask questions about scone recipes. She then
tailored them for her dad's taste, which is not easy. She
customized them for her various users' preferences and created
a new dataset that combines unstructured data across portable
document formats (PDF)s, web content, and her own recipes to
create new information, new recipes, and is now in the process
of using Generative AI to create her own app to take mobile
phone orders. If an 11-year-old can do this for something as
basic as creating recipes, imagine what our veterans and VA
clinicians can do with the same technology to address health
outcomes and other issues of much greater importance.
In my written testimony, I have provided numerous examples
of patients and clinicians as AI trainers, AI creators, and
empowered users. I am happy to review these examples in Q&A or
post this hearing.
How do we create this new, empowering future of bottom-up
innovation? Based on our experience so far with the PLOT
program in creating empowered patient advocates and
researchers, we have some proven best practices in place. Here
are four things we need to do together.
One, recognize AI fidelity, which is the value of health AI
being determined by its user, veteran, clinician, or
administrator in the context of its use.
Two, recognize and encourage the fact that AI use cases can
come from anywhere, both within and outside of the four walls
of any VA facility.
Three, we need to empower veterans to develop the tools
they need to address their personal problems. We need to create
public-private partnerships with appropriate data, tools,
upskilling and deployment option underscored by a veteran-first
AI ownership of their assets.
Finally, the personal and provider health AI that I
describe in my AI collaborative are new ways of bringing AI to
life in the VA. Hence, veteran-created models and AI apps
should be treated with minimally prescriptive regulations and
should encourage the use of open source.
Thank you again for inviting me to testify. I look forward
to working with you, the VA clinicians, and our veterans to
solve longstanding challenges and create new opportunities.
[The Prepared Statement Of Prashant Natarajan Appears In
The Appendix]
Ms. Miller-Meeks. Thank you very much.
Mr. Velasquez, you are now recognized for 5 minutes to
deliver your opening statement.
STATEMENT OF GARY VELASQUEZ
Mr. Velasquez. Thank you. Thank you Chair Miller-Meeks,
Ranking Member Brownley, and the esteemed members of the
committee. I appreciate the opportunity to come and speak this
morning on the use of AI at VA.
I possess advanced technical degrees with over four decades
of experience operating large healthcare analytic companies,
national health plans, large medical centers, and an
international clinical research organization. I also want to
acknowledge the Federal Government, including the VA, for its
AI initiatives, which have leaned into the use of ML and AI to
improve the health of Americans. My company was privileged to
participate in the early stage of ML programs for Operation
Warp Speed. While we specialize in precision health, we perform
at great speed and scale. However, I would say the path from
diffusion to operations has been somewhat clouded.
We, my company, when we take on a project, we intend to
deploy our solution, not a wish to deploy our solution. I think
that is a little bit of a challenge we see. Everything we do
has an intent to deploy in the private sector.
Today, we stand on the precipice of transformative ML and
AI possibilities to empower VA beneficiaries, reduce stress on
providers, improve patient outcomes, and deliver personalized
healthcare. However, we must ensure we do not become blinded by
following the next shiny AI announcement. We should focus on
the right use cases for AI and, more importantly, use cases
that can rapidly improve, meaning today, healthcare for our
veterans. I want to describe two use cases that can bring to
life immediate benefits of machine learning for veterans.
The first use case covers completed work at the VA, where
our models have been tested to predict beneficiary level
disease progression. My second use case covers the ability to
predict clinical conditions related to genetic mutations,
polymorphisms, that may have resulted from toxic exposures, so
actually using mutated DNA to predict future clinical
conditions.
An American hero raised me. My dad enlisted in the Army at
age 15. If you saw his pictures, he looks like he is 12. He
received two silver stars at age 17 for his service in Korea. I
got to see him when he came home, both with the physical and
invisible wounds. He raised me, and he has dealt with that
invisible wound of post-traumatic stress for many, many years.
Then I got to see him age, and he wrestled with his post-
traumatic stress. At the end of his life, he wrestled with my
mom's cancer while he was trying to manage his Chronic Kidney
Disease (CKD) and his diabetes.
As we all know, combat veterans are selfless. I think being
in combat makes you selfless, and he was selfless with my mom.
My dad chose to focus on my mom's health instead of his, and,
unfortunately, we did not know the actual state of his health.
He passed from the unseen, unknown complications related to CKD
and diabetes.
That choice my dad made does not need to be made today. We
have the tools. We have the machine learning tools to help VA
providers, his provider, to identify, predict, and communicate
that disease progression, not just to the beneficiary, but to
the providers, and to the family. When we have this tool
deployed to the VA, we can help a vet and their family navigate
the conditions of age.
As you all know, the average age of a veteran is increasing
every day. It is now 68. This machine learning capability can
assist these older vets, their family, and the providers with
specific insights into their conditions. These insights can
also reduce the number of touches of a medical chart, relieve
administrative burdens, and reduce the cost of higher acuity
care.
With toxic exposures we recognize the pressing concern of
adverse health effects and stressors from toxic exposures among
our veterans and the families. By leveraging ML techniques, we
can unravel this complex interplay of genetic mutations and
illness. We can enhance our understanding of how these factors
influence health outcomes and enable timely, earlier diagnosis
and treatment.
For example, my company's chief medical officer, his son
and daughter fly jet fighters. Both acknowledge they have been
exposed and they have three simple questions for us today. What
are my risk probabilities for future medical conditions? What
type of diagnostic testing should I get for these conditions?
What is the frequency of those tests?
They know they signed up for the military for those risks,
and they are fine with it. They have these three basic
questions, and with machine learning, we can quickly answer
these questions.
With the support of Congress, VA can be a cornerstone in
delivering AI-enhanced services to improve human health, not
just veteran health. Given VA's mission, operations, and rich
data repositories, few other organizations can deliver this on
this objective better.
This concludes my remarks, and I am pleased to answer
questions you may have.
[The Prepared Statement Of Gary Velasquez Appears In The
Appendix]
Ms. Miller-Meeks. Thank you. Mr. Rockefeller, you are now
recognized for 5 minutes to deliver your opening statement.
STATEMENT OF CHARLES ROCKEFELLER
Mr. Rockefeller. Great, thanks. Thank you, Madam Chairman.
Good morning, ladies and gentlemen. My name is Charles
Rockefeller and I am the cofounder and head of partnerships for
CuraPatient. It is a real honor to have been included today in
this very important discussion.
By coincidence, I happen to feel more historically
connected to the VA because I heard about it being discussed at
the dinner table since age 12. My father sat on the Senate VA
Committee for 30 years, either as a member or its chairman. My
other two cofounders of the company are Long Nguyen, who has
been supporting the U.S. Government and its AI endeavors since
its very inception 20 years ago; and Dr. Siddhartha Mukherjee,
a Pulitzer Prize-winning oncologist.
First, some information about our platform and our
technology. Its features mostly fall into three categories and
have been designed specifically to be able to support patients,
providers, and administrators to deliver care efficiently and
seamlessly. These features create seamless support for veterans
and reduce worker burnout as has been discussed many times
before. One of our first successes came while working with
Operation Warp Speed, where we helped provide equitable access
to critical care while also allowing our brave frontline
workers relief to focus on the job at hand. I am proud to say
that we received a Red Cross Heroes Award for this service.
With all that as a foundation, I would like to shift my focus
to our work with the VA in particular and more specifically.
CuraPatient, our company, first came into contact with the
VA in 2019 via the NAII tech sprint, as you all know what it
is. We won that and I am proud to say we were deemed the future
of healthcare, although they might have been generous with the
title since it was the first tech sprint. I think we are.
Today I would like to highlight five key topics from our
experience with the VA, and each creates the foundation not
just to innovate, but to do so responsibly and at scale, which
I know is two continuous themes.
Number one, data privacy and security. We have dedicated
over 2 years and thousands of hours, along with significant
resources, to gain FedRAMP certification. As part of this
commitment, we have implemented 421 National Institute of
Standards in Technology, NIST, security controls and those, of
course, the highest in the industry. The effort has been led by
a collaborative and cross-functional endeavor significantly
propelled by the leadership of Charles Worthington, Angela Gant
Curtis, Dr. Carolyn Clancy, of course, and first before all of
them, the initial direction and support from Dr. Paul Tibbits.
Our commitment to FedRAMP reflects our dedication to protecting
our veterans' sensitive data.
Number two, seamless, integrated, and veteran-centric
experience. Our work is centered on creating a seamless and
user-friendly experience for both veterans and VA staff. Then
they will both work together better. We are thrilled to report
that we have successfully completed five out of our six
targeted integrations, granting us the bidirectional ability to
both read and write to patient records.
Number three, clinical application of AI. Our collaboration
with the VA facilities in Long Beach and D.C., Long Beach,
California, has been a cornerstone of our efforts, where
established AI oversight committees and policies are already
enhancing our work, these committees.
Our technology's integration starts with addressing long
COVID, which, as you know, is in the news recently for being
much more prominent now. This condition, with its broad impact
on the body, provides a unique opportunity for wide-ranging
engagement using our solutions. There is a benefit to this as
well, because our solutions are designed to tackle other
chronic diseases and on a larger scale as well.
Number four, responsible AI. These pilots will be deployed
at NAII centers and will be available later across the entire
VA. The Long Beach and D.C. VA Medical Center teams led the
work. It enforces compliance with trustworthy principles as
defined by Executive Order 13960, incorporates NIST, AI, Risk
Management Framework (RMF), and all nonbinding principles
within the White House AI Bill of Rights.
The team has stated that the AI system we created, our name
CuraPatient, shall only move forward and can only move forward
with the full approval of these bodies. The more it is used,
which is important to realize, the more it is used, the smarter
it becomes.
Number five, contracting. We are optimistic about the
benefits of enhancing our contracting approach, which promises
to be a positive change. As technology, especially AI, advances
rapidly, navigating the complexities of traditional contracting
becomes a growing challenge. I am nearly done.
Often, by the time firm fixed-price contracts are executed,
the technology is maybe it is 3 years later, the technology has
already been replaced or advanced. It is vital to consider
alternative contracting methods, and I would call upon Congress
to make this a priority, as well as funding to turn these
opportunities into real benefits for veterans.
The leadership's works of the VA has resulted in a soon to
be mission ready system that can greatly apply advancements in
AI, not only in theory, but directly to our veterans and
support staff.
Thanks very much for your time, and I am happy to take
questions. Thank you.
[The Prepared Statement Of Charles Rockefeller Appears In
The Appendix]
Ms. Miller-Meeks. Thank you. Dr. Newman-Toker, you are now
recognized for 5 minutes to deliver your opening statement.
STATEMENT OF DAVID NEWMAN-TOKER
Dr. Newman-Toker. Thank you. Chairman Miller-Meeks, Ranking
Member Brownley, and distinguished members of the subcommittee,
thank you for the opportunity to address Congress on this
critically important topic of artificial intelligence in
healthcare at the VA in support of our veterans.
My name is David Newman-Toker and I am a physician
scientist with doctoral level training in public health and a
research focus on improving medical diagnosis, including the
development and deployment of novel diagnostic technologies
such as AI. I have been a faculty member at the Johns Hopkins
University School of Medicine for more than two decades, where
I am currently a professor of neurology and director of our
Agency for Healthcare Research and Quality (AHRQ)-funded Center
for Diagnostic Excellence. I am also a past president of the
Society to Improve Diagnosis in Medicine.
My testimony today will focus on opportunities and
challenges for AI in healthcare from a public health
perspective, with a special emphasis on AI to improve medical
diagnosis. I will tailor my remarks to the VA context as
appropriate, but I believe what I share here today is broadly
applicable to healthcare both within and outside the VA.
