Central Line

Episode Number: 169

Episode Title: AI from A to Z (Part Two)

Recorded: August 2025

(SOUNDBITE OF MUSIC)

VOICE OVER:

Welcome to ASA’s Central Line, the official podcast series of the American Society of Anesthesiologists, edited by Dr. Adam Striker.

DR. ADAM STRIKER:

Hello, everyone, and welcome back. This is Central Line, and I'm your host, Dr. Adam Striker. This is the second and final episode we are presenting in our series on artificial intelligence and the many ways AI touches our work. We posted the first episode on this topic at the end of March, and today we are sharing four additional short conversations with members of the Committee on Informatics and Information Technology, or CIIT. Some of our guests for this episode are either members of the US military or work for the US government, the opinions and exertions expressed herein are those of our guests and do not reflect official policy or position of the United States government.

We're going to start with an issue that colors how we understand AI and its limitations, the thorny challenge of addressing bias and ethics in AI algorithms. To help us work through this issue. I spoke with Dr. Christopher Goldstein.

Dr. Goldstein, if you don't mind, lay out the landscape for us just a little bit. What is the scope of this particular problem?

DR. CHRISTOPHER GOLDSTEIN:

Sure. As medical professionals, we are taught to be aware of our own human biases to not let them negatively affect our care or research studies. And there are about 150 to 250 human biases described depending on the source. Regarding our own specialty, we have been reminded by several articles in the last couple of years that pulse oximeters can also give us unreliable values depending on skin color. So it is an issue that we have had to be aware of, even pre AI, that examples of biased algorithms keep making the news is unfortunately not surprising then as many AI biases have human biases as underlying cause. Like designing and completing well conducted research studies, it's difficult to ensure AI models are trained on high quality and representative data, and that their learning pathway is steered in a desirable way. It becomes even harder when multi-layered deep learning algorithms and neural networks are employed, and in some circumstances, not to get too technical, but so-called model drift can lead to issues down the road that were not even apparent during validation or early use. So it's definitely a complex issue for developers.

From the clinician user side, I think what we must increasingly watch out for as AI applications become more ubiquitous is automation bias. Using machine intelligence tools implies that the decisions are rational or objective, basically near perfect, which can make it hard for us to question them. Having to deal with automation bias on top of the other time pressure tasks in a fast paced environment like the OR, can add yet another layer of stress and potentially reverberate negatively into patient care.

DR. STRIKER:

Take us through a real-life example of how a biased algorithm may impact patient care, and maybe even broaden that out to an example of the danger that that could present.

DR. GOLDSTEIN:

There's growing excitement about the rise of AI tools in perioperative medicine, for sure. So let's zoom in on a couple of examples.

Preoperative AI agents that call patients autonomously, going patiently over questions and providing personalized answers and pre-op instructions, even after hours and weekends, are becoming a reality, and it's exciting to envision a future and a time consuming pre-op evaluation and risk stratification process will become more streamlined thanks to this AI agent. But what if, in a worst-case scenario, a biased algorithm labels let's say a cancer patient not fit for potentially lifesaving curative surgery, and this doesn't get questioned or overruled. That completely changes this patient's care trajectory and probably the outcome.

Intraoperative and post-operative, we will also see more AI tools augmenting us. But imagine a semi-autonomous hypertension avoidance and pressure fluid management system, strongly recommending certain actions in an unstable patient. What if I disagree? Despite the machine alarming and suggesting immediate action? Will I doubt myself in the moment and possibly let automation bias insure me? Maybe thinking what am I missing? What if the situation turns even more unstable if I deviate from the machine suggested path? Now put a concerned surgeon wondering why anesthesia is not following these urgent recommendations on top of this. This scenario adds a significant layer of pressure, stress, and maybe even fear of potential medical legal consequences. Some might even think, why did you think the algorithm was wrong and decided to deviate from the recommendations, doctor, that they are being asked that during an M&M or even during a court of law?

So based on the confidence and experience level of the team involved, it's possible that care suggestions are being followed in the heat of the moment that turn out to be suboptimal later. There are already anecdotal reports from radiologists who had a hard time convincing their E.R. colleagues that the preliminary IAI read was wrong and no treatment was needed. So the stakes are definitely high.

