AI in Health Care: Foundations, Myths, and Practical Applications

Ira Klein, MD, MBA, vice president of medical affairs for Tempus Labs, provides insight into the future of AI in health care and how it will help cut down on mundane tasks and increase oncologists’ efficiency.

Transcript:

Ira Klein, MD, MBA: Ira Klein, vice president medical affairs for payer relations, Tempus Labs.

There was a lot of anticipation around the session for this conference. How do you feel the session went? 

Dr Klein: I'm very pleased at how the session really rolled out, because one of the issues we discussed with the panelists was how to introduce an audience who had a variable level of understanding of artificial intelligence, what were the foundational principles of artificial intelligence, so that when we talked about things like, is the data clean, how are the queries, things like that, we wanted to ground the audience in some of the technical background and we had some really great panelists in that regard in terms of Will. Will, who cut his teeth in computers and AI and had a major role at Spotify in their AI product and then also Zach, who came from Sloan Kettering and has worked with AI in big data for clinical trials and so between Will and Zach, they gave the audience a great grounding in artificial intelligence going through neural networks, deep learning, what is natural language processing, what is machine learning, rolling up to how does artificial intelligence really work for both generalized data sets and he went into what the transformer operations were and then also what happens with curated data sets. So, we had a great technical ground. And beyond that we had a fantastic discussion on applications because our panelists came from a varied set of clinical backgrounds where responsibilities range anywhere from Dr. James Hamrick doing work at Flatiron for clinical trials and data mining, which is fascinating because we have the Mac Parallel at Tempus Labs where I talk to my people who do what James does and it's the same operation and then we had Amy Valley, who came from the GPO world and she sees population health data across multiple disease states and that was great and then we have Ed Rogers, who's leading the Pathways product for Elsevier in decision support and so you have a lot of different viewpoints on the uses of artificial intelligence as a tool to manage problems that we have right in front of us that involve data and decision making.

What are some of the more pervasive fictions concerning AI and health care?

Dr Klein: I think because there's been so much hype about AI in the general press, people tend to think that AI is a tool to manage problems that involve data and decision making, on the idea that AI gives fake answers or hallucinations for many problems and that is an issue that is one that requires a little bit of foundational knowledge in how AI works. The tool uses deep learning to associate ideas and then come to weighted concepts and that's the trend process and the summary of those weighted concept answers is really dependent on the data that it draws from and so if you're using, for example, in the public chat GPT, and you ask a question for which the data isn't complete or there's inaccurate data in the public space in a generalized language model, you might get an answer that isn’t fully accurate. However, that can be dramatically reduced if the data you are using is highly curated and reviewed and so you know that the assumptions and the weighted decisions that will be made by the AI are dependent on factual information, not information that is assumed or associated because there’s a gap in information.

Can you talk a little bit about the practical utility of AI in oncology in the next year and 5 years after that?

Dr Klein: Yes, I think this really is a great segue from the fiction discussion, because one of the fiction discussions is that AI will take over for mankind and we will all be displaced in our jobs. That is totally not true and the reason it’s not true is because, as someone recently told me, AI won’t take your job, what will take your job is a person who knows your job but also knows AI and that’s the key for what you want to think about for applications today and tomorrow. In the near future, we’ll see AI, take over, and assist in many of the mundane tasks in health care that we associate with friction generation, inefficiency, poor process management, operational nightmares, AI is actually, if we deploy it correctly, the right tool to deliver us from this administrative nightmare, which is really a lot of what’s bogging down health care, because I think that if you do the thought experiment of understanding where we could be if you reduced 30% of cost that is administrative waste and friction, re-deployed that money, and the people whose lives are being ruined because they are doing things like for example, physicians entering EMR data at night, the infamous pajama time misery. If you re-deploy all that and put it towards great patient care, you achieve what we all want which is the triple aim goal. You can get better individual care and outcomes, better patient experience, better population health outcomes, and if we add the fourth goal in there, we get better clinician satisfaction as well, so AI can do that and those mundane tasks that I think we’ll see being done is patient check in with complete insurance information and demographics, which can then fill in those fields, which can then be portished over to insurance type of needs on what’s your network, we can then develop specific tools for disease requests for services and procedures and prior authorization and pre-certification, we can use it as a summary tool if we’re careful in the curation, about sending over appropriate specialist notes, I spent my early career in clinical medicine as an assistant professor, and I ran a hospital program, and also did a lot of high risk surgery consults, the problems were always around grabbing the data, understanding the data, and summarizing the data. This is what artificial intelligence is great at, you just have to understand how to have the backend data curated or if need be, digitized, and that leads into the genomic side. So I think in the present moment, we’re gonna see a lot of mundane health care tasks, delegated to AI, with human overseers, but with a lot less friction in the system.

Is there anything else you would like to expand on?

Dr Klein: I would just like to say that the future is really boundless in what AI can do, so while today we are gonna deploy AI, to manage the pain points of administrating health care, I think the future is when we understand how to do the mechanics at the ground level, learn from it, and then start deploying AI in more population health solving of big issues type task management and that may become really more important as we think about the Medicare trust fund running out in 2030 or 2031, and then the government as the biggest insurer, has to worry about what it’s going to do and that may lead to really global and regional capitation, which is population health for a captive population, and what does it require to do that? Understanding the subgroups with big data, and now we have AI, so the future will hopefully be that AI helps us solve this crisis of paying for health care, and making sure it remains at high quality.