I would like to state for the record that the opinions I
express here today and in my written testimony are my own and
do not necessarily reflect those of the Johns Hopkins
University or Johns Hopkins Medicine.
AI is the branch of computer science concerned with
endowing computers with the ability to simulate intelligent
human behavior. The most complex cognitive task in medicine is
the act of diagnosing a cause of a patient's symptoms. Errors
in diagnosis account for an estimated 800,000 deaths or
permanent disabilities each year in the U.S., including,
obviously, our veterans, more than 80 percent of which are
associated with cognitive errors or clinical reasoning
failures. This creates a unique quality improvement opportunity
for AI-based systems to save American lives at public health
scale.
Potential benefits of AI include better health outcomes for
patients at lower costs; greater access to and efficiency of
care delivery, especially for those currently underserved or
disadvantaged or in rural settings; and decreased healthcare
workforce burnout. However, none of these benefits will be
realized without tackling foundational data challenges facing
AI.
The rate limiting step for developing and implementing AI
systems in healthcare is no longer the technology. It is the
sources of data on which the technology must be trained. There
are multiple facets of healthcare data quality problems which I
address at greater length in my written testimony. However, in
plain language, they boil down to the problem of garbage in,
garbage out. If we train AI systems on faulty data, we will get
faulty results. AI systems that learn on faulty data will
generally make the same mistakes that humans make, or worse.
Put simply, if available electronic health record datasets are
used to train AI systems, the best we can hope for is AI
systems which replicate existing safety failures or implicit
human biases. The worst we can expect is AI systems that are
frequently wrong in their recommendations. If AI-based systems
are deployed without adequate testing, the quality of
healthcare will drop, not rise.
The VA healthcare data environment is better suited than
most to delivering high-quality data that might train AI
systems. Key attributes include the VA's commitment to
healthcare quality and safety, a large national network of
providers and patients, a unified health record offering
greater potential for standardizing data capture, independence
from financial reimbursement-driven problems in healthcare
encounter documentation, and addressing a patient population
that tends to stay largely within the VA system so outcomes can
be better tracked over time. These attributes give the VA the
opportunity to take a leading role in building high-quality AI
systems.
For AI and healthcare to maximally benefit the health of
all Americans, including veterans, the following are essential.
First, AI systems must be trained on gold standard datasets
that are unbiased and include complete information on both
clinical inputs and care outputs. Two, AI systems must be
effectively integrated into clinical workflows, leveraging the
strengths of computers and humans together to produce a better
result than could be achieved by either alone. Three, wherever
AI is used, systems to monitor, maintain, and even enhance
clinician skills should be codeployed so that clinicians and AI
systems will continue to fact check each other.
I have three primary recommendations for the committee with
regard to implementing AI at the VA, with an emphasis on
diagnosis. First, the next decade must focus on constructing
gold standard datasets for diagnosis. The promise of AI will
not be realized without quantifying bedside evaluations.
Two, AI systems must be held to a high diagnostic standard.
They must be demonstrated scientifically to improve safety and
quality over current care and then monitored closely over time.
Three, the impact of AI on human clinical diagnostic skills
must be monitored and managed. Clinical deployment of AI should
be explicitly designed to enhance, rather than reduce,
clinician skills by applying educational and human factor
science.
Thank you for this opportunity. I would be pleased to
answer any questions you may have.
[The Prepared Statement Of David Newman-Toker Appears In
The Appendix]
Ms. Miller-Meeks. Thank you very much.
We are now going to proceed to questions. Ranking Member
Brownley, you have 5 minutes.
Ms. Brownley. Thank you, Madam Chair. Appreciate it.
Mr. Natarajan, I am not sure that I agree with your
hypothesis that if 11-year-olds can create AI, imagine what
veterans can do. Perhaps younger veterans, I would agree, yes.
Older veterans like me, I am not so sure. Hopefully, we will
all have our children or our grandchildren to help us out.
Appreciate your testimony. Thumbs up.
Mr. Natarajan. Can I respond to that, Congresswoman?
Ms. Brownley. Sure.
Mr. Natarajan. Congresswoman, give me 1 hour of your time
and I will prove you wrong and have you doing, using, and
creating AI.
Ms. Brownley. Well, I have heard that AI is, you know,
going to tell you how to do it all anyway, so perhaps you are
right. I have to be convinced.
Mr. Rockefeller, I understand that CuraPatient is certified
through this sounds like very extensive process of the FedRAMP.
Even I think Mr. Worthington made comments about how expansive,
I guess, it is and may need to be kind of looked at and
evaluated from the government's perspective. Tell me a little
bit more about your experience becoming certified.
Mr. Rockefeller. Certainly. Thanks.
Ms. Brownley. Yes, speak into the microphone. You have to--
there you go.
Mr. Rockefeller. Thanks very much for the question, and
certainly there is a lot about that.
Overall, and then I will get to a couple particular points,
I think that the--and I would recommend to this committee that
the FedRAMP process, maybe the approvals time, I think there is
a fair amount of backlog in the system to review all these, and
there is several stages of review, as you know, and there
might, I think, be a backlog. I do not know for sure, but I
think there might be. If somehow Congress could fund additional
people to work on these approvals or to focus on it more, I
think that would be beneficial, because when we were getting
it--and we were very lucky, right? We took, you know, 2 entire
years and possibly more. The reason that I mentioned this is
that, you know, we made it through. Right? In fact, I do not
have a motivation to say what I am about to say, which is that
I am concerned that because the process takes so long that the
VA might be missing out on other medium-or small-sized
companies who want to pursue it, and they just cannot last that
long. Right. They just have to make more money on their own or
something.
Fortunately, we are, you know, well-funded through our
investors and other investments, and they all knew that they
were investing in us getting FedRAMP, which would then sort of
lead to other things. I am concerned that a lot of other
companies might sort of start the process and then hopefully,
you know, throw up their hands.
Ms. Brownley. Thank you. I have just a little bit more
time, and I have another question.
Mr. Rockefeller. Sure, please.
Ms. Brownley. I appreciate your response.
Dr. Newman-Toker, I wanted to ask you if you are aware at
all of partnerships between Johns Hopkins and the VA that is
going on.
Dr. Newman-Toker. Thank you, Congresswoman, for the
question. I apologize, I do not--excuse me. Thank you,
Congresswoman. I am not aware of those specific partnerships to
which you refer.
Ms. Brownley. Well, just a partnership around AI between a
university teaching hospital and VA with--in terms of using AI
applications.
Dr. Newman-Toker. Certainly, as Dr. Clancy mentioned
earlier, there is a tight relationship between many academic
medical centers and the VA system. It happens that the
affiliate in Baltimore is with the University of Maryland
rather than with Johns Hopkins. Some of those connections are
tighter in that space.
Ms. Brownley. I see. You talked about some of the risks,
and I appreciate that testimony, because I think we have to be
eyes wide open on that. Knowing sort of the VA and its
operation, what steps can the VA take now to avoid some of
those pitfalls?
Dr. Newman-Toker. I think you are taking them in the
trustworthy AI framework that you have delineated. I think
three of the six pillars are absolutely crucial, effective and
safe, fair and equitable, and accountable and monitored. If
those are followed, the others have some more technical
attributes to them. but those three deal directly with this
issue of the safety of delivery of the service. If they are
handled well, I think you will be in a good position, better
than, I think, many other places that have not put that kind of
framework in place.
Ms. Brownley. Thank you. Happy to hear that.
I yield back.
Ms. Miller-Meeks. Thank you very much. It is been a very
insightful testimony, and as a physician and a veteran, Dr.
Newman-Toker, I can wholeheartedly agree. It is not just what
data is available to put in, but what the clinician observes,
whatever level that clinician is, because that data, whether it
is verbal data, whether it is observed data, nonverbal
communication, and then actual physical findings, that data
goes into that system, which will then help with the diagnosis.
If that data is poor or bad, then the result will be equally
bad, which brings up another question, and that is the VA does
have an opportunity, because it is a relatively closed system,
to have a great input of data, but we have Health Insurance
Portability and Accountability (HIPAA) regulations.
Has there been a thought to allowing a voluntary waiver of
HIPAA for deidentified data that could go into that matrix and
be utilized to further help with both machine learning and
smarter augmented intelligence?
Dr. Newman-Toker. Thank you, Dr. Miller-Meeks. I am not
aware of any specific action that has been taken toward the
idea of HIPAA waivers for this specific purpose. I like the
thrust of your question. I think it is on point.
There are times where the inability to follow a patient
over time or to acquire information prospectively in a given
encounter in order to capture the sort of full diagnostic
journey, for example, may be challenging because of the HIPAA
constraint. I do believe that your suggestion to give patients
the opportunity to assist us in providing better care through
AI is a good one.
Ms. Miller-Meeks. It is imaging as well, imaging, blood
work.
Mr. Velasquez, and I can tell that you are winning, but in
your written testimony, you spoke to the ability of AI to help
with capacity and resource management, specifically with
aligning medical staff levels, optimizing wait times across the
direct and community care networks, rationalizing the use of
direct and community care, and efficiently tying those options
is a major concern, both for access and cost. Perhaps if we can
save money on one or spend money wisely on another, we will
have money that can go to, i.e., I am thinking of tech sprints.
Why are we giving a million-dollar prizes if we need people to
be able to solve a backlog on FedRAMP?
Mr. Velasquez, can you talk about how AI would do this,
particularly with a decade worth of community care data the VA
has, and what some of the obstacles would be?
Mr. Velasquez. If I can weave it into your first question
around HIPAA waivers and consent. We spent most of my work for,
the company's work is in the private sector. We have curated a
dataset of anonymized patients, but they are linked, so they
are hashed out of about 200 million Americans and about 100
million Americans' EMRs, they are linked. I do not know who
they are. They have a hash. It is literally, I would say, if
you leave out Wyoming and Montana, sorry, Senator Tester,
wherever you are at, we pretty have a healthcare view of where
people live. From a data perspective, whether it is clinical
capacity, practice patterns, supply, these datasets exist not
just to apply it in the VA and obviously bring it in the VA
datasets to look at future demands.
To me, it is an issue of not so much supply. It is where is
the demand and, frankly, where is the need? Trying to predict
those two using rules based methodologies or regression models
it is trying to predict the weather, but the clouds have
basically their behaviors, their agents, they change their
opinions, and they interact and talk among each other. They
emote reaction, because that is trying to manage healthcare. If
you think about it, how the patients interact with physicians,
physicians interact with each other. It is a very complex,
dynamic system. If we are going to really get our arms around
understanding supply and demand healthcare, that is a perfect
use for machine learning.
Ms. Miller-Meeks. Now I am going to ask a million-dollar
question. It is something that former Speaker McCarthy brought
to our attention on a visit to Massachusetts Institute of
Technology (MIT).
We are Members of Congress. I have a science background as
a physician, but certainly when it comes to technology, and
especially augmented artificial intelligence, our knowledge
base and foundation may be lacking. Yet we are making decisions
on how to both fund, implement, regulate, both the promise and
also the pitfalls of AI.
My question, if you all can just briefly answer it, how
would you recommend Members of Congress be able to educate
themselves so that--Ranking Member Brownley is saying it is
impossible. Very quickly, what would you advise Congress to do
so that we can, you know, adapt technologies rapidly, perform
the proper oversight, the proper protection of data, and to
legislate in a way that is most appropriate, that allows us to
really effectuate the promise of AI in healthcare, which can be
transformational?
Mr. Velasquez. Let me take a shot. In my company, I think,
and I will keep it short, I focus on the use case. Back to your
point, Ranking--the technology changes, it just changed. It
literally moves that quick. There is some kids in Cal or MIT
doing something that just blows us away. We are not going to
keep up with those. To me, we need to focus on the use case and
start there, or the challenge we are trying to address, then
back up.
