Artificial intelligence, COVID-19, and the future of pandemics [PODCAST]

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes

“Machine learning is only as good as the information provided to train the machine. Models trained on partial datasets can skew toward demographics that often turned up in the data—for example, Caucasians or men over 60. There is concern that “analyses based on faulty or biased algorithms could exacerbate existing racial gaps and other disparities in health care.” Already during the pandemic’s first waves, multiple AI systems used to classify x-rays have been found to show racial, gender, and socioeconomic biases.

Such bias could create a high potential for poor recommendations, including false positives and false negatives. It’s critical that system builders are able to explain and qualify their training data and that those who best understand AI-related system risks are the ones who influence health care systems or alter applications to mitigate AI-related harms.”

Richard E. Anderson is chairman and chief executive officer, The Doctors Company and leader, TDC Group of companies.

He shares his story and discusses his KevinMD article, “Artificial intelligence, COVID-19, and the future of pandemics.”

Did you enjoy today’s episode?

Please click here to leave a review for The Podcast by KevinMD. Subscribe on your favorite podcast app to get notified when a new episode comes out!

Do you know someone who might enjoy this episode? Share this episode to anyone who wants to hear health care stories filled with information, insight, and inspiration.

Hosted by Kevin Pho, MD, The Podcast by KevinMD shares the stories of the many who intersect with our health care system but are rarely heard from.

Transcript

Kevin Pho: Hi, and welcome to the show, where we share the stories of the many who intersect with our health care system but are rarely heard from. My name is Kevin Pho, founder and editor of KevinMD. Today on the show we have Richard Anderson. He’s an oncologist and chairman and chief executive officer of The Doctors Company, which is the nation’s largest physician-owned provider of insurance, risk management, and health care practice improvement solutions. He wrote the KevinMD article “Artificial intelligence, COVID-19, and the future of pandemics.” Richard, welcome to the show.

Richard E. Anderson: Thank you very much. Very nice to be with you.

Kevin Pho: So we’ll get into your article in a little bit, but first off, can you share your story and journey to where you are today?

Richard E. Anderson: I’m an oncologist. I practiced medicine in San Diego for 25 years at Scripps Hospital, where I served for a time as chief of medicine and senior oncologist, and then as a clinical professor of medicine at UCSD. But for the last 20 years or so, I’ve been chairman and chief executive officer of The Doctors Company, which is headquartered in Northern California. It is a national company insuring nearly 100,000 health care professionals all across the country.

Kevin Pho: What would you say are some of the biggest challenges transitioning from a clinical setting into the CEO role?

Richard E. Anderson: My journey is perhaps a little unusual in this regard, in that it was pure serendipity. I was very happy in clinical practice. I truly was one of the last of the dinosaurs: 22 of my 25 years were in solo practice of oncology, although I had a very collegial practice environment. The transition to The Doctors Company was really unplanned. I’ve been insured by The Doctors Company for professional liability insurance since 1977. I was asked to serve on some committees, and then I was elected to the board, became chairman of the board, and ultimately became CEO. But none of that was a planned evolution. All of my TDC activities occurred in conjunction with my full-time practice until I became chairman and then CEO, when the practice became part-time, and then I was really unable to continue an effective clinical practice given the responsibilities of the company.

But I’ve always thought that I was uniquely fortunate in the sense that the job I have at The Doctors Company is utterly informed by my years as a physician, my experience in practice, and a real understanding and indeed a love for clinical medicine. Obviously, The Doctors Company is about advancing, protecting, and rewarding the practice of good medicine. It’s hard to know what that is if you haven’t practiced it. So I feel uniquely fortunate in my transition to the company. I had been on the board for more than a decade before I became chairman, and even longer before I became CEO, so I had the great luxury of being able to learn intimately about the company and what my responsibilities at the company would be, and should be, and could be over a period of time. It’s been a wonderful journey for me, and I do feel uniquely fortunate to have combined both halves of this career. On the other hand, having said that, there are others who might look at my career and say, “Gee, you spent your life in cancer and malpractice? Really?” So there we are.

Kevin Pho: With the pandemic over the last year and a little bit, it’s affected so many areas of health care. Now, from your lens as the CEO of a medical liability company, how has the pandemic affected the medical liability business?

