"Clinical, Ethical, and Legal Considerations of AI in Healthcare Innovation"

 Welcome to Season 5 of The Big Unlock. This season aims to carry forward Paddy’s legacy. I’m Rohit Mahajan, Managing Partner and C.E.O. at BigRio and now, Damo. Joining me for an exciting discussion in this episode, are my colleagues from Harvard Medical School. 

N. J. Udochi: I’m J. Udochi, a Board-certified family physician, Geriatrician, HIV specialist, and Addiction specialist. I’m also C.E.O. of the Summit Medical Group, a multi-specialty group in Columbia, Maryland.

As a physician who’s been practising for 35 years and a physician leader, I’m in the trenches, daily. While I’m knowledgeable about the pain points that hamper the practice of medicine, I’m also really passionate about finding solutions that will make our lives easier as we deliver the best care to our patients. I’m also obsessed with technology and how it can be used to improve access to and quality of care for everyone, especially underserved populations here in the United States and in my home continent of Africa. In fact, my entire project during our training at Harvard was focused on this subject matter, so, I sought other physician leaders whose goals aligned with mine. And voila, here we are! I’m really excited to be part of this first recording focused on AI and its importance in healthcare. 

Craig McLeod: Thank you, for having me today! I’m ecstatic to be a part of this podcast with my classmates from Harvard Medical School. I’m Craig McLeod, Founding Member of the McLeod Law Group, a healthcare attorney, and Founding Principal along with Tarul at Ellipsis Healthcare Leadership. I reside in Lexington, Kentucky. I have  20 plus years of experience operating in the healthcare space, delivering healthcare-related services to the elderly and the physically and mentally disabled. Our healthcare enterprise currently employs around 1,100 people on a daily basis and serves approximately 1,050 patients per day around the state of Kentucky. I’m here to offer legal input on AI and Machine Learning. 

Tarul Kode Tripathi: I’m Tarul. I’m a Pharmacist. I received my Doctorate in Pharmacy in 2002, but over the last ten years or so, I’ve been very passionate about and interested in healthcare leaderships. What brought us together was the Harvard Medical School Global Healthcare Leaders Program. It was around that time that I decided to leave my job of 15 years to focus on not only leadership, but effecting meaningful change in the healthcare ecosystem. So, we’re really excited to talk about the implications of AI from clinical, legal, ethical, and design standpoints, and why this matters in this moment. 

Q: I’d like to start with N.J.’s standpoint as a practicing MD. There are a series of questions that I will go through on why this matters in current practice. My first question for you, N.J., is how is the healthcare landscape changing for primary care provider practices? What are the drivers? How is value-based care impacting things as they stand? What motivated you to start your practice? 

N.J.: Thank you, Tarul! In recent years, the U.S. healthcare industry has been transitioning to a value-based healthcare system. That’s one that puts people and their health first, much before profits and shareholders. 

Largely, this change has been driven by patients being more proactive in their approach to healthcare. Also, it’s led a lot of health providers now to begin making changes to the way they deliver care and the platforms they work with. This is especially important in the chronic care space. 

CMS has seen the value of a collaborative approach to healthcare delivery, and they have pilots and demonstration programs in several states encouraging this approach to care delivery. In fact, in Maryland, we have the MDPCP program—the one we are participating in. We’ve just applied for that.  

These trends have been further accelerated by COVID-19. More recently, the issues of workforce shortage, mental health, physician and helpers’ workforce burnouts have sparked a lot of interest in trying to make sure that care delivery is modified such that we not only enhance it but also decrease physician burnout. With technology and innovation, especially Gen AI, ChatGPT, and LNMs, the future is in blending technology as an augmentor in healthcare provision. These are the drivers in healthcare today.  

For me, what motivated me to form my group was what I saw coming—the administrative burdens already placed already on primary care practices and the dying out of solo practices. We would want to use technology to improve healthcare, ease administrative burdens, and enable collaborations to provide and deliver the best care for patients. 

Q: Thank you, N.J.! What are the top concerns of primary care providers, physicians, and business owners in this landscape? Would they rather be acquired by hospital systems or private equity players? What does that look like for you?  