Having these hearings, having the discussions, and just
asking the questions, what is that challenge we are trying to
solve and then start back up from there, I think is probably
more appropriate use of Congress' time, rather than trying to
keep up with the kids in the garage coming up with new models.
Ms. Miller-Meeks. Dr. Newman-Toker, and then I will go Dr.
Rockefeller--or Mr. Rockefeller.
Dr. Newman-Toker. I think, very briefly, I think you are
doing this by bringing in expertise. I think the most important
piece is the diversity of that expertise in order to make sure
that you have all the relevant perspectives on the
implementation of the technology.
Ms. Miller-Meeks. Mr. Rockefeller.
Mr. Rockefeller. I would say that the first step, because
it is an accurate question, yet with a vague sort of response,
I would say the first step is to become familiar with the
products and services that are being offered by the private
sector, with the VA. Right? This is what the tech sprints
enable, to sort of bring it forward to you. They are all sort
of, during that process, we became very familiar with the inner
workings of the VA, learning about the systems, how to do the
integrations, all of that is good groundwork of knowledge to
share with you. I would almost say the best way to break the
cycle is simply look at the products and request it through
whoever.
Ms. Miller-Meeks. Thank you. Mr. Natarajan.
Mr. Natarajan. Thank you, Congresswoman. Just a quick
couple of things.
We have experience in taking people across various age
groups, various education profiles, and converting them into
patient researchers where they are applying for their own
grants and getting funded. We are doing that with AI. One of
the things I would like to offer, the same thing I offered
Ranking Member Brownley, is for this entire subcommittee, allow
me to come and do a workshop for you. Give me 4 hours of your
time, and I will have all of you creating some AI or not that
is useful to your lives.
Ms. Miller-Meeks. Sounds like a topic for a roundtable.
Ms. Brownley. Can I just clarify my impossible statement?
Ms. Miller-Meeks. Ranking Member Brownley.
Ms. Brownley. I just wanted to clarify my impossible
statement. I do think that Members of Congress, most Members of
Congress, can wrap their heads around AI applications as it
relates to healthcare, but all of the risks involved in
national security and other kinds of things and how AI is going
to sort of penetrate the world, I just feel as though Congress
is--I mean, we have not figured out how to regulate social
media and privacy issues and so forth and so on. AI is just,
you know, way out there compared to dealing with Facebook.
I just think that many have recommended, I think have made
recommendations to Congress that what the government really
needs to do is provide an entirely new agency around technology
with a lot of smart people within that agency that can advise
Members of Congress, you know, how to wrestle with these
regulations and so forth, in particular our national security
issues. Thank you for letting me explain.
Ms. Miller-Meeks. I am now way over time. Ranking Member
Brownley, would you like to make any closing remarks?
Ms. Brownley. You know, I would just like to say this is--I
wish we had, you know, a lot more time, because I thank the
chairwoman for bringing this forward as a topic, and I think it
is a really important topic that we need to really focus on
more. I hope we will have additional hearings as we move
forward on this.
I really do think at the end of the day, we should probably
have a hearing with a full committee on it as well, so we can
really spend more time and drilling down on it.
I really do thank all of you for being here, and I am very
impressed with your testimonies and very impressed with the
work that you are doing for the VA, but all of the work you are
doing outside of the VA to move forward with this technology. A
lot of gratitude to all of you. Thank you very much. Again,
hope we will spend more time drilling down on all of this.
I yield back.
Ms. Miller-Meeks. Well, again, I want to thank our panel. I
want to thank the VA panel, appreciate all of the expertise
that was here. Perhaps my comment on how we can best assist is
my own deficits.
AI is a powerful technology with great promise. From
automating tedious tasks and saving time for clinicians and
administrative staff, to aiding in diagnosis of disease and
tailoring treatment, AI will alter the delivery of healthcare.
As we have heard, there are concerns that must be addressed. I
would like to bring Representative Rosendale to my district,
where one of the first AI-directed devices that was approved by
the FDA was developed, and that is for diabetic retinopathy, a
screening tool for diabetic retinopathy, and he would see the
power of AI and how that is going to lead to access,
prevention, and affordability.
These concerns, of course, have to be addressed in how the
VA uses AI and in how the VA acquires and implements AI. This
subcommittee will continue to exercise oversight of the VA as
it moves to assess, acquire, and implement AI, and also to
educate ourselves and our members, as well as the public. I
think continued hearings on this topic would be very
beneficial.
If AI needs authority to do things differently, this
subcommittee will proactively assess the need and the impact.
As the VA moves forward, it must do so with a plan and the best
interest of veterans in mind, and I look forward to the pillars
that come forward. This subcommittee will do its part to ensure
that those goals are met.
I would like to thank all the witnesses for their presence
and their testimony. It has been of tremendous value. The
complete written statements of today's witnesses will be
entered into the hearing record.
I ask unanimous consent that all members have 5 legislative
days to revise and extend their remarks and include extraneous
material. Hearing no objections, so ordered.
I thank the members and the witnesses for their attendance
and participation today. This hearing is now adjourned.
[Whereupon, at 11:29 a.m., the subcommittee was adjourned.]
=======================================================================
A P P E N D I X
=======================================================================
Prepared Statements of Witnesses
----------
Prepared Statement of Charles Worthington
Good Morning, Chairwoman Miller-Meeks, Ranking Member Brownley, and
distinguished Members of the Subcommittee. Thank you for the
opportunity to testify about Department of Veterans Affairs (VA)
exploration of current and future possibilities of Artificial
Intelligence (AI). My name is Charles Worthington, and I am the Chief
Technology Officer and Chief Artificial Intelligence Officer in VA's
Office of Information and Technology. I am accompanied by Dr. Carolyn
Clancy, Assistant Under Secretary for Health, Veterans Health
Administration (VHA), and Dr. Gil Alterovitz, Director, VA National AI
Institute (NAII) and VHA's Chief AI Officer.
VA is committed to protecting beneficiaries' data while responsibly
harnessing the promise of AI to better serve Veterans. While AI can be
a powerful tool, we must adopt it with proper controls, oversight, and
security. VA is taking a measured approach as we begin to scale AI
solutions to ensure we are adopting these powerful tools safely and in
a manner that aligns with VA's mission. Adopted in July 2023, VA's
Trustworthy AI framework outlines six principles to ensure AI tools
are: purposeful, effective and safe, secure and private, fair and
equitable, transparent and explainable, and accountable and monitored.
This framework aligns with various AI executive orders, Office of
Management and Budget memoranda, other Federal guidance, and VA-
specific regulations and policies. VA has created the foundational
guardrails it needs when considering AI tools that have significant
potential to improve Veterans' health care and benefits.
VA Is a Federal Leader in Artificial Intelligence
As one of the pioneering Federal agencies to adopt a national AI
strategy, VA has a head start on developing policies and procedures to
govern the use of AI in production. VA seeks to align these policies
with broader Federal guidelines and requirements covering privacy and
data protection, ethical use of AI, interoperability and standards,
procurement and acquisition, and research and development, VA will
ensure consistency and accountability when we implement AI technologies
while also safeguarding the security, privacy, and well-being of
Veterans. This clarity on our expectations will be critical for
entities in the private sector who are creating much of the AI
technology VA and other Government agencies seek to use.
VA has long been a leader in health care research and at the
forefront of technology, leading the way in various innovations such as
the development of the first electronic medical record, early adoption
of telehealth, 3-D printing, and more. To support VA's adoption of AI
in the health care setting, VA has established the NAII AI Network, a
collaborative effort among field-based AI centers pioneered by Dr.
Alterovitz and his colleagues in VHA. The network brings together data
scientists and clinicians to enable translational AI research and
development, accelerate the application of AI in health care
operations, and test AI quality control systems. The current locations
of the network include Washington, DC, Long Beach, California, Kansas
City, Missouri, and Tampa, Florida.
VA's Data Security and Privacy Safeguards
VA has a robust privacy policy for information technology (IT)
contracts that explicitly controls how others may use VA data. When a
vendor needs access to VA data to perform its services, its handling of
the information is limited to the strict confines of the contract, and
the vendor is prohibited from using or disclosing the data for any
other purpose. If vendors violate any of the information
confidentiality, privacy, and security provisions of an IT contract or
non-disclosure agreement, their penalties can include contract
termination, withholding payments, additional Federal Acquisition
Regulation remedies and measures, and Health Insurance Portability and
Accountability Act of 1996 sanctions. Additionally, any serious misuse
of data will be referred to the VA Office of Inspector General, the
Department of Justice, law enforcement, and other oversight bodies for
civil investigation and criminal prosecution.
VA's Current Efforts in Artificial Intelligence
As reported in the VA 2023 Agency Inventory of AI use cases, VA has
tracked over 100 AI use cases. Forty of those cases are in an
operational phase, and examples of the cases span from speech
recognition for clinical dictation to computer vision for assisting
with endoscopies and customer feedback sentiment analysis modeling.
Some highlights of our efforts so far include the following:
VA is incorporating AI technology into Veterans' health
care to enhance diagnostic accuracy and efficiency, and to predict
cancer risks and adverse outcomes. This includes using predictive
analytics for early and personalized interventions, streamlining
administrative tasks, and accurately identifying appropriate health
care providers for care.
VA's Recovery Engagement and Coordination for Health--
Veterans Enhanced Treatment (REACH-VET) initiative uses AI to identify
Veterans enrolled in VA care with the highest level of suicide risk.
Since its inception in 2017, the initiative has successfully identified
over 117,000 at-risk Veterans. VA evaluation of the REACH-VET indicates
that the clinical program has been associated with increased attendance
at outpatient appointments and proportion of individuals with new
safety plans, reductions in mental health admissions, emergency
department visits, and suicide attempts.
VA launched the suicide prevention initiative, Mission
Daybreak Grand Challenge in 2022. Among the winners was ReflexAI, an
AI-powered training simulation in which crisis responders at the
Veterans Crisis Line can build and practice skills to improve their
ability to provide response to Veterans in crisis. The VA Office of
Mental Health Suicide Prevention is collaborating with Oak Ridge
National Laboratories in the Department of Energy to develop new models
that enhance REACH-VET by incorporating community data and geospatial
data. The goals are to enhance the precision of predicting Veterans at
highest risk of suicide, reduce bias, and enhance equity in vulnerable
populations.
The VA Stratification Tool for Opioid Risk Mitigation
(commonly referred to as STORM) is a clinical decision support tool
that uses predictive models to assist in identifying patients who
require targeted monitoring and intervention for adverse outcomes.
The Food and Drug Administration-authorized GI Genius
system has been successfully deployed in 106 facilities for over
100,000 colonoscopies. Its primary purpose is to enhance the detection
of precancerous polyps in the colon in real-time during a colonoscopy.
VA has invested approximately $19 million in purchasing GI Genius over
the past few years, with the goal of completing deployment by this
fall.
VA has also recently launched the ``AI Tech Sprint'', a requirement
of Executive Order 14110, Executive Order on Safe, Secure, and
Trustworthy Development and Use of Artificial Intelligence. This sprint
has two tracks focusing on how VA could use AI to address provider
burnout by streamlining administrative tasks such as clinical note
taking and the processing of paper medical records. VA has allocated
nearly $1 million for contract and software license costs to facilitate
the AI Tech Sprint and plans to offer $1 million in prize money for
participants in the sprint.
These are just a few examples of the AI projects and initiatives
currently underway within VA. By investing in these projects, VA aims
to leverage AI technologies in the future to improve health care
outcomes, enhance patient experiences, and optimize resource allocation
for the benefit of Veterans.