Richard E. Anderson: That’s a hard question to answer, and I’ll tell you why in just a second. But I think the dramatic thing that the pandemic has done is that it’s refocused everyone in the country on how critically important health care is and how extraordinary the value of great frontline medicine really is. With all of the frustration that’s attendant to our system of health care, or even our non-system of health care, and all of the concern, legitimate concern, about cost and access, which are very important issues, that debate tends to supersede the need for health care that we all have. And the pandemic, I mean, this is extraordinary. We’ve all heard these analogies before, but hundreds of thousands, really millions, of frontline clinicians marched into the fire and continue to march into the fire, putting their lives and even the lives of their families at risk in order to serve their patients. I think that’s a very, very important refocusing of the country’s intention on what the true meaning of health care really is.

In terms of the professional liability situation, the pandemic caused a decrease in the amount of liability as measured by the number of suits. In other words, we had roughly 20 percent fewer medical malpractice suits than we would have anticipated. Some of that undoubtedly was aided by the immunities, which were quite broad, but much of it was also caused by a slowdown. The legal system slowed down, just as everything else did. So the great unanswered question about liability is what happens this year and what happens next year. How many of the claims that we expected to see in 2020 but didn’t were merely deferred into 2021 and 2022, and how many of them actually were in fact avoided? We’ll know the answer to that, but we really won’t know it until we get closer to the end of this year, and perhaps at the beginning of next year.

Kevin Pho: All right, let’s transition and talk about the KevinMD article that you wrote. It’s titled “Artificial intelligence, COVID-19, and the future of pandemics.” Now, for those who didn’t get a chance to read that article, can you just walk my audience through it and share the story of why you decided to write it?

Richard E. Anderson: Artificial intelligence is everywhere. It’s become a buzzword, and it’s everywhere. I think there’s little doubt that there’s unlimited promise associated with AI, but there’s also, one could argue, an unlimited risk associated with AI. There’s an old saying that goes back many decades, which is that to err is human, but to really screw up requires a computer. Well, AI takes that to the next level, because AI is a black box. Even the designers, the engineers who design the algorithms, don’t ultimately know what the algorithms are going to produce, because even if they have a good statistical handle on what the outcomes would be in a defined set, the whole point is the continued learning, machine learning, and there’s no guarantee in what direction that evolves.

So it’s hard to know, even if the FDA, for example, approves a particular algorithm for one thing or another, and at the time of approval that makes perfect sense, good results, and so on and so forth. There are two issues that we don’t know about. Number one, what was the data set on which the algorithm trained? Was the data set appropriate? It’s all about the data set. Again, old cliches, but still relevant: garbage in, garbage out. It’s not so much garbage here, but the point is, if you’re looking at an unrepresentative cross-section of the population, for example, you may get an algorithm that’s blind to racial differences, or blind to socioeconomic differences, or blind to other things that it’s simply blind to out of inadvertence. So that’s definitely a concern, the concern that I think generically exists with AI.

Again, the promise is unlimited, but lots of things have unlimited promise, and usually the world tends to be more complicated than that. If we go back to recent precedents, electronic health care records are something we have needed in this country, a good electronic health care system, since at least the 1980s. We’ve all been begging for a decent electronic health care record going all the way back four decades or so. Well, we now have them. They’re now ubiquitous, and yet for most physicians they have failed utterly to meet their promise. They’ve not made practice easier. They’ve not granted better access to lots of clinical data, at least not universally.

I think in large systems that have invested literally hundreds of millions of dollars and perhaps a decade of time in designing a system, they have some really very fine medical records that are really meeting the promise. In most individual practices, I hear, not universally but almost universally, from our members how frustrating they are, how they’re working at night to feed the electronic medical record. That’s not what it’s supposed to be. We see it from every point of view: Unless the EHR is very sophisticated, the volume that’s produced by the records is incomprehensibly large, poorly organized, and endlessly repetitive, and actually, despite all the promise, it makes it harder to find the clinical threads that are necessary to provide care than where you used to be. Now, nobody’s nostalgic for paper records, right? We all lived through paper records. There’s no going back.

But my point is the promise of the EHR: We know we can get there, because there are great EHRs, so we know it’s achievable. But the point is, it’s taken an awfully long time, a lot of treasure, and a lot of sweat to get EHRs to where they need to be. I have no doubt that we’re going to run through the same kind of iterative journey with AI. We look at our surveys: One-third of our members report using AI to some degree in their practice already. But it’s hard to know what that really means, and it’s also hard to know if it’s true AI or if it’s just, “We have a program that helps us with triage or acuity determinations,” and so forth, and how fundamental that is to the practice. It’s very difficult to know where it goes.

