▶ 0:11:54The subcommittee will come to order. The chair now recognizes himself for five minutes for an opening statement. Today's hearing gives us the opportunity to continue the energy and commerce committee's leadership on artificial intelligence, or ai, by examining current applications of ai across the healthcare sector. Last congress, the subcommittee held a similar hearing on ai and machine learning.
▶ 0:12:16It is critical that we continue these types of educational hearings to understand the evolving health ai landscape and ensure that congress keeps up with the many advances in this space. Applications of ai and machine learning have increased across the health care sector in recent years, and will only play a more pronounced role in the daily lives of all americans moving forward in the healthcare space today.
▶ 0:12:38Ai is being deployed by innovators to empower patients along their personal healthcare journey, support healthcare providers, and reduce unnecessary administrative burdens. I look forward to learning more about these real world applications from our panel of experts today. I also believe as ai applications advance, it is critical that congress continues to examine this landscape to ensure proper safety and proper oversight.
▶ 0:13:05These ai applications can be hugely beneficial to patients and providers, but they are to assist. They are to assist in not to replace the clinical workforce today. I want to briefly highlight a few examples of how ai is being used to improve patient experiences and outcomes in the market today. Pharmaceutical companies are using ai to help improve core scientific research functions and develop life saving treatments and cures, as well as using ai to expedite clinical trials to bring safe and effective medicines to market quicker.
▶ 0:13:37Insurance companies are using ai to process claims in order to get care to patients quicker, but this is an area where oversight is needed to make sure that the ai is not being used in an inappropriate way. Uh, and we've seen that on occasion. So we've got to be careful. Physicians in hospitals who've been dealing with documentation burdens are using ai to assist in writing up and consolidating post-visit records, which has helped reduce documentation time by roughly one third in some cases and allowed for doctors to spend more time.
▶ 0:14:06That's right, they get to spend more time with their patients. What a novel idea. Companies who develop medical devices are using machine learning to better understand certain diseases and help advance innovations to deliver more clinically appropriate and effective care interventions. The trump administration has also been forward leaning on advancing ai in the health space and streamlining regulations to increase its application.
▶ 0:14:33I applaud the work of the current administration to incorporate ai in a responsible manner that can help improve care. To date, the cms innovation center is working to utilize ai and machine learning to identify waste, fraud, and abuse in federal healthcare systems and to root out improper taxpayer spending. The fda has incorporated the use of ai to drastically shorten the time needed to complete some tasks in their review process.
▶ 0:15:00Researchers at nih have developed an ai algorithm that modernizes the process of matching potential clinical trial volunteers to suitable trials, cutting down administrative time by 40% while maintaining the same level of placement accuracy. And one would hope that would even get to be a better placement accuracy. These are only a few examples of the many ways the administration is integrating ai, and streamlining the way our american healthcare system operates.
▶ 0:15:27With all these innovative advancements being leveraged across the american healthcare ecosystem, it is paramount that we ensure proper oversight is being applied. Because the application of ai and machine learning is only going to increase. We must ensure that these tools continue to empower and not replace the providers that serve our communities across the nation.
▶ 0:15:49These tools should help improve the patient experience and ultimately access to care, particularly in rural areas like the one that I represent in virginia's ninth congressional district. I hope we have a constructive conversation today about the opportunities and the risks that come with ai, and how our committee should be thinking about the role we can play in helping shape the future of ai in healthcare. I'm looking forward to hearing more from our witnesses and the members on this subcommittee on the application of ai in healthcare.
▶ 0:16:22With that, I yield back, and I now recognize the subcommittee's ranking member, my friend diana degette, for her five minutes for an opening statement. >> thank you so much, Mr. chairman, and welcome back. If you judge from this subcommittee schedule, um, you'd think that nothing pressing has happened in the healthcare arena in the last five weeks. And certainly I think it's important, and all of us think it's important that we look at the role of ai in healthcare.
▶ 0:16:48And I hope at some point when we're not in a crisis mode, we can do a thorough investigation, because it's certainly a concern of mine. But my friends, right now, rome is burning. Rome is burning.
▶ 0:17:00More important than this is the peril that the trump administration's policies have put my constituents and the constituents of every single person on this panel in, into all of our constituents are at risk because of the risky and dangerous and unscientifically based principles of rfk jr and and and his and his advisors. Last week, Mr.
▶ 0:17:29Chairman, president trump and secretary kennedy fired susan john rose, the new cdc director who had taken her post just a month before her offense was refusing to rubberstamp secretary kennedy's evidence, free vaccine policies and fire seasoned, respected senior cdc staff as cdc doctors and scientists are removed for following science and telling the truth.
▶ 0:17:56The rank and file people who defend america against disease are no longer going to trust the cdc. And in addition, ever since donald trump became president, an estimated 20,000 people have been fired from hhs critical research into cancer, into pediatric um diseases and many other diseases have been put on hold or shelved.
▶ 0:18:23My constituents, your constituents, and everyday americans across the country are going to suffer. Secretary kennedy's dereliction of duty and incompetence are underscored by over a thousand and growing hhs employees, who signed a letter demanding his resignation today. Mr. chairman, I'd ask unanimous consent to put that letter into the record. >> without objection. So ordered. >> I agree with those thousand professional researchers.
▶ 0:18:54Secretary kennedy must resign immediately. And if he does not, president trump should fire him for doing such damage to the institutions that have, up until now, been the paragons of research in the world and have kept americans healthy and safe. And Mr. chairman, we need to do our jobs.
▶ 0:19:15This subcommittee must bring rfk, junior, doctor and others in to shed light on this administration's unprecedented interference in public health and science. Mr. chairman, I looked at the calendar. I think we have time in this subcommittee next week, and I would urge that we bring this up. Let's use it to get the answers that our constituents deserve.
▶ 0:19:37The democrats on this panel and I have been asking for months for this committee to do its job and provide rigorous oversight of hhs under this administration. We've been asking to move critical legislation that we all agreed on at the end of the last year. A number of extensions of critical programs that are bipartisan.
▶ 0:19:59Elon musk killed those programs with a tweet when we were doing the continuing resolution, and those bills still have not come up to this day. And we've been asking for this committee to rediscover its sense of duty and initiative. Congress, us. We are not a coal branch. Equal branch of government. We're the first branch of government. Article one.
▶ 0:20:24The american people's most direct expression of their will and where they constitution that vast by far the most power up until now. And we're languishing between asking the president for permission to take the smallest action and refusing to ensure that the executive is faithfully executing the laws passed by this congress. Everybody in this room, on both sides of the aisle should be embarrassed about congress's work.
▶ 0:20:50I know I am, and we've only had just one hearing, a budget hearing with a presidential appointee of this administration. During this entire congress, inaction in the face of incompetence and malice that is complicit complicity. And we've got to stop that. We've got to investigate the misdeeds of this administration. We've got to follow the investigations where they lead.
▶ 0:21:17We have to insulate scientific judgment from politics and ideology. And as the health subcommittee of energy and commerce, we must base our decisions on science. Mr. chairman, I hope we can start doing that as of today. As you can tell, this issue is important to the american people and our constituents, and we're not going to stop talking about it until we restore these important protections for our constituents health, I yield back.
▶ 0:21:50>> I got it. Gentlelady yields back. And now recognize the chairman of the full committee, Mr. guthrie, for his five minutes. >> thank you. Thank you, chairman griffith. And thanks to all of our witnesses for being here today. We're here today to continue the discussion surrounding artificial intelligence and specifically the meaningful impacts this innovative technology has had on american healthcare. Uh, the american health care system, as well as ai as a whole.
▶ 0:22:15This committee has been a leading voice in the discussion on how to leverage ai across all of our subcommittees, focused on unleashing innovation, innovation while safeguarding consumers to improve health of all americans while ensuring its use safely and responsibly. I would also like to commend the administration for their leadership on these issues.
▶ 0:22:34We support the administration's national ai action plan that will accelerate ai innovation, build american ai infrastructure in both, and bolster us leadership in ai development and deployment. I've also been encouraged by progress across hhs to maximize these new technologies, and I look forward to continuing work with cms, fda, and everyone else, uh, as that work continues to evolve. Today we're here. Just a few examples of ai technologies are being applied across the health care system.
▶ 0:23:04I want to emphasize that I in no way is intended to overtake the jobs of hard working americans. It is instead an opportunity to enhance the work conducted by humans across our health care system, improving the quality and efficiency of care, reducing time consuming and costly administrative tasks and burdens, enhancing the provider patient relationship, and expediting the discovery of new treatments and cures.
▶ 0:23:27Recent news stories have also highlighted potential safety concerns with improper personal use of ai, and the committee is working toward stronger guardrails to prevent tragic events from occurring in the future, especially in healthcare. It is important to understand where ai technologies are creating significant positive impacts on efficiency and research and development of health products, along with the overall delivery of care.
▶ 0:23:53But not to forget the importance of appropriate safety precautions. The committee remains steadfast in its efforts to improve the use of ai technologies, while ensuring ai is used in a human centric, responsible and safe manner. I thank all of our witnesses for being here today and for your participation in today's hearing, and I look forward to how we can all work together to have the most optimal use of ai in the healthcare field. Thank you, Mr. chairman, I yield back. >> gentleman yields back.
▶ 0:24:23Now recognize the ranking member of the full committee, Mr. pallone, for his five minute opening statement. >> thank you so much. Today the committee republicans are holding this hearing on ai technologies and healthcare. But unfortunately, this is happening at the same time that the trump administration and health and human services secretary robert kennedy junior are dismantling the department of health and human services and waging a war on public health.
▶ 0:24:49For nine months now, the trump administration and secretary kennedy have created chaos in our nation's health care system. Last week, it hit a breaking point when the administration fired the director of the centers for disease control and prevention and three key scientific leaders resigned after refusing to sign off on watered down and misleading and unscientific recommendations on vaccines. The trump administration's reckless actions leave the cdc with a critical leadership vacuum.
▶ 0:25:15As we head into the fall and the flu season, coupled with the mass layoffs of over 25% of the workforce, this administration's path of destruction threatens america's health. The cdc must be guided by leaders who govern above politics and remain focused on the agency's core purpose, which is safeguarding and improving the health of our communities, preventing infectious and chronic diseases, and protecting all americans.
▶ 0:25:40I'd like to submit for the record the powerful resignation letter of doctor daskalakis, the former director of the cdc's national center for immunization and respiratory diseases, which details the widespread deception at hhs over changes to the vaccination schedule. Last week's actions followed the trump administration's decision to restrict access to covid 19 vaccines for millions of americans.
▶ 0:26:05This unprecedented action by the fda means that healthy adults under 64 will no longer be able to receive the covid vaccine. Americans will die as a result of secretary kennedy's actions, and that's why it's well past time for this committee to finally begin demanding answers. In april, we called on chairman guthrie to schedule a hearing with secretary kennedy, specifically on his illegal layoffs and reorganization.
▶ 0:26:30Then in june, I called for chair guthrie to join me in conducting a bipartisan investigation of secretary kennedy's actions to restrict access to vaccines. I have also called for this committee to examine the administration's devastating cuts to the nih, and the administration's dangerously inadequate response to the growing measles epidemic. These requests have been met with silence. Republicans would rather sit on the sidelines as kennedy moves ahead with advancing his dangerous pseudo science agenda.
▶ 0:26:57Secretary kennedy must testify before this committee to answer questions and be held accountable for threatening public health by abusing his authority, propagating disinformation, and politicizing science. It's time that republicans take these threats seriously and join us in holding secretary kennedy accountable. Turning to today's hearing, I has the potential, if used appropriately, to provide innovative approaches and efficiencies in health care.
▶ 0:27:25The goal in using this technology should be to provide better direct patient care and health outcomes. But I'm concerned that the expanded use of ai in healthcare has generated significant risks that we simply cannot ignore. Congress must recognize and address the complex ethical, legal, economic, and social concerns raised by using ai in our healthcare system. Without adequate oversight, these new technologies can lead to devastating consequences for patients.
▶ 0:27:50Medical care could be delayed and insurance coverage denied, preventing americans from getting the care they need to live. There's also the threat of personal health information and data privacy breaches. While ai tools can support healthcare providers, their advice and recommendation should not serve as a substitute for the nuanced judgment of healthcare professionals and should not come at the expense of patient safety. And I'm very concerned about the impact of ai on mental health.
▶ 0:28:16The tragic stories of teens interacting with ai chatbots that encourage them to take their own lives is devastating. The speed at which ai technologies have been incorporated into healthcare has so far outpaced the regulation and oversight. That's why it's concerning that trump administration rescinded a 2023 executive order aiming to strengthen ai governance in high risk domains, and issued a separate executive order to remove barriers and revoke policies that they view as hindering ai innovation.
▶ 0:28:45I'm also extremely concerned by the administration's announcement to implement a pilot program in new jersey and five other states that will allow for nonprofit companies to utilize ai to review prior authorization requests for seniors in medicare. I also believe it's critical that safeguards are in place to protect the privacy and security of patient data. I cannot. I cannot function without large quantities of data.
▶ 0:29:09And we must ensure that the increased data which ai relies on does not come at the expense of consumers right to privacy. Again, this is a very important hearing, but we have to think about the consequences of what's going on with ai, which can be positive but often can be negative without proper regulation. Thank you, Mr. chairman. I yield back. >> gentleman yields back. That concludes my opening statements. The chair would like to remind members that pursuant to the committee rules, all members opening statements will be made a part of the record.
▶ 0:29:38We want to thank our witnesses for taking their time to testify before the subcommittee today, although it is not the practice of this subcommittee to swear in witnesses, I would remind our witnesses that knowingly and willfully making materially false statements to the legislative branch is against the law under title 18, section 1001 of the united states code, you will have the opportunity to give an opening statement followed by questions from members. Our witnesses for today are Mr. tj parker, lead investor in general medicine.
▶ 0:30:07Andrew toy, chief executive officer, clover health. Doctor ibrahim, chief clinical officer at. Viz. I if I said this, I guess I'm getting that right.
▶ 0:30:23Doctor michelle mello, us professor of law at stanford law school and professor of health policy at stanford university school of medicine, and doctor c vaile wright, senior director, healthcare innovation, american psychological association. Her committee custom each witness will have the opportunity for a five minute opening statement, followed by a round of questions from members. The light on the timer in front of you will turn from green to yellow when you have one minute left, and I will now recognize Mr.
▶ 0:30:52Tj parker for five minutes to give his opening statement. >> thank you. Chairman. Guthrie. Ranking member. Pallone. Chairman. Griffith, ranking member, degette and distinguished members of the subcommittee. I'm tj parker, a second generation pharmacist and the founder of general medicine, which is a brand new health care store and a partner in matrix, a 50 year old venture capital firm. It's an honor to be here today to discuss the important role ai is playing. In general, medicine's ability to offer price transparency and a better customer experience.
▶ 0:31:21Previously, I was the founder of pillpack, a pharmacy that made it easy for customers to get and take medications as prescribed. I sold that company to amazon in 2018 and spent four years building amazon pharmacy and amazon clinic and putting amazon on the path to expand broadly into healthcare. As an investor, I see myriad new healthcare ai startups.
▶ 0:31:41Most, including some I'm involved in, are building administrative tools, ai scribes, prior authorization, automation, ai phone calls, and other administrative tasks. It's important work that will have a downstream impact on customers. However, today I want to talk about improving the customer experience in healthcare. Americans are frustrated. They want healthcare to be easier.
▶ 0:32:05At general medicine, we're using ai to directly improve the customer experience across all care, from basic needs to complex cases. General medicine is a health care store, a one stop shop for telemedicine, prescriptions, imaging labs, specialists, or anything you need. We've made it as easy to shop for healthcare as it is to shop for anything else. This means a clear choice of providers upfront price, with or without insurance, and a simple way to actually book the care they need right on our platform. We take insurance or customers can pay with cash.
▶ 0:32:36This approach eliminates the opacity and quagmire of complexity customers experience while trying, and often failing to access healthcare today. Our mission is to get every customer the best care for their needs. So far, customers love it, and it wouldn't have been possible without ai. Let me give you two examples of how we've used ai to reinvent the experience. The first is pricing. In healthcare, people rarely know what care will cost. In the end, they can't possibly shop without a clear price. We set out to change that.
▶ 0:33:03Behind the scenes of general medicine, we analyze the customer's insurance information, including the 80 to 100 page coverage of benefits. We use large language models to turn these pdfs into structured data, and combine that with the open pricing files and other sources to give customers a clear upfront price for any service, including procedures, labs, imaging, referrals, and pharmacy at any location. This is the first time comprehensive pricing has actually been available in us healthcare. Another example proactive care plans.
▶ 0:33:33When you go to a doctor today, it's almost always for an isolated reason. You leave with a prescription or one follow up task. It's all our current system is currently set up to handle. Today, general medicine is using ai so that patients after their visit have a comprehensive, personalized, actionable plan covering not just what brought them in, but all their conditions and any overdue preventative care.
▶ 0:33:52With patient permission, we use ai to pull together their medical history, labs and prescriptions, then flag what's missing maybe a colonoscopy, a cholesterol check, or a follow up lab for blood sugar. The doctor reviews it, and the customer sees it all in one place, where they can easily take action on each step via general medicine. It's the equivalent of adding to cart and checking out, truly making healthcare as intelligent and easy to shop as everything else. Looking ahead, we're building the ability for patients to directly request these insights themselves.
▶ 0:34:21People can ask questions back and forth with the ai to clarify the recommendations based on their needs and preferences. Preferences then seamlessly book with a provider to discuss and take action. This innovation allows patients and collaboration with providers to take informed control of their health. At pillpack and amazon, we learned that when customers feel in control and actually enjoy using a service, they are more likely to take the right steps for their health. The same pattern is proven true at general medicine.
▶ 0:34:47In closing, I also want to sincerely thank the members and staffers who have worked diligently over the past decade plus to enable broadly available price transparency files, improve provider directories, and enable data interoperability between providers. Without these efforts, very little of what we do today would be possible. By combining this infrastructure with ai, we can offer a vastly better healthcare experience for every american, regardless of location, insurance, or background.