Artificial Intelligence Is a Generational Shift in Technology
VA believes AI represents a generational shift in how our computer
systems will work and what they will be capable of. If used well, AI
has the potential to empower VA employees to provide better health
care, faster benefits decisions, and more secure systems. Similar to
other major transitions, such as cloud computing or the rise of
smartphones, VA will need to invest in and adapt our technical
portfolio to take advantage of this shift. With the strategies,
policies, and programs already in place, VA will continue in its
mission to protect the security and privacy of the data entrusted to us
by the Veterans we serve.
Madam Chair, Ranking Member, and Members of the Subcommittee, thank
you for the opportunity to testify before the Subcommittee to discuss
this important topic. My colleagues and I are happy to respond to any
questions that you have.
______
Prepared Statement of Prashant Natarajan
Chairwoman Miller-Meeks, Ranking Member Brownley, and Esteemed
Members of the Subcommittee--thank you for the opportunity to testify
today, on current state and future uses of AI in the VA. I am honored
to appear before you today as an author, a health AI practitioner, and
as a tireless advocate for transformative change in healthcare. My name
is Prashant Natarajan, and I am here to testify on how we can demystify
health data, supercharge the potential of artificial intelligence (AI),
and empower humans.
As lead author or co-author of four books on AI and data-driven
decision making, I have dedicated over twenty years to practicing,
researching, and documenting the complexities of innovation and change
management in the health and life sciences sectors \1\. With the
Nation's leading physicians and health technologists as co-authors and
case study contributors, my books demystify data science and digital
transformation for patients, caregivers, physicians, nurses,
administrators, and policymakers alike.
---------------------------------------------------------------------------
\1\ Demystifying AI for the Enterprise (2021), Demystifying Big
Data and Machine Learning for Healthcare (2017), Multidisciplinary
Approach to Head & Neck Cancer (2017), and Implementing Business
Intelligence in Your Healthcare Organization (2012)
---------------------------------------------------------------------------
I work as Vice President of Strategy and Products at H2O.ai--an
open-source generative AI company, with responsibilities for health
systems, pharmaceutical companies, and public sector health. Before
joining H2O.ai, my professional career included global stints as
product leader and consulting principal at Oracle North America,
Deloitte Consulting Australia, Unum Group, McKesson Health Services,
and Siemens.
I commend you for convening today's pivotal hearing on ``Artificial
Intelligence at VA: Exploring its Current State and Future
Possibilities.'' With Generative AI, we have a ``once in a lifetime''
opportunity to solve long-standing challenges and create
transformational health and economic opportunities for America's
veterans and their families, and the clinicians who serve them.
This cause is deeply personal to me beyond professional expertise
and interests. Since 2016, I have been volunteering as Industry
Advisor, Data Science & AI at San Francisco VA Medical Center (SFVAMC)
and University of California at San Francisco (UCSF), where AI is being
developed to improve the speed and quality of brain imaging as well as
automatically extracting clinically useful information from brain CT
and MRI, especially for veterans with Traumatic Brain Injury (TBI). Our
efforts include the development of deep learning, and more recently,
generative AI technologies such as transformer neural networks and
denoising diffusion models. As a result of this work, we expect AI to
improve diagnostic accuracy and reduce medical errors; drive cost-
effective equipment utilization; and increase physician empowerment.
AI has enhanced our collective knowledge in the enterprise,
expanded commerce, and elevated productivity in the workplace. Health
systems, academic medical centers, life sciences and biotechnology
companies, health and disability insurers, and public sector entities
have already brought hundreds of AI use cases to life.\2\ While there
are diverse AI success stories, our veterans still face inconsistencies
related to healthcare access, knowledge, and care gaps. Based on my
experience with patient-and clinician-focused AI products, generative
AI provides ways to address these gaps and inconsistencies.
---------------------------------------------------------------------------
\2\ Health AI Use Case Catalog: https://health.h2o.ai/h2o-ai-
health-usecase-catalog/full-view.html
---------------------------------------------------------------------------
Generative AI
A new era emerges with Generative AI with the coming
democratization of health knowledge, new innovations in care delivery,
and most importantly, personal health outcomes. This new AI is not
merely artificial or automated, but amplified and augmented
intelligence. It empowers and benefits individual veterans directly and
measurably when designed for shared decision making.
Generative AI is a new and powerful equalizer for patients and
clinicians. It transcends barriers and empowers individuals by
democratizing the language of computing. No longer do we need armies of
technologists, data engineers, and data scientists to accomplish the
generation of actionable insights. Any veteran or clinician - with the
need for answers, access to data, and access to an AI sandbox
environment - can now analyze complex multimodal data (text, images,
videos); build analytics tools using plain language (English, Spanish,
etc.); finetune Large Language Models (LLMs), or design personal
generative AI applications.
The following example is a real-life illustration of how generative
AI can empower users in their regular tasks and daily lives.
Scones AI
Our 11-year-old daughter, Shivani, bakes as a hobby. Previously,
she used her mother and Google for advice and recipes until she heard
about ChatGPT. After we trained her for an hour, we left her to her
devices until she surprised us with delicious savory scones a week
later. These scones were a new creative first for her but with a recipe
where AI was a co-chef and more. Using an LLM-powered chatbot, she
created new recipes, collaborated with her friends on packaging; and is
now in the process of creating her first AI app for other novice
bakers. More importantly, she did this on her own with her new AI tools
simultaneously serving the roles of expert chef, chemistry mentor,
taste tester, and a collaborator who is more helpful than her parents.
How did this achieve this fluency and what did she do with generative
AI?
In short, Shivani used LLMs and chatbot interfaces (ChatGPT and
h2oGPT) to
Ask questions about baking and discover existing scone
recipes, or Prompting
Provide feedback on the AI results and help the AI
improve itself based on her instructions, or RLHF (Reinforced Learning
with Human Feedback)
Use the answers in her subsequent prompts and taught the
AI to play distinct roles (as food critic and content creator), or
Prompt Engineering
Add public PDF documents on nutrition data and macros to
create a custom dataset, and query the combined unstructured data and
documents using RAG (Retrieval Augmented Generation)
Labeling to improve the quality of the labeled results,
or Finetuning
Creating an autonomous agent to refresh buyer
requirements and feedback, or AI Agent Development
Developing the new Scones AI workflow that will allow her
to accept mobile phone text orders from the community, or Generative AI
App Development
Personal Health AI - designed, built, and used by Veterans
If an 11-year-old with no prior knowledge of or exposure to AI
could do this in a few days, imagine the possibilities in front of
veterans, their family members, and clinicians in the VA. If we expand
AI in the VA to meeting each veteran's or clinician's unique
requirements/expectations - beyond the organizational needs that
already exist - we will have enabled veterans to
Better manage daily living - diet, activities,
appointments, prescription refills
Become informed participants in the determination of
health outcomes - and active contributors to the treatment plan
Forge deeper connections with physicians and caregivers
Develop, deploy, and use disease-, environment-, and
task-specific AI assets
Create peer-to-peer best practices
Explore/establish ``on demand'' collaboration spaces,
monetization channels, and entrepreneurship opportunities
Our work in the Stanford-Pfizer Public Led Opportunity Training
(PLOT) program \3\ demonstrates that not only are these goals
achievable but also provides a framework for education, reskilling, and
mentoring for patients to become patient-researchers, prompt engineers,
data analysts, and obtain grant funding. In the last 12 months, we have
trained 14 patients to become informed patient researchers and help
them create AI products that are specific to their disease/s,
demographic background, and other realities of their lives. The result
is Personal Health AI, where the individual creates the AI they need -
as compared to personalized AI where the organization or institution
determines what works best for a broader cohort of people with similar
characteristics.
---------------------------------------------------------------------------
\3\ GMG--2022-HOS-G--SupportingPatientPoweredResearch.pdf
(pfizer.com)
---------------------------------------------------------------------------
Provider Health AI - designed, built, and used by VA Clinicians
The COVID pandemic has created new challenges for physicians and
nurses in the VA and beyond. Post-pandemic, the needs of our veterans
have increased but the health and wellness of those who serve them has
gone into a steep decline. VA clinicians report reduced job
satisfaction, increasing health challenges, precipitous burnout, and
reduced face time with their patients. Despite best intentions,
technology modernization and administrative simplification programs
have delivered suboptimal results for our clinicians - even as we
redesign systems and workflows frequently.
Generative AI - as described above for veterans - can provide
similar benefits for doctors, nurses, and allied health professionals
by allowing them to create scientific AI assets and workplace assist
agents, such as
Systematic review of hip and knee replacement procedures
for orthopedic surgeons
Patient-friendly discharge notes generator
Guidelines-based AI agents for diverse specialties
Smart appointments manager
VA clinicians are looking for solutions that will bring relief to
their work, reduce medical errors, and improve the quality of care.
Even as we recognize and support ongoing efforts by the VA to reduce
clinician burnout, relief for an individual physician can be as simple
as using AI tools to find and enjoy 15 minutes - to relax, decompress,
or smell the roses.
Veterans' AI Collaborative
A Veterans' AI Collaborative is one validated approach to support
both Personal Health AI and Provider Health AI perspectives as outlined
previously.
Connecting veterans, VA clinicians, and veteran groups to data
sources, training programs, and AI resources is the need of the hour.
Bringing these stakeholders together and creating the opportunities for
them to experiment on data and collaborate with each other is the most
optimal way to create sustainable, bottom-up, and cost-effective AI
innovations. Public-private partnerships like the California Initiative
to Advance Precision Medicine (CIAPM) are proven efforts that enabled
collaboration between, and brought verifiable value to, researchers,
physicians, and patients. The recently launched National Artificial
Intelligence Research Resource (NAIRR) pilot is a commendable effort
and can serve as an invaluable foundational resource for our
collaborative.
We must learn from the successes and failures of similar
partnerships & pilots - while keeping in mind the new capabilities
coming from access to generative AI resources and new user types.
Providing access to and training non-traditional users on data science
and AI competition platforms, open-source AI software repositories, and
natural language-based data science experimentation sandboxes will
increase data literacy and improve health outcomes.
Prescriptions for Health AI Success for Veterans and VA
Recognize AI Fidelity: like the concept of data fidelity,
AI Fidelity is about the value of health AI being determined by its
user (veteran, clinician, or administrator) in the context of its use.
AI use cases can come from anywhere--especially beyond the four walls
of any VA facility.
Regulation & Validation Flexibility: administrative,
operational, research and care delivery AI are important and must go
through external AI validation as outlined in President Biden's
Executive Order. However, one size does not fit all. AI created or
managed by veterans - for their personal and peer uses - within the AI
Collaborative must be treated with an appropriately light regulatory
touch. Enforcing any new AI regulations, especially the ones that apply
to healthcare organizations and business entities - to veteran AI
creators and their Personal Health AI, or VA Clinicians and their
Provider Health AI - is counterproductive to bottom-up and user-first
value creation.
Veterans-First AI Ownership: the intellectual property
and monetary rights of personal health AI models, applications, and
agents must either be
with their veteran and/or clinician creators, or
distributed under Apache 2.0 licensing as individual
veteran's or clinician's preference. Private companies
including cloud service providers, AI vendors, or data
providers must be restricted from using data or insights from
the Veteran's AI Collaborative to train any proprietary data/
LLM/AI agents, or extensions. These protections will ensure
that veterans' data/model privacy and economic interests are
reinforced.
Encourage Open Source AI: even as we are just getting
started with Generative AI, there are incipient efforts at regulatory
capture using compute, storage, the number of LLM model parameters, and
exaggerated fears of safety and/or AI omnipresence. Independent of our
opinions and biases, we need personal and provider health AI to have
access to viable open source platforms, so that veterans and VA
clinicians can contribute to them in a trustworthy fashion.