And the rubber meets the road when we have to say, well, let me take one step back. The notion of AI is that it can provide better care than standard care, right? In other words, care can be enhanced because of the introduction of AI into analyzing more data than an individual clinician, or even an institution, is able to bring to bear out of human intelligence alone. OK, that’s the promise. But the analogy would be that AI is essentially a consultant, right? You’re still the doctor. The machine is not the doctor; you’re still the doctor. So the AI is basically a form of consultation. If the AI supports your notion of the standard of care, then great. You feel good, you can go ahead, and there’s no real disruption.

But if the AI comes up with a conclusion that’s different from the standard of care, that’s where the issue of liability and concern comes up. Because if the AI suggests a path that’s different from the standard of care and it turns out to be right, well, of course that’s great for the patient, and there’s no liability. But if it turns out to be wrong, who’s liable for that? Is it the algorithm, the developer of the algorithm? Is it the doctor who deviated from the standard of care, the recognized standard of care, in order to follow the black box? And how do you sue a black box? So all of these things remain to be worked out.

I think the pandemic underlines some other things as well. Certainly in our lifetimes, I don’t think we’ve ever really had a true sense of having potentially the whole country as a patient, which means there’s an enormous amount of data that needs to be crunched. We see the confusion, even with the new administration, which is infinitely more lucid than the prior administration in terms of recommendations, the approach, consistency, and using data to make reasonable suggestions. It changes all the time; at least the recommendations change all the time, at least as reported in the general press. And just the New York Times, in the course of the last week, says we’re on the cusp of an apocalypse, this is terrible. Today the news is, “Oh, it looks like we’re really turning a corner. We have 140 million Americans who are vaccinated. Everything is coming back down. It looks great.” Well, that’s very good news, but it’s in sharp contrast to what we were hearing as recently as last week about the threat of the variants and everything else.

So it would be great to have an encompassing system that could analyze this infinite amount of data and come up with more coherent recommendations. That would have the potential of looking at more data, number one, and number two, the algorithm presumably would be consistent in the way it analyzes it, so that we don’t have shifting recommendations day by day, or week by week, or month by month. I don’t think the virus really cares what region of the country it’s in, and yet we have 50 states and even hundreds, if not thousands, of counties issuing individual recommendations for their counties. It’s very, very confusing, and only half of it is half rational. So it would be great if we could have machine assistance in managing these kinds of global risks.

Kevin Pho: Sure. Whenever I think about that intersection between AI and medicine, I think about IBM’s Watson about five to 10 years ago and how it was used in a diagnostic realm in oncology. Now, today we’re talking in May of 2021. What are some specific ways that AI is currently being used in medicine?

Richard E. Anderson: Well, there are a number. For example, it’s being used in triage. It’s being used in emergency rooms to say, “Well, here’s a symptom constellation. What do we do with it? Here’s the case presentation. What should you be thinking about? What’s the triage for this patient? What’s the acuity for this patient?” In oncology, it’s being used actively, not universally, but actively in a number of practices, to constantly review all the clinical data from the EHR in the practice about individual patients and say, essentially, “This patient’s doing well,” or, where the algorithm is concerned, “This patient has to follow, this patient looks like they’re not doing as well, and they need to be seen before their next appointment, scheduled two weeks in the future.” So it’s being used to track the practice.

Ultimately, the promise of AI is that it can develop better therapeutic pathways. Right now, its main use seems to be an intense focus on where patients are along recognized clinical pathways, including, of course, diagnosis. Ultimately, the promise is that we’ll get more sophisticated, more consistent, or more personalized clinical pathways for individual patients based on more data. The promise of AI is to look at data outside the purview of the physician’s office, to integrate personal data from health fitness apps and health apps. So these social determinants of medicine, all those kinds of things, can be integrated potentially into an AI that’ll give a much more nuanced notion of who and what the patient actually is. This whole notion we’ve had for decades and decades, to focus on the whole patient, not just the clinical illness. That’s too much data right now, but the promise of AI is that we can actually use much more of that data than we’re using now to make better health care recommendations.

Kevin Pho: We’re talking to Richard Anderson. He’s an oncologist and also chairman and chief executive officer of The Doctors Company, the nation’s largest physician-owned provider of insurance, risk management, and health care practice improvement solutions. He wrote the KevinMD article “Artificial intelligence, COVID-19, and the future of pandemics.” Richard, you brought up an interesting point. When the clinician deviates from the suggestion that the AI program has, from a liability standpoint, what normally happens in those situations?