DR. STRIKER:

Well, are all patient populations at risk, and are there ideas out there that might make AI safer?

DR. GOLDSTEIN:

I would consider anyone at this point at risk, depending on the situation. As we discussed before, high quality representative data is of course fundamental, but there are often quite unpredictable factors that determine what an AI learns depending on the complexity of the algorithm deployed.

Take mass casualty events or ICU bed triage in shortage situation like a pandemic as an example. Here AI offers the tempting benefit of fast, data driven prioritization within seconds at a scale and speed that humans just can't match. Again, an exciting use case of AI augmenting humans. But what if the algorithm was trained on data that does not adequately represent all victims? Maybe a US trained model deployed in a different country to render humanitarian assistance. Good intentions can still result in harm, especially when we defer to autonomous systems because we are overwhelmed by the scope and pace of an event.

To answer your second question, to mitigate risks of bias and more generally, facilitate ethical AI, various frameworks have already been developed. Core strategies usually include maintaining human agency; being able to veto and request human review, like in the cancer pre-op patient example that we had earlier; adequate human supervision as anesthesiologists still being there, ready to intervene in the ORS, for example; and transparency. To what extent do we have to, or should we, disclose the use of AI tools to our patients? And importantly, what should manufacturers disclose to us as end users? I like the concept of AI nutritional labels, a sticker on a device, or a pop-up window declaring key metrics such as data origins, performance, known biases, and the intended use cases and limitations. For instance, label statements--like algorithm trained on US adults 18 to 79 use on pediatric patients, patients at the extremes of ages, and non-US populations, is off label--can allow for quick safety check if the AI tool is appropriate to use.

DR. STRIKER:

Let's zoom in on clinicians. What can we, as anesthesiologists, do to help move the needle in the right direction and protect patients?

DR. GOLDSTEIN:

I'm hopeful that we have what it takes to lead in the age of AI. Important is to remind ourselves that despite AI our tools becoming very good and extremely confident, it remains just another tool that comes with its own set of problems like the discussed bias, hallucinations, or a better, maybe a less anthropomorphic term, AI misinformation, and even cybersecurity problems like adversarial attack vulnerabilities.

Being vigilant, patient advocates, adaptable and lifelong learners is at the core of our profession. These traits will remain pertinent when it comes to AI. A cornerstone will be fostering awareness and providing ongoing education. The ASA and ABA should continue to take an active role here by adapting educational offers like this podcast; postgraduate training AI certificates may be modeled after the successful POCUS pathways; and modifying residency in exam content outlines accordingly; and possibly even offering AI fellowship pathways. Some anesthesiologists will continue to be closely involved with industry and R&D, while others will focus on advocacy, policy making and representation similar to the aviation industry. The risks of automation bias, deep scaling, and overreliance on automated systems can be tackled through courses and simulation, facilitated again by the ASA and ABA. Overall, we have to ensure that we adapt and become dual competent physicians. I would like to call them anesthesiologists in the loop, capable of harnessing AI to the maximum benefit of our patients while remaining in the loop, ready to intervene when needed.

DR. STRIKER:

Well, Dr. Goldstein, thank you very much.

DR. GOLDSTEIN:

Yeah, thank you very much.

DR. STRIKER:

Dr. Matthew Wecksell shared some thoughts on cybersecurity and how AI is impacting the safety of care.

Dr. Wecksell, if you don't mind, first, talk to us a little bit about the risks here. When we're talking about AI, we're talking in part about large language models and the impact these models have on patient Information. So distilling so much information across disciplines is great, but there's got to be some security risks, right?

DR. MATTHEW WECKSELL:

Absolutely. Large language models are really great at amplifying what people can do, especially the people that are below the top of the bell curve and skill. So what you'll hear in the popular media is that LLMs aren't going to replace lawyers or programmers, but they're going to make the lawyers and programmers who use them significantly more effective. And unfortunately, this is also true for the bad guys in healthcare IT we've got a real lack of diversity. Epic's got about 35 to 50% of the market. Oracle Cerner has about 25%, Meditech about 15%. And the rest of the IT stack that healthcare is running on tends to be a windows environment, and with very few potential targets, it becomes a lot easier for anybody to say, okay, what are some of the problems and vulnerabilities with this? It used to take a lot of great skill to identify those problems. Now, it's as easy as asking a large language model. Hey, how do I exploit vulnerabilities in this vendor's platform? How do I exploit lower in the technology stack? And then, okay, now how do I add a ransomware module to that? So in some sense it almost becomes like plugging Legos together and the bar for a bad actor becomes significantly lower. Ideally, you don't want a nation state to fund bad actors to target your hospital for ransom. But at the same time, you also don't want drunk teenagers or disgruntled employees also being able to target you. And I, particularly with a large language model behind it, really lowers the skill bar for somebody launching those kinds of attacks.

DR. STRIKER:

Are there real world examples? I mean, have American anesthesiologists or healthcare organizations faced issues with security and and what like that look like in practice?

DR. WECKSELL:

Absolutely, they have faced those issues. In April of this year, Yale-new Haven Health System determined that an unauthorized party had gained access to its network and copies of its data--copies of its data, including patient names, birth dates, phone numbers, addresses, email addresses, Social Security numbers. Fortunately, Yale's electronic medical records weren't involved in the breach, so the incident didn't impact their ability to provide care. What we have to understand is in health care and everywhere else, there are different levels of security. It's one thing for somebody to deface the website of a bank. It's another for somebody to steal a bank's client list. But it's really, really bad if they steal the money. So too, in health care, you don't want somebody doing things that are going to shut down your EMR because then you're dead in the water delivering clinical care. This attack against Yale, the EMR, wasn't affected. But obviously, even if somebody just gets your patient list, if they know that somebody is a patient of an HIV clinic, you really can make some obvious assumptions about why they might be going to that clinic and be able to extract some fee from that without actually having access to the EMR.

On the other hand, AI helps us play defense, and I noted that Yale determined that an unauthorized third party had gained access. Well, how do you do that? You get to have an AI looking at your server logs and your computer logs. And so instead of waiting for humans to say, hey, we've got unusual logins, we've got unusual data transfer, you can have an AI tool that's sitting there constantly going through the logs and saying, what's out of the ordinary? What's spurious? Do we need to do something about this? Do we need to alert the humans have then take a look and do something about it so I can help with compliance monitoring. It can help with automated incidents response. So it's not all bad.

DR. STRIKER:

Well, obviously everyone's at risk here from individuals to the organizations. If you don't mind, take us through how those risks differ depending on the particular target audience, and then specifically how we as anesthesiologists might be affected.

DR. WECKSELL:

Sure. It absolutely depends on the risk that you're talking about. Individuals are mostly at risk of seeing their data leak out of the health care system. You know, we mentioned earlier, the bad guys can know you're a patient at a cancer center, you're a patient in an HIV center, and extract information just from the patient list, just from the data, that's not actually part of the EMR. Obviously, if the EMR information gets breached and leaked, there's significantly more risk to the individual about all of their PHI being shared, either with bad actors or being shared widely. Fortunately, we really haven't seen attackers altering patient data or the behavior of therapeutic devices. Um, security wise, that's the nightmare. Either, you know, changing lab values in the computer so that patients receive care that's not indicated so that they receive emergency care that's not indicated, or modifying therapeutic devices where now that you have CPAP machines on Wi-Fi, insulin pumps, pacemakers, uh, the more networking tools and networking hardware we put into those kind of devices, the greater the risk that somebody has the opportunity to compromise them and engage in some really bad activity against patients. Uh, fortunately, that hasn't happened so far. Or if it has, it really hasn't gained traction in the media.

For institutions, the risks are going to be both legal and operational. Legally, HIPAA spells out penalties for data breaches. There's just a clear legal responsibility to prevent that from happening. But operationally, ransomware can shut down a hospital. Other attacks that compromise an EMR can shut down a hospital as they go through and verify that they still have a functional EMR, that nothing's been changed, and that they can be secure in the tools that they have. And forcing an institution back to paper for an extended period of time carries with it its own patient risks. We can provide anesthesia records on paper, but obviously the pharmacy is not going to be able to automatically check for drug interactions when people order that. All the safety features we've built into the electronic medical record and our health care workflows go away when we move back to paper. For the operating room, the other risks are just extended downtime. Again, if the HMP is in the EMR just because you can provide an anesthetic, give the drugs and chart on paper. If you can't read the history and physical. If you can't read any consult notes if the lab values aren't available. Um, that's not an environment in which you probably want to be practicing medicine unless it's emergent. So there's risk there as well.

DR. STRIKER:

Well, how do we protect ourselves and our patients? If you don't mind, share a little bit of advice with our listeners to help them engage in some modem of protection from these kinds of security risks.

DR. WECKSELL:

Sure. Most of the safeguards that we have are going to exist at the institutional level, which is where the HIPAA security rules focused, So we need to have safeguards about who can access what data, when and how. Um, HIPAA requires institutions to have security policies, and clinicians really need to be aware of that and understand why hospital IT and healthcare IT departments are doing what they're doing. There's a tendency to want to throw our weight around, say, I'm the doctor, I need this. But it's important to understand why the hospital IT department is often going to say no. We've got a lot of web-based resources that are valuable in everybody's day to day practice. You might have a call schedule that you purchased through an online vendor… And all of that stuff on the web is great, but it becomes a bad idea to have internet access on any computers that patients, their families, and visitors have physical access to. So we become limited. You can access those resources in your office, in the operating room. But in the holding area, you've got computers in the patient bays where the patients are sitting, they’re left alone. Those are places where we probably shouldn't have access to stuff, even if those are resources that are valuable for clinical care, even if they're valuable for the administrative running of the department and you're paying for it. There's a time and a place for that. So too, IT is going to want to mandate different services the EMR, the PACs, whatever, automatically log out after a certain amount of time that they're not being used. And even if it's more convenient for the clinicians to have, you know, you log in once and stay logged in forever, there's good reasons why IT is doing what they're doing in order to, at that institutional level, protect everybody.

DR. STRIKER:

Dr. Wecksell, thank you so much. Appreciate it.

DR. WECKSELL:

You're welcome. Thank you.

DR. STRIKER:

We wanted to learn how AI can be used to boost revenue in anesthesia practices so we turn to Dr. Jonathan Tan.

Dr. Tan, when it comes to perceptions of AI, I think many of us have some mix of hope and fear. Let's focus on the hope. As the specialty grapples with various monetary challenges, do you think AI represents some opportunity to win back some revenue?

DR. JONATHAN TAN:

Yeah. This is a great area of focus for our specialty. Now, if we think about how complex billing can be, the complexity of understanding how to properly bill, how to efficiently bill, and just the whole administrative structure around it, it's a great opportunity for disruption. And, you know, what we're seeing is artificial intelligence as a potential tool to be able to streamline this processes for practices in large groups. I think there's a lot of hope--a little bit of hype, but there's a lot of hope with artificial intelligence to help address billing and revenue cycles.

DR. STRIKER:

Well and talk a little bit about how we can partner with AI to create higher reimbursement rates for the anesthesia community, how it might be used for coding, etc..

DR TAN:

Yeah, I see there's two different sides and it's a really wonderful approach. After you think of, you know, two sides of the coin here. The first is ensuring and optimizing billing capture. So for artificial intelligence to be able to use, to scan our charts, to scan the cases, and to understand if we're optimizing our billing practice from the front end. And then in the back end, the entire administrative support around billing and the time it takes the bill and the staffing required, there's a huge cost right there of real dollars for practices. And if you could either reduce that overhead from an administrative standpoint or make it more efficient on the front end. Those are two really great opportunities for artificial intelligence to make a huge impact on the finances of clinical practices.

DR. STRIKER:

Well, I imagine these tools exist now. And if they do, do you think the anesthesiology community at large is aware of them and the financial upside, or if they do exist currently, is it under the radar?

DR. TAN:

Yeah. You know, I think it's a mixed bag. I think it depends who you're talking to. Um, certainly there are early adopters that are out there in our specialty, just like early adopters in any technology domain we're talking about. Those are the ones that are going after it. They're looking for the best tools to help address their real-life problems. And in this case, it's billing. And the great tool is artificial intelligence.

On the flip side, I do think certainly that our field and our specialty of anesthesiology and all the subspecialties around it, really need to come around and we need to move really fast as a specialty to grow and learn about these tools. It's incredibly hard to do so. You know, I think anesthesiology in itself and the practice of medicine is busy enough. It's something that we spent our entire careers and educational life investing into. And artificial intelligence is another just large and amazing domain. And it's going to take some education for our specialists, those in training, those out in leadership in our field, and also collaborating with AI experts to bring them into our field to help us. So I think that while there are early adopters, the majority of the field still need to learn about what AI is outside of things they might use on their phone. And then, of course, there's a lot of hype around AI as well, and a lot of just conversations, a lot of headlines, but not necessarily like the real proof is out there yet for the value of artificial intelligence. And, you know, I think that has to be shown. And I think that takes time. It's going to take integration into practices to see the return on investment.

DR. STRIKER:

Now, two parts here. One, how can our listeners or anesthesiologists at large tap into this? What are some things that they can do to to reap some of these benefits? And then secondly, as a specialty or as a collective, what are things that we should be doing to tap into this?

DR. TAN:

Yeah. You know, I think I think to start with individuals, all of us really need to just read and learn and see what's out there. You know, to see beyond some of the failures of artificial intelligence or at least some of the fear of it, the lack of transparency and the fear that I might take our jobs. I think that's always been this kind of in the ether of the conversation of AI and anesthesiology. You know, a lot of people really talk about AI really being more about instead of artificial intelligence, AI actually might stand better for augmented intelligence, where we more view artificial intelligence as a partner in our field. And I think we have that mindset as individuals, and seeing how it actually already helps us with our decision-making day to day in non medicine fields, I think it allows us to kind of see the opportunities as an individual in the medicine field.

But then as a society, I think that's where the real opportunity is. As a larger group like the ASA, the ABA and many other organizations. I'm on the board, for example, for the Society for Pediatric Anesthesia. And one of the things we're committed to in our new strategic plan that we're going to release pretty shortly, is integrating artificial intelligence to every facet of our society. The educational sessions, you know, wouldn't it be great if we're having a conversation about medication safety and all the errors that result from it? So we have the classic conversations about medication safety problems. But then one of our speakers or, you know, as a part of it, will always bring in -- what are the technological tools that exist out there with artificial intelligence that's on the cutting edge, that might help reduce our risks of medication errors in the operating room. So I think there's a huge opportunity for conferences, societies just to to build artificial intelligence into their educational curriculums and conversations. I think that's going to really move the needle forward for our field.

DR. STRIKER:

Excellent. Well, Dr. Tan, thanks for joining us.

DR. TAN:

Thanks for having me. Dr. Striker, it's a pleasure.

DR. STRIKER:

Finally, because AI is an ever-shifting landscape of opportunities and challenges, we asked Dr. Hannah Lonsdale to share her thoughts on future trends.

Dr. Lonsdale, what does the future look like? Let's start with what we can look forward to. Is there something you expect AI to help us tackle or any potential future solutions you're particularly excited about?

DR. HANNAH LONSDALE:

Hi, thanks very much for the question. It's really useful to look forward to what is going to be beneficial about AI, because I think at the moment we're in this, uh what's known as a trough of disillusionment from the Gartner hype cycle, which is where there's been a lot of publicity around AI, people have heard a lot about it, they've got very excited, and then there's nothing. Like for most people in their clinical practice, they will not see a lot of AI clinical applications yet. And AI is in this weird place between oh, look how silly AI is. It doesn't really do anything useful. But the flip side of it, some people are thinking, oh my God, AI is coming for my soul. Um, so it's this really weird balance at the moment.

In terms of what I'm excited to see in the near future. Large language models have been game changers, so things like ChatGPT, um, are going to be able to give us assistance, for instance, with patient communication. I would love to spend as much time as I possibly could with my patients in the pre-op assessment phase, but with the increase in clinical demands, that's pretty challenging. And so seeing curated large language models that are able to answer patient questions like chat bots at all times of day, whenever the patient has a question, or if they have an hour's worth of questions, seeing a large language model that can really help with that. They can also summarize past medical histories, and so that will streamline our pre-operative assessments. They can give us literature summaries--that's already out there. So if I have a patient with a rare complex condition, I can go online and look for a AI guided literature summary and also AI augmented teaching. Then more clinically, AI is going to come into clinical decision support. Those tools that are going to provide us with knowledge and patient specific information that is more intelligently filtered and presented at the right time so that we can enhance our decision making and also hopefully improve patient outcomes. And clinical decision support gives us suggestions and summaries rather than mandates. A PEW study recently showed that patients are very open to doctors using clinical decision support from AI, but they are uncomfortable with AI as replacing the physician decision support. I'm also uncomfortable with that. It might give us personalized care. So there's an idea of the human digital twin, where we take all the information we have about a patient and construct a model of that patient. So that includes genomic data, laboratory data, past medical history. And it means that we can get increasingly personalized treatment plans for patients, which antiemetic might work best, or whether we should avoid TiVo because there'll be a longer wake up than there would be with inhalational induction, and many other decisions like that. I foresee that AI will help us forecast complications and patient outcomes, enabling us to better direct or pre-operative optimization. And also it may, for instance, help us decide appropriateness for ambulatory surgery centers by screening patients much faster than any human can. It's also coming in for our workflow and patient flow enhancement. So what medical center doesn't need more bed capacity or more O.R. capacity? Using AI to smart schedule to make the best use of the resources we have. And in time, we will see more intraoperative applications. Although they're the toughest because they are the highest risk and they need the most testing. Things like automated closed loop IV anesthesia are in development, and I'm really excited to see where that will take us.

DR. STRIKER:

What concerns do you harbor when it comes to how AI might impact the future of healthcare? I know we're always concerned about what the potential negative consequences of of AI, but what do you see on the horizon when it comes to negative issues?

DR. LONSDALE:

Absolutely. We I mean, we have so many real-life examples from, for instance, Silicon Valley, where the mantra of move fast and break stuff causes all kinds of problems. But as doctors, we can't do that because move fast and break stuff means harming people, which is absolutely not what we want.

One of the major concerns is data and bias in data. So AI doesn't add new bias to data, but the data we use to train AI models is taken from the real world. And so in real life, bias is baked into a model unless we take steps to take that out. And that may perpetuate inequalities and care that currently exists, for example, in marginalized groups. Another big concern is clinician well-being. Some of these new tools may save us time or give us more information. But as a specialty, we need to protect ourselves to ensure that we aren't forced to accept the decisions of these tools without clinical scrutiny, that we lose our clinical autonomy, and that any savings we make in time or cognitive load aren't simply repurposed to further stretch us beyond the current stressors of workforce shortages and ever growing patient complexity.

Large language models have their own set of concerns. They can present incorrect information, which is known as hallucinations. The internet has always contained misinformation, and I think we regularly see this in our patient conversations, that a patient has read something on the internet that is perhaps not well screened or written by a person who is very knowledgeable. And all of that data from the internet has been used to train publicly available large language models. So it then becomes baked into the answers, and they may not give completely accurate answers without curation by knowledgeable clinicians.

Another concern is that a study in radiologists examine the diagnostic performance of doctors who received either correct or incorrect support from AI. And giving these radiologists accurate suggestions drove a modest increase in their correct diagnoses. However, when they were given incorrect suggestions from the AI, their decision making accuracy decreased from 78% to 28%, though it appears that putting something in text adds an element of legitimacy that may be unwarranted, and we really need to be careful with that.

There's a lot of snake oil out there, a lot of solutions to problems that don't exist. And we can mitigate that by focus on problem solving and not solutions. Selling, maybe commissioning of ineffective or untested applications by people who don't necessarily understand the nuts and bolts and the limitations of AI, that is, those many examples from Silicon Valley. And it's my feeling that early premature implementation of tools that are not appropriate for what we need will actually harm trust, and ultimately delay implementation of the right AI based tools. My advice to listeners who are asked to look at these new tools is to treat a clinical AI tool as you would a new drug or a new anesthetic technique. Look at the evidence. Ask for large scale randomized clinical trials. At the moment, these are few and far between, although there are many excellent physician scientists working to change that. And it will happen.

And my final concern comes from academics. It's becoming increasingly common to see fabricated articles and hallucinated citations in submissions to journals, and without the appropriate skills to be able to detect those things that may be incorporated into the academic literature. And that's very concerning as well.

DR. STRIKER:

Let's talk about anesthesiology specifically here. I can't imagine our jobs will not look differently ten years, even 30 years, down the road for a number of reasons, but particularly because of AI. How do you see this specialty evolving?

DR. LONSDALE:

I think we have to think bigger picture even than just our specialty here. So the Pew study I mentioned earlier that suggests patients are open to clinical decision support, also suggested that they are open to partially autonomous surgery. And so a bigger picture is how will that change our practice? Even today, everyone loves a nice long robotic case, and partially autonomous surgery when it is introduced is not going to be fast either, so we will need to adapt to that big picture. I think it's going to be a slow evolution, though, because compelling evidence of effectiveness needs to exist before we can adopt these tools in a widespread fashion. It's not going to be a sudden and dramatic change. Will things look very different in 30 years. Yes, without a doubt. But every sci fi novel, every distant future prediction I've ever read, looks kind of adorably quaint when viewed from the present time, so it's hard to see where we're going. But a lot of interesting AI tools and development have been explored in details by other guests on this podcast. I expect to see more clinical decision support, real time dynamic adverse event prediction; using machine vision, so cameras and and perhaps maybe even autonomous robotic support for things like ultrasound guided regional anesthesia, invasive lines, intubation and bronchoscopy. But more importantly, I think we're going to see a reduction in those low-level repetitive tasks. And that means that they'll free more time to focus on complex tasks, things that require expert level cognition. And then we can focus on education of the next generation of anesthesiologists, and in keeping the human in the process of caring for each patient rather than by replacing the human clinician.

DR. STRIKER:

Finally, when it gets right down to it, our top priority is always patient safety. And so distilling all this down to its essence, do you believe AI will benefit patients?

DR. LONSDALE:

Absolutely. And the reason for that is there is still scope for improvement in safety. There was a 2020 study showed a poor rate of preventable in-hospital mortality of over 3%. And reducing that is just the tip of the iceberg. By personalizing care and reducing the low-level tasks, allowing physicians to do what they do best, focus on the patient, we're going to get more right, first time. We're all currently overwhelmed by data: academic literature, vital signs, baboratory results, complex past medical histories, genomics, scheduling challenges. And somewhere in there, as a patient who might be scared or have questions or be able to tell you something vital that you can't glean from that chart review. Any human brain, yes, including those of attending anesthesiologists, can only retain 3 to 5 chunks of information at any given time. And so using AI based clinical decision support to reduce the conflicting demands on our cognitive processes and help us to dedicate more time to that human connection for our teams, our trainees, and most importantly, for our patients.

DR. STRIKER:

Dr. Lonsdale, thank you so much for joining us.

DR. LONSDALE:

Thank you very much.

DR. STRIKER:

And thanks to all of you, our listeners, for tuning in to this special short series on AI. If you missed the first AI episode, you can find that. Tt's episode 157—157--on ASA's website or your favorite podcast platform where you normally get centralized. We appreciate all the various members of Asa's Committee on Informatics and Information Technology sharing their expertise, and we certainly look forward to seeing you here again soon.

(SOUNDBITE OF MUSIC)

DR. JEFF GREENE:

Hi, this is Dr. Jeff Green with the ASA patient safety editorial board. Medical malpractice statistics are sobering. 1 in 3 providers will be sued over the course of their career. Anesthesiologists experienced an annual rate of paid malpractice claims of 11.7 per 1000 physician years, with 10% of paid malpractice claims reaching over $1 million. There are increasing anesthesiology claims for Mac, NORA, and office-based surgery procedures. Anesthesiologists can reduce exposure to these increasing risks for medical malpractice claims by one. Understanding the risks of Nora and Mack anesthetics, and educating stakeholders to recognizing and reducing the incidence of respiratory events in Dora and Mack anesthetics. Three monitoring and deterring complications in the postoperative setting before rescue is required. And finally, by thoroughly examining and documenting dental disease prior to anesthesia. With these safety tips in mind, it is hoped that anesthesiologists can reduce the incidence and severity of malpractice claims in the future.

VOICE OVER:

For more patient safety content, visit asahq.org/patientsafety.

Get to know your new knowledge assistant, BeaconBot, just for ASA members with a focus on anesthesiology content and information not widely or publicly available. Ask a question today at asahq.org/beaconbot.

Subscribe to Central Line today wherever you get your podcasts or visit asa.org/podcasts for more.