I would like to thank Chairwoman Miller-Meeks and Ranking Member
Brownley for this opportunity to testify today, and all members of the
Subcommittee for prioritizing such a critical issue. The VA has no
greater priority than ensuring that our veterans receive the best
possible care, and this imperative can only be met with AI that
addresses veterans' needs where they receive care and where they live,
work, pray, and play.
______
Prepared Statement of Gary Velasquez
Chair Miller-Meeks, Ranking Member Brownley, and members of the
House Committee on Veterans' Affairs Subcommittee on Health, I
appreciate the opportunity to come and speak this morning on the use of
AI at VA and future applications of these transformative technologies.
I possess advanced technical degrees with over four decades of
experience operating national health plans, large-scale care-integrated
delivery medical centers, and an international clinical research
organization.
I also want to acknowledge the federal government, including VA,
and its initiatives, which lean into the use of AI and ML to improve
the health of Americans. My company had the privilege to participate in
these early stage programs, from identifying the most clinically
vulnerable resulting from COVID-19 infection to predicting
beneficiaries with high clinical risk due to deferred care and untoward
events of VA ICU patients.
However, before I begin my testimony, I believe we must use a
standard definition of Artificial Intelligence (AI) compared to Machine
Learning (ML); while closely related, they differ in many ways.
AI is a broad field that uses technologies to build systems that
mimic cognitive functions associated with human intelligence, such as
seeing, hearing, understanding, and responding to spoken or written
language or visual cues, analyzing data, and making recommendations or
taking action. AI is a machine or system that senses, reasons, acts, or
adapts like a human.
ML extracts knowledge from data and learns from it autonomously. ML
leverages algorithms to analyze enormous amounts of data, learn from
insights, and make informed predictions, analyses, or recommendations.
Machine learning algorithms improve performance over time as they are
trained--and exposed to larger, diverse data sets. Generally speaking,
the more varied data used, the better the model will get.
Today, I speak before the committee with two voices as the CEO of
Cogitativo, a Berkeley CA based artificial intelligence company, and
with a second voice as the son of a retired Master Sergeant, a Korean
War combat Veteran, who was awarded two Silver Star medals at age 17
when serving in 1st Ranger Company, 2nd Infantry Division.
Over nine years ago, I co-founded Cogitativo with a single purpose
to advance the use of AI to serve as a beacon to identify our most
vulnerable individuals and their families while enabling the delivery
of effective personalized care - our initial mantra was and will always
be ``making the unseen, seen.''
An excellent example of our ethos is our work during COVID. On
March 7, 2020, my co-founder and I wanted to help the Country by using
AI to minimize the pain, suffering, and loss of life from COVID-19.
Based on lived experiences dealing with SARS, we could foresee the
scale of devastation from this virus.
We quickly determined that several universities had built strong
predictive positivity models that track the movement of the virus
through our communities. At the same time, the federal government was
predicting mortality rates. We decided to select a unique endpoint to
predict--what if we could accurately predict which individuals would
have the highest risk of being admitted to the ICU post-infection of
the virus? We believe that predictive endpoint would enable government
agencies, healthcare organizations, and other community organizations
to encourage the most vulnerable to stay sheltered in place.
Fortunately, we found two large California healthcare payors who
sponsored our AI COVID work to develop and deploy this model within
their organizations. Their support and efforts allowed us to validate
our model while enabling these payors to bring food and medications to
their most vulnerable members.
These efforts led us to Operation Warp Speed, where, in November
2020, we received a contract through the Department of Health and Human
Services to use this ML model to score over 200 million Americans for
the probability of ICU admission resulting from infection. The outputs
from this work were used to develop distribution plans for the initial
vaccine shipments.
However, being raised by a Ranger where ``end results'' are
measured, we knew we had to get jabs in arms. So, we collaborated with
several religious organizations and Drew University to establish
vaccination sites at local parks in South Central Los Angeles. We
vaccinated over 2,500 individuals over four weekends.
Today, Cogitativo's AI/ML capabilities have been deployed in the
VA, HHS, and private sector clients such as Kaiser Permanente, Blue
Cross Blue Shield plans, Cigna, and Molina Health. We offer a unique
fusion of nationally recognized healthcare operators, complex systems
researchers, and world-class data scientists who address some of our
most complex healthcare challenges. Our projects within the VA include
predicting disease progression, identifying the most clinically
vulnerable, and predicting clinical deterioration for ICU patients.
Why VA and Cogitativo?
As I previously mentioned, my father was a Korean veteran with
combat-related injuries. However, he did not use VA for most of his
medical care - like many other Veterans, my father believed that other
Veterans needed these precious resources more than he did--he did not
want to ``take'' from other Vets.
However, he dealt with PTS for over 60 years, for which he did use
VA for treatment----fortunately, his mental health counsel would gently
nudge my dad to get an annual physical from a VA provider. This nudge
saved his life!
Unbeknownst to our family, VA had been using these visits to log
his biomarkers (lab values) into VistA for over a decade, creating a
detailed temporal continuity of care view of his health status.
Seven years ago, my dad was admitted to a private sector ICU with
severe pneumonia, including an extensive volume of fluids in his
lungs--standard treatments were not working.
The ICU physician was about to order Lasix with an angiotensin
inhibitor. As we were awaiting the preparation of treatment, my dad's
cell phone rang. It was a San Diego VA patient advocate calling for his
annual appointment. I told her what was happening, and she took the
initiative to find his Primary Care doctor immediately, who viewed his
medical chart and identified a negative GFR ``trend'' line even though
he had not been diagnosed with chronic kidney disease. The VA clinician
asked to speak to the ICU physician and warned her that the
administration of the proposed treatment could irreparably damage my
father's kidneys.
While this is not a true example of ``machine learning,'' it shows
the value of human (or machine) learning and analyzing temporal data,
incorporating previous knowledge, and then making an informed decision-
this is the foundation of machine learning.
In our family's case, we were divinely lucky that VA called at the
moment of need, but we should not have to rely on luck; given the
current state of technology, VA can effectively and safely deploy ML/Al
solutions that serve the mission of the best care anywhere.
VA AI success
I have witnessed VA's commitment to advancing healthcare through ML
and AI, which is evident in its proactive approach to research
initiatives and the exploration of diverse advanced analytical
techniques. VA has invested in groundbreaking research endeavors,
ranging from predictive analytics for personalized treatment plans to
integrating AI in medical imaging, significantly improving diagnostic
capabilities. This commitment to innovation extends to the nation-
leading expansion of virtual and augmented reality throughout the VA
network, bringing a state-of-the-art approach to a variety of use
cases.
Furthermore, VA's partnership with Cogitativo on deferred care and
telecritcal care advanced analytics underscores a commitment to
advancing healthcare through Machine Learning. These ML algorithms can
predict patient deterioration across various conditions, including
prevalent Chronic and ICU clinical endpoints, enabling early
intervention and more effective clinical resource use.
I applaud VA's efforts and early successes in exploring the use of
AI in healthcare delivery and administrative functions. However, there
is an extensive greenfield of use cases that could immediately benefit
VA and its beneficiaries. Some of these use cases include targeting ML/
AI in processing disability claims with higher accuracy and speed,
accelerating the diagnosis of diseases, revealing underserved Veterans,
and reducing provider administrative tasks.
Now, I would like to turn to more immediate opportunities for the
use of AI at VA.
VA AI opportunities
VA has an immense opportunity to make substantial advancements in
using advanced analytics. Immediate opportunities include the
deployment of VA-proven, validated, and human-in-the-loop supported
solutions for enhancing national access, availability, and outcomes
while improving effectiveness.
For example, VA can apply ML/AI to the following challenges:
1. Identifying the most clinically vulnerable from deferred
care induced by changed behaviors resulting from the COVID-19
pandemic.
2. Solving the escalating challenges of capacity and prolonged
wait times with the direct and community delivery systems.
3. Uncovering health risks resulting from toxic exposures
4. Understanding, preventing, and providing comprehensive
support to Veterans at risk of suicide.
Deferred Care: The issue of deferred care has become increasingly
prevalent throughout all healthcare delivery sectors, with disruptions
caused by the pandemic leading to delayed or postponed healthcare
treatments. In this context, ML emerges as a powerful ally, capable of
proactively identifying beneficiary-level clinical vulnerabilities and
intervening to avoid adverse outcomes resulting from deferred care.
Through VA support, my company tested and refined four chronic
condition-specific algorithms that scored all 8 M+ beneficiaries for
clinical risk resulting from deferred care. The central office and two
VISNs have validated these outputs. We are currently in dialog with
several VISNs regarding the operational deployment of this capability.
Further, an ML-driven approach to combat this surge in service
demand could focus on a proactive approach, allowing the VA to identify
vulnerable patients early, enabling more efficient use of available
clinical resources, and a significant opportunity to reduce community
care costly in-patient, ER admissions as well as lowering cognitive
demands on provider practices.
Capacity and Resource Management: Addressing capacity planning and
access challenges within VA requires an approach that builds on these
techniques. Any solutions must use advanced ML and AI methods to
identify at-risk individuals, clearly account for current state service
demands, and predict future demands with specific needs across
specialties and geographies. The goal is to align medical staff levels
with beneficiary care needs, optimizing wait times across the direct
and community care networks while decreasing costly acute healthcare
costs. Furthermore, predicting future patient demand across regions and
specialties helps mitigate the potential cost overrun from referring
beneficiaries from VA care to community care. With a massive workforce
of over 450,000, we advocate that AI is central to addressing this
complex, dynamic challenge.
Toxic Exposures: We recognize the pressing concern of adverse
health effects from toxic exposure among our Veterans and active
military personnel. Despite the successes of ML in predictive
toxicology, there is a significant gap in understanding, predicting,
and managing the health impacts of toxic exposures.
The PACT Act is a transformative enabler representing the largest
benefits expansion for Veterans in a generation. While VA has
necessarily focused on the health care and benefit needs of Veterans
who are ill today, we submit that ML/AI can be a powerful tool in
helping to identify veterans at risk of longer-term or latent
manifestations of various exposures.
Genetic polymorphisms play a pivotal role in influencing health
outcomes post-toxic exposures, as evidenced by conditions such as Gulf
War Illness (GWI). An ML/AI-driven analysis allows us to analyze large-
scale datasets from projects like the Million Veteran Program (MVP) and
VA Corporate Data Warehouse (CDW), providing invaluable insights into
these factors' intricate, presently unseen relationships. By leveraging
machine learning techniques to unravel the complex interplays between
genetic polymorphisms and chronic illnesses resulting from toxic
exposures, VA can enhance its understanding of how these factors
influence health outcomes and, consequently, enable timely, earlier
diagnosis and treatment.
Suicidal Tendencies: Addressing the prevalent issue of suicide with
the VA beneficiary population demands a comprehensive approach. As
reported by VA, 6,392 Veterans died by suicide in 2021-an increase of
114 from 2020. We applaud all VA efforts in this area; however, we must
continue to bring new approaches and tools to prevent suicides. We
assert that we can rapidly bring a new capability to address this
national crisis by employing AI. Through AI, we can capture and curate
clinical, audio, and visual data to predict an individual's risk of
suicide. The VA's unique position, with access to extensive datasets
and robust systems, positions them at the forefront of research and
design of targeted suicide prevention strategies.
Other areas of ML and AI demonstrate exceptional potential in
various critical healthcare domains, including critical care and
telecritical care, remote patient monitoring, opioid use disorder,
mental health, and operational domains such as claims processing and
medical coding.
In critical care scenarios, AI/ML algorithms can analyze thousands
of patient data, from vital signs to lab results, to swiftly identify
deteriorating conditions and prompt timely interventions. Remote
patient monitoring, facilitated by AI, allows continuous tracking of
patient health metrics, enabling early detection of subtle changes and
reducing the need for hospital admissions. In the realm of opioid use
disorder, ML algorithms can analyze gut biomes to predict the
predisposition of addiction, thereby enabling the use of new, less
addictive therapies.
These AI/ML-powered solutions promise to improve patient outcomes,
optimize resource allocation, and improve stewardship of our precious
healthcare resources.
Today, we stand on the brink of transformative possibilities with
the potential to empower beneficiaries, reduce stress on providers,
improve patient outcomes, and deliver genuinely precise healthcare. We
must swiftly embrace innovation and harness AI's capabilities to uplift
our providers, streamline processes, and ensure every Veteran receives
unparalleled care.
How can Congress help?
Improving human health through innovation is not inevitable, nor is
it dependent on divine intervention, as in my father's case - Improving
human health comes through a continuous struggle of use-case ideation,
disciplined experimentation, validation, and thoughtful scaling.
As the committee is aware, several clinical research and
development studies already suggest that AI can perform as well as or
better than humans, such as diagnosing disease. Today, algorithms
outperform radiologists at spotting malignant tumors and guiding drug
researchers in constructing cohorts for costly clinical trials.
AI has clear transformative potential, but its success goes beyond
technology. Unlocking progress requires a deep understanding of the
clinical domain and healthcare delivery. Trustworthy and ethical AI
solutions necessitate integrating human-in-the loop clinical expertise
and the dynamic nature of medical decision-making. VA should combine
these novel technologies with deep domain expertise, world-class data
scientists, and hands-on workflow experience that targets impactful use
cases.
With the support of Congress, I believe VA can be a national
cornerstone in delivering AI--enhanced services that improve human
health while deploying AI workflow tools that enhance the efficacy of a
provider's daily administrative practices and clinical interventions.
Given VA's unique mission, operations, and rich data repositories, few
other organizations can deliver on this objective better. I am
confident that AI will provide essential capabilities for improving
human health and that the VA can be central in delivering these
capabilities.
I look forward to discussing with the committee the opportunities
to deploy AI/ML in safe, appropriate ways that benefit the health and
life of our Country's most precious heroes, our Veterans. These remarks
conclude my statement, and I would be pleased to answer any questions
you or the Committee members may have. Thank you
______
Prepared Statement of Charles Rockefeller
Good morning Ladies and Gentlemen.
My name is Charles Rockefeller, and I am the Co-Founder and Head of
Partnerships for CuraPatient. It is a real honor to have been included
today in this very important discussion. By coincidence, I happen to
feel more historically connected to the VA because I heard it being
discussed at the dinner table since age 12. My father sat on the Senate
VA committee for 30 years, either as a member or its Chairman. My other
two co-founders are Long Nguyen, who has been supporting the US
Government in its AI endeavors since its inception 20 years ago, and
Dr. Siddhartha Mukherjee, a Pulitzer-Prize-winning oncologist who has
long been a thought leader in healthcare. Together, we bring a well-
rounded and unique perspective to the very real challenges healthcare
professionals face every day. To help us realize our vision, we have
assembled a team of veterans, scientists, artists, technologists, and
healthcare professionals from all walks of life.
To the extent that it's possible, I'd like to express the level to
which I appreciate how important, difficult, and Herculean this
committee's work is. Even the very title of this hearing: ``Artificial
Intelligence at the VA: Exploring its Current State and Future
Possibilities,'' accurately captures the nature of the field--the fact
that it is a dynamic and developing technology. Even more important are
the policy guidelines that the VA has implemented. These guard rails
can ensure that the VA's AI follows the Executive Orders on the use of
AI. In the following testimony, I will contribute to this topic from
the perspective of an emerging high-tech company and from the
perspective of spending the past five years in working with the VA to
enable this vision.
To start, I will provide some background on our work--the primary
impetus behind my co-founders' and my decision to start the company was
that the WHO had just declared worker burnout an occupational
phenomenon--in its 11th Revision of the International Classification of
Diseases. This has been particularly evident in the healthcare field,
where we have seen the challenges and tremendous burdens placed on our
front-line workers, ultimately compromising their mission and support
for their community. While this has been an age-old challenge, we are
excited by how the use of AI in this space can dramatically impact the
workforce and improve patient engagement and support.
Some information about our technology and our platform:
Our platform's features mostly fall into three categories and have
been designed specifically to be able to support patients, providers,
and administrators to deliver care efficiently and avoid it when not
needed at healthcare units and beyond the traditional walls:
1. Assisting patients with a virtual patient companion to guide
them through their care plan and help them recover beyond the
walls of traditional healthcare settings.
2. Empowering providers to be able to do their jobs and
understand the unique needs of the organizations while being
able to focus on an individual patient with the relevant
information at their fingertips
3. Enabling administrators to set up, plan, and schedule
resourcing and enable programs to be implemented consistently
and at scale
These features extend patient care beyond the hospital's four
walls, creating seamless support for veterans. With recent advancements
in AI, we have overcome some of the traditional hurdles of digital
healthcare where it overwhelms our providers and staff--we are
deploying it to help summarize the enormous volume of data into
digestible formats and nudge patients and providers without
overwhelming them. Again, this is directed toward the dual, hand-in-
hand goals of increasing veterans' access while reducing staff burnout.
One of our first successes came while working with Operation Warp
Speed, where we helped provide equitable access to critical care while
also allowing our brave front-line workers relief to focus on the job
at hand. Because everything was automated, the workers could go home
exactly when their shift ended to be better rested for the next day.
For the sites that didn't use our platform, the workers spent, on
average, an hour and 43 minutes manually entering data into
spreadsheets. Ours was uploaded immediately into the state health
registry system. I'm proud to say that we received a Red Cross Heroes
Award for this service.
With that as a foundation, I want to shift my focus to our work
with the VA. It goes without saying that our veterans are beloved
around the country--we also think that the VA itself should be a
beloved entity. CuraPatient was first introduced to the VA NAII in 2019
when we won that year's Tech Sprint. I'm proud to say that we were
deemed ``The Future of Healthcare.'' Today, I would like to highlight
five key topics from our experience with the VA, although there are
many more, and each creates the foundation not just to innovate but do
so responsibly and at scale:
1. Data Privacy and Security: We cannot discuss the VA and our work
with them without discussing privacy and data security. The focus and
emphasis here from the VA have been nothing short of amazing--even in
the face of tremendous pressure to rush milestones, the steady hand and
continued discipline to ensure patient safety and privacy are
admirable. Together, we have implemented 421 NIST security control--the
highest standard in the industry, with independent 3rd party assessors
and ongoing continuous monitoring. We now have a fully operational
national High Impact Authorization to Operate (ATO) with a native
patient app connected to wearables, a suite of machine learning tools,
and bi-directional integrations into the six core VA systems--more on
later. It has been a cross-functional effort with Charles Worthington,
Angela Gant-Curtis, and their teams to move this down the field. Dr.
Paul Tibbits and his team at OIT first helped us get started and
navigated us in the right direction.
2. Seamless Integrated Veteran-Centric Experience: Our work is
centered on creating a seamless and user-friendly experience for both
veterans and VA staff, streamlining everything from branding to single
sign-on for hassle-free data management. We're thrilled to report that
we've successfully completed 5 out of our 6 targeted integrations,
granting us the bi-directional ability to both read and write patient
records, thus ensuring our technology is perfectly in sync with VA
operations nationwide. These enhancements not only foster greater
engagement between visits but also ensure that clinicians have the
relevant information they need for every patient encounter, optimizing
the flow of information. This progress significantly enhances the VA
ecosystem by adding intuitive, easy-to-use features that improve
efficiency without increasing the workload, demonstrating our
commitment to advancing technology within the VA.
3. Clinical Application of AI: Our collaboration with the VA
facilities in Long Beach and DC has been a cornerstone of our efforts,
where established AI oversight committees and policies are already
enhancing our work. Our technology's integration aims to extend veteran
care beyond hospital boundaries, starting with addressing Long Covid.
This condition, with its broad impact on the body, provides a unique
opportunity for wide-ranging engagement using our solutions. Moreover,
our technology is aptly designed to tackle various chronic conditions,
and we're paving a path toward addressing cardiovascular diseases,
diabetes, behavioral health, and cancer. The abundance of well-curated
content and literature tailored for veterans has been particularly
impressive, simplifying our task. It allows us to leverage our
technology to maximize the benefits derived from these veteran-specific
resources.
4. Responsible AI: These pilots will be deployed at the 4 NAII
centers and will be available across the entire VA. The Long Beach VAMC
and DC VAMC teams led the work on it. It enforces compliance with
trustworthy principles as defined by EO 13960. It incorporates NIST AI
RMF and all non-binding principles within the White House AI Bill of
Rights. The team has stated that the AI system we created, CuraPatient,
shall only move forward with the full approval of these bodies. They
will prove that our technology can scale while emphasizing the
importance of engagement and interaction--the more it is used, the
smarter it becomes. As I mentioned earlier, AI is the most profound
technology that has come to bear in the last 25 years and is the
culmination of generations of scientists, mathematicians, etc. It also
represents the greatest opportunity to address the challenges of
burnout and access.
5. Contracting: We're optimistic about the benefits of enhancing
our contracting approach, which promises to be a positive change. As
technology, especially AI, advances rapidly, navigating the
complexities of traditional contracting becomes a growing challenge.
Often, by the time Firm Fixed Price contracts are executed, the
technology has already advanced significantly. To stay aligned with
these fast-paced technological changes and avoid administrative delays,
it's vital to consider alternative contracting methods. Such strategies
will keep the VA at the cutting edge, ensuring we deliver responsible
and effective solutions. Moreover, gaining Congressional support for
the necessary funding is critical to transforming these opportunities
into real benefits for our Veterans. Inspired by the adaptable nature
of AI, we aim to make our contracting processes equally flexible and
responsive.
These 5 topics summarize our work and experience with the VA. In
reflection of the past 4 years, we have a true appreciation of what it
takes to be in the role of the VA leadership. The ability to be
steadfast in the mission, while adapting and innovating to drive new
technologies into the echo system. More importantly, it has resulted in
a soon-to-be mission ready system that can greatly apply advancements
in AI not in theory only, but directly to our veterans and the staff
that supports them in their journeys. While change may not always come
quickly, the breadth of our impact is undeniable, reaching millions of
veterans and staff. Dr. Clancy's leadership has been a guiding light,
and we're energized by our current state and look forward to getting
our hard work out into veteran's hands across the country.
In closing, I urge the committee to recognize the critical
importance of directing additional funding toward operationalizing AI
and turning these groundbreaking ideas into tangible actions. Such
investment will not only reinforce the VA's pivotal role in advancing
innovative technology. Still, it will also significantly enhance the
care and services provided to our veterans, benefiting our nation as a
whole. It's important, however, that we continue to approach this with
a mindset geared toward responsible implementation and scaling. Our
experiences underscore the VA leadership's commitment to thoughtful
action, having set a strong foundation that enables us to pursue our
goals effectively and on a broad scale, thus ensuring a widespread
positive impact.
The very fact that we are all, in a small way, helping to carry out
the words of Abraham Lincoln on the plaque outside the entrance of VA
HQ at 811 Vermont Avenue is humbling to us.
Along the way we have witnessed some of most effective aspects of
the inner workings of the VA as well as some of the less effective
ones--some of the examples we have seen are:
-Bilateral partnerships between its divisions have been very
effective.
-The efforts that have been put in place will be a big force
multiplier (across medical conditions and diseases.)
-We do think that some of the administrative items and the
various silos to be navigated have been too burdensome but we
always work through it.
-We have made cuts in the past to accommodate for these
periods, but I am concerned that the VA could be depriving
itself of new input and opportunities from other private
companies who simply can't wait as long as we did. There was a
time or two when we even wondered if we could keep it up. Too
many others would just throw up their hands and walk away, and
the VA could have lost good talent and ideas.
-When industry falls in line, others will follow and want to
work with the VA--so that their work can be done in the right
way. This would allow us to continue innovating and working on
the best product instead of re-doing a lot of applications.
-The role that public/private partnerships play is utterly
important and should be facilitated as much as possible. I have
seen from other hearings that the VA is very open to new ideas,
and even the occasional critical remarks. We at CuraPatient
have also taken some criticism, but the important thing is that
we are both in the mindset that we want to improve so we can
serve our veterans as well as possible.
-And within the VA, a renewed focus should be on making the
process faster to get from the pilot stage to operational
stage. We actually see that as an important part of our role
and hope that the work we have done will be implemented going
forward. Part of my job today as a witness is to emphasize that
point to you and make sure you have the very latest commentary
from the field. To simplify and clarify everything we do, we
always put it through the lens of ``How do we get this into
more people's hands?''
-An idea to consider: There should be someone assigned as a
central project coordinator or watcher/observer to make sure
that the various different departments within the VA are all
synced up on the status of applications. At times we felt on
our own or isolated and we didn't know who to call or write to
ask basic things.
-An idea about how our and others' contracts might be more
efficient: For example, the VA leases Microsoft Office. An
aspect of that lease is that it is possible to add/contract
line items along the way. Make fixes and additions along the
way instead of redoing the entire contract.
Ladies and gentlemen, thank you very much for listening today. I
appreciate your time, and, if you wish, please feel free to contact me
going forward. Whatever I can do to further this cause, consider me to
be at your service.
Prepared Statement of David Newman-Toker
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
Statements for the Record
----------
Prepared Statement of Pratik Mukherjee
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
Prepared Statement of Society for Human Resource Management, (SHRM)
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
Prepared Statement of North America Siemens Medical Solutions USA, Inc.
Chairwoman Miller-Meeks, Ranking Member Brownley, and distinguished
Members of the Subcommittee, on behalf of Siemens Healthineers, our
17,000 employees in the U.S., and approximately 71,000 employees in
over 70 countries globally, thank you for the opportunity to provide a
statement for the record in response to the House Committee on
Veterans' Affairs Subcommittee on Health hearing on ``Artificial
Intelligence at VA: Exploring its Current State and Future
Possibilities.''
Siemens Healthineers is a leading medical technology company with
more than 120 years of history and experience bringing breakthrough
innovations to market that enable healthcare professionals to deliver
the best care for patients--from prevention and early detection, to
diagnosis, treatment planning and delivery, and follow-up care. Our
core portfolio includes imaging, diagnostics, comprehensive cancer care
and minimally invasive therapies, augmented by AI. We focus on
addressing the deadliest diseases impacting the United States (U.S.),
including cancer, neurovascular, neurodegenerative, and cardiovascular
diseases. We partner with more than 90 percent of providers in
healthcare and in addition to the medical devices we provide, we also
work to address population growth and chronic disease prevalence,
healthcare workforce shortages and lack of access to care in
underserved areas throughout the U.S., and globally. Given the depth
and diversity of our product portfolio, we have the distinction of
being the only medical technology company in the world capable of end-
to-end cancer care - from diagnosis and screening to treatment and
survivorship. This is a responsibility we take very seriously, and we
keep patients at the center of everything we do.
Our U.S. headquarters is in Malvern, Pennsylvania. Our global
headquarters for diagnostics is in Tarrytown, New York, and we have
laboratory diagnostics manufacturing facilities that serve customers
worldwide in both Walpole, Massachusetts and Glasgow, Delaware. Our
global headquarters for molecular imaging is in Hoffman Estates,
Illinois. Cary, North Carolina is home to our training center, where we
train thousands of engineers annually, including active service
members. Our AI research and development team is housed in Princeton,
New Jersey. Our Varian business is headquartered in Palo Alto,
California. We also have manufacturing, engineering and research and
development sites in Washington, Indiana, Tennessee, Nevada, and
Colorado.
Siemens Healthineers Partnership with the Veterans Affairs
Administration (VA)
Siemens Healthineers is committed to providing outstanding products
and services to veterans through the VA and the Veterans Health
Administration (VHA), the largest integrated health system in the
country. We are a proud participant in the Military Friendly Companies
list. Receiving this award displays our dedication to serving the
military and veteran community by creating sustainable and meaningful
career paths, community outreach, and enduring partnerships. We also
partner with a diverse team of service-disabled veteran-owned small
businesses (SDVOSB) who provide critical services on behalf Siemens
Healthineers to veterans and our military servicemembers.
Siemens Healthineers AI Experience & Algorithm Development
Data, digitalization, and AI to improve patient care is at the core
of the work we do every day, and who we are as a company. Each day, an
estimated five million patients, including veterans, benefit from our
600,000 cutting-edge technologies and services worldwide. Siemens
Healthineers has been working on applying AI into medical technology
for more than 20 years. At our Big Data Office in the U.S., we created
and maintain one of the most powerful supercomputing infrastructures
dedicated to developing algorithms. This infrastructure allows our
research scientists to collect, prepare and organize correct and secure
medical data - including more than 2.1 billion curated images from more
than 200 clinical providers and partners - needed to train and deliver
accurate AI. From its inception, we created and maintain a quality
assurance process, which involves clinical validation to both
understand the treatment outcomes associated with the curated data as
well as guarantee the data being used to train our algorithms is
accurate for diagnosing and treating disease. To ensure we develop
reliable algorithms that are reflective of the patient populations they
will be applied toward, we continually maintain a holistic view of the
patient with high-quality training data. This training data is based on
a balanced cohort of people of different ages, genders, ethnicities,
healthy people, and those who are sick. From the inception of data
collection, we work to build algorithms that are reliable, accurate,
unbiased, and protect the patient.
We take great pride in the work we do to develop reliable AI and
have company-wide guardrails for AI that I have included in an addendum
to this testimony. In addition, we have recently partnered with the
American College of Radiology (ACR) to improve transparency and patient
care through the launch of the Transparent-AI program. We disclose
detailed product information, including training data demographics and
machine specifications, to help radiologists choose tools that meet
their specific patient population needs. ACR's public website includes
comprehensive information on our FDA-cleared AI imaging products.
Partnering with physicians is essential to the adoption of AI, and its
ability to be a powerful clinical tool to drive better patient
outcomes.
Regulation
Our algorithms go through a regulatory approval process with the
Food & Drug Administration (FDA). We follow all AI/Machine Learning
(ML)-enabled medical device regulatory requirements for premarket
review and post-market surveillance to ensure the safety and efficacy
of our devices. We also engage with the FDA regularly on AI/ML and
provide feedback on ways to ensure the continued safe and effective
application of these technologies. In this regard, our AI is distinct
from unregulated AI products.
With the rapid acceleration in development and innovation of AI,
the need for the regulatory environment to be able to balance safety,
effectiveness, as well as update and improve functionality, without
hampering innovation and adoption is critical. While we believe the
current regulatory framework is sufficient to support AI innovation, we
support the continuation of flexibility in the approval process, as a
one-size-fits-all approach could seriously inhibit the potential of AI,
as well as efforts to facilitate global harmonization and the
development of appropriate international consensus standards.
Additionally, Siemens Healthineers recognizes the importance of
continuing to address unintentional potential bias in AI. We feel that
these concerns are currently addressed for applications in medical
devices and mitigated under existing risk management processes, quality
systems, and compliance with regulatory requirements from the FDA and
other regulators.
Algorithm Based Healthcare Services (ABHS)
AI in health care can take two dominant forms - AI for operational
or workflow improvements that help reduce physician burden and improve
patient experience, and AI for clinical services. We refer to clinical
AI as Algorithm Based Healthcare Services (ABHS), which are analytical
services delivered by FDA-cleared devices that use AI, machine learning
or other similarly designed software to produce clinical outputs for
physicians to use in the diagnosis or treatment of disease. They
provide quantitative and qualitative analyses, including new,
additional clinical outputs that detect, analyze, or interpret data to
improve screening, detection, diagnosis, and treatment. ABHS are
developing rapidly and represent an additional service provided to the
patient to deliver the best care possible. These are clinical uses of
AI that have a separate and distinct place within the healthcare AI
conversation.
Siemens Healthineers has over 80 FDA-cleared products on the market
that represent groundbreaking innovations for patients. One of our
cleared products, AI-Rad Companion \1\ is our dominant AI platform that
highlights, characterizes, measures, and reports clinical abnormalities
to aid the clinician in formulating a diagnosis and treatment. This
ABHS supports physician decisions in diagnosing disease based on
imaging scans. We support separate and distinct payment for this new
and innovative health care service to ensure adoption of it to benefit
all patients, including veterans.
---------------------------------------------------------------------------
\1\ General Availability Disclaimer for AI-Rad Companion: AI-Rad
Companion consists of several products that are medical devices in
their own right, and products under development. AI-Rad Companion is
not commercially available in all countries. Its future availability
cannot be ensured.
---------------------------------------------------------------------------
The Patient Journey
The patient journey is at the heart of Siemens Healthineers AI
work. ABHS are already improving care for veterans. Siemens
Healthineers is proud to be part of the VA-PALS program to increase
veteran access to lung cancer screening. According to the VHA, lung
cancer is the second most diagnosed cancer within the veteran
population, with approximately 8,000 veterans diagnosed annually and
approximately 5,000 deaths each year. We work with Phoenix VA Medical
Center, who is providing comprehensive CT lung cancer screening
management to over 1,500 US Veterans, to integrate AI tools, including
ABHS, into their advanced CT lung cancer screening management system.
This includes providing quantitative and qualitative clinical results
generated by Siemens Healthineers AI-Rad Companion Chest CT in the
identification of potential cancerous lung nodules and sharing these
clinical findings with physicians and nurse navigators managing the
veteran. The use of our AI-guided computer software as a companion to
the clinician to identify small nodules and other abnormalities
includes the ability to measure the density and characterize the size
of suspicious nodules that were previously not possible to visualize
without the assistance of ABHS.
Suspicious lung nodules diagnosed to be cancerous by the clinician
can potentially be treated by radiation therapy. To minimize the risk
that healthy tissue around the cancer is not unnecessarily radiated,
radiation physicists create a radiation treatment plan, which includes
the tedious task of manually drawing the unique contours of the
cancerous tumor. This manual contouring potentially delays the time to
treatment for the patient. Our AI-enabled auto-contouring software can
automatically detect these contours of the cancerous area,
significantly speeding up the patient's time to treatment and
potentially eliminating extraneous treatments.
Utilizing AI or ABHS at each point in the process to screen,
diagnose and treat lung cancer can reduce the time to treatment. This
allows for a reduction in patient stress and anxiety, more precise and
faster diagnosis, and more specialized treatment that we believe will
improve patient outcomes.
Another example of the benefit of ABHS is particularly relevant
when discussing prostate cancer. According to the VHA, prostate cancer
is the most prevalent cancer diagnosis (29 percent) among the veteran
patient population. Traditionally, a urologist identifies suspected
areas of prostate cancer by manually reviewing written reports and
pictograms of the prostate provided by radiology and then, as needed,
acquires tissue samples from the areas in question using ultrasound-
guided biopsy. We are developing an algorithm that is planned to be
part of the AI-Rad Companion product family, which will automatically
segment suspect areas of the prostate and characterize and measure
suspicious lesions in the prostate from MRI images. This qualitative
and quantitative analysis may support the urologist's decision on
whether a tissue biopsy is additionally required for diagnosis or if
such invasive procedure can be avoided, which is significant in
managing a prostate cancer patient's well-being and minimizing
unnecessary costs within the health system. This ABHS takes much of the
grey area involved with prostate cancer, particularly when it comes to
active patient monitoring, and provides a health care service through
data that the physician would not otherwise have to allow a more
informed diagnosis and treatment decision. These Siemens Healthineers
AI healthcare services provide clinicians with otherwise unavailable
quantitative and qualitative clinical data that allows them to make a
more informed decision, resulting in better patient outcomes.
The Future of AI in Healthcare
AI has enormous potential to improve access to care, diagnose
disease faster and more precisely, and enable physicians to make
treatment decisions based on comprehensive access to patient data in
real-time. Siemens Healthineers is researching a patient companion tool
to synthesize this data and apply AI to look for patterns and detect
the potential for disease much earlier. In addition, we are working to
create a digital twin of the patient that would allow a physician to
perform an interventional procedure, say for a heart procedure, on a
digital replica of a patient's heart to test how that patient will
react and respond to a specific course of treatment before it is
applied to the individual. The digital twin will minimize unintended
consequences and provide more personalized, precision medicine for the
patient.
We are excited about what the future holds for AI in healthcare and
are committed to continuing our work with the VA as a trusted partner
to ensure veterans have access to health care innovations. As such,
Siemens Healthineers has sponsored and participated in the Department
of Veterans Affairs (VA) National Artificial Intelligence Institute
(NAII) International Summit for AI in Health Care, where Siemens
Healthineers scientists and engineers contribute annually as speakers
and panelists in discussions around artificial intelligence and the
impact to veteran care. The most recent event brought together over
1,000 registrants and over 100 speakers across government, industry,
and academia, including remarks from Honorable Denis Richard McDonough,
Secretary, US Department of Veterans Affairs. A scientist from Siemens
Healthineers provided expert insight during a plenary session focused
on the future of AI in medical imaging, and the barriers to research,
development, and translation into clinical practice.
Conclusion
While there are many forms of AI applications in health care to
reduce physician burnout and streamline operational complexities, we
believe the highest value of AI in health care comes in the form of
ABHS, and that this will revolutionize health care services for
patients and veterans. Siemens Healthineers is a market leader in
researching and training AI in medical technologies and welcomes the
opportunity to continue this discussion. It is critical that we all
work together to ensure we create trust with consumers and build
ethical, transparent, and accessible AI in health care to improve
patient outcomes, particularly for our veterans. Again, thank you for
the opportunity to provide a statement for the record in response to
the House Committee on Veterans' Affair Subcommittee on Health hearing
on ``Artificial Intelligence at VA: Exploring its Current State and
Future Possibilities.''
Addendum
We use a set of guardrails to guide the way we develop and
implement AI in healthcare:
We believe that healthcare professionals, backed up by AI
solutions, make a strong team.
Our AI solutions learn from the best: Siemens
Healthineers collaborates with a huge network of world-class
clinicians, where we combine our research and development (R&D)
capabilities with our customers' clinical expertise. The
results of this collaborative process are powerful, clinically
proven AI companions for decision-making that help to provide
better patient care at lower cost. Humans and artificial
intelligence have vastly different abilities. We believe that
the future of medicine lies in combining the strengths of these
capabilities. Such systems will provide healthcare
professionals with tools to meet the rising demand for
diagnostic imaging and actively shape the transformation of
radiology into a data-driven research discipline. Moreover, AI
algorithms are expected to help speed up clinical workflows,
prevent diagnostic errors and reduce missed billing
opportunities, thus enabling sustained productivity increases.
We believe the level of autonomy of AI solutions
needs to be balanced with ethical expectations and human
values.
Societies are currently discussing the extent to
which AI solutions could be a vital part of everyday human
life. Depending on the area of life, society allows and strives
for lower or higher levels of autonomy. In this regard,
healthcare is a special area, as patients benefit from and rely
on the trusted doctor-patient relationship. A high degree of
autonomy of an AI solution substantially impacts this
relationship. In healthcare areas, where the personal and
trusted patient-doctor relationship is key to the success or
course of the treatment, we believe that the autonomy of AI
solutions needs to be well-balanced. Therefore, we develop AI
solutions only for areas where they are ethically acceptable
and beneficial to humankind and society.
We develop AI solutions to support patients' desires for
more personalized medicine.
An increasing choice of personalized therapies is
leading to significantly improved outcomes in oncology, but
personalized medicine is also gaining traction in other
application areas. For physicians, however, it is becoming more
and more challenging to keep abreast of the constantly
expanding treatment options. With our AI solutions, we enable
physicians to make more accurate diagnosis and treatment
choices, based on comprehensive patient data and the ever-
advancing wealth of medical knowledge. With our vision of the
``Health Digital Twin'' as a constantly updated virtual model
of the human body, we strive to develop the next generation of
systems for personalized medicine.
We believe data handling in healthcare needs to focus on
the individual.
We support patients, so they can share their health
data safely and securely with physicians in health systems. Our
e-health solution creates a decentralized electronic health
record that enables patients to make their longitudinal health
data accessible to physicians. The patient is in control and
decides who to share their data with. We promote the vision of
a ``Health Digital Twin'' in healthcare, which models and
represents a human body based on a multitude of datasets like
body composition and vital parameters. For both patients and
healthy people, their digital twin will help physicians to
diagnose complex systemic diseases earlier and find the best
treatment available for the patient's given condition.
We strive to develop AI solutions for both healthy people
and sick people.
Our current portfolio focuses on diagnosing and
treating patients. Yet, we believe that stewardship for a
patient starts with prevention, and the predictive power of AI
offers a wealth of opportunities for us to help people stay
healthy. In the future, we want to extend our portfolio to
support health systems in their transformation from caring for
the sick to proactively caring for the well.
We work passionately to make AI solutions accessible to
patients everywhere.
At Siemens Healthineers, we believe that every human
being has the right to access high-quality healthcare,
regardless of location, age, and social circumstances (in line
with Art. 27 (1) Declaration of Human Rights ``right to
progress''). Thus, we support the United Nations' 3rd
Sustainable Development Goal (SDG), which ensures healthy lives
and promotes well-being for all at all ages. By providing
powerful AI solutions, we contribute to better and more
personalized healthcare that is accessible around the globe.
We believe AI development needs to be transparent.
We openly communicate insights into underlying
technology, training/test datasets, and quality assurance for
our AI solutions. We carefully compile training and test
datasets which we document to allow traceability and
transparency. Specifically, we strive to free our data from
bias and prejudice to enable equal treatment for all people.
We measure ourselves against the highest scientific
standards.
We aim to improve clinical outcomes with state-of-
the-art technologies. We do not fuel technological hype;
instead, we invest in science to improve technology and
establish new standards. Our world-class scientists therefore
critically evaluate and thoroughly assess our AI solutions with
carefully designed evaluation studies for the respective target
populations.
We speak honestly about the capabilities of our AI
solutions.
We are aware of the capabilities and limitations of
our AI solutions and share these insights with our customers
and users in order to promote the setting of realistic
expectations. Expectations of any technical system need to be
realistic to prevent false hopes, misunderstandings, and errors
in judgment. Healthcare professionals need to be aware of the
capabilities of an AI solution, so that they can make an
informed decision in line with applicable best practices and
guidelines and advise patients accordingly.
Data Privacy--we believe that to fully realize the potential of digital
transformation, people need maximum confidence in the processes,
institutions, and technologies used.
At Siemens Healthineers, our data vision is, ``we use data
responsibly to develop innovations in healthcare to help people live
healthier and longer lives.'' This vision has given rise to a set of
data principles that guide our handling of very sensitive health data
and the development of today's and tomorrow's digital health solutions:
We use data for the benefit of the individual.
The purpose of our company is to advance human
health. People should benefit from data-driven medical
innovations through the prevention of sickness and best-in-
class procedures and treatment. We invest in data-driven health
solutions because we support the patient's desire for
personalized high-precision medicine to live a healthier and
longer life.
We use data to drive healthcare innovation.
Data will become the key enabler for innovations in
digital healthcare. Data-driven innovations are essential for
medical research and progress. Our tailored and responsible use
of data enables us to fill our innovation pipeline, push data-
driven medicine and develop innovative procedures for patients.
We are trustworthy and ethical in our handling of data.
We only use data in a purpose-bound manner to develop
medical innovations and to enable our data-driven products to
perform according to their specified performance capabilities.
We treat data responsibly, reliably, and securely.
We apply proven and high data privacy standards
worldwide.
We believe that trust and accountability are basic
pillars for responsible data privacy management. Consequently,
we apply high data privacy standards worldwide. Fundamental
legal principles of the GDPR - including the legitimacy and
lawfulness of data processing, purpose limitation, the need-to-
know principle, data avoidance and data economy - are mandatory
for Siemens Healthineers worldwide based on internal
directives. In addition, we apply proven technical standards
and organizational measures to ensure data security,
authenticity, and confidentiality. Our ISO-certified
cybersecurity management system follows a holistic approach and
integrates information security management (ISO 27001) and
privacy information management (ISO 27701).
We support the advancements that enable individuals to
have sovereignty and transparency over their data.
Every person should have sovereignty over their own
health data. This includes transparency on what data is used on
what basis and for what purposes, and the right to grant or
revoke consent to the use of one's own data. This right should
also include the freedom to donate one's personal data for the
purpose of conducting research, advancing progress, and
improving healthcare solutions. The processing of health data
in private-sector research and development work also
contributes significantly to advancing medical and technical
progress. To safeguard this valuable contribution, we believe
that private-sector research is also subject to the privilege
of research, and that the development of medical devices or
artificial intelligence that facilitate(s) improvements in the
early detection or treatment of illnesses, for instance, also
serves the public interest and public health. We promote trust
throughout society and among all patients for the application
of digital technologies and support the exercising of their
rights accordingly.
We leverage data as a strategic asset.
Driving digitalization and promoting value creation
from data are essential to advancing medical progress and
providing efficient, high-quality healthcare. Leveraging this
potential of data is strategically important to us. Besides
developing data-and software-driven solutions for supporting
decision-making, we continuously pursue efforts to further
develop our portfolio by automating devices and workflows and
expanding our use of predictive maintenance. The
interoperability and connectivity of our products and solutions
accelerates this development into a platform-oriented business.
We use state-of-the-art technology to protect data.
We offer a state-of-the-art portfolio of secure
products, cybersecurity services and consulting that helps to
ensure optimum protection. We continuously improve our systems
and processes and train our teams in aspects of cybersecurity
and data protection to maintain a consistently high level of
threat awareness is. Our engineering practices include a secure
development lifecycle (SDL) to ensure that high cybersecurity
standards are implemented for every product and solution.
Examples of our core development principles are the
implementation of privacy by design and privacy by default.
We support open standards for data interoperability.
The key to data-driven healthcare innovations is the
ability to interconnect various health datasets. It is only
through data integration and data interoperability that the
value of data can be fully utilized. We strongly support the
standardization of healthcare data and data sharing. When
designing our solutions, we aim to systematically include
standardized interfaces such as DICOM5, FHIR6, and increasingly
uniform APIs7.
We invest in trustful partnerships to access data.
Efforts to improve medical knowledge and to advance
data-driven healthcare solutions depend on having rights to
access health data from diverse, genuine sources. We believe
that providing fair access to relevant data by all healthcare
stakeholders and using this data responsibly to our mutual
benefit will contribute to advancing medical progress. We
therefore build our data-related partnerships on fairness and
transparency.
Prepared Statement of Johnson & Johnson
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
[all]