Richard E. Anderson: We don’t know, to be honest with you. In other words, we haven’t seen enough cases where that specific conflict has been at the crux of the case, and so I don’t know how the courts are going to respond to it. It’s one of the great unanswered questions. But I can say what we would expect, at least today, is that at the end of the day, the doctor is responsible for the care. In our system of litigation, whether it’s malpractice or not, we tend to litigate about adverse outcomes regardless of whether or not they’re malpractice, which is why we win 82 percent of all claims, winning being no indemnity paid, because most adverse events have nothing to do with malpractice.

So my guess is that for the foreseeable future, if the physician follows an algorithm that turns out to lead the physician in the wrong direction, the physician’s going to take the fall for that and is going to be held responsible for that. Ultimately, the law is going to have to understand more about algorithms, understand more about what an appropriate response to an algorithm is, and where the physician’s responsibility ends and where the algorithm starts. But I think the best way to think about it, at least today, is that AI is a consultant. The AI is not writing orders. The AI is not taking over for that physician. It’s not moving the physician out of place. It’s providing inputs for the physician to integrate. So I would think of AI as a consultant. We have to hope that that consultant’s giving you good information.

But again, not to denigrate the unlimited promise, the unique challenge of AI is that, yes, AI is a consultant, but it’s a consultant that you don’t know, with a disembodied voice, who can’t tell you what his reasoning is. That’s a tough way to rely on a consultancy, and it’s quite different from the way clinicians integrate consultants’ advice today. So it’s a real challenge. There’s going to be a decade, probably, even at the speed of technology, there’s going to be a decade of further sophistication of AI, so that promise gets greater and greater, as well as integration of AI, both into medicine first. And once it gets more integrated into medicine, it’ll have to be integrated into legal pathways as well.

Kevin Pho: So I’m an internal medicine physician, and I do primary care. Now, in your estimation, what would be the first effect of AI in my daily clinical life?

Richard E. Anderson: Well, it really depends, because there are so many individual AI programs. You could have an AI program that’s dedicated, let’s just say, to chest pain, or to all your cardiology patients or all your cardiac patients, looking at how your patient presents with chest pain and what you should do. Or alternatively, taking your whole practice, all your cardiac patients or all your diabetic patients: How are they doing? What medications do you have them on? What are their glucoses? What’s the status of their chest pain or exercise tolerance? And are there things that you should be concerned about that are not evident in your record at this point?

So, putting all that together, I think the major impact today is on diagnosis, number one, and number two, asking questions. For example, at UCSF, and at other institutions as well, a single endocrinologist can look at all of the diabetics hospitalized at UCSF and their glucoses and their insulin orders, and with the use of AI and the EHR, can have signaled any individual patient whose glucose and insulin orders seem to be out of balance. Now, again, the algorithm is not going to change the orders, but it’s going to bring it directly to the attention of a single endocrinologist, who can be looking through the algorithm at hundreds of patients at once. So again, the promise is extraordinary. On the other hand, as we alluded to earlier, if, God forbid, the algorithm gets it wrong, that’s a real problem, not for one patient. It’s a real problem for lots of patients.

Kevin Pho: And my final question: What’s your take-home message that you want to leave with the KevinMD audience?

Richard E. Anderson: Well, my take-home message is this is an exciting time to be a physician. The pandemic obviously put enormous stress, and continues to put enormous stress. We were all concerned about burnout before the pandemic. It was a constant subject, and if you look at various surveys of burnout, there’s no specialty where we’re less than 30 to 40 percent, no specialty where clinicians don’t report themselves as having significant symptoms of burnout. Then you put in the crucible of the pandemic, with the heroism and the remarkable accomplishments of clinical medicine in the face of this previously unknown threat to human life, and extraordinary pressure. So the burnout is worse than ever. That’s a real challenge.

On the other hand, we’re in an accelerant phase, a catalytic phase, of technology. EHRs are one example, but they took a long time. AI is going to be another example that’s extremely promising. Help is on the way. It has the promise of relieving physicians of a lot of the routine aspects of care. I could imagine, just as a simple example, you can have an AI thing with an insurance company that says, “Of course this chest X-ray is indicated. Don’t ask them.” That would save an awful lot of wear and tear on physicians’ offices. “Of course, yes, obviously we needed an MRI. Of course we do, so don’t need it; it’s approved.” Those kinds of things. So there’s potential, there’s an enormous amount of potential, but the potential is potential. The burnout and the pressures on clinical medicine are real, and they’re there today. So we need to close that gap. We need to make real the potential benefits of AI, and if we do, medical practice will be better for the doctor, and especially for the patient.

Kevin Pho: Richard, thank you so much for sharing your time and insight, and thanks again for being on the show.

Richard E. Anderson: Thank you very much. Pleasure to be with you.

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes