How AI Is Changing Health Care - from Diagnosis to Discharge

Erka Amursi: We see AI as the light at the end of a dark tunnel. It will be AI doing the first diagnosis, Who's responsible for the safety side. it is essential to understand the AI tools that health care organizations use.

Steve Odland: So Erka, you've written a couple of papers on AI and especially as it affects the health care market. Talk about those papers and where can our listeners find them?

Erka Amursi: Yes. Actually, we are continuing to write about health care. We have already published two papers in our website, tcb.org under the Human Capital Center. In the first two papers, we focused on how AI can help organizations with their workforce shortages, all the positive impacts that it brings to the organizations.

In the second paper, we focus more on the implications, the concerns, the risks. We are now focusing on what health care organizations do currently to implement and scale AI, and we will continue to write because this is a very interesting topic.

Steve Odland: Yeah, and health care is a huge portion of the US economy, what, close to 20% of GDP. And it's growing like crazy because of our aging population but also technological advances in medicine, which allow more access. It's also continuing to struggle with workforce shortages. How is AI going to affect all of this?

Erka Amursi: Yes, workforce shortages have been a big problem for health care for a long time, and predictions of health care organizations show that it can be worse in the coming years because of all the reasons-- aging population and all of that. But we see AI as the light at the end of a dark tunnel that can help in this regard because AI can actually do some of the tasks that physicians or nurses or administrators in health care actually do and save them the time and save them the capacity and staff in order to see more patients, do more clinical work and specialized work.

Currently we are seeing different use cases of AI in health care. AI has been used to, they call it ambient documentation. AI has been used to record, transcribe, summarize, and document the meeting between a physician and a patient, which relieves a lot of energy and time for physicians because they had to write everything on the computer after the meeting or during the meeting with the patient.

After using AI, now this is automated, saving around one to two hours per day per physician, according to some studies. That and many other use cases: AI reviewing radiology tests, AI being used in triage in hospital. Robots have started to be used in hospitals help with giving medications and food to patients or collecting garbage and all of that.

There are different use cases in different parts of the health care organizations.

Steve Odland: Yeah, you mentioned radiology, and that appears to be a big opportunity. Everything's digital now, so all scans can be digital and then radiologists can sit, frankly anywhere in the world, and AI can help diagnose and screen and turn that into almost an instantaneous diagnosis on things, right?

Erka Amursi: Yes. Yes, and there are some studies showing AI doing sometimes a better job in diagnosing illnesses and diseases. And it also it's very helpful in prioritizing which cases are more critical and need more urgent care.

Steve Odland: Yeah, and there are some aspects of health care which require hands-on, particularly elder care where you're trying to help somebody in their lives and so forth.

AI is not going to impact every aspect of the health care industry evenly. But it does add capacity and productivity, which allows us to do with fewer resources even as population grows.

Erka Amursi: Yes.

Steve Odland: Give us some more examples if you can. We talked a little bit about diagnostics but give us some more examples of where it's already delivering value.

Erka Amursi: Yeah. One other use case is, for example, during the discharge of patients from hospitals and all the medical education that needs to happen at that moment. That has started to being be covered by AI tools. Also, monitoring chronic patients or patients after the discharge.

AI tools can help detect if the patient is stable or deteriorating and be able to prevent this patient from going to ER and be able to help physicians monitor them better remotely.

Steve Odland: In the office workplace we've talked about the deployment of AI and how you can't just plug it in and have it take over. You really have to redesign workflows and that's the case in health care as well, isn't it?

Erka Amursi: Yes. Like everywhere, we are seeing a lot of benefits where AI is implemented for specific tasks, for specific workflows. But where the big potential of AI is actually in the operating models, in changing how these workflows are integrated together.

That is more complicated. It's complicated for every industry. Imagine health care. But the future is there. The highest opportunities and benefits are there.

Steve Odland: If you divided health care into sort of the care side and then the administrative side, because it takes a lot of administration to run the health care industry. Where do you see AI playing the biggest benefit, on the care side or the administrative side?

Erka Amursi: I think it's both. It's also critical to have AI as much and as better implemented in the administrative side of the organization. We know health care, especially in the US, has a lot of administrative burden, and AI can help with finding pre-authorization.

It can help with coding. But also with other tasks, like predicting staff needs, for example, for part-time nurses on specific seasons, etc. So yeah, I think both are critical.

Steve Odland: In any evolution in the digital age, jobs change. You see some jobs lost but you also see jobs created, and that was the case with the rollout of electronic medical records, where people said, "Oh, it's just going to kill all these jobs." And in fact, it did, but it created a lot of jobs. Do you see that happening in AI as well?

Erka Amursi: Yes. In health care and AI, we see a lot of need for clinical staff to better understand AI tools and the technical side of the business now.

We are going to need more professionals who have both clinical and technical understanding and knowledge so that they can integrate the two aspects. And also the security departments integrating understanding of the security and risks and all the problems of AI and the technical tools with patient safety, with privacy, with all of that.

Steve Odland: Yeah. It sounds like it's going to alleviate some workforce shortages but it also will create new work, right?

Erka Amursi: Mm-hmm. For sure.

Steve Odland: Yeah. And therefore, we'll see how this goes, but it's hopefully it's higher value added work and more rewarding work.

Erka Amursi: Yes.

Steve Odland: Yeah. As you look at it, what are the biggest risks here in AI development?

Erka Amursi: Yes, it comes with risks, and there are risks everywhere. There are risks for providers. Patients may face risks, the organizations, etc. One of the main risks all are talking about is for physicians especially: over-reliance on AI.

We know AI can be convincing. It can give you very detailed analysis of radiology test, etc. Sometimes it has done a great job in identifying risks and diagnosis or special treatments for rare diseases, etc. We have read about those cases, but it comes with risks. We have heard about AI mirages, where AI just produces a diagnosis without even seeing any test at all.

It's all hallucination and AI creation. So these cases can be very risky if physicians take them for granted and don't double-check and don't use their critical thinking in reviewing them. This over-reliance that may happen over time is risky.

Steve Odland: This is the same thing everywhere else, where it's still early stages of rollout and you can use this as a tool, but you still have to then again amend your work processes. It's just part of the work process but it's got to be validated and checked by human beings.

Erka Amursi: Yes. And the concern long term for physicians is that it will be AI doing the first diagnosis, and then you have to double-check and control the output. The motivation for new generations to go through all the school years and all the training in order to only double-check what AI is diagnosing or the treatment that it is providing, it's less motivating.

It is that concern about long-term education and being able to keep the critical knowledge and thinking of future physicians.

Steve Odland: Yeah. We Are in a litigious society and there are people who are waiting in the wings sometimes to sue doctors or hospital systems and so forth.

If you're in the industry, if our listeners are in the industry, the deployment of AI requires risk and making sure that it is just part of the process and not the whole process. And I think that's the point of your comment, Erka.

Erka Amursi: Yes. Risk Management is huge.

We talked about the risks that impact physicians, but patients may be impacted if AI outputs are overlooked. There are risks about privacy because AI can be used to collect the information about symptoms and issues, but it can overdo it and go into a lot of patient details.

So keeping that privacy is important. Also, another concern with patient care is informed consent when AI is used. We know informed consent is difficult because physicians have to translate what they are doing into simple language for the patient to understand and consent.

With the implication of AI, sometimes even physicians may not be able to actually understand why this AI tool went into this direction or that other direction. That why it is essential to understand the AI tools that health care organizations use. And also it is essential to co-design them with the involvement of the physician because there have been experiments out there when an AI tool has been implemented and no physician was involved, and then it didn't work out because no one took into consideration the workflow, the realities of the work and that those tools were not effective and efficient, so they have been replaced.

But on the other side, when AI tools have been co-designed with physicians, they have been more successful based on what we have had.

Steve Odland: So Erka, we were talking at the beginning about these two new papers that you've written and more to come. Talk about your second paper, your second article, which talks about AI's impact across all layers of health care.

Erka Amursi: Yes. In this second paper, we focused on, as I said, on the implications of AI, the risks, the concerns, and we wanted to be as comprehensive as possible, trying to understand all the implications. And it goes from the physician, we mentioned some of the risks out there, but going to the patient safety and patient care level, we know there are biases that impact AI tools that may affect different populations and minorities.

Patient privacy may be impacted if these tools are not implemented in the right way. But then going to the organization level, of course, implementing AI at scale and getting the benefits from changes in workflows and operating models will require a lot of coordination and collaboration from the clinical team, the administrative side, IT, legal, patient safety-- all the departments and functions involved in health care so that they are able to monitor any AI tool that they are implementing.

But in health care, we go into another concern, which is the paying system and the market. We know that especially in the US, you need to align with different insurance policies and paying systems. So being able to align with that is another concern. We have currently a system that is pay per service, where health care organizations are paid for any visit, tests that they provide for the patient.

Now, if some of these are done by AI, how is this going to be paid by the payers? And we are seeing some movement there into new systems, pay per value, pay per outcome, so that health care organizations are paid for the outcome. How good are they at managing a certain chronic condition or what's the outcome for the patient, not how many tests have been done, how many visits have been done, etc. But the coordination with the payment systems will be concerning, and there are risks there that need to be clarified and both sides be aligned.

Then we go to the regulating system, to policy, to government. And we know that there are many regulations that come for health care organizations from the federal government or the states. Being able to align with those regulations will be a challenge because AI is evolving day after day. In many cases, health care organizations will need to use tools, or they will think that it's beneficial for them to use new AI tools for which there is not a regulation yet. So they need to be able to define their own internal policies and document them in order to be aligned with the coming regulations from the government.