▶ 0:35:13Thank you all for the opportunity to share the story of general medicine, and I welcome any questions you may have. >> I appreciate your testimony. Thank you so much. And we'll now go to Mr. toy for his five minutes. Thank you. >> chairman griffith, ranking member degette and members of the subcommittee, thank you for the opportunity to testify today. My name is andrew toy, and I'm the ceo of clover health.
▶ 0:35:37We're an ai technology company offering medicare advantage plans and helping doctors take better care of seniors. We serve over 100,000 seniors, many of whom live on fixed incomes or in underserved communities. I'm not your typical health plan ceo. I have both a bachelor's and master's degree in computer science from stanford university. I built a startup that was acquired by google and later worked on android and google cloud.
▶ 0:36:03I'm also a patient with marfan syndrome, a rare genetic condition that took my father's life as he was diagnosed too late for his condition to be managed. His death may have been prevented if his doctors had the right data and coordinated care. That drives my mission using technology, so no patient slips through the cracks the way he did. So how can I help? Doctors today are asked to do the impossible.
▶ 0:36:32Care for patients while juggling mountains of data across disconnected systems. This is where I can make a difference. It can pull together information from labs, pharmacies, hospitals, and electronic records, then highlight what matters most right at the point of care and in the doctor's existing workflow. Our tool, clover assistant, does exactly that. Now let me be clear. I should never be used to deny, care or replace physicians.
▶ 0:37:01It should be used to empower physicians, helping doctors identify diseases earlier, personalize treatments, and spend more time with their patients. I see three big healthcare ai opportunities. Number one, bringing sophisticated care to every community. Number two, making care truly personal. And number three, reducing inefficiencies and costs. So starting with bringing care everywhere.
▶ 0:37:30For too long, advanced data systems have only been available inside large hospital networks. Independent doctors, especially in rural and underserved areas, haven't had access. That has real consequences. Misdiagnoses and opportunities for care. Duplicate or incorrect tests, and dangerous delays.
▶ 0:37:48It's what happened with my father, with the ai powered capabilities in clover assistant, a solo physician in iowa or a small clinic in urban new jersey, has the same data driven insights as a major academic center. By integrating records from national networks, geography and practice size no longer determine care quality. Turning to personalizing care. Every patient is unique.
▶ 0:38:14As someone with a rare condition, I know I'm not a statistic. I'm an n of one. Ai helps doctors move beyond generic population management to truly personalizing care, and the results are real. For example, doctors using clover assistant start diabetes patients on treatment three years earlier on average. They identify kidney disease one and a half years earlier.
▶ 0:38:41Preventive screenings also go up, 11% more colorectal cancer screenings and nearly 5% more breast cancer screenings. Among heart failure patients. Clover's ai enabled care has been tied to 18% fewer hospital stays and 25% fewer readmissions. And because of this, doctors using clover assistant have significantly lower cost ratios over 10% better.
▶ 0:39:05We use those superior economics to help patients afford care and lower out-of-pocket costs with $0 premiums, the lowest copays for specialist visits, and $0 co-pay. Access to both in-network and out of network primary care. Ai is powerful, but must be used responsibly. Three things matter most, I think. Number one, empowering providers. Ai should empower, not replace clinicians.
▶ 0:39:34It should make their jobs easier and their time with patients more meaningful. Number two ensuring interoperability. Ai is most effective when patient data is accessible and secure. We must protect patient privacy without creating burdensome mandates. And number three accelerating care. I should make it easier and faster for patients to be diagnosed, receive personalized treatment, and get that care paid for.
▶ 0:40:02Ai is here. It's helping doctors identify diseases earlier, deliver more personalized care, and lower costs. Used responsibly, I can give every senior in america the personalized and effective healthcare they deserve. Thank you and I look forward to your questions. >> appreciate the gentleman yielding back and now recognize doctor ibrahim for his five minutes.
▶ 0:40:30>> chairman griffith, ranking member degette and members of the committee, thank you for the opportunity to testify. My name is andrew ibrahim. I'm the chief clinical officer at viz.ai, a practicing general surgeon at the university of michigan and a research scientist with over 200 publications focused on how we improve the delivery of care. Over nearly two decades, I've seen firsthand how workforce shortages, rising costs and mounting complexity are pushing our healthcare system to the breaking point. Doctors are asked to see more patients in less time, with information scattered across multiple systems and logins.
▶ 0:40:59Patients may need to make 3 or 4 calls just to schedule a single appointment, only to sometimes be sent to the wrong clinic or lost to follow up altogether. And the heavy documentation burden takes away from doctors what matters most caring for their patients. Artificial intelligence offers one of the most viable paths forward to save us healthcare and reverse these trends. There are many types of artificial intelligence, including machine learning, deep learning, and large language models. But the technical distinctions are not what matter most in healthcare.
▶ 0:41:29What matters most is not the algorithm in isolation, but how these tools are integrated into real clinical workflows to solve real high stakes problems. At viz.ai, that has been our approach. Viz.ai is a healthcare technology company that uses artificial intelligence to empower clinicians with actionable information within seconds, so they can make timely decisions to provide high quality care by getting the right information to the right doctor at the right time. Viz.ai helps hospitals treat patients faster, save lives, and reduce costs.
▶ 0:41:58Importantly, it offloads burdens from busy clinicians and improves diagnostic accuracy. We began tackling one of the most devastating conditions in medicine stroke. These events are often caused by a blood clot to the brain. Every minute untreated cost a patient nearly 2 million neurons. Without immediate action, a stroke can leave patients permanently disabled or even lead to death. Approximately 800,000 americans from all walks of life, urban and rural, suffer a stroke each year. This reality is personal to me.
▶ 0:42:27When I was in medical training, my father, a university professor, called me from his office and he was experiencing vision changes and trouble finding words. I immediately recognized it may be a stroke and I called campus security on him. They went to his office and facilitated getting him to the hospital, and while he was in route, I called ahead to the emergency room. I also called in personal favors to make sure that the neurologist and interventionalists were available.
▶ 0:42:52Because of these extraordinary steps, he got timely treatment and days later walked out on his own feet, able to speak and see clearly. But not everybody has a medically trained son ready to coordinate their emergency care. We need systems that ensure patients get the right treatment without relying on luck or personal connections. That is why I joined viz.ai. Our ai platform helps coordinate timely care for patients just like my dad.
▶ 0:43:14It automatically analyzes cts to identify those with life threatening conditions, alerts the stroke team in real time, and assembles a relevant clinical information to support urgent decisions. Independent studies show that this reduces treatment time by more than 30 minutes. Insurance hospital stays by as much as three days. The result is simple but profound. Patients walk out of the hospital who otherwise may never return home. Our platform is scaled to more than 1800 hospitals nationwide.
▶ 0:43:41Every five seconds, a patient's care in this country is supported by our platform, from leading academic centers like the cleveland clinic to critical access hospitals in rural america, where ai helps identify who needs to be transferred quickly for emergency care. And stroke is just one example. We have applied the same model to other serious conditions, such as hypertrophic cardiomyopathy, a disease where the heart muscle becomes abnormally thick and makes it harder to pump blood.
▶ 0:44:08Left undiagnosed, it can lead to sudden cardiac death, particularly in younger adults. It affects 1 in 500 americans, yet most don't know they have it. The cleveland clinic studies show that our ai tool cut time to diagnosis from years to just three months. Pulmonary embolism is another time sensitive condition where a blood clot can go to the lung, resulting in straining to the heart. It affects 900,000 americans each year and is responsible for over 100,000 deaths annually.
▶ 0:44:34A trihealth study used our platform reduce time to treatment from hours to just six minutes, with a measurable reduction in in-hospital deaths. Finally, brain aneurysms affect nearly 1 in 50 americans, which should be at least 1 or 2 people in this room. But most don't know they have it. Left untreated, it also can lead to life threatening consequences. We've expanded these approaches to other areas, including oncology, cardiology, vascular medicine, and trauma.
▶ 0:45:02I mentioned all these to underscore that the real world potential of ai in healthcare is not hype or theoretical. For us, it's been a decade of hard work that's translated into fda cleared cms reimbursed tools embedded in frontline care today. But the road has not been easy, and we face barriers common to many innovators, including regulatory challenges, reimbursement, and data accessibility.
▶ 0:45:24I'm deeply optimistic about what I can do for american patients and clinicians, and I believe the united states can lead the world in this space with the right policies, incentives, and commitment. We can ensure that every patient, whether in san francisco or rural montana, receives timely, life saving care. And we can restore doctors to what they entered medicine to do care for patients. Thank you. >> gentleman yields back now recognize doctor melo for her five minutes of questioning or excuse me statement.
▶ 0:45:55>> thank you, Mr. chairman. I appreciate the opportunity. Chairman griffith, ranking member degette and others in the subcommittee to speak with you today. I'm part of a group of data scientists, physicians and ethicists at stanford university that helps govern how ai tools are used in healthcare. At our stanford facilities, and I would like to share some of the learnings that we've had at stanford and my own professional experiences over two decades studying patient safety and health policy. I'm enthusiastic about healthcare ai.
▶ 0:46:24There are so many intractable problems that it will help solve, but ai presents both a historic opportunity and a serious risk. And although properly improperly designed, regulation may hinder innovation, there is a critical role for the government in ensuring the conditions for innovation to translate into ai adoption.
▶ 0:46:47The key problem isn't that there is a lack of innovation in the ai space. As you've heard today, there is much to be proud of, but the uptake of these innovations is relatively low, and a major reason for that is what experts call a foundational trust deficit.
▶ 0:47:08There are four areas where I think experts agree that policy interventions could help build trust in healthcare, ai, and spread innovation and adoption system wide. First, policymakers can help ensure that the entities that develop and use ai adequately assess, disclose, and mitigate the risks of these tools.
▶ 0:47:32Surprisingly, most healthcare organizations and insurers really don't do that much vetting of ai tools before they use them or monitor them after they deploy them. Nothing requires them to do more. Ai developers aren't required to make any particular disclosures to customers when they pitch their products, and the law also permits them to disclaim liabilities and warranties.
▶ 0:47:54As a result, developers have little incentive to reveal weaknesses of their ai tools and little consequence when things go wrong. This rule free space has left hospitals, clinicians and patients apprehensive about the risks of ai, and that fear is chilling. Adoption.
▶ 0:48:16It's no coincidence that the largest area of ai implementation today isn't the amazing, exciting, life saving tools that are being developed. It's tools that perform simple administrative tasks like taking notes during clinic visits. We need a supportive infrastructure to build trust in ai tools. For starters, healthcare organizations and insurers should be required to show they have a governance process in place that meets certain standards.
▶ 0:48:44Fortunately, there are many good models for institutional governance to draw on. At stanford, for instance, a c-suite level committee decides which tools will be deployed in the hospitals. Drawing on an evaluation of the tools clinical utility, its financial sustainability, its fairness and ethical considerations. That's done by an interdisciplinary team.
▶ 0:49:05And we have found that with just a modest investment of effort, we can spot and address risks, mismatched assumptions and unfounded expectations. The federal government already requires research organizations to prove that they monitor human subjects, research and protect participants using institutional review boards.
▶ 0:49:28Such a concept could be exported to the ai realm, requiring that participants in federal health programs have a process for vetting any ai tool that affects patient care, and a plan for deploying and monitoring it. The joint commission, which accredits hospitals for medicare, is already developing voluntary certification standards for facilities use of ai. But institutional governance is only effective if developers feed information into it.
▶ 0:49:53And that's why developers should be required to disclose information about ai design and performance. Work on model cards is a great place to draw from here. A second step the government can take is to support independent research on how ai tools perform in practice that can provide critical information about benefits and risks of the tools. Once they're in the hands of clinicians and insurance companies, and it can ensure that that information disseminates out to the public.
▶ 0:50:23Third, healthcare reimbursement policies should support adoption and monitoring of effective ai solution. Many ai tools don't enhance revenue for healthcare organizations and won't generate significant cost savings, and they're costly to monitor. Adapting medicare and medicaid reimbursement to help cover costs would bring the benefits of these tools to a wider spectrum of hospitals, including those in underserved communities. Finally, the fda should be empowered to be a more constructive partner in ai development and deployment.
▶ 0:50:53There's only so much the agency can do with a structure that dates to the ford administration. Many tools aren't subject to fda jurisdiction at all, and many of those that are are faced with cumbersome processes that weren't designed for software evaluations. Modernizing the fda authority could help build confidence in ai, while fixing areas where current regulation makes it hard to bring products to market. With these steps, congress can help close the trust deficit that hinders innovation from reaching the bedside. Thank you.
▶ 0:51:25>> thank you very much. And now recognize doctor wright for her five minutes. >> good morning, chairman griffiths, ranking member degette and distinguished members of the subcommittee. My name is doctor wright. I'm a licensed psychologist, and I serve as the senior director for healthcare innovation at the american psychological association. On behalf of the apa and our 173,000 members, thank you for the opportunity to testify today.
▶ 0:51:49I'm here to discuss the critical role of psychological science and researchers in shaping the oversight of artificial intelligence in healthcare, and the role psychology can play in the ethical and safe development and deployment of ai technologies. Ai is not just a technology. It's a tool built by humans to be used in human systems.
▶ 0:52:14Therefore, a deep understanding of human cognition, behavior, and emotion must be central to its deployment to ensure it serves people effectively, ethically, and equitably. The apa recognizes ai's ai's immense potential to revolutionize health care for consumers and patients. It can enhance diagnostic position and expand access to treatment and preventative care, especially in behavioral health.
▶ 0:52:44When we know that there is a huge workforce shortage for providers. Ai can alleviate administrative burdens that can lead to burnout. For example, ai powered ambient scribes, as we've heard today, are already being used to automate progress notes, which frees up clinicians to focus on direct patient care. We are also seeing the rise of digital therapeutics, software based interventions that deliver evidence based psychological treatments under the oversight of a licensed provider.
▶ 0:53:12These regulated tools represent a responsible pathway for innovations to address the mental health crisis. However, this promise is matched by significant uncertainty. Public trust is fragile, with polls reporting that the majority of americans are uncomfortable with ai being used in their own healthcare. Currently, this discomfort is not unfounded. First, ai risks amplify existing health inequities.
▶ 0:53:40For example, one deployed algorithm determined a patient's level of illness based in part on their total health care cost. But we know that certain patient populations have historically spent less on health care as a result of systemic issues. So what this lead to is unfairly concluding that they were healthier, making them less likely to be identified for high risk care management programs.
▶ 0:54:07This problem can impact patients based on their gender, age, race, ethnicity, and socioeconomic status. Second, the direct to consumer market is flooded with unregulated chatbots, making deceptive and dangerous claims. One entertainment chatbot, presenting itself as a psychologist engaged in millions of chats. And in another documented case, the chatbot appeared to validate a user's violent thoughts toward his family members.
▶ 0:54:36This is unacceptable. The apa has formally requested investigations by the ftc and the consumer product safety commission to address these potentially harmful products, to realize ai's promise while protecting patients, the apa urges the subcommittee to advance legislation and oversight built on the foundation of ethics, equity, and evidence the core mission of healthcare to help and do no harm must be our guiding
▶ 0:55:07Principle. We recommend five key actions. The first, establish clear regulatory guardrails. We need a robust federal framework that prohibits the misrepresentation of ai as licensed professionals, and mandates transparency and human oversight over clinical decisions. Second, we need to prioritize equitable access to care and mitigate harm.
▶ 0:55:31We must require that ai models undergo rigorous, independent testing for harms across diverse populations before they are deployed in the marketplace. Third, we must protect vulnerable populations, especially our youth. Adolescents are in a critical developmental period, as outlined in apa's recent health advisory. We must require age appropriate safeguards and robust data protections to support their healthy development.
▶ 0:55:59Fourth, we need to invest in research in ai literacy. Ai development is far outpacing the research. We need a significant federal investment in independent research with psychological scientists and developers in the same room to understand api's impacts. Paired with comprehensive ai, literary education for the public and for providers. Last, we must enact comprehensive data privacy legislation.
▶ 0:56:26A strong federal privacy law is essential. We must establish a right to mental privacy by safeguarding biometric and neural information, like data from wearables, that can be used to infer an individual's mental state without their consent. Ultimately, we must ensure that a human remains in the loop. Ai should be seen as a tool to augment, not replace, the clinical judgment and therapeutic relationship that are the benchmark of quality health care.
▶ 0:56:53The apa believes ai holds the potential to create a more accessible and equitable health care system, but only if we intentionally embed psychological science into its entire life cycle. We are eager to collaborate with the subcommittee to develop enact legislation that advances these principles. Thank you so much, and I look forward to your questions. >> thank you very much for committee custom. Each witness will have did.
▶ 0:57:17Your opening statement will now move on to questions, and I will ask members not to begin a new question to our witnesses as their five minutes expires, and would encourage members to submit written questions for the record. I now recognize myself for five minutes. Questions? Let's go back to you, doctor wright. I thought you made some really good points. And so let's make this clear for everybody at home. You need to have somebody in whatever you're looking at, whatever you think your disease might be or condition.
▶ 0:57:47If you're just online and you're using ai online, you need a healthcare professional to help you interpret that, to make sure that you haven't gotten ahold of some rogue, untested ai. Is that basically part of what you said? I know it's not the whole five minutes, but you took about a minute on that. Is that correct? >> yes. That's correct.
▶ 0:58:05>> and so we need to make sure that we have the clinicians involved because and uh, originally I was going to ask doctor ibrahim this, but I have a article I read a couple of years ago coming out of the london, uh, times related to, um, a diagnosis of lung disease from x rays. And the radiologist actually did better than the ai because there were so many different factors that had to be looked at. And in a different study.
▶ 0:58:30Part of the problem that that was identified was they had gotten a lot of the input data from sub-saharan africa, where the conditions were very different than the conditions in europe, and they were given a lot of false positives. And so what we have to do is we have to figure out how much data has to go in before it's effective. And if you could tell us if you have it. And I know there's no magic bullet, but roughly, I mean, how much data has to go in.
▶ 0:58:54And I guess the more complicated it is, the more data you need, but you need a fair amount of data in order to make the ai work correctly. Isn't that a fair statement? >> thank you, chairman griffith. I agree with your point, and maybe would add to it that it's not just the amount of data, but it's the correct data. And ideally, to get these models to function correctly, you want to train them on a population that's similar to the one you're treating. In our experience with stroke, a lot of our initial data came from the southeast united states because that's where the stroke belt was.
▶ 0:59:23And so training our models there was relevant, um, and that's been a learning curve of the field to know to do that. Um, so I appreciate you raising that point. It's incredibly important. >> yeah, I appreciate that. I am excited about the potential of ai. Uh, Mr. toy, you raised the issue, and, uh, marvin, if I said it correctly, and I hope I did, uh, but I have two cases at home that I know about. One was a friend of mine who, uh, had a form of als, but apparently there are 4 or 5 different types.
▶ 0:59:51And to get the proper treatment, it took them several years to figure out, uh, exactly what version of the disease he had. And when I, I mean, I grant you got to get a lot of data in, but wouldn't ai help shorten the time period to eliminate or to, uh, identify which type that he might have had he since passed away? Uh, the treatments just weren't there. And by the time they diagnosed it, it was it was not possible to get him what he needed.
▶ 1:00:18But, uh, wouldn't I help with making that identification shorter? Mike? >> absolutely. Thank you. Um. Uh, definitely. Obviously, we look after a lot of als people in our population. And within marfan, I have type four, and there's like many, many, many different types, exactly as you just said, which, uh, understanding protocols of care are treated actually very similarly.
▶ 1:00:41I think with ai you can match symptoms, you can match genotypes, you can match phenotypes much more precisely in that n of one personalization I was talking about, so that you can then match care protocols, drug protocols directly to the kind of disease that your friend. Unfortunately. >> yeah. And this can also help in the uh, population that doesn't have a family history. Uh, I know of a case where an individual who was adopted had no clue that that they, uh, might suffer from huntington's.
▶ 1:01:12Uh, they it turns out that's what they had, but for. But they spent a year or more looking at Ms. and other muscle diseases before they, they said, wait a minute. Maybe it's something we haven't looked at and I might help identify based on the symptoms. It might help quicker identify. And that's not to say that the doctors were doing anything wrong. It's just that they've got hundreds of options and I can help cut through those options. Can it not? >> I think absolutely.
▶ 1:01:35When we talk about earlier identification, this is a huge area that we're seeing already happening with ai, where a doctor will be doing the right thing and looking at this set of symptoms, but then not thinking as much about something that's a little bit rarer or not obvious. And I can notice something there in the symptom set and bring it to the doctor's attention. And the doctor then says, oh yes, I should consider that. And that's a big, big area. We see that doctors are saying it's helpful to have from ai. >> yeah. All right. Last but not least, my district is very rural and economically stressed.
▶ 1:02:06Most of my hospitals and community health centers run on very thin margins. What type of upstart software is needed for rural health entities to use advanced ai and machine learning? And is it expensive? >> uh, that's a great question.
▶ 1:02:20I think that what's really exciting is that the infrastructure we're building on ai and the infrastructure of the country is built on cloud means that we can deliver these capabilities very quickly to more rural areas, to smaller towns, and they can enjoy the same benefits that a larger, more dense urban center would have and the same coordinated care as well between different sites of care. So that local independent pharmacist in the small town would be coordinated with the local family doctor in a way that they've never seen or been able to before. >> I appreciate it very much.
▶ 1:02:48I yield back and now recognize Mr. leggett, ranking member for her five minutes questions. >> thank you so much, Mr. chairman, and thanks to all of our witnesses for their very illuminating testimony. There's been tremendous innovation in the machine learning and generative ai spaces in recent years, and the health care sector really can see the potential benefit of this innovation. Um, and, and, um, all of you gave excellent examples of how this can happen.
▶ 1:03:17Uh, one thing that struck me, Mr. toy said that I should assist doctors, not substitute for them. And I think all of you are nodding. So I think none of you disagrees with this statement, and we all agree with that. Um, so, doctor mallow, I want to ask you, um, you testified that the uptake of the innovative tools is low, in part because of a foundational trust deficit. Is that right? >> yes, congressman.
▶ 1:03:44>> so in general, how would you characterize the evidence base supporting the use of novel ai tools? >> it's thin, you know, and what we have predominantly comes from the people selling the technology. >> and I also suspect there are not well established, industry wide best practices for the use of ai tools in the health space. >> we have best practices emerging. The difficulty is why adopt them if there's no incentive or reason to do that, use them.
▶ 1:04:12>> and at present, which types of institutions are able to measure the impact of ai tools and to optimize their use? >> the ones in the best position to do that are the ones with the most money and the most scientific expertise. >> right? So is it likely that institutions with fewer resources are going to be able to pull off that sort of robust evaluation and thoughtful implementation in the current environment?
▶ 1:04:36>> I don't believe so, certainly not with help from the reimbursement improvements that would enable them to staff up and gain that capacity. Right. >> so those organizations with the fewer resources that we're talking about, like rural hospitals, community health centers and other healthcare providers, they all serve our constituents. They're facing tighter and tighter resources to do their jobs. And as as all of you know, and, of course, my colleagues here know, earlier this summer, h.r.
▶ 1:05:04One, which I call the big bad bill, was signed into law. What that bill will do, it will kick 15 million people off of their insurance now. So that's 15 million fewer people who are going to benefit from an ai revolution in healthcare and hospitals, who need resources to care for our constituents and to work on improving care, including with new ai tools, are facing an additional $400 billion in uncompensated care
▶ 1:05:36Over the next ten years. So all told, h.r. One is slashing over $1 trillion from our constituents health care. I've got to say, this is what what I meant when I said in my opening statement that we're fiddling while rome burns, all of us agree. I can really work. It should be used. We should use best practices.
▶ 1:05:58But when we're taking away the money to treat our constituents, then the idea that somehow they're going to find a way to do a robust and appropriate ai system is is a dream at best. Impoverishing the health care system is just not going to help these improvements. And there's no, as we all agree, there's no ai solution that can replace the doctors and nurses. That hospitals, which will serve the poorest americans, can no longer afford to employ.
▶ 1:06:27And I want to say this committee just simply has to stop acting like it's business as usual and ignoring the real problems that are facing our healthcare system and our constituents. If we really want to use ai as a good tool, we're going to have to reverse the cuts to medicaid before they take effect and devastate millions of americans.
▶ 1:06:51And we're going to have to stop the administration's damage to our public health system before the emergency lays open our vulnerabilities. I mean, this is really serious. I would like to be able to find a way to to, um, do some innovation at the fda. I did that in 2016 with congressman fred upton, then congressman from michigan.
▶ 1:07:16We did 21st century cures, which which we restructured a lot of things we do at the fda and the nih and the cdc, and enabled us to have operation warp speed, which I guess now the president doesn't like, but which saved millions of lives with the covid vaccine.
▶ 1:07:34So so I would like to be able to help spearhead a bipartisan effort to put the protocols in place and the funding in place at the fda to do it, to do to do, um, artificial intelligence folks, that's not where we are right now. And people better know that I yield back. >> gentlelady yields back. Now, recognize the chairman of the full committee, Mr. guthrie of kentucky, for his five minutes. >> thank you.
▶ 1:08:04And I thank you all for being here today. So for Mr. Mr. toy and doctor abraham, uh, clover and I have both launched successful ai technologies in healthcare. Speaking from your company's own experience. So I'll start with Mr. toy and then Mr. speaking from your company's experience, how can ai tools be leveraged to empower and not replace providers in ways that can help them better manage and deliver care? >> thank you for the question.
▶ 1:08:31I think the key here is, as we've all said, that the physicians in question need to be able to feel like that they are able to do their jobs better than they were before, that the friction is being lowered and that they are feeling effectively empowered to do a better job. There should be a force multiplier that comes from ai. So what they were going to do, perhaps anyway they can do faster, quicker and easier, including getting reimbursed for that, that, that treatment.
▶ 1:08:57So I think that I we're showing in our own plan is already doing that with our tool clover assistance. We're putting that in the hands of primary care physicians who are diagnosing diabetes earlier, who are managing chronic kidney disease earlier. And that's by synthesizing that data and putting it at their fingertips. >> thank you, doctor abraham. >> I'll just add to that and I'll maybe just give perspective from my own practice.
▶ 1:09:20I have a clinic every tuesday afternoon, um, where I may see a dozen new patients, and monday night I spend 2 or 3 hours to read about them, because to get the 3 or 4 data points that I need to make a decision about their care, I log into 3 or 4 different places. I read through scanned pdfs from an outside hospital to try to piece the whole story together. And I know many of my colleagues do the same thing, including for time sensitive things in the hospital to harness our technology.
▶ 1:09:45And it's one of the things we've done in our company to help find that information quicker so that the physician can get to what they want to do is make a treatment plan for that patient. Maybe one of the most valuable contributions of ai. >> okay, so follow up for the two of you as well. How should policy makers think about insuring more of these innovations can reach patients and clinicians without overly burdensome regulations getting in the way, and we balance the need? How do we balance the need between the appropriate guardrails and patient safety, and getting these deployed?
▶ 1:10:17>> I'll take that first. Thank you. That question as well. I think that the way we consider it is by having that clinician looking, being accelerated by the ai, knowing that that clinician is available to review its recommendations, that serves as an important guardrail in and of itself, and so that we are accelerating what's in their own mind. We are relying on the training that they have. We're just making their job easier.
▶ 1:10:38So anytime we're making that, uh, expanding the possibilities of how we do that with a clinician, a physician, we feel like we're in a pretty good place from a guardrail perspective. >> thank you, chairman, and I'll maybe connect my answer to the ranking member to gets previous comment. Um, I have a lot of empathy and sympathy for the fda at the moment where they're being asked to evaluate a new technology that they're not really designed to evaluate.
▶ 1:11:05Right now, they're using laws that are decades old, or statutes or authorities or frameworks that don't really fit the pace or the way that ai is being used. So support from congress that would help fda modernize, whether it's another version of 21st century cures or something similar, would be helpful so that the fda can efficiently figure out which of these products are snake oil and should not be approved, and which of these are approved. And once they're approved, they can then be supported through reimbursement.
▶ 1:11:35So I think there's some tangible policy there that would be incredibly helpful. >> okay. Thank you. And so, Mr. parker, this committee has in the trump administration has has made health care price transparency a top priority. Is that bipartisan? How are price transparency and reporting policies enabling your company to empower american patients in their healthcare journey? And can you discuss the role ai plays enabling general medicine to do this work?
▶ 1:12:02>> yeah, I should state very clearly that the work we're doing to show consumers prices up front for procedures, labs, imaging, anything that they need would not be possible without all of that work. Um, we are using the open pricing files to figure out the the sort of reimbursement and their specific pricing for the consumer, but the it really also wouldn't be possible without ai.
▶ 1:12:22You know, the example I shared in the testimony is we're literally taking down your full 100 page coverage of benefits and using ai to read those coverage benefits and turn it into useful data. Um, and then munging that together with the open pricing files to give patients a clear out of pocket price. So it really is a by product of all the work that's been done over the last decade to, uh, to require providers to disclose these, these pricing files.
▶ 1:12:48>> and so it it's the providers, insurance companies, whatever pbms, you have to have the price data publicly available for ai to be able to call out and get people the best answer. >> that is correct. >> yeah. Because if it's hidden data then ai is not going to find it. Yep. Right. Thank you. And I yield back. >> gentleman yields back. And I recognize the ranking member of the full committee Mr. pallone for his five minutes of questions. >> thank you, Mr. chairman. I'm trying to get in two questions to doctor mello and one to doctor wright.
▶ 1:13:17So bear with me here. Doctor miller, we're witnessing a significant increase in ai driven medical product development. Given the rapidly expanding technologies, we must ensure that we also better understand and evaluate how these ai enabled devices function so they don't put patients at risk. So my question or first question to achieve this, do we have the appropriate regulatory framework in place?
▶ 1:13:40And does the fda's current statutory authority allow the agency to be nimble in its oversight of ai enabled devices, if you will? >> thank you for the question. I'll answer the second part first. I think the agency has been remarkably nimble given what they have to work with, but what they have to work with does not allow a great deal of maneuvering room. So absolutely, there needs to be modernization to solve two problems.
▶ 1:14:04One is that they can't do enough, and one is that the statutory framework requires them to do too much or the wrong things, given the type of technology that they've confronted. Now, as doctor ibrahim has said, okay. >> and I'm also concerned that there are insufficient guardrails in place to protect patients from bias in ai and coverage denials for necessary care.
▶ 1:14:26In july, the trump administration announced the wasteful and inappropriate service reduction model, which would impose prior authorization and traditional medicare and allow for profit companies to use ai to perform prior authorization reviews. And I'm concerned that this ai model will result in denials of life saving care and incentivize companies to restrict care.
▶ 1:14:48So my question is, how can we ensure that these algorithms are not being misused to delay and deny life saving care for patients, if you will? >> right. So the wiser program really raises two questions. One is when we take a process that is fundamentally and for a long period of time, flawed and put it into traditional medicare, what would we expect to happen? And then the second is when you take that process prior authorization and amp it up with ai, what will change?
▶ 1:15:14Uh, I think we have pretty good evidence that prior authorization as a process itself is, is fraught, uh, that there are high rates of wrongful denials, low rates of appeal, high rates of overturn on appeal. So we should have some concern about any effort to expand its use, while at the same time recognizing that in some areas, including, I will say, the areas that are currently targets, it is a necessary cost control. And the question is how you do it. And then the second issue is what happens when you bring ai into the game.
▶ 1:15:44And there we don't know. We don't know because there are no publicly available information that would enable somebody like me to be able to tell you, does using ai make prior authorization better for patients or worse? Because I think both outcomes are very possible. >> all right. Let me go to doctor wright. Another area of concern is the intersection of mental health and ai. Um, if I could, Mr. chairman, I'd request that the washington post article titled what is ai so psychosis?
▶ 1:16:12And how can chatgpt affect your mental health be submitted into the record, if you will? Mr. chairman. >> the date on that. >> was that. >> we've been. >> I think you have it. >> we have it. Okay. Without objection. >> all right. >> thank you.
▶ 1:16:28>> so, doctor wright, the question is the american psychological association issued a health advisory on ai and adolescent well-being in june, calling for robust protections for young people interacting with ai, urging developers to implement safety features, and for educators and parents to foster comprehensive ai literacy. So can you share information on what prompted apa to issue this advisory, and what your members are seeing in their practices that would relate to that? >> thank you for the question.
▶ 1:16:58Yeah, apa really felt a sense of urgency to look at what the psychological science has to say in this space because as we've said, ai development is just outpacing the research and the guardrails. And we've been hearing from both clinicians who are seeing patients that are coming in with their chatgpt advice.
▶ 1:17:18And we've heard the public stories like the washington post story you just mentioned, about the harms that seemingly are happening, in part because to this point, we have a broken mental health system, and so people are seeking out help wherever they can find it. Um, and that while very human, might not be the best idea.
▶ 1:17:40And so I see a future where, you know, we will have, um, mental health options that are more appropriate that do utilize ai, but that's just not what's on the commercial market right now. Um, and so helping parents, uh, teachers, educators, technologists understand what is the psychological science say about how to do this.
▶ 1:18:00Well, and what are the considerations that need to be factored in, is a huge part of why we put out that advisory and why we're also standing up a second advisory, more specific to the use of these wellness chat bots and these general chat bots. >> thank you. Thank you, Mr. chairman. >> gentleman yields back. Now recognize the vice chair of the subcommittee, miss harshbarger. >> thank you, Mr. chairman. And thank you to the witnesses for being here today.
▶ 1:18:29Uh, I'm excited about this line of questions, and I hope you are, too. Um, I'll start with Mr. parker. Uh, you know, I'm excited about the fact you you got a one stop shop for telehealth, basically telemedicine. If you can do the labs, the prescription, give them the price and do all that and bill insurance or let them do a cash model, kind of like a concierge service.
▶ 1:18:51You know, when you walk into a brick and mortar, uh, I guess, uh, enhancing the role of the clinician is a key win for technology. And as part of the expansion of the over-the-counter drug class, the fda promoted something called additional conditions for non-prescription use.
▶ 1:19:06And what this essentially means is consumers can answer some questions on a phone, app or another, um, technology platform to determine if a drug is right for them and then access a prescription drug therapy. So how do you see technology and automation as a platform to expand and reinvigorate the practice of community pharmacy? Since that's what I am a compounding pharmacist and I have a community pharmacist. >> yeah, it's a great question.
▶ 1:19:35And I think, as we all acknowledge, pharmacists are are underutilized compared to their. Absolute capacity. Um, and I do think if you take some of the work that we've done at general medicine, where we've worked really hard to codify these intakes for specific medications. So it's very the the clinician, which in this case could be a pharmacist, has all the information they need to make a judgment on whether it's appropriate for the patient in that moment would apply very well in that setting, I think wouldn't just be constrained to a traditional clinical setting.
▶ 1:20:03And so I think a lot of the the logic infrastructure and work we're doing would be would be a great opportunity for pharmacists to participate more in that. >> absolutely. Thank you for that. We need to practice at the top of our license, which is something I'm working on too. Uh, Mr. toy, you mentioned that as far as utilizing your community provider in rural areas, in your pharmacist, can you what do you think about this? >> absolutely.
▶ 1:20:28I think that and you probably see this in your own practice, but, um, there's not a lot of coordination right now between the pharmacist and the physician. Oftentimes the physician uses their mind. They think about things, they write a script, and then they're like, okay, where would you like that filled? Right. Um, that's the question they ask the patient. And so what we want to make sure is that there's coordination there and make it very easy for the pharmacist to have the same information that was available.
▶ 1:20:52And the thing that I can do also is take that same information and customize it not just to the patient, but customize it to the clinician in question. So it'll show a certain version of that to the pcp, and it'll show a pharmacist relevant version of that to the pharmacist. But they're seeing the same data but two different facets of that same data. So they can then apply their clinical. >> you work with collaborative agreements between a physician and pharmacist is how you would do that. And every state is different.
▶ 1:21:18So, you know, that's why I'm looking at an icaps bill and some other things associated with that. So yes, sir, I got it. I'm glad you're on my team. Thank you. Uh, doctor ibrahim, uh, while some applications of ai in healthcare, such as basic chatbot, uh, may appear relatively low risk, uh, others carry higher um stakes, including tools targeting mental health minors or carrying coverage of determinations.
▶ 1:21:44Uh, should congress begin to define high risk categories of generative ai in health care? And if so, what safeguards like, uh, such as independent validation of clinical claims or patient safety protocols for sensitive use cases, maybe labeling and disclosures for consumers? And what would best protect that, uh, patient without stifling that innovation. >> thank you. Congresswoman. Um, innovation moves at the speed of trust.
▶ 1:22:11That is something we've all sort of been trained and brought up in, um, in our practice, anything that we've tried to bring to patients requires a partnership with hospitals first, where we map out what do we have the capabilities to do, what are their needs? We pilot things several times in historic data before even trying it in real clinical practice, and then have rigorous milestones to meet before it actually gets scaled to any other, um, settings.
▶ 1:22:37And that's been our practice, in part because that is necessary for our business if we do not establish that trust and maintain it, um, we will lose our clients. So it's required for us to, to do that. Fundamentally, um, the fda has some precedent for devices, which is what much of this has been mirrored on. And there likely is a similar tiering of class one, two, three devices that would mirror an ai technology.
▶ 1:23:05Um, I think that would be welcome guidance to help create some roadmaps. So we understand where we need to spend a lot of time and be methodical and rigorous, and where things that we can accelerate faster. >> yeah, outputs only as good as the input. And it better be correct. Right. Okay. Thank you sir. And Mr. chairman I yield back. >> gentleman yields back. Now recognize the gentleman from california, doctor ruiz for his time of questioning. >> thank you, Mr. chairman. Artificial intelligence has potential for expanding access to healthcare and improving patient care.
▶ 1:23:33However, as with any new or developing technology, we need to make sure that we are evaluating any potential harms and ensuring appropriate safeguards as the technology continues to expand, it is crucial to directly address the risks, including mental health risks that ai tools may pose for our children. Doctor, right. In your testimony, you discussed the need to protect vulnerable populations like youth. You stated that, quote, a.I.
▶ 1:24:01Systems designed for adults are not necessarily appropriate for youth. Could you please explain what specific dangers and potential harms ai technology used for healthcare can have on children and their development? >> thank you so much for the question.
▶ 1:24:17Um, so I think when we think about the model, right, it's really about representativeness and ensuring that whatever training model we're employing has been, um, trained on the data that's appropriate is critical because you do end up seeing these harms when you've got systems that were developed for adults. And children are not just little adults. They have very different developmental trajectories. What is, um, helpful for one child may not be helpful to somebody else.
▶ 1:24:43Um, not not just based on their age, but based on their temperament and, um, how they have been raised. So, you know, at worst, some of the harms we're seeing have been, you know, alleged completed suicides and also suicide attempts and much of and suicidology discussions with different types of chatbots.
▶ 1:25:05Um, you've seen, as I mentioned in my testimony, there was a case in, uh, in texas where a young individual, um, following the validation and advice from a chatbot, attacked his family and had to be hospitalized. Um, so the harms are real. Um, we also have you positive use cases on top of it, right? So I hear stories of children who use a chatbot to practice their social skills so that they can make a new friend at school.
▶ 1:25:33It's not an all good or an all bad thing. It's about how we use the tool appropriately, how we safeguard it, how we test its effectiveness, and to ensure that children who are going to continue to use these ai tools because they're not going away will be protected. >> you know, I, too, am alarmed at stories of individuals turning to ai chatbot as a source for therapy, including the tragic story of teenager, uh, california teenager adam rain.
▶ 1:26:01Chatgpt provided mental health resources when adam shared his suicidal ideations. Maybe a pop up or something, which is inappropriate in a clinical setting. Imagine you're admitting your suicidal thoughts, and the clinician just holds up a poster board with a hotline number, and then continues with more deep conversations about how to actually complete suicide. So it's it's completely not clinically appropriate. Uh, or or that helpful. Um, doctor.
▶ 1:26:34Right. If ai systems can behave unpredictably, what enforceable guardrails are needed to ensure that they do not worsen crisis or harm individuals, especially those seeking mental health resources? And what benchmarks must be met before I can be responsibly scaled as a tool for mental health support, especially for youth.
▶ 1:26:54>> well, I think at first we have to ensure that these tools do not misrepresent themselves as licensed professionals, that they cannot call themselves psychologists or psychiatrists or social workers, because that gives a sense of credibility that doesn't exist.
▶ 1:27:07I think on the back, the real, um, expertise for these products around the back end or how they're being coded, and we could encourage companies to make them less addictive in their coding tactics and make it so that user engagement isn't the sole outcome that they are trying to achieve. Um, I think some of those things would also help.
▶ 1:27:29We could also ask for specific audit and reporting requirements, where these companies had to actually disclose anytime they detected suicidality or, uh, attempted or completed suicides. We could have better age verifications and age restrictions. Um, I think there are some pretty low level things that we could do that would actually ensure that these tools are used better. We also just need better tools. >> and what can we do for, uh, to educate clinicians on how to use these?
▶ 1:27:58>> well, I think for sure, clinicians need a lot of education in how to use these. Um, I think part of why you haven't seen adoption is the anxiety and the fear that these tools bring, um, and, and the lack of understanding of how they work. Um, so I think helping providers know what works, what the limitations of these tools are, um, is first step towards helping them decide whether or not they want to choose to implement them in their practices. >> thank you I yield back. >> gentleman yields back.
▶ 1:28:28Now recognizes gentleman from florida. Mr. bilirakis for his five minutes. Thank you. >> I appreciate Mr. chairman. >> uh, this is a great hearing. I thank the witnesses. One area that I'm interested in is the role of ai in drug development, particularly in the rare disease space. Scientists have developed ai models to identify drug candidates from existing medicines that could be applied for rare diseases with unmet needs.
▶ 1:28:57Uh, the innovation is just the beginning, and we should continue to explore similar applications to improve our healthcare system. Most people agree with that. The question is from, uh, Mr. this is for Mr. troy. Mr. troy, uh, your testimony discusses your personal experience as a rare disease patient. Thank you so much for sharing. Your story really means a lot to the community.
▶ 1:29:23Can you share further how I will continue to accelerate the patient diagnosis journey, please? >> absolutely. And thank you for talking about rare disease here amongst the committee. Um, so speaking as someone there, I'm also serve on the board of the marfan foundation. Um, it's a rare disease. It's a genetic based disease. And so something I would point to is rare.
▶ 1:29:44Diseases are, uh, rare individually, but across, like, the entire country, many people have a family member who has at least one form of rare disease, so they're not rare in collection. The other thing I would say, and we've discussed it here and we're talking about clinicians, I think the key thing is most clinicians are in their training, their chance of having a clinician who has the training, the cme, the continuing medical education on your rare disease. It's actually quite low because they can't take training on everything.
▶ 1:30:11They're what I can do is help them, uh, learn more about the condition. Also help them identify, uh, practitioners in the area who have more expertise in those and collaborate with those practitioners and get a given patient closer to a site of excellence for that rare disease. >> very good. Excellent. Appreciate that information. Uh, doctor. Right.
▶ 1:30:35Uh, and again, I'm just following up with doctor ruiz on, on this particular subject, uh, doctor wright as chair of the conference, uh, the commerce, manufacturing trade subcommittee, I'm very interested in protecting children and teens online, as we all are. And that space, there's a lot of talk about the how children and teens using online services may face greater risk of suicide, self-harm and other dangers.
▶ 1:31:04One discomforting new trend is the rise of ai chatbots, which we've been discussing, which I'm concerned may exacerbate existing problems of loneliness and social isolation among young people. While still early days.
▶ 1:31:24The prospect of children and teens socializing primarily with chatbots and building supposed relationships and even romantic relationships, uh, with these new online services is troubling to many parents. Uh, can you speak to how loneliness and social isolation impacts children and teens development?
▶ 1:31:50And what could the long term impacts be of children and teens interacting primarily with ai chatbots instead of their peers? >> thank you for the question. Yeah, um, obviously we are very concerned as well on the role that all emerging technologies will have on our children, adolescents.
▶ 1:32:14In our recent stress in america survey, the majority of all adults that included children said that they feel like they can't talk to people about their stress because they don't want to burden them. We also have a study that says that adolescents would prefer to speak with an ambient source, like a chatbot, than a person, because they fear being judged. This is a real challenge in our culture. And yes, it is spurring increased loneliness and isolation.
▶ 1:32:40So what do we need to do as a culture to help people understand that coming together and having empathy and social and having social connections is actually the better solution than turning to these technologies as a replacement for people. They can be a huge tool as long as you use them and then go back to the people that you're trying to connect with.
▶ 1:33:04And so if we as parents need to model good use of these technologies for kids, we need to help them understand what the risks of them are, what the business models are, and how that influences how they interact with them. And part of that's going to have to be how we teach this in schools to kids, because again, this ai is not going away, but how do we help people remember that it's personal connection that makes us the people that we are?
▶ 1:33:35>> absolutely. Thank you very much. I appreciate I yield back, Mr. chairman. >> gentleman yields back and now recognize the gentlelady from michigan, miss dingell, for five minutes. >> thank you, mister chair. And thank you and our ranking for holding this hearing on the use of artificial intelligence in healthcare systems. Look, ai is transforming every sector.
▶ 1:33:59And we've seen that this emerging technology has the potential to transform lives and improve health care incomes. But let's be honest, with rapid advancements come serious challenges as well. Any potential ai related legislation must consider potential data privacy, which I worry about deeply.
▶ 1:34:23Uh, bias, displacement of jobs and in the case of healthcare misdiagnosis. And as our witness also said, the humanness of patients. And you need compassion and empathy. It's also important to highlight the context in which we are having this conversation.
▶ 1:34:44Um, my republican colleagues in congress and the trump administration have already enacted and seek to advance further reductions in staff and funding of our health care system. Adequate resources are critical to ensuring that the usage of ai technology in the healthcare space is subject to guardrails that protect the safety of the providers, the patients and their loved ones.
▶ 1:35:10So, doctor melo, as providers grapple with the increasing presence of ai in their workplace, many worry that this technology jeopardizes patients right to high quality care and displaces our talented doctors and nurses. How can we ensure that ai serves as a tool that physicians use to enhance care, not one that replaces person to person care?
▶ 1:35:34>> it's an important question in the context that you specifically raise of cuts in medicaid reimbursement just intensifies both the need for ai to augment capacity in rural and low income hospitals, and also the constraints around their ability to monitor it properly. Uh, I think there is a lot of promise in use of ai to address the needs of populations in those facilities, especially those with complex needs, but they will need help.
▶ 1:36:01And assuming, as many organizations do, that simply because there's a human in the loop, we don't need to provide support. Training and institutional level monitoring is a grave mistake. Let's not forget the whole reason that organizations are implementing many of these technologies, like scribes and and note summarizations is the burnout that afflicts those facilities. So these people in the loop, this is not superman. They need support and it has to be at the institutional level.
▶ 1:36:31>> so let me take that one more step down. Many patients, particularly seniors and individuals with disabilities and children, as we're talking about, have unique needs that require knowledge and care of a human provider. What specific considerations are necessary to preserve the level of patient care for those with complex health conditions, as ai plays a larger role in their care?
▶ 1:36:56>> well, as I think about that question, I have to start from where we are now, which is that those patients are in many ways really poorly served by the system. We have in its extreme fragmentation of both care and information. So I actually think to to some of the testimony that other witnesses have given here that those are some of the best opportunities for ai to make things better, but there are reservations that I have as well, in part because patients in that position are not themselves positioned to be good overseers of the ai that's being used in their care.
▶ 1:37:26For example, whereas I might be inclined to review an after visit summary that I know an ambient scribe generated, somebody who's dealing with multiple complex conditions will not. And so we need to be extra careful about monitoring for them. >> I have some thoughts, but I'll put them in another question. I'm going to have time for one more. I want to turn to the potential mental health impacts of ai, particularly on women and girls.
▶ 1:37:50We know that the victims of abusive ai generated content, such as deepfakes, feel isolated and distrustful around them. We've already talked about depression, anxiety, suicide, etc. I was one of the proud leads of the take it down act. But doctor wright, because I'm almost out of time, how can our mental health care system be bolstered to better respond to the impact of ai enabled abuse? >> well, thank you for the question.
▶ 1:38:20I think in part it starts with helping providers, clinicians and others, um, really know how to evaluate the tools and what they should be looking for if they're going to incorporate these tools into, um, any kind of setting that they're doing. Um, and I think it's about helping clinicians and providers know what questions to ask of their patients about their ai use. Uh, I think, uh, a lot of providers have no idea where to even start to ask that question.
▶ 1:38:45And so you can't even sift out whether abuse is happening if you don't know what the right question to ask. And so it's these sorts of things, these whether they're continuing education trainings that that providers get or we think about how do we train individuals within the school setting. So our future providers, because they're the ones who are really going to be up against it. >> thank you, Mr. chair. I'm out of time. So. >> yes, ma'am. Gentlelady yields back now. Recognize gentleman from florida.
▶ 1:39:14Doctor dunn for his five minutes. >> thank you very much, Mr. chairman. And thank the witnesses for. >> being with us here today. As a medical doctor, I'm certainly excited to see the advances that ai can make possible in healthcare from improving efficiency at the administrative level to assisting providers in clinical practice. I'm sure it's it's going to be great for all of us.
▶ 1:39:42You know, we clinicians are drowning in paperwork. And I think that that could be the low hanging fruit for, uh, for, uh, ai. Frankly, it's essential that the united states remain at the forefront of ai, and I firmly believe the fda should be continuing to work with pioneering companies to ensure that patients benefit from the full scope of ai technologies.
▶ 1:40:06Uh, it can also be, you know, valuable in complementing the work done by physicians to provide better health care outcomes. Uh, but by reducing administrative burdens and costs, I think it can give us far more time to focus on our patients and save money. And by assisting in clinical decisions, I can bolster the human expertise.
▶ 1:40:29Uh, in, in medicine, doctor ibrahim, uh, patients know that physicians are essentially irreplaceable in the system. At this point. We are the linchpin of human responsibility. You and I are both physicians. We know we can be more efficient. Uh, how do you see the earliest applications of ai in clinical medicine as increasing the efficiency of frontline clinicians or replacing your administrative burdens?
▶ 1:41:00>> thank you, doctor. And congressman, I appreciate your question. And that strikes personally close home to me. Um, I think if I could go home to my family and say, instead of reading charts monday night for two hours, all that information would be ready for me in clinic and I could just spend time with my family. I think that would be great. There are a number of things that we need to do in the hospital that require a significant amount of manual abstraction, whether it's quality reporting, whether it's documentation or billing.
▶ 1:41:28Many of those things require very highly trained people to spend hours of time literally, sometimes checking boxes for information that we could readily extract. So I think there's enormous potential to get people actually doing the things that they want to do. Um, and doing less of the things that we could probably automate now. Um, it certainly would require, as mentioned by the panel, that we have the right safeguards in place and the right checks to make sure it's being done appropriately. But I think there's a lot of potential there. >> could agree with you more.
▶ 1:41:58To that end, doctor ibrahim, uh, can you envision a world where most imaging studies pathology slides, radiology, pattern recognition and whatnot are pre-read by software? >> uh, thank you again for that question. I think pre-read is challenging.
▶ 1:42:15I think the idea that it could triage and identify high risk conditions, to say this imaging likely shows a high risk stroke, or this shows a life threatening heart condition, still needs a human to really solidify that from our own data. We know it's not near perfect yet, but it helps identify the highest risk ones. Um, I don't know. It'd be too hard to speculate if we would get to a time in my career where that could be entirely automated.
▶ 1:42:45>> well, I agree with you, by the way, on that, but I think that we may need some updates on the cms payments model to make that world possible. Can you see how that might happen? >> um, cms currently does have an opportunity, um, through ntap for new technologies, um, are able to be temporarily reimbursed for 3 to 5 years that assist with either ekg reads or radiology reads.
▶ 1:43:09I think it would be great to have that a more permanent pathway to reimbursement. Um, but that would still be in the assistance category. >> thank you. Thank you very much for your thoughtful answers, and I look forward to working with my colleagues to ensure that ai in healthcare is utilized to its maximum potential, and ensuring patient safety remains at the forefront. With that, Mr. chairman, I yield back. >> gentleman yields back.
▶ 1:43:36And now recognize representative kelly for her five minutes of questions. >> thank you, madam chair and ranking member, for holding today's hearing. And thank you to all the witnesses. It is important that we do not ignore how in this last week, secretary robert f kennedy jr and others in the trump administration continue to undermine congress and the public health system's ability to function cohesively with the firing of senior staff and experts.
▶ 1:44:05Secretary kennedy is putting americans health and well-being at risk by actively disseminating hhs. Unfortunately, too many of my dismantling excuse me, hhs. Unfortunately, too many of my colleagues remain silent on this devastating public health issue and other harmful policies caused by the big ugly bill. But our work continues, and that's why I'm proud to co-chair the bipartisan digital health caucus with my colleague, rep troy balderson.
▶ 1:44:31I encourage my colleagues on both sides to join us as we aim to educate and learn more about these issues. The integration of ai in the healthcare system offers a potential to be a transformative solution to address long standing disparities and access issues. As you guys have talked about, many in both the healthcare and technology fields have promoted ai to create a more accessible and equitable healthcare landscape, particularly in minority, underserved, and rural communities.
▶ 1:45:01However, we must remain vigilant of the promises as well as the limitations. Doctor melo ai tools are only as good as the data they're built on. When the data is incomplete or skewed, the results may lead to inaccurate suppositions.
▶ 1:45:16We already seen examples where algorithms underestimated the needs of black patients, because healthcare costs were used as a standard for health status, and dermatology tools trained mostly on lighter skin tones, misdiagnosed conditions on darker skin. These cases show how quickly I can replicate the inequities already in our health care system.
▶ 1:45:40What steps should developers, health systems, and federal regulators take to make sure ai tools are tested on and validated for diverse populations before they're deployed? >> it's such an important question, and I know you've heard already about the need for large data sets that include sufficient representation from a variety of subgroups, including all subgroups that are not usually on the map for anti-discrimination worries like kids, people in rural areas, people with rare conditions.
▶ 1:46:08Um, and fda, of course, should be doing more in the area of performance testing. But let me add a wrinkle that I think doesn't get talked about enough. And that is that even when you have a well-performing algorithm that performs well for all of those subgroups, bias and inequities arise at the point of care because of the way that it's deployed. And I'll give a simple example. We've got tons of algorithms at stanford that screen our patients using information in their medical record to identify people who would benefit from additional services.
▶ 1:46:36But our ability to get those patients in the door is dramatically different, depending on what those patients look like and where they live. So unless we're willing to spend more money doing outreach and creating more capacity in our clinics for those patients, they can benefit in theory, but not in reality.
▶ 1:46:55>> interesting you say that as chair of the digital digital health caucus of co-chair, one of the things I'm looking to better understand is how ai and data science can help with efforts such as clinical trial recruitment and retention. I'm also chair of the congressional black caucus health brain trust. What are some ways I can be used to get more people into clinical trials? And what ethical or policy guardrails should congress consider as we support this responsible use?
▶ 1:47:24>> well, I think the main area where I can help is in the identification of potentially eligible patients. We heard already about the nih initiative at stanford. We have a new tool being deployed across our system now that enables a researcher to query the medical record to find all patients who have a particular health condition and meet other inclusion criteria for the trial, and that could be hugely helpful.
▶ 1:47:44I will say, though, that the major, a major reason why we don't have representation in clinical trials is not that we can't identify the patients, it's that we can't convince them to enroll, and ai is never going to get us there. That's where we really need the human touch, reaching out to those communities and explaining the benefits of trial participation along with the risks. >> and my office has been doing a lot of work around that, because we know all the past issues that people have and the the lack of trust, but we know there's still a lot more work to do. Thank all of you so much.
▶ 1:48:16And I yield back. >> gentleman yields back. And I'll now recognize my friend doctor john joyce for five minutes of questions. >> thank you, madam chair. >> much like other industries. >> the growth. >> rate that we're seeing. >> in ai within healthcare over the last few years has dramatically expanded, offering the potential to fundamentally alter how care is developed and delivered. A key area of this growth is in the fda approved, ai enabled medical devices.
▶ 1:48:48Just over a decade ago, there were just six approvals by the fda for ai enabled medical devices. Today, that number exceeds 600. Cms has begun to offer some level of coverage for software based technologies to support clinicians, and is currently seeking input as part of the proposed ci 26 physician fee schedule and outpatient payment rules.
▶ 1:49:14However, I believe that more can be done here to provide a sustainable and consistent pathway for these devices under medicare. This is why I've been working with senators round and heinrich on introducing a house version of the health tech investment act, which would do just that.
▶ 1:49:33Doctor melo and then doctor ibrahim, do you feel that the lack of a stable environment for reimbursement for these new devices makes it much more challenging for rural and small hospitals like those that I represent in pennsylvania's 13th congressional district, to invest and ensure that rural patients and communities have access to the latest innovation. First, doctor melo. >> I do.
▶ 1:49:59There's no question that when you're trying to just get through the day thinking about innovation and staying on the cutting edge of future care is not possible in the thought, space and resources that you have. So there are many things we can do to provide direct assistance to rural hospitals, both in the academic sector but also in the policy sector. And the reimbursement piece, as you said, is critical.
▶ 1:50:19I think the, you know, the fda example that you you began with is really illustrative because that is an area where despite regulation, despite clearance processes, we've gotten a lot of those innovations out broadly to hospitals. These are overwhelmingly radiological tools. Of the nearly 1000 applications approved by the fda, overwhelmingly these are radiology tools, and we've done a good job getting them out into practice. >> and you see that mostly those are in major metropolitan areas and not in rural areas.
▶ 1:50:51>> you know, again, it depends on whether there's a mechanism for billing for that extra. And it's really critical to remember the extra is not just the cost of buying the thing. The extra is the cost of training radiologists on how to use it and making sure that it's working well for everybody. And that's where rural hospital might be willing to adopt, but not do the work of keeping people as safe as they could be. >> doctor ibrahim, could you please comment? >> thank you so much for the thoughtful question, doctor, and I appreciate you bringing up rural health.
▶ 1:51:17Some of the most impressive improvements we've seen in stroke care from our technology have been in rural work. Um, that one of the references in interventional neuroradiology in my testimony was from texas showing that in rural texas, we were able to improve some of the time to treatment by 80 minutes in stroke patients. You are correct to identify the tension there, that when hospitals make a decision about whether or not to adopt the technology, the ability to get reimbursed for it is front of mind.
▶ 1:51:46If we are able to identify ways to reassure hospitals that there will not only be temporary payment, but sustained payment for technologies that have already gone through the rigor of the fda and gotten approved as safe and effective. Um, that would do a lot to help improve the adoption of these technologies. >> thank you. In my remaining time, I want to pivot to doctor toy. Um, we recognize the opportunities that ai has provided.
▶ 1:52:11It is loosely regulated usage in prior authorization. That is my concern. In fact, in march of 2025, the ama uncovered that over 60% of physicians are concerned that the expanded use of ai in prior auth is exacerbating the rate of incorrect care denials. Over 60% of doctors, given this concern of from physicians.
▶ 1:52:35Do you feel that I should be leveraged in the ai authorization prior authorization, decision making process? And if that answer is yes, what guardrails should be put up? >> thank you for the question there. So definitely in the case of prior auth, I think that right now at clover ourselves, we do not use ai in any form within the prior authorization process. >> but are you aware that many, many, many medicare advantage plans utilize and medicaid managed care plans do as well?
▶ 1:53:06>> absolutely. And I think that from our position, they should not be used in any kind. I should not be used to review, to deny. In any case, like that is not a good use of ai. I do not think ai is ready to do that right now, which I think is the problem you're addressing. There are use cases where it can reduce burden, accelerate getting to yes, faster. Those are good you places you can use ai, but not to deny care that. >> you feel that if ai does cause a denial or refusal of care, that there should be a person to person, a physician to physician evaluation?
▶ 1:53:36>> absolutely. A peer to peer is always appropriate. And that should be in all cases, maybe not even a denial if requested or if appropriate. A human clinician should always be the one making those decisions. >> thank you for your comments, madam chair. My time has expired and I yield back. >> gentleman yields back and I recognize representative from california for her five minutes of questions. >> thank you, madam chair.
▶ 1:53:59Um, this is a great conversation to be having, although to do it in a context of what's going on in this country is to turn a blind eye. You know, the trump administration has caused chaos at the centers for disease control. Under secretary kennedy's leadership, thousands of dedicated public health workers have been fired. Scientific experts like former cdc director, have been forced out, and hundreds of millions of dollars in critical research funds have been canceled.
▶ 1:54:26This week, eight former cdc directors who served under presidents from jimmy carter to donald trump took the extraordinary step of sounding the alarm in an op ed. Now this is it. The title says it all. It says we ran the cdc. Kennedy is endangering every americans health. Think about that for a moment.
▶ 1:54:53That the secretary of hhs is endangering the health of every american. They write that the administration has done is, quote, unlike anything we've ever seen at the agency. And unlike anything our country had ever experienced. End quote. They warned that americans are now less protected from cancer, heart disease and infectious disease. They note the cdc is being directed to downplay vaccines and pursue unproven treatments.
▶ 1:55:21I want to ask for unanimous consent to enter the former cdc director's op ed into the record. >> without objection. So ordered. >> I think we should be taking up legislation to protect workers at the cdc. My colleague, representative mcclellan, has a bill to prevent mass firings at hhs.
▶ 1:55:37Um, and I recently introduced a bill that would protect our public health workforce act to bring back the cdc staff that trump has fired instead of oversight hearings or legislative action to hold this administration, administration accountable. Republicans continue to pretend that nothing has happened. And so you can't go on without saying something. Um, at this hearing, doctor ibrahim, I want to turn your attention to ai and the topic at hand.
▶ 1:56:07Today, ai has shown real promise in healthcare, such as helping doctors catch devastating diseases like cancer earlier. But as of last year, over 600 ai applications approved by the fda out of over 600 applications approved by the fda, less than ten were eligible for reimbursement from the centers for medicare and medicaid services. That's less than 2%.
▶ 1:56:33Does that number concern you, or is that a good thing? >> thank you so much for raising that that point. And it has some historical context that the, um, role of the fda and the way it's been statutorily designed has disproportionately focused on safety and effectiveness and deferred the, um, decision of reimbursement to cms, which unfortunately creates almost reinventing the wheel, because a lot of the material they have to review and evaluate ends up being redundant.
▶ 1:57:02There have been some efforts to address that in parallel path reviews. Um, ultimately, I think what would be more efficient is if we focused the rigorous evaluation upfront by the data and if it got fda approval, that it would then be reimbursed by cms. >> well thank you. You know, first, when you look at the number, you think like, oh, geez, people are not getting access to these ai tools that could be, you know, life saving.
▶ 1:57:27But of course, we've heard today about other concerns about trust and regulation and whether, um, it's helpful or whether you need to have a plan in place. Um, so, um, doctor mello, what can congress do to make sure I healthcare tools are accessible to more people? >> yeah. So both of the things that you just said are true at once, right? That we are concerned that there's not enough access and that there are real, uh, you know, policy barriers to getting there?
▶ 1:57:55Uh, again, where I land is if I am a hospital, I'm doing radiological scans. What's going to move me to spend more on ai if there's no reimbursement? Maybe it's a substantial improvement in quality of care, but that sometimes is not documented. Maybe my radiologists are clamoring for it, but they know they're on the hook if anything goes wrong. So this trust deficit really matters.
▶ 1:58:19And the steps that I've outlined, including both modifications of fda review, modifications to reimbursement, but also institutional governance, that's the pathway forward. >> thank you, doctor mello. The use of ai can reduce workloads and improve efficiency in our health care system, but it can also make mistakes.
▶ 1:58:36Earlier this year, hhs secretary kennedy released a report that was based on incorrect information and made up sources because of ai, how can we ensure that ai is used responsibly and with full accountability in the healthcare system, especially when it affects people's trust in science and their health? >> I think it only begins by saying there's a human in the loop that's supposed to be reading that stuff. You have to create the conditions to make it possible for him or her to do that job in the setting where they're working.
▶ 1:59:05And right now, hospitals are not doing that. >> great. Thank you all. With that, I yield back. >> gentlewoman yields back. And I'll now recognize my friend from ohio representative balderson for his five minutes of questions. Madam chair. And thank you all for being here today.
▶ 1:59:23My first question to, uh, in your testimony, you mentioned that ai platforms like clover assistant can very quickly bring together fragmented patient data and integrate it into existing clinical workflows, even in practices that are still paper based. From your experience, what steps do you believe congress can do can take to support a wider adoption of ai enabled tools like yours, especially in rural areas where providers may lack access to modern ehr systems.
▶ 1:59:53>> thank you for the question there. I think that most important thing I noticed discussed at the congressional in congress is making sure that there is connectivity to the internet within any practice within america. Like, I know that's a focus already, but once we can connect physicians to the internet, we can then connect them to cloud based ai, and then we can also connect them to the vast amounts of data that we will bring online via interoperability. So I think those things all train together.
▶ 2:00:19And, uh, if we'll solve the connectivity problem, um, today by giving out devices to people like a connected ipads will help them work with their local telcos to connect. But if that could be solved, we can pretty much get rid of, uh, fix the rest. >> thank you. Follow up, Mr. toy. In your testimony, you also provided a good example as to how a lack of coordinated care can lead to miss warning signs. Much of my district is rural ohio, and so many of my constituents have complained about fragmented care.
▶ 2:00:46How can congress help accelerate connectivity and data sharing, especially between independent specialists and primary care providers, so that ai tools can deliver the full picture of patients health in real time, regardless of where care is delivered. >> yes, thank you for that.
▶ 2:01:02I think that a huge part of what we can do here is that connectivity to the internet and then within the healthcare ecosystem, making sure that the interoperability rules that are being passed right now and future rules are all being accelerated and enforced. We want more ehrs talking to each other. We want to collect the lab system to that data stream. We want the pharmacies connected, that data stream, and all of that can come together.
▶ 2:01:27And once we have that, and I think we're very close to that, we want to make that to be physician mediated, which it is right now. But I also encourage us to consider making it member, uh, patient mediated as well. A lot of the networks right now are for physicians to request data. And we also should allow patients to request their own data off those same networks. >> agree. Thank you very much. Um, my next question is for doctor ibrahim. Um, thank you for being here today.
▶ 2:01:53Um, doctor ibrahim, um, this ai platform, interogate, integrates ai to speed up diagnosis and enhance care coordination across specialties and physicians. Given your background in both clinical practice and health care delivery, how do you see this benefiting both physicians and patients, especially in rural and critical access hospitals? >> thank you for the question.
▶ 2:02:18One of our, um, important studies around improvement in care for rural patients comes from trihealth in the cincinnati area, um, in ohio. Um, one thing I've noticed in my own practice, we often have rural patients that may have a concerning sign on one of their imaging studies, and they get transferred 3 to 4 hours to a big academic center. We may not have the imaging, we end up rescanning them. Or if we have the imaging, we can look at it and say, actually reassure them and tell them it's okay.
▶ 2:02:45Um, it's hard to estimate the number of patients who've been transferred, only to give them reassurance and have them drive back home, um, at no small burden. So the ability for us to leverage this technology to share information, readily get expertise, eyes on some of this imaging before asking a patient to take on the travel burden would be significant.
▶ 2:03:04Um, there are also many conditions where the care can stay local, but if the expertise can be disseminated by sharing information, it could also help maintain the volume of care that stays local in rural communities. >> thank you. Is there any concern for er in using these technologies for patient care?
▶ 2:03:25>> I think we need to be incredibly humble that all of us here, including the people on the front line of doing this, are always learning something new about it and that as good as our models get, as we get more experience, they get better and we learn, um, we try to put several safeguards in place to try to mitigate and anticipate that. Um, you heard multiple times today about having humans in the loop.
▶ 2:03:45We also have additional monitoring safeguards that also alert us that if there's some signal change in how many things we're detecting, we start to go back to the providers and say, there's something different about the patients you're screening. Is there something different about the information you're giving us? Because the signals start to look different acutely. So we try to put multiple safeguards in place to prevent that. >> okay. Thank you all very much, Mr. parker, I apologize. I was going to have you next, but I'm out of time. So, madam chair, I yield back. >> okay.
▶ 2:04:15The gentleman yields back. And I recognize doctor schreier from washington for her five minutes. Questions? Thank you, madam chair. Uh. >> as a pediatrician, I have always relied on trusted experts at fda and cdc to help me make the best decisions possible when I am taking care of my patients and they are being purged.
▶ 2:04:37Um, madam chair, that is why I'm joining all my democratic colleagues in calling on this committee to hold an oversight hearing to investigate the firing of cdc director and the subsequent departure of several cdc officials who were unwilling to abandon science and bend a knee to our conspiracy theory driven hhs secretary rfk jr.
▶ 2:05:00We have in this committee oversight jurisdiction, and we cannot stand by while secretary kennedy discards decades of american medical advancement and jeopardizes all of our health for the sake of his own fringe anti-vaccine agenda and a very profitable anti-vaccine industry. Our children deserve better, we deserve better, and we have jurisdiction and responsibility for oversight.
▶ 2:05:31Speaking of children, I'd like to discuss ai tools that are currently used on children. According to the american college of radiology, over 200 fda approved ai tools are used for medical imaging, but only six are actually marketed for pediatric use after review by the fda. That is a staggering disparity in innovation between adult and pediatric care, and partly that's because it's much easier to develop ai imaging software for adults.
▶ 2:06:01It's trickier to do it for kids because they're not just little adults, as you said. Um, bone heart thymus images look different in infants, children, teens and adults, and normal lab results are also different ranges. When, uh, when you look at different ages.
▶ 2:06:19So we need just a lot more data to generate the volume of age specific, uh, uh, data that we'll need to generate this, uh, doctor mello, you know, that packard children's, where I trained is connected by a hallway to stanford medical center. And that kids and adults go to the exact same emergency department. And I just believe that children in that, er, should not get the short end of the stick.
▶ 2:06:47And, uh, we not only need more tools for pediatric use, but we also need the existing tools to be clearly labeled so that providers know if the software they're using is actually intended and safe for kids.
▶ 2:07:02Uh, doctor ibrahim, given the variability and complexities of pediatric care, um, but also acknowledging that we need this cutting edge technology that makes doctors better and the doctors make ai better, um, what can be done to incentivize and increase the development of ai tools for pediatric patients? >> thank you so much, doctor and congresswoman, for your question. And highlighting the differences between pediatric and adult patients.
▶ 2:07:31Our current product has been only on adult patients, recognizing that it is an entirely different specialty to be able to apply those tools. Underscoring much of it is the sharing of data. Um, I'm encouraged that the field of ai development has progressed in this space where we're able to share not the actual individual data, but the weighting of our algorithms from our data across institutions in a way that we can benefit from sharing the data without actually exchanging the individual
▶ 2:08:01Patient's data. Pediatrics. That's particularly important because just the sample size is much smaller. So incentives and even national efforts that would help institutions aggregate this information, that they could share it in a way that maintains privacy without the individual records being dispersed. But being able to share the weights of the algorithms would be incredibly helpful for adults, in fact, as well.
▶ 2:08:24>> and and do you have ways of not only incorporating the the pediatric radiologists at children's hospital, but also specifically pediatric radiologists at outside hospitals so that everybody is involved? >> that's a great that's a great point. I think one of the hidden benefits of ai is that the it infrastructure has had to improve significantly to enable the ai technology to do what it does. One of the things has just been the exchange of radiologic images.
▶ 2:08:51So we've had a number of hospitals say, thanks for being able to just share the images. We're not sure if we want to use your ai tool, but you at least got us to that part. >> so thank you. I'm going to just quickly, I don't have time to ask you the question, doctor mallow, but I just want to bring up the issue of physician liability that when you have ai giving one answer and a physician giving another answer and something goes wrong, I think we need to think hard about, uh, where the liability is.
▶ 2:09:17Is it on the ai company or is it on the physician and the toll that that takes on physicians also to be a checkpoint and also to just be a reviewer of, uh, loads of data? Thank you all for your testimony, I yield back. >> thank you. The gentleman yields back, and I'll now recognize, uh, my dear friend from iowa, doctor miller-meeks. >> uh, thank you, madam chairman. And I want to thank the witnesses for testifying before the subcommittee today.
▶ 2:09:46Um, three years ago, uh, we were, uh, asked as, uh, as a as a conference, uh, to be on certain task force. And so there was a healthy future task force, and I chaired the subcommittee on modernization, and with the chair of the subcommittee on health, and we focused on ai.
▶ 2:10:04Uh, data, data, interoperability, devices, technology, and how it can help with both access to care, affordability, prevention and better outcomes. So I'm delighted for this hearing today and for your testimonies. And earlier this summer, I hosted a bipartisan briefing to spotlight how artificial intelligence is transforming clinical practice and patient care.
▶ 2:10:32Um, and thank you, doctor toy, for assisting the provider, augmenting the provider rather than replacing providers or physicians. During the briefing, we heard from doctor michael abramoff from the university of iowa about how his company, digital diagnostics, is using ai to tackle diabetic retinopathy. In fact, it's the, I think, the first autonomous ai device to be approved by the fda. Uh, doctor abraham and his diagnostic device is now used in numerous healthcare systems around the country.
▶ 2:11:02And it underscores that ai powered diagnostic platforms can be deployed safely while improving patient outcomes and healthcare productivity. So, um, doctor toy, your testimony mentions the importance of interoperability and standardized data access for effective ai performance.
▶ 2:11:20How can congress help accelerate accelerate interoperability across federal and private healthcare systems in a way that encourages innovation but avoids imposing rigid, one size fits all technical mandates? >> yes. Thank you for the question there. Um, I think the key on interoperability is ai runs on data. And the better, more, uh, relevant the volume of data.
▶ 2:11:46And also like the more personalized that data is, the better the ai is going to be able to run. So it's critical, critical that we have built infrastructure around healthcare interoperability. The current rules that we have on the books right now, I think that, uh, it would be wonderful if we could focus on modernizing those particular areas. I would point everyone to, for example, hipaa, hipaa does, uh, you know, good work.
▶ 2:12:12The intention of it is very good, but it was originally developed in a almost pre-internet world, let alone pre ai world. And so the intentions of portability and the intentions of accountability and hipaa must remain, but we must update some of those regulations. I think otherwise they will start to hinder, uh, interoperability. >> so I think that ai has tremendous potential, uh, in healthcare, as I have previously mentioned, I'm going to discuss some of the dangers of ai.
▶ 2:12:39We know that older individuals have less trust, but that younger individuals who have grown up in a world connected remotely or by the internet have greater trust in these systems. Um, and we know that they're looking towards at the same time, uh, the internet and social media platforms to address loneliness and isolation and connectivity exacerbated by the pandemic. But they also it's a source of bullying and threats to them.
▶ 2:13:07I'd like to enter into the record an article august 26th by the new york times. A teen with suicidal chatgpt was the friend he confided in, um, the tragic story of adam rein, as reported by the new york times, has ignited a difficult but urgent conversation about the role of ai in mental health support. I'm not going to list everything because my time is running out. But his parents are now suing openai, alleging that the chat bot became his closest confidant and ultimately a contributor to his death.
▶ 2:13:36And after reading your testimony and we have very little time direct to consumer chat bots are making deceptive and dangerous claims, with one presenting itself as a psychologist that validated a user's violent thoughts. And where should the boundary lie? If you could be brief in your comments. >> well, again, I think that the boundary really lies in ensuring that these types of chat bots a don't misrepresent themselves, that they're not allowed to call themselves a licensed anything.
▶ 2:14:05We wouldn't want them to call themselves a licensed lawyer. I mean, we just we would not want that. Um, because it gives a sense of credibility that doesn't exist. I think that we also have to, um, build better chat bots. I mean, I think everybody deserves therapy. Everyone deserves human one on one therapy. That might not be what everybody needs. So we need to think about how can we actually leverage emerging technologies, personalize them in a way that actually reaches people where they're at?
▶ 2:14:34We have a system that waits until people get in crisis. >> so I'm going to I, I'm running out of time. But you mentioned the human connectivity and everybody that interaction and parental oversight. And I just want to emphasize that groups, programs, schools that are pushing to exclude parents prohibit informing parents of children's activities when at school. It actually undermines the very human connectivity and support from those groups and those parents and those families that our youth desperately need.
▶ 2:15:06Thank you. And I yield back. >> gentleman yields back. And I recognize representative trahan from massachusetts for her five minutes of questions. >> I want to thank the chair and certainly the witnesses here today. Uh, in normal times, a hearing on artificial intelligence can be, um, uh, used to advance health care access would be a welcome and an important conversation.
▶ 2:15:29And in those times, I'd be asking how we can harness ai for powerful uses, like improving medical imaging to detect diseases earlier and more accurately. But these aren't normal times.
▶ 2:15:43Uh, just weeks ago, this congress passed the largest health care cut in our country's history, and our federal public health agencies are in chaos with mass resignations, dismantled programs, and an exodus of scientists that leave us less prepared for outbreaks, less secure as a country, and an erosion of public trust in government and medical guidance. And it's all of this unfolds. You know, the secretary of health and human services is busy posting videos of himself drinking raw milk and competing in push up contests.
▶ 2:16:13Against this backdrop, my questions today will be about ai in the clinical setting, but just pointing out some patients won't even make it there because they're going to lose their coverage. If we want technology to improve health, we must start by ensuring that people are covered and history shows technology can either wrongfully take that coverage away or help keep people eligible and enrolled.
▶ 2:16:37By february 1st, 2024, more than 16 million americans had lost medicaid coverage as states sunset the continuous enrollment protections that were in place during covid 19. Nearly 70% of these losses were procedural, not due to changes in eligibility, but to missed paperwork or state software glitches.
▶ 2:17:02These cases are a stark reminder that technology can either amplify or improve systemic problems, and it's up to policymakers to direct responsible uses of technology, design appropriate safeguards, and conduct oversight when necessary.
▶ 2:17:19Now, with states required to implement burdensome medicaid paperwork requirements thanks to the republican big ugly bill, it is critical that technology is used to protect access to care for as many folks as possible. We know that these new requirements have led to the removal of tens of thousands of americans who are eligible for coverage in the past, and that's when these requirements were implemented for only a short time and for a small number of people.
▶ 2:17:48But republicans just took that model of burdening people with unnecessary paperwork to keep their health coverage, and they expanded it nationwide. States must use every tool to keep eligible people covered and ensure that automation safeguards coverage rather than triggers wrongful terminations. In tennessee and other states, automated eligibility systems have wrongfully cut medicaid for thousands who were still eligible.
▶ 2:18:15Tenncare connect, that which cost over $400 million, was meant to use income and health data to determine eligibility, but instead often mis loaded data assigned people to wrong households and made incorrect determinations, according to a federal court ruling.
▶ 2:18:32Doctor mello, as we consider ai's role in health programs, what safeguards should be in place so that technology doesn't duplicate the procedural disenrollments we saw during the medicaid unwinding? >> thank you for the question. The kind of situation that you described is, frankly, inexcusable, that there would be failure on that larger scale really raises fundamental questions about the amount of testing that was done before that system was rolled out and who was watching.
▶ 2:19:01So job one, I think, is going to use automation to enforce these disenrollments that there are humans monitoring, not particularly individual disenrollments alone, but also what is happening at the population scale. But as you say, I could be used to improve some of these problems. And so I my hope is that there will be innovation in the private sector that develops apps that enable enrollees to be put on alert before they are disenrolled.
▶ 2:19:28I think ai has a lot of promise for making sure people don't miss filing deadlines, and are aware when the paperwork that they've submitted is not going to be sufficient. >> I appreciate that. Um, and building on that, we've also seen that even when technology is working as intended, it's often placed on an on top of an overly complex reporting system. It can still cause harm.
▶ 2:19:49Um, automation is often sold as a silver bullet for eligibility determinations, but in practice, when it is layered on top of complicated reporting rules, it can speed up coverage losses for people who still qualify. So I appreciate the alerts. I'm wondering if we can if you could expound on how we design systems so that they clear barriers out of the way instead of building new ones for those who need care the most, I. >> think were the incentives there are. Collaborative design is the key there.
▶ 2:20:18If you involve advocates who understand what the major pitfalls and errors are that keep people wrongly off of medicaid rolls, people who are eligible under the rules that we've set up, that goes a long way to building ai that's designed to avoid replicating those problems. >> I appreciate your answers. Thank you I yield back. >> gentleman yields back. And I recognize, um, representative bentz from oregon for his five minutes. Questions? >> thank you, madam chair. Thank you all for being here.
▶ 2:20:47This is an extraordinarily interesting to me as a lawyer for a hospital for many years, uh, I obviously became extremely familiar with the standard of care for doctors. So, doctor melo, where would you put I were you on the witness stand? And we were asking you, um, if if the doctor involved in the malpractice case should have, could have did used I and go both directions on it, they did use it and it turned out wrong or they didn't use it and it turned out wrong. Tell me what? Tell me where you think it fits.
▶ 2:21:18>> thank you. It's an important question. And although I'm not a medical doctor, I'm a scholar of medical malpractice law. And we know that malpractice law, reasonableness of action is always the lodestar, even more so than custom. What your colleagues. >> thought it was, I thought it was whether who the lawyers were. That was generally the never mind. Go ahead. >> so the question is when is it reasonable for a physician to depart?
▶ 2:21:37And the problem with being able to answer that question in the courtroom is that physicians are provided with so little information by the tool that enables them to understand whether it's recommendation, which makes sense at a group level, applies well to this particular patient. So that's why I think there needs to be shared liability between physicians, hospitals and developers, because sometimes the errors that doctors make as a result of output could not reasonably have been foreseen by them, and therefore are not unreasonable for the physician. >> let's let's take it.
▶ 2:22:06Let's take it back to a point that was made earlier in the testimony you folks have given today, and that was the lack of accountability or the promise by those who are providing ai that is going to work or that meets certain standards. Don't you think that that the law of, uh, of you and I are now discussing these standards? Don't you think that creates some sort of accountability for those who are creating these ai systems?
▶ 2:22:34>> well, the difficulty will be, as you'll understand as a hospital attorney, that when you contract away that responsibility at the licensing stage, it eliminates a lot of avenues for recovery. And many of these licensing agreements that hospitals are now executing with developers have disclaimers up and down that hospitals have to accept in many cases. >> so are you suggesting that we do something about that here at the congressional level, that we say that's not something that they should be able to do?
▶ 2:23:00>> I, I don't understand why ai is treated differently from other products, where we don't allow makers to disclaim ordinary warranties. Why should that be? And if the answer is, there isn't a good reason for treating it differently than we ought to treat it the same. >> thank you. I want to shift to to Mr. toy and you indicated or someone forgive me, because it's been a long time since I've been waiting to ask my questions. I may have confused who it was that said what, but at what point do you think that the human becomes less analytically able than ai?
▶ 2:23:28Do you see that in the next two years, ten years? >> when I think that it depends on what kind of reasoning and what kind of information like, I think humans are already, um, the ai is probably already better at searching a large corpus of data and finding something like relevant to show to a human being. So searching through data, I think ai is very, very good at that. I think you're asking about reasoning in particular. I do not think that most human thinking will be replaced by ai anytime soon.
▶ 2:23:56I think that what will be replaced will be more, uh, perfunctory kind of tasks. And that's where we should be focused right now. >> thank you. Doctor. Right. You you said a phrase I thought was very interesting. Personal connections make us who we are. I have all these younger adults coming into my office, and I ask them how many hours they're spending on their cell phones, and it runs. Frankly, the average is about 12 hours a day. It's frightening.
▶ 2:24:20I would just ask you if you think algorithms have now replaced the human connection and and if so, why isn't this kind of a wish that we go back to some previous time, which I see. Unlikely. So where do you think this, uh, this human connection is? Personal connection is going to come from now that have we haven't we given it away? >> it's a really interesting question. Um, I would argue no, that we have not given it away.
▶ 2:24:50Um, I think that, um, what makes relationships unique are the reciprocity of it. There is no reciprocity when it comes to ai. That's partly why people like it, because they don't have to care about the ai and give back. But that's not a genuine relationship. And it does not build true intimacy. Ai knows things, but it doesn't understand. It doesn't understand you or why. >> I agree with all that.
▶ 2:25:19What are we going to do about it? >> I think we need to be much more intentional about ai literacy and technology literacy within schools. I think we need to train parents. I think that we need to incorporate ai literacy and the impact it can have developmentally within healthcare.
▶ 2:25:39I think we need to have, uh, public, uh, campaigns helping people understand how to use these tools appropriately and when to walk away. >> and I very much appreciate your testimony. And the entire panel's very interesting. Yield back. >> gentleman yields back. And I recognize representative from texas for his five minutes of question. >> uh, thank you, madam chair.
▶ 2:26:04Um, and, uh, before I get into the I want to begin by saying that I'm deeply disappointed and and frankly, fearful about the reckless direction that hhs is taking. Uh, under secretary rfk jr's failed leadership and the decision to include pulling the rug from under, uh, cancer vaccine research and putting america's lives directly at risk, in particular in illnesses like cancer, quite frankly, don't care about what political party you may happen to belong to.
▶ 2:26:32It's something that has touched literally every one in this committee and probably, uh, in this body. Uh, and yet the secretary is abandoning the millions of americans who will face a cancer, who will face cancer in the years to come. And we're entering a very dark chapter in american history, and I hope that this committee will devote some serious time to demanding some answers. Uh, but today, uh, hearing is also a matter of life and death. And we've seen some very disturbing reports.
▶ 2:27:00And when people talk about what they see as the existential threat, uh, and talk about things that scare them, I scares me more than any anything else, quite frankly, uh, we've seen disturbing reports about ai chat bots leading to real harm. Uh, and some stories are so shocking, they can honestly be hard to believe. Uh, I'd like to enter into the record for, uh, a washington post article titled instagram's chat bot uh, helped teen accounts plan suicides and parents can't disable it.
▶ 2:27:30Um, this article discusses, uh, instagram's, uh, chat bot, uh, meta ai that actually walked teens through how to kill themselves. And kids are turning to ai when they feel they have no one else to talk to, and they're being told that these bots are companions. Uh, and instead of guiding them towards parents, uh, teachers or doctors who could help them, the chat bot is actually helping them go through with the suicide. Uh, and we have to stop this. This is not rare.
▶ 2:28:00Uh, and it's not just children. Uh, it's vulnerable adults who are at risk. And chat bots have been have reinforced paranoid delusions, validated dangerous thoughts, and in some cases, contributed to self-harm or even violence.
▶ 2:28:14Uh, and we know that these, uh, tools, uh, that there are millions of americans already using them, uh, on the other hand, as doctor wright described in her opening testimony, there are ai programs deceptively marketed as mental health providers, and some even masquerade as licensed psychologists. And they're racking up millions of conversations with people, uh, while handing out reckless, harmful advice to people in crisis.
▶ 2:28:41Uh, again, this is deceptive, and it's a danger, uh, and it is a threat to public safety. And yet every single republican on this committee voted just a few months ago to prevent states from regulating these danger bots, uh, including the companion bots that kids can access, uh, and programs marketed to adults suffering from mental illness. And so we have to do something. Uh, we should not be discouraging states from saving kids lives.
▶ 2:29:05And, doctor wright, in your testimony, uh, you spoke about the danger of unregulated products given harmful advice to people searching for help and in some cases, treatment for their mental health. Why are children and adults in crisis, particularly those that are susceptible to harmful, uh, outputs of, uh, why are children and adults in crisis particularly susceptible to the harmful outputs of ai chatbots? >> I thank you for the question.
▶ 2:29:37Um, I think they're particularly vulnerable for a couple of reasons. One, adolescents and children are at a very different developmental stage. They're not, um, have the life experiences to be able to listen to their gut and know when something seems off. Um, and individuals in a vulnerable position, um, who are in uncertainty, want to seek out answers. And these chatbots are coded in a way to give them the answer that they want.
▶ 2:30:02They are unconditionally validating and reinforcing even harmful or unhealthy behaviors. Um, and they told you exactly what you want to hear over and over and over again. And so when people engage in these prolonged chats, so not a one off chat, but hours and hours and hours, we know that the algorithms a lose accuracy and become less safe.
▶ 2:30:28So what we need to do is somehow disincentivize these companies from a user engagement business model, where they are incentivized to continue to program in an addictive way. They don't have to do that. There are ways to not code these things addictively so that they could be more helpful. But unless there's some regulation that's encouraging them to do that, I don't see them doing it on their own. >> well, thank you very much. Thank you, madam chair. I yield back.
▶ 2:30:58>> we'll add your document to the record. Without objection. Thank you. Okay. I now recognize Mr. langworthy from new york for his five minutes of questions. >> thank you, madam chair. Uh, I is no longer a theory about the future. It's already transforming the way we deliver care and how we diagnose disease and how we use data to improve outcomes.
▶ 2:31:20Uh, in my district in western new york, in the rural southern tier, families are counting on a health care system that's more efficient, more affordable, and more responsive, uh, ai driven tech innovation and technology can help us meet those expectations by reducing administrative burdens, strengthening clinical decision making, and unlocking discoveries that once took years. In a matter of months. Uh, and with that, uh, Mr.
▶ 2:31:46Toy, as you know, medicare advantage has been at the forefront of developing ai tools to streamline beneficiary enrollment, manage costs of care, and reduce administrative burdens. And as medicare advantage continues to grow, both in new york and nationwide, more than half of new york's medicare beneficiaries about 2 million people, have chosen medicare advantage over fee for service medicare.
▶ 2:32:15And for them, medicare advantage is medicare. Uh, Mr. toy, can you speak to how these innovations in are being used in medicare advantage and even medicare more broadly by reducing administrative paperwork for providers and improving outcomes for patients? >> absolutely. Thank you for the question. I think the advantage of the, uh, no pun intended, the advantage of the medicare advantage model is that we can bring a lot of, uh, better outcomes.
▶ 2:32:44We can provide better data to doctors, we can actually provide better reimbursement to doctors as well. So a lot of the things we're discussing here at the, at the committee are things that we can, uh, actually put into production right away. And we have put into production right away.
▶ 2:32:59So people in medicare advantage can go to a doctor, they can have a doctor use an ai tool safely, be trained on that ai tool, know that that ai tool is helping them with not, uh, sort of like, uh, with the diseases that they actually have in their own personal life.
▶ 2:33:16On the main diseases of aging, such as, uh, pulmonary disease, such as kidney disease, such as diabetes, and know that their doctor is being reimbursed by that, that ma plan for the time that they're using to use that ai tool. Because that's something we can do. We must reimburse on the cms schedule. We can also reimburse above the cms schedule.
▶ 2:33:37And so we're already reimbursing primary care physicians above the regular medicare rate for the extra time that they're taking to use an ai tool. So their doctors are using it getting better outcomes and being reimbursed better. >> Mr. how does clover's use of ai inform your thinking on network design and delivering the best product that you can to seniors choosing to enroll in medicare advantage plan? >> uh, thank you for that as well.
▶ 2:34:05So as I said before, um, when we improve outcomes and when we're improving, uh, accelerating diabetes identification and accelerating diabetes management, when we are keeping kidney function at a higher level because management begins earlier, all of that results in a better medical loss ratio, which means that we are actually more efficient per unit dollar when we allocate dollars out to the medical network.
▶ 2:34:30What that also means is we can reinvest those, uh, that efficiency back into lowering out of pocket costs for patients. So our intense focus is to reduce those financial barriers to care, making care more accessible, more available. And that's powered by ai improving the care that's being delivered to the membership as well. >> thank you.
▶ 2:34:53Uh, there's also been a lot of critiques over the past year or so about medicare advantage plans, leveraging ai to make prior authorization determinations. Uh, does clover leverage ai in their utilization management process, and if so, is there a clinician or a human that is a check in that process? >> so all of our prior authorization is being done by a clinician who is working with the providers in our network in order to make those determinations.
▶ 2:35:22We are not using ai, uh, to to make any kind of utilization management or prior authorization, uh, determinations. I do think there's an opportunity to do that. But to accelerate getting to, yes, faster to to lower friction for our provider network to make it easier, as you were mentioning, to, to reduce the amount of time they're spending preparing documentation to apply for prior authorization, I think ai is going to have great use cases there, and that's where we should be heading. >> well thank you.
▶ 2:35:49I is not a replacement for doctors, nurses or caregivers, but a tool to help them deliver better care for more people. Uh, if we embrace it responsibly, uh, and have the right framework in place, we can strengthen our healthcare system and ensure those across the country see the full benefits of this innovation in this great country. And with that, I yield back, madam chair. >> the gentleman yields back. And I recognize representative fletcher from texas for her five minutes of questions. >> well thank you.
▶ 2:36:16Uh, thank you, madam chair, and thank you to all the witnesses for testifying today. Um, I do think that the opportunities, as well as the concerns about ai are important for our committee and the subcommittee to consider and for the congress to act on. And I appreciate your testimony today. Um, but like my colleagues, I am deeply concerned about the context in which we're having this hearing.
▶ 2:36:39Um, and this this conversation we really do have to have, in the context of 2025, both the rapid advancements in ai and the massive reductions in access to healthcare that the trump administration has set in motion with a rubber stamp from this congress since january, the administration has dismantled our government agencies and public health infrastructure. And this congress hasn't just ignored it, although it has.
▶ 2:37:08It has enabled it. This committee is included in that problem. From the mass layoffs to the firing of the vaccine advisory board, to the twitter firing non-firing of the cdc director and the mass resignations at the cdc just last week, uh, to the gutting of the substance abuse and mental health agency. Um, this administration has weakened our public health agencies.
▶ 2:37:32And at the same time, it has demanded that congress reduce access to care and efforts to make care affordable for all americans. And this committee has done so. We've sat through those markups and those votes. This committee has done no oversight on the issues. I just identified, um, and has held hearings and markups, reauthorizing programs that the administration is gutting.
▶ 2:37:57And so while there is an important opportunity here, and I really appreciate your testimony and your being here to help inform us, um, in this work, it's really important to know that it's happening in this environment and that this committee and this congress need to get serious about the things that you all are sharing with us. And with that said, uh, with kind of the time that I have, um, doctor melo, I do want to follow up on your exchange with ranking member pallone about prior authorization. Um, and on some of the issues we just heard about, um, as well with Mr.
▶ 2:38:27Toy in the last questions. Um, I represent a lot of physicians in houston, texas, and they and their patients tell me that prior authorization slows down access to appropriate care by forcing physicians to spend time, as we've heard about, um, and also just leading to wasted time, worse health outcomes. It's documented.
▶ 2:38:49And for years there's been bipartisan agreement, I think, in congress that, uh, medicare advantage prior authorization practices need to be reformed. We also heard about that from doctor john joyce. Um, and particularly when it comes to the use of ai to deny claims.
▶ 2:39:06And despite the consensus, as you all know, the trump administration announced the cms pilot program, the wasteful and inappropriate service reduction model, um, that implements prior authorization for a list of services for people enrolled in traditional medicare in six states, including my home state of texas, as well as arizona, new jersey, ohio, oklahoma, and washington. And, um, the way cms plans on carrying out this pilot is to contract with companies that will use ai to determine whether a procedure will be covered.
▶ 2:39:33Cms has said that the ultimate decision making authority will be a clinician, but it has also included financial incentives for these ai companies to reduce spending. And one way they can reduce spending is by denying care. And so they're going to receive a percentage of the savings from denying care to patients.
▶ 2:39:53And so instead of improving these prior authorization policies in medicare advantage that have been under scrutiny for years for denying critical medical care, the administration's really doubling down, um, as far as I can tell, and creating the same problems, um, for the traditional medicare program. And so doctor melo, in your view, is current federal law sufficient to regulate the use of ai in prior authorization? >> no, it is not.
▶ 2:40:24>> do you want to expand on that a little bit on how and what specific things that we could do on this committee and in congress with federal legislation to address, um, and ensure that I used in coverage determinations, doesn't deny or delay medically necessary care. >> well, I want to acknowledge that cms did some very good work on the regs in the final rule for use of pa in medicare advantage plans. Much better. Uh, there are two gaps.
▶ 2:40:49One is that it specifies that there needs to be human review before it is issued. No, no one argues about that. Everybody does that. But the question of what does that mean? There are no standards for what that means. And in my opinion, when you present a human with a packaged up, predetermined denial, nicely curated with all the reasons that human approaches the decision in a very different way than they would with a fresh pair of eyes and a non curated file.
▶ 2:41:15So it should mean that human isn't primed by a computer to to validate the ai decision. And the second gap is commercial plans. So aside, people who are in employer sponsored plans, commercial insurance plans who are non medicare don't have any of these protections that medicare advantage plans do. And those plans have no little or no reporting obligations, particularly for erisa plans.
▶ 2:41:39So finding ways to penetrate the commercial market and understand what they are doing, where ai is being used to get to yes, faster and where it's problematic, that requires additional action. >> okay. Well, thank you so much. I've gone over my time. I have one more question for you all. Submit for the record. And with that I will yield back. >> thank you gentleman yields back. And I recognize my good friend, another pharmacist in congress, uh, representative carter from georgia, for his five minutes of questions.
▶ 2:42:08>> well, thank you and thank all of you for being here. Very interesting discussion. In case you're wondering why I'm at the top of the dais and at the bottom of the questions, I was 40s late, so that explains that. But nevertheless, that's the rules. We have to abide by them. Mr.
▶ 2:42:22Parker, thank you for being here, having heard a lot from you, but I I've got a lot of questions for you as a fellow pharmacist and certainly want to know how I can help us, particularly when it comes to, to patient compliance and particularly when it comes to helping patients and making it easier for them. The process. Um, can you tell me how I help? Healthcare platforms can empower pharmacy.
▶ 2:42:49America's vast pharmacy providers, particularly our independent pharmacies, to serve patients. >> sure. I mean, I think maybe helpful to talk through the patient journey at general medicine in that context where, you know, prior to a visit with a patient, we are able to collect their full historical medical record. You use ai to analyze that and figure out where they might have care gaps and recommendations the clinician can provide.
▶ 2:43:14But there's no reason that information couldn't also be presented to a pharmacist to help them identify similar levels of care gaps. Vaccinations, other opportunities to help the patient improve their health. Um, and I think, you know, with general medicine, the ambition is that we're able to empower providers, not just our providers, but providers in the community, to provide that level of care. And I think it's a combination of robust access to to medical records based on all the work that's been done here, coupled with the intelligence of I.
▶ 2:43:45>> so obviously we'd have to have the patient's permission to use those medical records and that that would that would go without saying, but can you see it? Um, you know, one of the things that I've seen over my years of practicing pharmacy is how insurance companies and third party payers have, have really taken part, I can remember, and I'm telling my age here, but I can remember when, you know, everything was cash and and you just, you just got the lowest price you could get if you were a consumer.
▶ 2:44:13But now it's it's so much of it depends on your insurance and what's covered and what's not covered. So I can see it helping in the sense that before you even go, you would, um, before you even visit the pharmacy, you ought to know whether that this is covered or not by your insurance company. >> it's exactly how we think about it. So in reality, you should be able to see all of that information before you ever get to the pharmacy counter.
▶ 2:44:40Both the insurance price, the lowest cash price, whether it requires a prior authorization. So as a consumer, you're not sitting there trying to wrestle through this at the pharmacy counter. I think one interesting anecdote from my previous experience building out the pharmacy business at amazon is that by just showing them a clear insurance and cash price upfront, half the time the consumer was choosing to use cash because it ended up being cheaper than their benefit. And I think you see a very similar dynamic across the medical benefit.
▶ 2:45:07I don't know that the ratio is specific to each category yet, but it really is empowering to the consumer to have that information upfront and to be able to make an informed decision. >> and that that's going to improve patient compliance. Yep. You know, I can think of so many situations. One of my pharmacies, I had three pharmacies, but one of them was kind of in a rural area where people had to travel a long distance.
▶ 2:45:29And whenever they bring me a prescription and they want to get it filled, I have to tell them, I'm sorry, I got to call you doctor, because I got to get a prior approval. A lot of times they never come back, never come back and get the medication. And so patient compliance I would think would increased tremendously with this as well. >> yep, I totally agree. I mean I think you there's often times in the, in the amazon experience where something was $14 and required a prior authorization, like the amount of wasted energy that was going into that far, far exceeded the $14 cost that they could just pay out of pocket.
▶ 2:45:58Um, and so I think just pulling this forward into the customer experience as a tremendous impact. >> absolutely. Well, I'm glad to hear this, because this is this is something, obviously, that I've experienced over my career, my professional career, and in ways to improve it and make it, um, more streamlined, if you will. I think will will really help, will really result in better healthcare for americans. Um, real quickly, doctor ibrahim, I wanted to ask you, how can I accelerate innovation in drug development?
▶ 2:46:29>> thank you so much. I think, um, one of the themes that you have heard today is, um, ai helps in a lot of the informational challenges. So much of the drug development historically has been empirical. We've almost been lucky. We just try hundreds or thousands of things to get the one. That's right. Um, ai does a good job in pattern recognition, and to be able to predict of the thousand things that we were going to try, these ones may be more likely. So there's enormous potential in the early discovery phase to make it more efficient. >> good.
▶ 2:46:59Well, I certainly hope and I'm looking forward to this, particularly with, um, with ai because, um, you know, in my years of practicing pharmacy, I've seen nothing short of miracles as a result of research and development and drug development. And I hope that that's going to be the case with ai. Very good, very good. Panel. This has been very interesting. Thank you all for being here. And I yield back. >> gentleman yields back.
▶ 2:47:22And I recognize miss um ocasio-cortez from new york for her five minutes of questioning. >> thank you so much, madam chairwoman. >> and thank you to our witnesses. >> uh, for being here today. Uh. >> and I want to kind of dig into a topic that's come up several times over the course of this hearing, uh, which is this area of prior authorization.
▶ 2:47:45Now, generally, I like to take a step back so that folks at home can kind of understand what it is exactly that we're talking about here. I think most people hear the term prior authorization, and understandably, our eyes start glazing over in this bureaucratic language.
▶ 2:48:01But in the way that this affects our lives is that if someone is diagnosed with a condition, an illness, a disease, they require medication, treatment of some kind, they will their doctor may recommend it to them, and that doctor has to very often, um, depending on their insurance, ask their insurance for permission so that their insurance can say, we will cover this treatment or we will
▶ 2:48:32Not cover this treatment. And that is the process known as prior authorization. Is that in a fair description of it, doctor mello? >> it is. And I would just add that it also applies to procedures and surgeries. >> yes.
▶ 2:48:46Um, and prior authorization oftentimes can be a really big headache and pain point for patients and doctors, and especially if you have interacted with a for profit health insurance company, although all sorts of health insurance companies use prior authorization, you've likely dealt with this.
▶ 2:49:06And there have been famously, I think, um, videos that have circulated on social media of doctors trying to haggle with an insurance company, of everyday people trying to get navigate their chemotherapy or any other sort of condition with, um, the prior authorization system. And if you have a medicare advantage plan, which is the for profit version of medicare, you've definitely dealt with prior authorization.
▶ 2:49:33In fact, in 2023, medicare advantage for profit health insurers filed over 100 times more prior authorization requests than traditional medicare and medicare advantage plans, also deny care to patients at significantly higher rates than normal medicare advantage plans, deny as much as 16 times the normal rate.
▶ 2:50:01Does this sound consistent with some of what you've seen, doctor melo? >> yes, I believe that's what numerous reports have indicated. >> and this is all to restrict patient care because we have a profit margin to maintain. And many of those for profit insurers use unregulated and unsupervised ai models to review prior authorization requests.
▶ 2:50:26And I think it's very important that people understand that this is happening, that ai is being rolled out in the industry, so that when a doctor who has gone through four years of medical school, in addition to their additional training, say that a patient needs something and they submit a prior authorization request many times in ai model may deny them that care to a human being who
▶ 2:50:57Has gone through extensive medical training. Now, Mr. toy, you've said earlier in this hearing that you are the ceo of clover health. Um, but you've said in this hearing that clover health does not use ai in your company to implement prior authorizations. Correct? Correct. And but you have seen this happen in your industry. >> that's correct. >> and you disagree with this usage? Why? >> I disagree with it? Because it should always be a clinician who's making this decision.
▶ 2:51:27So I myself am as I said, I'm also a patient. I have had my own scans be denied through prior authorization, and I fortunately have the sophistication to navigate my own appeal, as you said, to to get there. But that is not a reasonable thing for, for for it to happen.
▶ 2:51:47>> and for a person that is diagnosed with a condition in which time is of the essence and an ai model has denied a trained clinician the ability to provide them care, this could potentially threaten a person's life, correct? >> absolutely. Timely care is critical in many cases. >> and so we are seeing certain healthcare companies.
▶ 2:52:11And on top of that, the trump administration has planned to launch a new program to expand the use of ai in prior authorizations, known as wiser beyond medicare advantage and into regular medicare recipients. Um, Mr. toy, it is not your company's plans to participate in the wiser program, correct? >> that's correct. We are not participating. >> and you have no plans to. >> and we have no plans.
▶ 2:52:41>> to thank you very much. And I yield back my time. >> thank you. Gentleman yields back. And I recognize my good friend from california, representative obernolte, for his five minutes of questions. >> well, thank you very much, madam chair, and thank you to our panelists. I've really enjoyed this panel on one of my favorite topics. Um, Mr. parker, I want to start with you. Uh, I really enjoyed your testimony about the impact that ai is going to have on healthcare delivery.
▶ 2:53:05And indeed, it mirrors the findings that we made in the ai task force that I had the honor of chairing last year. When we issued our report in december, we had a whole chapter on healthcare because honestly, there isn't a segment of the economy in the united states that's going to be more dramatically impacted by ai than healthcare.
▶ 2:53:23Uh, one thing I wanted to ask about, though, is that, uh, you testified that as you, uh, see these startups come through, that the most promising of them right now are the ones that use the application of ai to bring down administrative costs. Uh, and that actually mirrors what we found in our, uh, our task force report as well on healthcare. That's the seemingly the low hanging fruit that ai is going to bring first.
▶ 2:53:53Uh, but you have to admit that if you are a the average american, that's a little disappointing, right? You know, we've been telling people for years how healthcare is going to ai is going to aid in drug discovery, how it's going to aid in diagnosis, predictive diagnoses, where there's pattern recognition and talking about risks you didn't even know existed, uh, tailored drug therapies that are tailored to your genome. You know, all of these things are what we've been promising people that ai is going to do.
▶ 2:54:19And we're coming now to them with it's it's going to reduce your administrative costs. And that's what we're doing with it right now. Should people be disappointed about that? Or is this just the first of many things to come? >> yeah. Thank you for the question. And I think that is why I, you know, we're spending all of our energy on improving the customer experience and using ai in ways that create real, tangible value to the end patient. And I think I'm optimistic that those those new experiences will help people be healthier, help them save money.
▶ 2:54:49And it's why our implementations are so pragmatic, right? I think everyone's always talked about how you can't get a price for anything in healthcare. So solving that using ai is a very useful utility for the consumer. I think similarly with the care plans enabling consumers to be able to manage their health, understand what they should be doing as next steps is, to me, what gets me up in the morning is helping patients realize the value of ai in their everyday life. It's not to diminish the value that would be created on the administrative side. I think that's low hanging fruit.
▶ 2:55:20There's a lot of opportunity there. Um, but I'm personally most excited about the potential to impact the end. Patient. >> right. Well, let's not lose sight of that optimistic vision, because I fear that when we start talking about how we're using ai today to automate note taking, to automate billing, you know that that's going to distract away from what I could do. Uh, doctor mello, I was really interested in your testimony.
▶ 2:55:42You were talking about how do we get americans to trust the use of ai, especially in a domain as important to them as their health care? And you talked about a foundational trust deficit, which I very much agree with. That was one of our key findings, not just in healthcare, but over and over and over again, as we did our task force hearings last year, was the fact that americans really don't trust this new technology. And in some ways, they're right not to trust it because it is new.
▶ 2:56:11But, um, another thing that you said that I'm not sure I agree with is that the solution to this is government certification, uh, standards, settings, disclosure requirements. Uh, and I'm not sure.
▶ 2:56:27I mean, my understanding of why that trust deficit exists is because americans don't have a very realistic view of what ai is and what it isn't, what it does and what it doesn't do, the way it works, and what the real risks of ai are. And I don't think that, uh, short of educating them on those issues, that just telling them that, well, the government says it's safe and so you ought to trust it is going to do the trick.
▶ 2:56:54Would you, would you agree with that or would you disagree? >> I would, but to be clear, what I'm suggesting is that there be requirements that institutions do governance because hospitals and physician practices, they're the ones that are best situated to understand their patients, their staff and what can go wrong. I spend a lot of time interviewing patients as part of my work at stanford, including patients who are very sophisticated about ai.
▶ 2:57:13They're part of a learning community, and the message I consistently get from them is, we expect you to be looking at this stuff, especially when policymakers turn to things like consent and disclosure requirements in lieu of requiring things that will actually keep them safe. Can the government do this directly? No. But does it need to provide additional incentives and nudges for organizations to do it? Yes. >> yeah.
▶ 2:57:37I just want to make sure we don't conflate the two, because I think with absent that, that, uh, education, I think that foundational trust issue is going to continue to exist and it's going to be an impediment to deploying ai, not just in healthcare, but in a lot of different domains. Anyway, I could talk about this all day. Thank you very much to our panelists. I've really enjoyed the hearing I yield back. >> gentleman yields back. And I recognize my good friend from ohio, representative greg landsman for his five minutes. Questions.
▶ 2:58:06>> thank you, madam chairwoman and ranking member. Uh, degette for today's hearing. Uh, on the use of ai in our health care system. Obviously, as we've heard, there are so many opportunities to improve care and to lower costs. There are also enormous dangers, uh, associated with this. And I want to talk about the medicaid medicare piece.
▶ 2:58:35And, uh, this, uh, potential new program that the administration has announced. Uh, because ohio is one of the states. So this this is urgent for us.
▶ 2:58:48Uh, and I want to make sure that this committee, which I think will appreciate the need for some bipartisan pushback and guardrails here, uh, will act because it is going to be our seniors in ohio and in texas and the other states. Uh, that may get hurt. And we don't we don't want that to happen.
▶ 2:59:11Um, but as you all know, for for decades, tens of millions of americans have seniors have have paid for and relied on on medicare, uh, for their health care needs. Uh, the administration announced this new effort that will require pre-authorization pre-approval, uh, for procedures, uh, under medicare.
▶ 2:59:36Uh, and it's been referred to as the ai death panel in that you're taking what private health insurance companies do now in terms of denying care, care that ultimately leads to the loss of life, and you're handing it over as I, as I understand it, to a tech company that is going to use an ai model that has not
▶ 3:00:07Been vetted. Um, and to make matters worse, uh, the incentive is there to deny the claim you get more money if you're that ai tech, uh, company, if you're if you're able to deny more and more claims, that is going to lead to people getting hurt. Uh, and I do again, think there is a bipartisan desire to to ensure this doesn't happen. Uh, the.
▶ 3:00:36The doctor and the patient should determine what care is needed. And we all agree on that. Uh, and while I could provide some additional insight, I think doctor melo, you you talked a little bit about what needs to happen.
▶ 3:00:55Um, but I wanted to get your thoughts on the legal, moral and health challenges that this new program, uh, creates. >> as you've pointed out, the reimbursement model for these tech companies really stands the concept of shared savings on its head.
▶ 3:01:20You know, shared savings evolved in the medicare program to give health care organizations incentives to take better care of people. And then when they got better, if they saved money on expensive procedures that the health care organization got to save that money, this is flipped. Now they're going to make more money by denying care. And yes, there are safeguards there, procedural safeguards to prevent the worst abuses. But that is a very weird reimbursement model for what they're trying to do here.
▶ 3:01:44The other thing I would point out is there is some political genius in the architecture of this program in that they have chosen to begin with a small set of procedures that most people who do what I do for a living and assess health services would say we're providing too much of they're expensive, and there's a very thin to no evidence base. The question is, where does this program go in the future? And the problem is going to be that we don't have national coverage determinations to expand it in ways that can protect beneficiaries.
▶ 3:02:15>> yeah. Until we have those answers, until we have those assurances, I think the, the, uh, the pilot should be stopped and, and hopefully there'll be support to stop it. Uh, because you mentioned I want to just do this very quickly and others have agreed or reiterated that there should be disclosure of the product, full disclosure of the products, uh, being used seems smart enough that, uh, there's full vetting of those
▶ 3:02:45Products that an independent review board and maybe the joint commission, as you, as you mentioned, can serve as, as that review, uh, board deal with the liability questions and ensure that this doesn't touch, uh, quite frankly, anything that could harm patients. Uh, and until we have that, I think the, the, the program should be shut down. Should should they. And I'll help lead on that. Uh, sorry.
▶ 3:03:16My time ran out I yield back. >> gentleman yields back. And I recognize my good friend from florida representative for her five minutes of questions in a new mama. By the way. Congratulations. Thank you. >> a new mama who's actually got to go pump. So I'll make this quick. So thank you, madam chairwoman. And thank you to my wonderful colleagues for letting me jump the line and to our witnesses for being here.
▶ 3:03:42This is really important as we're grappling with all the new technology and all the opportunities, but also some of the challenges along with it. So, uh, we know that healthcare and I have incredible opportunities and potential. We're seeing it speed up drug development, cutting down on paperwork and giving the doctors and patients more time, which is really the upside, the good side that has been talked about at length here today. I really do appreciate a lot of the commentary on data privacy as well.
▶ 3:04:10So when we're talking about these potentials, we need to also examine the downsides, which has been if you have been highlighted here today, we all agree that I should never replace a doctor's judgment, and it shouldn't become another way for bureaucrats, uh, or swamp creatures, as I call them, or big corporations to pull strings. Patients deserve to know that their data is safe, their privacy is respected, and that their care is being guided by people, not just algorithms.
▶ 3:04:36So in my district, the university of florida, aka the everything school, uh, we are testing some of the most exciting applications of ai, from digital twins of icu units to patient level models that help predict the most effective treatments and outcomes in tough cases like cancer. We're showing what's possible. So I want to highlight that, but also talk to you guys about some data privacy issues that we're wrestling with. So I'm going to start with you, doctor ibrahim. As I mentioned, my district, university of florida gator nation.
▶ 3:05:05I keep having to say that, you know, lots of national titles, five in the country. Just putting that out there for my fellow sec friends here on the committee, uh, researchers are using digital twins to not just replicate hospital units like the icu and operating rooms, but also to model individual patient trajectories in complex cases like cancer or intensive care.
▶ 3:05:28How close are we to seeing this type of patient level digital twin technology guiding real time treatment decisions nationwide and beyond? Direct care what role do you see digital twins playing in education and training, particularly for providers in smaller or rural hospitals who don't have access to some of this cutting edge technology? >> thank you so much for the question, and I'll keep my big ten thoughts to myself.
▶ 3:05:53Um, the importance of this technology in rural communities is incredibly important. And in our own program, we've done discounted rates to help ensure that rural hospitals can participate. About 25% of our hospitals are in rural communities. Um, and they're often partnering with large communities or large, um, hospitals so they can have some coordination of care and triaging and sharing of expertise. Your question about our ability to identify better treatment plans for patients is super important.
▶ 3:06:23One of the things we've spent a lot of time on is finding diagnoses that potentially could have been caught sooner or weren't identified initially. One example was hypertrophic cardiomyopathy, where, um, we've looked back in old ekgs and rerun our models and found that for many patients, we could have diagnosed years earlier. So I think in the current state, we have the ability to diagnose things faster. Um, I think the future state you described that we could actually tailor a unique novel treatment.
▶ 3:06:52I think that's in its infancy. It's certainly in the pipeline in horizon. But I think even today we can point to that. We are able to diagnose people faster. Whether how well we can tailor the treatment is still, um, nascent, but coming very quickly. >> how soon do you expect the digital twin technology utilizing someone's medical history, their chart, their dna, biometrics? How how soon do you see that being regular and commonplace?
▶ 3:07:18>> I think for common conditions, because the the hitch of these models is having enough patients to train on. So for common conditions, I think that's very plausible within the next year or two. And there's probably some pilots that already exist for more rare conditions, nuanced novel cancers and things. I think we're more than a couple years away. Okay. >> I appreciate that. Thank you. I'm going to go to you, Mr. toy.
▶ 3:07:43Um, as we talked about, uf is pioneering immune system, digital twin to personalized treatment and developing synthetic data that allows ai to train without risking patient privacy. From your perspective, can synthetic data and patient level digital twins serve as a foundation for fair coding and care management across payers, and also what safeguards are needed to ensure that they don't become tools that privatize patient data or constrain physician judgment. >> yes, absolutely.
▶ 3:08:10So a very interesting question. So number one, doing the second part first, I think there's a there's a lot of privacy implications around digital twin technology. There's a lot of upside. But literally if you are making a digital twin of yourself, you probably want to know how that's going to be used. >> it's a little. >> scary right? So so it can be a little scary. So I think the answer is we haven't solved all of that yet. But there are technologies that will help with this.
▶ 3:08:34So I think that what's probably going to end up happening is you're not making a full twin, as in, it's literally my thing moving around and being used. That's going to have too many privacy implications, but it's going to be sort of like a collective model of people who are kind of like me that can then be analyzed and processed. So I don't have I have fewer privacy concerns. >> okay. And I know my time is expired, but I'll submit for the record some questions about de-anonymization of data and some of the honest broker questions that surround that. But thank you all so much for your time.
▶ 3:09:02And thank you to my colleagues again for your understanding. Thank you I yield. >> gentleman yields back. And I now recognize my friend from, uh, indiana representative hampton for her five minutes. Questions? >> thank you, madam chair. Um, I appreciate all the panelists testimony today. This has been a great hearing, and I think it's one that's much needed. I am particularly interested in hearing from doctor wright, and I want to just expound upon the written testimony that you've provided. You said ai is a tool built by humans for human systems.
▶ 3:09:32Therefore, a deep understanding of human cognition and behavior must be central to its deployment to ensure effective, ethical and equitable use of the technology. I couldn't agree more. I think the american psychological association probably will be playing a big role, hopefully, in ensuring that. And I appreciate that you've asked the ftc for, um, and the cpsc for an investigation into, um, unregulated products making claims of being a psychologist.
▶ 3:10:02Um, that's particularly dangerous. And, um, I joined your call for information from the ftc. Um, on the health disparities that you mentioned in your comments. I do think and I think Mr. landsman, um, alluded to this, too, that that that does create concerns for rationing care. Um, and it resurrects fears of death panels and other things if ai is in the mix. And I think that we have to be particularly concerned about that.
▶ 3:10:31But my focus of my comments is really on the impact on youth and children. Teens and children, um, you know, we know the reports and we've heard them today, the new york times, the wall street journal, uh, raising concerns about how children have been engaging with ai chat bots and companions.
▶ 3:10:47And while I'm encouraged that companies like, um, openai and meta have updated their policies, I think we have to be thoughtful and understanding how these technologies are being used, what risks they pose to children and teens, and and to our vulnerable americans. Uh, and, and what we should be thinking about as policy makers for those protections that need to be in place, uh, the risks, the benefits, how families can be equipped with, um, necessary tools.
▶ 3:11:15So, uh, you mentioned in your, um, in your comments that we should require age appropriate safeguards and robust data protections to support healthy development. When we passed the kids online privacy protection act in 1998, enacted in 2000, uh, we couldn't possibly have imagined the threats to children and teens that we see today through social media interactions and others.
▶ 3:11:41And I think that, um, this issue is, is exacerbated by access to chat bots, as we've recently seen. Um, so I guess several questions that I have for you. And then I'll let you expound on on this. It is, um, what should we be thinking about as policy makers to protect children from harm? Should we consider limits such as age restrictions on access to ai chatbot bots from children and teens?
▶ 3:12:09Uh, should we consider disruptors periodic interruptions within the chat bot that reminds the user that it's not a human, and that it's not a licensed mental health professional? What things are we not thinking about to further protect children from harm? Um, coppa 1.0 basically said an online standard that a 13 year old is an adult online.
▶ 3:12:32And, um, we know now that the highest rate of teen suicide among is among 15 year old males. Uh, so I think we have a lot of work to do in this arena. And I welcome your input. >> thank you so much for, um, that incredibly important question. Um, I think that there are probably several things we could try. We could try age restriction.
▶ 3:12:57We could try cutting people off the platform when they're on for a prolonged period of time. We could, um, roll out, you know, uh, parent programs where they link to the kids. We could, you know, as I mentioned earlier, have disclosures around, uh, detected suicidality. Really? The thing is, though, these are all empirical questions. What we actually need is independent research, looking at what the problem is and what the solutions ought to be.
▶ 3:13:26And those should be empirically driven, not just us kind of throwing spaghetti at the wall because it makes sense at the time. And so I think that really requires a greater investment in research of these products, and in particular research that happens before they go to the marketplace. >> yeah.
▶ 3:13:43I thank you so much for, for, um, for your note about that, because I do agree that we have to require independent testing on the harms, almost like sandboxing, um, some of these technologies before they go live to make sure that we're adequately protecting the public. I also agree with you that we need to be distributing educational materials to parents, uh, and maybe through our school systems about the dangers of ai chat bots. Thank you all for your testimony. I yield back. >> gentleman yields back.
▶ 3:14:11And I'll now recognize my friend from north dakota, uh, representative, for her five minutes of questions. >> thank you, madam chair. Oops. Chat box is talking to me right now. Sorry about that. Thank you all for your, um, participation today. It's been a great panel. Uh, there's been a lot of talk about context. I want to take an even bigger context than what's been referenced today and look more broadly.
▶ 3:14:39Um, 50 there's been a 50% increase in total national health expenditures since 2019. By 2028, projections call for health care spending to consume nearly 20% of our nation's gdp. Health insurance premiums are skyrocketing, and for all the additional money that we're spending, our actual health outcomes are not improving at all.
▶ 3:15:07In fact, many indicators suggest we're getting less healthy. So and what I've learned I've only been here for eight months is when you talk to the different players in healthcare industry, it's a lot of this. Somebody else is to blame. So enter I, which I think has very serious, you know, downfalls and potential risks, but also incredible upsides which many of you have highlighted today.
▶ 3:15:34So I want to drill into some of those in your conversation with Mr. landsman, he correctly outlined some concerns about the pre-auth. When I go home and talk to healthcare providers, this is their biggest issue, is how much time and expense and delays it takes to get through the pre-auth process. So how can we use ai and not just say ai is bad? It has to be, you know, it has to only be human beings.
▶ 3:16:03How can we use ai and provide the right guidelines and parameters to protect against the death panels and these blanket, you know, disapprovals, but use the best of it to help us improve the situation so that patients get get their care faster. And I'll turn that over to you, Mr. parker and Mr. toy. >> that's a great question. Thank you for the for the question.
▶ 3:16:28I mean, I think we've heard a lot of conversation today about prior authorization and automating both sides of that transaction. But we've heard a little conversation about how to actually help the consumer navigate that problem. Um, and very little sort of investment in energy into that. And so our frame of reference here is how do we enable the consumer to both understand what their options are, understand what the price is. And in the world where a prior auth is required, help them navigate that prior auth and ultimately get it approved if it's appropriate.
▶ 3:16:56And I think if we can reorient a bunch of this energy into that experience, and oftentimes that might mean that there's a there's a lower level of acuity, a lower level of care that's more appropriate. And the consumer might be happy to to utilize that level of care. They just don't have any of that information. They don't really understand, uh, how to navigate any of this. And so we're super focused on how do you help the consumer figure out the right level of care. And in a world where that does require a prior auth, help them get that prior auth approved. >> okay.
▶ 3:17:25And I'm going to come back to you with a follow up question after Mr. toy speaks. But the question will be what do we need in terms of rules, guidelines or incentives to make that happen? So Mr. toy. >> thank you for the question. So so something that, um, I think can definitely help is that, um, something I use my experience at clover for since we run a plan, is I actually advise on the marfan foundation for the rare disease. I advise our expert clinicians, and I advise patients on how to navigate their own prior off.
▶ 3:17:55So you can think about that as using expertise to say, this is how you talk to an insurance company to get them to cover your care. And so something we're looking at doing at on the clover side is providing actually that in a chatbot form so that people can say, hey, this is what I'm trying to get covered. What's the right way to get that covered? And then that will synthesize the data, all the documentation for you, package it up, phrase it in a way that makes it easy for the insurance company to say yes to, and I think that will greatly help. >> cool. >> Mr. Mr. parker, how do we incentivize this? >> I largely agree with that.
▶ 3:18:25I think we're we're building a very similar experience where we can take your full medical record, use ai to analyze it, help justify what procedure or medication you might need based on your clinical record. And then to Mr. toy's point package that up in a way that the insurance company can approve that. And I think this is a place where the ai is very capable and the the data is now available enough to make that experience really possible.
▶ 3:18:50Um, and I do think it's a place where if you can execute that customer experience, it will both be more convenient, more sort of understandable, and ultimately help them get these things approved. >> okay, I have 20s left. Does anyone else have short thoughts to offer? >> I'll just say there's enormous opportunity that we're trying to explore that the things that you need for prior authorization are predictable. They're public information. They're online. Wouldn't it be great to know that when a patient's in front of you?
▶ 3:19:19I just have to ask them these two more things, and that'll meet the ten things that I need to do. So we're exploring other real time ways to notify physicians of what else is left on that checklist so that before the patient goes home, and then you have to bring them back and require anything. So in the works. >> very good. Thank you all. Appreciate your insights today. >> okay. The gentleman yields back I ask unanimous consent to insert in the record the documents included on the staff hearing documents list and without objection, so ordered.
▶ 3:19:48And I'd like to thank all of the witnesses for being here today. We have talked about how good this panel has been, and members may have additional written questions for you all. And I'll remind members that they have ten business days to submit questions for the record. And I ask the witnesses to respond to questions promptly. Members should submit their questions by the close of business on wednesday, september 17th, and without objection, the subcommittee is adjourned. Thank you all.