[This transcript is auto-generated and unedited.]
[Kirsten Bibbins-Domingo] Hello. I'm Dr. Kirsten Bibbins-Domingo. I'm the Editor in Chief of JAMA and the JAMA Network. And I'm joined today by three of my colleagues and authors in our Diagnostic Excellence series that concludes this month in JAMA. Over the past year, JAMA has published more than 20 scholarly articles on the subject of how we make diagnoses now and in the future across a range of specific focus areas. These articles are centered on the theme of diagnostic excellence that is, the optimal process to obtain an accurate and precise explanation of a patient's condition. The optimal diagnostic process would be timely, cost-effective, convenient and understandable to the patient, and an accurate and precise diagnosis is important because it leads to better choices and treatments. My colleagues today are each authors in this series, and we wanted to bring them together to talk about why this subject is important and what's new on the horizon. Dr. Harvey Fineberg from the Gordon and Betty Moore Foundation kicked off the series in October 2021 and wrote the concluding viewpoint with his coauthors, Susan Song and Tommy Wang on the Future of Diagnostic Excellence. Dr. Urmimala Sarkar from the University of California, San Francisco, wrote with me about issues related to diagnosis and health equity with a focus on cardiovascular disease. And Dr. Jonathan Chen from Stanford University wrote with his coauthors, Doctors Dhaliwal and Yang about artificial intelligence and tools to augment diagnostic excellence. Thank you all for joining me today. I'm going to start with Dr. Fineberg, who started off our series to talk about why this is an issue that's important. As clinicians, we know diagnosis is important. What what prompted this specific view on diagnostic excellence?
[Harvey Fineberg] Well, thank you so much for this opportunity to join with our eminent colleagues and talk about a subject that is so critical to successful care of patients. Namely, diagnostic excellence. The reason that diagnosis matters is obvious to every doctor and patient. You need to know what is wrong in order to know how to make it right. The motivation for this series, in part, is about how often we fall short in making a timely, accurate and useful diagnosis. In fact, when patients have been asked about errors that they have experienced, more than half of the reported errors are shortcomings in diagnosis. The lack of a timely, accurate and useful diagnosis. So that was a big motive for taking our series on diagnostic excellence.
[Kirsten Bibbins-Domingo] What are some of the reasons we don't get there? We all went to medical school. We learned about making diagnoses. We do CME. We try to keep up to date. What what are the areas in which we fall short?
[Urmimala Sarkar] I think one of the challenges in advancing the field of diagnostic excellence is often when we are making a diagnosis, we haven't reached a specific disease system and that is how our funding mechanisms are organized. And there is a point at which diagnosis needs to be considered independent of a specific organ system or a disease type.
[Kirsten Bibbins-Domingo] So we're starting off with an area that's unknown and and we're trying to hone in on across multiple possibilities. And, and yet we study tend to study within organ systems.
[Harvey Fineberg] I would say the problem begins with access to care. If you don't get access to care in the right way with the right team taking care of you, you may not get the right diagnosis. In fact, you may not get any diagnosis. And then on the second side, there are the complications of or limitations of any one of us as human beings, as thinkers, as cognitive beings. The shortcomings in the way doctors or everyone else thinks about decisions and making in or inferences from evidence. So we know a lot about those cognitive limitations. And indeed that was part of what was written about in the series on diagnostic excellence. And then there, of course, technical limitations in what it is possible to discern in what time.
[Jonathan Chen] I have an echo some of that I think term the scarce resource in the health care system is access to a competent professional expert who can make these kinds of diagnostic and other decision making. And in part, that's because we've had amazing advances in science We can literally diagnose thousands of diseases and conditions, use different thousands of different diagnostic tests. But how do you use that? The reality is no one person can assimilate that. All without some kind of consistent support system, and it just opens up the problems of pathways. But how do we have a systematic approach to a very challenging, wide scope issue? And then I don't want to necessarily get into it, but also the incentives within our health care system are very awkward to drive diagnosis as a priority.
[Kirsten Bibbins-Domingo] So you talked first. I just want to underscore some of these points. You talk first about sometimes the technology might be there to to make a diagnosis, but for any one clinician to to keep track or stay up to date on all of the wealth of possibilities can be challenging. And then you just talked about the the reimbursement and the incentives there. What were you talking about there?
[Jonathan Chen] The very strange thing, we're still largely in a fee for service world. I get paid to make a diagnosis whether I'm right or wrong. And if I made the wrong one, you know what? I get paid to treat your complication of the misdiagnosis as well. So as doctors, we are inherently motivated to do this right. But it's just strange how the system doesn't necessarily drive us and support us in the way to do the things we need to do. For all the down patient care that we're trying to achieve.
[Urmimala Sarkar] When you ask clinicians what they feel would support timely and accurate diagnosis, one of the top answers that I have seen in my prior studies on the topic is continuity. Physicians strongly believe that if they know their patients, they are more able to make a timely and accurate diagnosis. We've all had the experience. I know as a primary care physician, I've had the experience of knowing when someone is not acting like themself and that is something that is not very well supported in our our current payment system.
[Harvey Fineberg] And then the amount of time that we can spend with patients. This is often also a limiter in the ability to make an accurate and timely diagnosis.
[Kirsten Bibbins-Domingo] So these are some of the challenges and and we launched then many articles addressing addressing some of the specific themes here. I want to turn to to Dr. Sarkar, who wrote about issues of equity, and she focused on cardiovascular disease diagnosis. And I want you to talk a little bit about how how you thought about organizing from with an equity lens, the challenges we just talked about in diagnostic excellence.
[Urmimala Sarkar] I think there are both. When we think about equity, we think about structures and we think about individuals. And if we start with the structures of care, I hope that we take it as a premise that the structures of care are deeply inequitable in this country, and cardiovascular diseases is a good place to make this case because they're common and consequential. Right. But I want to just start by saying that this is pervasive throughout diagnostic diagnosis. And I don't think we'll ever get to diagnostic excellence unless we get to health equity. And I'm thinking of cardiovascular disease as a case study. So if we think about structural determinants, we can think about physical location where people are and in neighborhoods that are of lower socioeconomic status, where there may be disproportionate numbers of people of color, we're seeing longer ambulance times, more distance to tertiary care centers in many cases. And so just structural challenges with people getting the right care at the right time because they cannot access it. Then if we think more on the individual level, who has access to primary care, right? Do you even have a doctor you can call if you're having chest pain or a stroke symptom? And there's obviously inequitable access to primary care. There are take care in general. So access is a huge site of inequity. And we can think about symptom recognition, which is on the structural level can be framed really as what is the what are the community assets that can support timely recognition of a cardiovascular acute cardiovascular event. And this there are thinking about who are the people in the neighborhood? Do you have a friend who's a nurse or a P.A. or someone that you can call and say, I don't feel good? And the access to these types of of community assets are very differential. Similarly, health literacy is an individual level determinant of symptom recognition for acute cardiovascular conditions, but really just broadly for disease and health. And then of course, bias is really important. And randomized studies show that people of color do not get the same treatment when they present with the same symptoms. And that is not we are not going to fix this problem unless we admit that there is a bias problem in the way that health care is delivered. Even when people show up to our door, we're not going to get there in terms of getting equitable outcomes. And then the result of all this bias that people experience is a breakdown in trust. And I remember you said to me, Kirsten, that we have to get to trust trustworthiness before trust. So whenever I hear people say, well, certain communities don't trust health care, my question is really why? Why should they? And what can we do to become trustworthy? Because that is also on the pathway to reaching equitable diagnostic excellence.
[Kirsten Bibbins-Domingo] I think the issue of trust is is sort of an overarching one because if if you're not sure that a symptom that you have, you can talk with a doctor freely about it or a clinician freely about it and that it's going to be respected and and that you're going to be acknowledged and figure out what the cause of that symptom is. That might be something you don't necessarily choose to go to a doctor to talk about.
[Urmimala Sarkar] And the cost of going to a doctor is of course, there is a cost, right, in in actual dollars. There's also a cost in time. There's also an opportunity cost. And when people have more competing demands and less trust, it creates a negative spiral and can really impede timely diagnosis. And that's why I think new tools are really important. I think if we keep trying to deliver care in the same way and try to just adjust things around the margins, we're not going to get substantively different outcomes.
[Kirsten Bibbins-Domingo] Okay. Well, that was a good lead in to new tools. I'm going to go to Dr. Chen to talk about some of these new tools. Okay. Dr. Chen, you wrote about artificial intelligence and these tools that are helping us with some of our diagnostic dilemmas or to be better diagnostician. But, you know, we're pretty suspicious of these tools because we can't look under the hood. So tell us about this and how do these tools how can you help us to understand why we should not be so suspicious of these wonderful tools?
[Jonathan Chen] You should be appropriately skeptical, but that doesn't mean you need to be cynical about it's there. We do need to be cautious of over ridiculous hype. 80% of doctors are replaced by algorithms. All radiologists and pathologists we fire. Like those are the wrong expectations. I would phrase it also in another way, you know, Andrew Beam and Isaac Kohane wrote a very nice perspective in JAMA a while ago, said, Be cautious. These algorithmic A.I. decision making tools, they offer no guarantee of fairness, equity or even veracity. And I thought to myself, that is totally true. You should be aware of that. You're smart to think that, but I'm not aware of any human Dr. Sarkar, who provides me a guarantee of fairness, equity or even veracity. So I would instead consider the standard of care where millions of people, tens of millions, the US alone have deficient access to specialty care. We're not even reaching them at all. So be cautious of the hype. Be appropriately skeptical, but realize there are plenty of real opportunities where my gross bias is that technological solutions are the only credible way we can scale and reach the unlimited demand of patients about for medical, diagnostic and other expertise.
[Kirsten Bibbins-Domingo] So one of the things you do so nicely in this viewpoint is that you help us to understand how these tools, these artificial intelligence driven tools, how they learn and you use the examples for how clinicians learn. So maybe you can just talk us through the ways these tools are developed.
[Jonathan Chen] Certainly. Right. Is this a giant umbrella term AI medical A.I.? It's really overly broad. And people have been trying to do this for 50 plus years that general ideas are not new. And I find that very helpful. And what we wrote about in our perspective, just forget computers for saying this as a human being, as a human doctor, how do we learn and how do we practice? You first start learning from an expert. Someone just tells you what the rules are. If fever and opacity and success rate, then pneumonia give antibiotics. And we just learn these guidelines. Right. And is a very practical way to get started. And that's how a lot of classic rules based expert A.I. systems for 30, 40 years, all your order sets, all your alerts, that's basically what they're doing and very helpful in the near term, but also very brittle. The amazing advances being published in JAMA every day are changing our guidelines every few years. So we have to be able to learn those rules but have a way to adapt. Nowadays, when people call, usually what they're actually selling you is supervised learning, machine learning. This is more a pattern recognition tools. Instead of learning from an expert, learn by example. Here's an EKG that shows a STEMI. Here's an EKG that's just early repolarization and you could try to verbalize what makes them different. The reality is just here's ten examples. Just look at them and you will start to figure it out. I can't even describe to you exactly what it is I'm seeing, but I can see that there is a difference and a lot of very powerful, modern, so-called A.I. tools these days. That's what they're taking advantage of, not rather expert knowledge, but massive data sets, label data sets, and doing the pattern recognition is turning out to be a very powerful ability among many domains, and medicine is somewhat catching up and integrating this more into our practice.
[Kirsten Bibbins-Domingo] How do you imagine in the future, as these tools continue to become better? How will we be using them? In practice?
[Jonathan Chen] It becomes another tool. We've actually been using tools like this plenty of times. What's your Meld score? A prediction about whether your liver patient is going to die soon? That's actually what it is. Where's your Framingham risk score prediction or how likely already have a heart attack assembly? How likely are you going to have a stroke and decide whether it's worth and cholesterol medicine? So we actually do have the experience that uses it just now. These types of tools can be much more common and pervasive. They're so much easier to produce than they were because the infrastructure is catching up.
[Kirsten Bibbins-Domingo] So in some ways, one of the things you write in your in your viewpoint is that it really ideally frees up the physician to talk about what they're uniquely poised to do to take a good history to to explain uncertainty to patients and to understand uncertainty and to think about the context which which is ultimately important. And it seems like for right now, we still will rely maybe on the computers to help with our diagnostic excellence. But but still, we're not going to replace physicians right away or any type of clinician.
[Jonathan Chen] Any time soon. And there's a famous quote from, I believe, Walter Slack from the sixties. Like any doctor who can be replaced by a computer should be. And the point of that is emphasize a good doctor, a good teacher cannot be replaced by a computer. A computer is a very effective calculator, but that is very different from being a good thinker and specifically getting a good history, counseling your patient and listing their values. Right. That's something a computer statistic can never tell you. Would you rather live for one more year of healthy or live for five more years disabled? No one can tell you what that answer to that question is. That's very personalized. And as doctors, we're always integrating how these different pieces of information apply to our individual patients characteristics and their life contexts.
[Urmimala Sarkar] I will say that I think clinician cognition is is part of the challenge of diagnosis, right? As a clinician, when you're trying to make a diagnosis, you're integrating a lot of inputs. And I think one reason diagnosis has gotten more challenging is there is more there are more inputs coming at clinicians, more tests, more specialty opinions. And what Jonathan is suggesting is this is another tool in our armamentarium. But and I, I see that as a benefit. I would just suggest that part of what needs to happen to advance diagnosis is we need to think about physician workflow and how it supports diagnostic cognition. Because in some ways, if the if the A.I. output is analogous to another diagnostic test, as a physician, that makes me wonder how am I going to integrate that with all the other information that's already coming at me? So I think that's an unanswered question.
[Harvey Fineberg] One of the potential contributions of machine learning and artificial intelligence is not only about reaching the diagnosis, but helping the process of making clear decisions along the way of how to reach the correct diagnosis. So embedded within this tool is an added complication, but perhaps also an added aid to solving that complication. If I could follow up on that.
[Jonathan Chen] In fact, we had another one of our pieces with Professor Abigail saying about diagnostic wayfinding and where I think to myself as as a doctor, how, how often if I have said I wish a computer would tell me the ICD ten code for this patient, I can't think of any time. That's what I needed to take care of a patient instead, it was just, what's the next step? Given what I know so far? Is it a blood test? Is it a CT scan is an antibody that actually is that process is such the key diagnostic really process to kind of help patients not come up with a label. It's how do you get there? And a lot of these systems can benefit that way. But I think it's a meeting of the minds of the clinical and the engineering practitioners to realize what each other needs versus guessing what the others are really going to benefit from.
[Kirsten Bibbins-Domingo] Yeah, that really resonates for me. The diagnostic way finding and sort of making a little bit more explicit because some of these processes can be long and meandering and and figuring out ways to make us more confident as clinicians in, in the next step. I think I think that will resonate for for most people. Certainly does for me. Dr. Fineberg, so your concluding piece on this series is about the future of diagnostic excellence. And you point us to a number of themes, whether we like it or not, that we as clinicians are going to be living with for the future and talk us a little talk to us a little bit about that.
[Harvey Fineberg] This is a bit speculative, it's true. But every thing that we wrote about in this final piece actually already have the seeds planted in what we're seeing now. For example, we've talked about how important continuity is in interpreting what's going on with the patient, knowing and seeing them over time. And one of the trends that we highlighted in this concluding piece, looking ahead, is a movement away from episodic, symptom driven testing to more nearly continuous monitoring of important indicators in a patient of their state of health. This is going to come from technologies that we where technologies may be we've ingested. It's going to be a way of gathering evidence in a much more continuous way as part of everyday living, not only just when we are visiting a clinic now along with that, we project that there's likely to be a shift in emphasis of individual tests and their results to a recognition that patterns of testing over time and interpretation of those data streams, which again is where machine learning and artificial intelligence can enter as important dates, that that is going to be another very important and insightful way of understanding the state of the patient. And this, in turn, leads to a different way of thinking about what is an abnormal result right now. When we do an individual test and we look at the results, we'll get a lab report that talks about whether it is within normal limits, meaning that if you test a group of healthy people, that this result is within the 95% range of what healthy people without any known diagnostic problem would produce as their result. We see that this could be replaced in a very constructive way by comparison of any individual pattern over that individual's personalized pattern in the past. So that what represents abnormal for you conceivably could still be within normal limits of a population. But for you and your history, it's indicating something is a little bit different or starting to change in your physiology. And so that shifting of what we recognize as abnormal is another potential important future refinement. And then going along with that, we believe will be increasingly precise diagnostic labels or understandings to go along with increasingly precise treatment. A diagnosis that's meaningful is one that leads to a adjustment or change in the way you're being treated. So that increasing precision of diagnosis, coupled with increasing precision of treatment, is going to be an important trend, we think, looking ahead in the future and finally thinking about the meaning of diagnosis, which we now think of as a classification of disease, we foresee a time over time in the future where your diagnosis stick label will not only be more nearly unique to you, but also indicative of your state of wellness. Not only the presence or absence of particular diagnostic disease labels in all of this. What we stress looking ahead is that diagnostic excellence has to begin and end with the interests of the individual patient, foremost at heart and everything we've been talking about, from access and equity to the application of technology to timely and relevant diagnoses. All of this is directed at improving the quality of care for our individual patients.
[Kirsten Bibbins-Domingo] That's really that's really so nicely laid out in your piece, hearing you sort of go through these these various themes that will be continuing to see in the future the use of continuous monitoring, the the multiple data streams to understand an individual patient that will be thinking through not just how you differ from the population, but how you differ from what you what your measurement was in the past. The individualization and how we think through next steps and we think through treatments. I think in many ways when I first read it, it sounds sort of very futuristic. Of course, I think we all recognize the seeds as what we experience today. In some ways, though, I also think of it as very sort of back to that. The old school way we think about the doctor who knew the patient intimately over a long period of time and who sort of knew when something was a little bit different from the way that patient usually is. And so what, what it strikes me that you're describing is really the use of these technologies and the multiple data streams to really help us to get back to understanding the individual patient and to be better in, in making the right diagnosis for that patient.
[Harvey Fineberg] All of this is directed at liberating doctors to be their patients physician. And that's exactly what you described. If we use these technologies and lessons in a right way, it means we can be better doctors.
[Urmimala Sarkar] That means that the entire way from development of the AI algorithms to the communities are included in studies. And where these studies are based, it all needs to reflect the diversity of of all of the communities that we want to have diagnostic excellence. And that is that is a sea change from the way things have been happening.
[Kirsten Bibbins-Domingo] Dr. Chen, another thought from you.
[Jonathan Chen] I think the advancing technologies that we want to improve our diagnostic processes are essentially amplifying tools to a large degree as practicing physicians with our hard fought experience, we kind of know what to do, whether we can do it at scale. For thousands of patients, that's gets pretty hard, pretty fast. With the support of technologies, we can amplify our ability to reach more, to do more. But as Dr. Sarkar points out, also be careful. These things are amplifying. So if you're doing something good, like monitoring your patients, getting to know them, giving them good access, it will help us do that. So that's great. If one of these things we maybe accidentally do is perpetuate existing social biases, you have to be self-aware of that, or else technologies will amplify your ability to do that as well.
[Kirsten Bibbins-Domingo] Wonderful. So I really appreciate this conversation. I appreciate the sets of viewpoints and the ones that you've been discussing with us today. Diagnosis is so important for doctors and for patients and I love the way that that each of these viewpoints take us through the the current state and what we could all be doing better, as well as the things that are on the horizon and and at the heart. This is really about putting patients first that that really is at the heart of a lot of what each of you have written about. And as you have come has come through so nicely in this conversation. I think allowing clinicians, doctors to be the best that they can be to doing the best for their patients. So, Dr. Fineberg, Dr. Sarkar and Dr. Chen, thank you so much for your viewpoints in JAMA and for being part of this discussion today. Kirsten Bibbins-Domingo I really appreciate it.
[Urmimala Sarkar] Thank you so much. It's really been an honor.