Everything goes to the highest level in our description, which is the moral layer. With all the AIs evolving, all the regulations evolving, there will be moments for the physicians to go and make decisions based on their ethics and their morals. And having them trained on that aspect is also important.

Steve Odland: In your paper you talk about a principle of co-pilot, not autopilot. It's a tool. But you just can't set it and forget it and just rely completely on AI in any part of the process.

Erka Amursi: Exactly. That's the future we believe in. Having clinicians, physicians, health care workers in control and using AI just to support them and help them with their work.

Otherwise, organizations are going to lose a lot in terms of trust from patients. It is hard to to accept to be treated solely by AI and to have only AI making your diagnosis and treatment plan. That is something that we don't see coming in the future.

Physicians and all clinical workers are in charge. But it is important for them also to understand from the vendors that are providing their AI tools, how are these tools working? Why are they coming up with those conclusions? How do they make their judgment, et cetera?

So the better they understand the tools that they are using, the better it is for the patients and the health care.

Steve Odland: I remember decades ago, when you go to a doctor, they would come into the room for an appointment, and they would say, "How are you feeling?" You know, "What do you want?" And they would do a physical exam. It was hands on. Now it seems, they come in, they open up their computer, they log on and they sit there and type on their computer, and you ask a couple questions. But it just seems like technology has become a bit of a barrier between the physician and the patient.

Do you worry about that AI could actually contribute negatively to that physician-patient relationship?

Erka Amursi: I think it is possible if it is not implemented correctly, but it is also possible, very possible that AI can help with the relationship between physicians and patients because as we said, AI can cover all the administrative tasks, all the things that doctors need to do: scribing the case in their computer. That is something now that AI can do. AI can record the conversation. AI can document the conversation and summarize it. That releases the physician and gives them more time to actually see the patient, communicate with the patient.

But it will also depend on how health care organizations are implementing AI and how they are using it, because if a health care organization is now saying that, "Oh, you used to see, I don't know, eight patients a day. Now with AI doing all the administrative work you have more time, so now you're going to see 15 patients," that is still concerning.

Aligning with payment systems is another issue that will need to be discussed in the future years. But yes, AI is removing administrative tasks, potentially helping.

Steve Odland: But spending more time with patients is a good thing. Particularly if scarce resources can be deployed across a broader number of patients.

So pros and cons to all of this, and that's, you know, I think that's your point. You have to take all of this into account. The other thing is if AI is recording every conversation, that's different than even what you have now. That creates a record. Every word that you say then has to be carefully adjusted and you have to think at some point it's going to show up in a deposition.

It could be negative in that sense because it puts physicians on edge. So lots of considerations here, ethical considerations, legal risk all the way around, and that's what your paper talks about.

Erka Amursi: Yes, exactly. Yes. The privacy issues, going too deep into all the details, recording them without consent, that is a problem.

Steve Odland: So what should health care leaders do to try to establish effective AI governance?

Erka Amursi: First of all, there is a need to have governance systems in place. There is a need to, especially in health care where the administrative side, with the clinical side, with the safety side are so much intertwined, it's important to have those governance systems that include leaders from different functions and roles.

In addition, it is important to have some allocation of responsibility. There is some need for clarity. Who's responsible for what in all this transformation? Who's responsible for the safety side. There are those roles, but when workflows are going to merge, because we believe it will go beyond singular functions and workflows, then responsibility will be more challenging to define. And defining it is the key.

Steve Odland: Yeah. Last question. You've got a regulatory environment out there, which is surrounding all of this and it's evolving. You've got health care professionals trying to implement AI in that evolving regulatory environment.

How should health care professionals deal with that?

Erka Amursi: Yes, that is another challenge for health care right now. Moving faster forward with AI without a clear regulations coming from government and states is challenging. But it is important for organizations to create their own policies and internal regulations when they use AI, focusing on their values, priorities, all of that, and be able to document every step of that AI implementation so that they can be able to demonstrate it when the regulation comes.

But also be agile and change things, because they will need to change things once a regulation goes somehow in a different direction or or new regulations are coming. So being able to create policies, create overseeing systems, but also tracking down in detail what they are doing so that they can demonstrate what they did and how and why. Be agile enough to change with any evolvement in AI and regulations.

Steve Odland: Any last thoughts that you want to share with our listeners today?

Erka Amursi: I am very, at least for health care, I'm very optimistic about AI, it helping with the workforce shortage, helping with the research side. Big gains come with big risks and big efforts. That's a challenge, but at least we have a lot of hopes in health care supported by AI.

Steve Odland: Erka Amursi, a voice of optimism. Well, thank you for being with us today and sharing your thoughts about the paper, which people can find at tcb.org, and look at our Human Capital Center and you'll find Erka's paper. Erka, thank you.

Erka Amursi: Thank you. It was a pleasure being with you.