N.J.: I think that primary care physicians and business owners really are very focused on trying to survive the administrative burdens that have been posed now when you deliver healthcare. The big problems for us are not just burnout but also, as I mentioned earlier—workforce shortage. Many of us had to look to outsource to help deliver care that we needed and also, profitability. We were having a very difficult time trying to survive with these payers and what had been delivered in terms of payments. So, a lot of the providers, health groups, and even solo physicians were looking to find a way to be acquired by hospitals or private equity (PE) groups just to help them in terms of delivering the best care possible because of these administrative burdens.  

Q: From a place of values-based leadership, how do you think this impacts patients and the quality of care?  

N.J.: This is something that is really dear to me. When I talk to other physician leaders in the space and patients, what really stands out is that the quality of care that they feel is being delivered is not where it used to be. This is where it’s really important to blend technology with care delivery in a way that will be a win-win situation for both, providers and patients. This is where the value or the value proposition in this lies—finding a solution that will help decrease the administrative burdens that occur while improving the quality of and access to care for patients. 

If you have a solution that really meets those needs, then, that’s truly what providers are looking for. In the short run, it may seem more expensive. However, in the long run, this sort of a solution is what will lead to profitability. It’s all about really taking a taking a long-term look at it.  

Right now, a lot of physicians and providers in the health space are really burnt out because of all the administrative burdens that we face. This workforce burnout impacts the quality of care delivered—which is less than you would want. Trying to find or use technology that will help solve these pain points is where we need to be. 

Q: You’ve actually answered my last question on how technology could help reduce the administrative burden and reduce burnout. Would you like to add anything else? 

N.J.: I would like to add a few things.  

It’s really exciting because what we have now is Gen AI and these Large Language Models. You’ve also heard about ChatGPT. I remember being in a conference in San Francisco early on this year, and I was able to test drive the Doximity GPT and its application in solving authorization problems for physicians. It was wonderful. A lot of the physicians there were very excited because this has been one of the pain points that in care delivery. Using technology to solve these administrative burdens is one way a solution can help improve care delivery—it will leave physicians and other health workers more time to actually deliver the care that patients need. 

There are so many ways in which technology can even add value at the front-end or the back-end, in terms of scheduling appointments or even enabling patient engagement. Rohit was speaking about trying to help with care management but there are just so many ways, even in virtual care—we’re all very familiar with telemedicine, whether it’s in audio or in video format—with reimbursement from the government and other payers. This has really taken off, and it’s just about finding ways in which to improve access to care for all populations, especially those living with chronic diseases.  

Remote Patient Monitoring and Chronic Care Management are all areas that really lend themselves to improvements in technology and care delivery. So I’m really excited about this. It is solutions that can actually combine both, that will be the best for the future. Any company that can bridge these two gaps will be the companies that really make it in future. 

Q: There are many providers who are trying to talk about how AI and technology can help benefit care delivery. What are the ethical standards? What should we be watching out for? What makes for a good program? 

Tarul: Where I’ll start, especially after you’ve given us some very thoughtful comments from the clinical lens is, from the high level with what is the current state of artificial intelligence in healthcare innovation and why is it such a big topic.  

We are definitely in a significant moment when it comes to healthcare technology specifically powered by AI. By  being aware of and up-to-date with the implications—clinically, ethically, and legally—we can ensure that we’re putting patients at the centre of our care which is obviously the MVP for all of us.  

AI is a term that’s applied to machine or software, and it refers to the technology’s capability of simulating intelligent human behavior, which includes instantaneous calculations, problem solving, and evaluation of new data based on previously assessed data. That is what we mean by Generative AI. 

As you mentioned, AI applications in healthcare have literally changed the medical field, including Imaging, Electronic Medical Records (EMR), lab diagnoses, treatment, augmenting the intelligence of physicians and providers in a variety of different settings, new drug discovery, providing preventative and precision medicine, biological extensive data analyses, speeding up processes, data storage, and enabling access for health organizations, at large. So AI and Generative AI are creating massive efficiencies for providers and patients, alike.  

With that, I’m going to turn it over to Craig for some of his perspectives, not only from a legal standpoint, but why we should all be sitting up and paying attention to this topic. 

Craig: I’m going to start from a 10,000 foot perspective and a historical perspective.  

Historically, the evolution of technology has been based on a concept or principle called Moore’s Law. In 1965, Moore’s Law was put forth by Gordon Moore, who was Co-founder of Fairchild Semiconductor and Intel. He was the C.E.O. of Intel—and everyone that has a computer, cell phone, PDA, knows or should know who Intel is.  

Mr. Moore observed that the number of chips would double each year and that would turn into double processing power. This was on a yearly basis from 1965. This law stayed pretty intact until 1975 when it doubled to reflect that the doubling of chips per integrated circuit and doubling of the resulting processing power would move to every other year. Moore also established his second law, also known as Rocks Law, which stated that, even though the size of the chips on the relevant integrated circuit was reduced and the processing power doubled, the cost of the physical plant and the necessary R& D related to the advancement would increase exponentially over the same time period.  

The reason why this is important is, it creates pricing issues which are inevitably passed on to the ultimate consumer which is either a business model or the personal consumer. Staying with Moore, his laws are based fundamentally on the historical norms of manufacturing, supply chain principles, as well as basics of economics and supply and demand considerations. There’s always physical work necessary to maintain the underlying basis of Moore’s law. However, the explosion of AI—and I use AI to include Generative AI and machine learning—over the past year or so, has made these traditional considerations less of a factor to predict and forecast the increase of computing power and evolving technologies. 

As AI and machine learning are not customarily limited by these physical factors as the processing manufacturing principles due to the advancement of self-learning that’s done internally and within the relevant AI machine learning systems, their elimination allows for the exponential increase in advancement of the capacity of the relevant AI and ML systems that’s assuming that the information fed into the AI and ML systems are accurate and factual. This requirement to effectively process information is being interfered with intentionally, specifically by the artistic community—and this is with regard to the Nightshade and Glaze programs that are being implemented by the University of Chicago. More about this coming up on my next statement and probably, future podcasts.  

Q: What are the best practices in AI innovation and implementation?  

Tarul: So AI is not going to replace clinicians, but it will advance caregiving capabilities. Clinicians must increasingly rely on human skills such as, empathy and culturally-relevant care. For example, we have to train our providers— MDs, nurses, all caregivers—in all different modalities in meeting patients where they are. This real world data feeds the algorithms that AI will influence and ultimately use to improve care and efficiencies. For example, in marginalized populations, we have to do a better job of understanding the factors and interventions that will close gaps in healthcare disparities.  

Another example is trauma informed care which needs unique ways to speak with and treat patients who have lived through gender-based or other forms of violence. There are some really interesting discussions going on all over the country on how to carry out this care. We all must sit up and ensure that we’re paying attention. These intentional practices will definitely influence health equity as AI technology advances. Otherwise, we’re at risk of leaving the same communities who have historically been underserved behind, again and again. 

If AI is viewed as critical to healthcare’s future, then, a diverse group of stakeholders must be engaged in the dialogue. So a group like ours—data scientists, engineers, clinicians, patient advocates, ethicists, economists, and policy makers—have to come together and hold honest and wider discussions around how to implement ethical and equitable AI. 

Q: What are the legal considerations and why it matters in terms of risk avoidance, management, aversion etc.?  

Craig: AI and machine learning are evolving very quickly but as everyone knows, the legal field and the resulting legal decisions and lawsuits are so slow. This is a very new topic that we as attorneys are trying to figure out how to speed our pace of evaluation and bring these to light to expedite these type of decisions. The key here in understanding AI and machine learning legal perspectives is comprehending where and how the protected and confidential information and or the data is stored, utilized, and integrated into local or and/or hosted on external systems or platforms. 

In law, usually confidential or protected information is just that until it’s not. Where you get to the “it’s not,” is voluntary disclosure, negligent disclosure, or for an authorization by an agent or principal of a business, or him or herself donating or allowing this information to be imported into an AI or ML system. That brings us to confidentiality and trade secrets. I’m using this generically to mean “any confidential protected information” includes protected health information under the U.S. Code. This states that an owner of a trade secret must take reasonable steps and measures to keep such information secret. What that means is, you can’t go out and claim you have a trade secret or protected or confidential information and then, put it on a billboard somewhere. That’s the historical viewpoint. But we’re not in the 1950s. We’re not in Cleaverville, anymore. This is 2023 where everything happens over a computer. 

Information is stored everywhere. What happens when an employee or an agent of this information—a steward or a custodian of a company or an individual—just happens to link an information source or a depository containing this great secret or protected health information or any other confidential information? It doesn’t even have to be intentional. It could be that they unintentionally and even negligently link this information or this platform to an AI or machine learning system. Typically, when you feed information into the AI and machine learning system, it doesn’t really care where the information comes from because it’s utilizing it. But once those processes become active, there’s no going back. Once the trade secret is input into this AI and machine learning system, it’s more than likely your trade secret will be considered public domain. Anyone can use it for any reason at all. There’s about four or five big ticket lawsuits going on right now around the United States. If you read the initial briefs and decisions, they’re pretty much in lockstep with each other. The courts are saying, “Hey! You gave it up. You fed this into and made it part of the public domain.” You’ve lost any protection that the courts can give  you.  

That leads us to patient and data privacy. So everyone knows how important the Protected Health Information (PHI) is from a patient’s healthcare and societal positions. However, one problem immediately arises when the AI systems become aware of this data. As I just brought to light, how does the AI system know that this information is still confidential? How does it know that some of this information that was confidential hasn’t been made un-confidential or public by release or by a court order? It doesn’t. This is why it’s so critical to safeguard information from an AI or machine learning environment.  

Another issue that I want to bring to light is the facial recognition technology that may be hijacked for improper, immoral, or unethical purposes. The first thing that came to my mind is it’s akin to the past concerns of DNA information and the sharing of the same on Ancestry.com and 23andMe because of the notion that this information really is the last line that police agencies can utilize to find and crack cold cases etc.  Ultimately, the jury is still out on how to utilize this DNA information and how it is being utilized after being deposited on a third party site.  

I mentioned earlier about the bad actors and the poison pills. There are some participants in the space now that are attempting to mask or thwart AI machine learning’s ability to learn to do just that. Specifically, I’ve been reading and studying the Glaze and the Nightshade programs from the University of Chicago. The baseline notion here is, these artists that are seeking protection of their work. They’re artistic and into music and things like that. They shouldn’t have their work stolen, which is serious and valid concern. I think, we can all agree on that. But AI doesn’t seem to steal this information. They’re just utilizing it to learn for future endeavors. My initial observation on this position by the artists utilizing Glaze and Nightshade was, “Did they or did they not go to art school? Did they or they not learn from Van Gogh and Renoir? Didn’t they, weren’t they influenced by those artists?” I doubt that they gave credit or anything to those artists, but they learned from them. They created their own style. They didn’t steal it. They just were influenced by their artwork. So it seems to me that the artists utilizing these two programs sort of want their cake and eat it too. 

Q: How does all of these that we’re discussing influences the design best practices.

Rohit: As N.J., Craig, and you mentioned, this is a revolution. Today, AI is becoming household name with ChatGPT and that has reached inflection point. Everyone’s using it for some purpose or the other. It took a long time. And AI has been around. In fact, I’ve been playing with AI since my engineering school days way back when, but it has now become ubiquitous to the point where it’s easily available. One can log in and use it especially those Large Language Models and the applications which are immense, very broad, and across industry segments. 

However, in the healthcare industry segment, there are obviously privacy concerns and patient lives at stake. So, when we design AI systems, we do have to keep in mind all the various aspects that we discussed here, today—the ethical, the moral, the patient-centric etc. 

I’ll illustrate it by an example of a project that we are currently doing for one of our clients. Hopefully, that will bring to life some of the design approaches. We are working with this very innovative digital health company that N.J. mentioned I’d shared with her. We are working on building a care planning tool and platform applicable to chronic diseases and patients with multiple disease conditions.  

The challenge was—how do you build a wall so that the Large Language Model is not infringing on any private data of the patients?  

There are now possibilities, especially with large tech companies which will allow you to create these walls so you can stay in the swim lane where you’re building a Large Language Model, which is custom and learning from your proprietary data. It is staying in that silo. It’s not getting externally released, at least outside of that wall. These are some of the design conditions when we approach projects.  

Typically, you look for a POC—a proof of concept—and then, as you build it out, you put it in front of the stakeholders so they can test it. Usually we’ve also found that even in this project, there is a clinician in the loop. So, we still need that certified, qualified human being—a nursing staff or a physician—in the loop to look at the output of this Large Language Model and like sign off on it before it goes into the next step of being shared with the patient.  

This whole landscape is changing very fast. There is literally innovation happening almost every day. It’s explosive. I don’t know which law—Moore’s law or AI law or what we should call it—but things are changing very fast. In the company, we have some very qualified data scientists who keep up-to-date with all the current literature. That is how we are able to bring to bear our expertise and technology to create these robust POCs and pilot projects for our clients. 

Q: As we wrap up our discussion, now is the time for the “so what?” What should our multidisciplinary stakeholders take away from this podcast? Where do we start? What actions should be taken? Rohit, I’ll start with you and then, N.J. can describe how advances in this space can improve the lives of providers and patients alike using a real world example. 

Rohit: I’ll chime in with a very small example. I think this stems from a recent discussion I was having with someone who’s at a very large healthcare provider system in New York. He was saying, “Rohit, I have so much data in my repository. I don’t know what to do with it.” I think, that is really the starting point.  

Organizations and providers have data sets and that is a gold mine, right? How do you leverage that data set is where AI and Gen AI comes into play because AI is good at making predictions. So, if you have data sets, especially those which are retrospective, longitudinal, and immensely powerful, these can be a base for building AI and Gen AI models which can truly help with patient care. In my mind, for any AI project, everyone knows 80 percent is data engineering and 20 percent is data science. That still remains true. 

Q: From a provider and a practicing MD’s standpoint, what do you think people in this space need to be doing in terms of action? 

N.J.: It’s really exciting because the healthcare industry and providers produce a lot of data like Rohit has said. These massive amounts of data from multiple sources is exactly where AI and machine learning can add value. This is especially in healthcare delivery. I’ve talked so many times about this that there’s no day I don’t wake up trying to figure out how to solve the administrative burdens and improve healthcare delivery here in the United States and around the world. So, it’s really about helping solve those pain points.  

In healthcare, AI can add value in healthcare diagnostics. It’s been very valuable for many of the providers in Radiology, Pathology, and inpatient monitoring. Also, we have AI that has been—and many tech companies have also helped in—improving care delivery in terms of predictive modeling. We talked about virtual care—telemedicine and wearable devices—there are so many areas in which AI can really add value.  

But really, the technological solution that we want is one that really can blend the pain points that we experience in care delivery. A solution that can solve those problems and at the same time improve patient outcomes, enhance access to care, and ensure profitability for health systems, provider groups, and organizations. That is what the healthcare community is looking forward to.   

Like I said earlier, in the conference I had attended in San Francisco, there was so much excitement with Doximity GPT because we could really use it right there and and solve the prior authorization problem on a daily basis. There are so many problems that we have, whether it’s at the front-end of care delivery or at the back-end. The value add will be in trying to use AI and LLMs to solve the problems that we have in care delivery when we see patients.  

We talked before about the back-end in terms of patient coding in the Revenue Cycle Management (RCM) process, too. That is another area where technological advancements are needed. Perhaps startups can focus on this in the healthcare space because when you combine both the front- and back-ends to solve those problems, there is where you’re going to see a lot of excitement.  

In the end, the problems that we’re experiencing as providers—the burnout, physician charting, in vitro scribing—are all areas where physicians are looking forward to using AI to try and augment the care that they provide on a daily basis. 

Rohit: Thank you for the excellent panel discussion, N.J., Tarul, and Craig! It is great to get together, virtually. I would like to end by giving a shout-out for a book, Quantum Care, that I recently wrote and which has been published. It is a deep dive into AI for health research and delivery.