First of all, I'd like to thank my co-presenters here this morning. Thank you very much for coming here early to RSNA and for your excellent talks. And also thank you to the live audience for being here, for engaging in discussion with us later on, and also welcome to the audience online. So I will start us off with an analysis of communication errors in radiology, and my colleagues will discuss ways to present them. We will spend a lot of time on why should we care about communication errors. Simon Sinek said in his book, How Great Leaders Inspire Everyone to Take Action, that we need to know why something is important so that we can become engaged. We will then discuss where communication errors occur, when they do occur, what they are, what is the root cause of them. And then I will add one personal perspective of what we can do about them, possibly. So there are at least four primary reasons why we have to pay attention to communication errors. So communication errors are the number one root cause of adverse events, causing an estimated number of deaths up to 100,000 per year in the U.S. And this is confirmed by data from the Joint Commission, which reported that 65 percent of SREs that are reported to them are due to communication errors alone. A Danish study from 2011 found that about 52 percent of SREs have communication errors as a cause, although this study terminated data collection after a while because they found that once you start looking for communication errors that could have prevented an error to happen, where communication could have prevented an error, they were always able to find one. The same sort of data we found at our own institution, where if we look at our hospital incident reporting system, about 56 percent of errors in radiology are due to communication errors. However, in our peer learning database, only about 5 percent of cases that are reported are due to communication errors. And that shows a general lack of awareness in radiology about how important those are. Now, the impact of communication errors on patient care is significant. Twenty-three percent of patients will have major complications, such as a delay in the diagnosis of a malignant finding or an emergent finding, unnecessary surgery, or even a complication after unnecessary treatment. But even a delay in a benign diagnosis is not a minor event. When we are dealing with a delay in diagnosis of an abnormality on a mammogram that was not breast cancer, it can cause great emotional harm to the patient who is experiencing this delay in diagnosis and is worried all along. Additionally, unnecessary follow-up imaging is not only harmful to the patient, but costly to society. The impact of the second reason to care is that communication errors cause dissatisfaction, and the effect on patients is severe and completely underestimated. Studies in primary care have shown that patients report emotional harm worse than that from a physical condition when they are experiencing a communication error. Patients feel disrespected and see health care providers as incompetent, not being able to communicate, which they firmly believe is part of our job. Or in case of transcription errors, that radiologists are not paying attention, not doing a thorough job. This leads to a loss of trust in the medical system and leaving the care environment altogether, and patients avoid medical care even years later. Our referring physicians also lose trust and send their referrals elsewhere, which leads to loss of revenue. Communication errors also impact our workflow efficiency. While many can be remediated in the moment, this requires extra effort and time, such as making additional phone calls, rescheduling patients, or trying to work a procedure patient in as an add-on. And this we can, at the moment, ill afford in the current times of short staffing. Much of the work falls also to dedicated QA staff or managers who are dealing with incident reports or patient complaints. And service recovery, while this is possible, it takes an inordinate amount of time. Lastly, communication errors are a malpractice risk. 80% of malpractice cases have a communication component. Remember this number if you're taking the CME test. And communication errors are in the top four single errors in radiology after diagnostic error, which is the most common, procedure complication, and failure to recommend for the testing. Interestingly, the total indemnity payments are 15% higher in cases of delay in diagnosis with a communication error than without. So documentation of communication is critical. Although one wonders why. Well, the reason is that jury members and patients understand well how a communication error happens, but they do not understand how communication errors can happen in medicine. And you can see here on the right, this is a quote from a judge during a malpractice trial. This is from the paper by Leonard Berlin quoted below. Jurors and patients feel that everybody can communicate. They know, or they think it is easy, that they can do it. So how come that we as physicians cannot do it when it is, in fact, part of our job? But is it really that easy? We shall see on the next slide. And let's see what happens when patients make their way through the department. So not surprisingly, it's actually in radiology, it's quite complex. So during the patient's journey through the radiology department, from ordering the examination to scheduling the examination, performing it, test reporting and interpretation, performing a procedure, multiple staff are involved inside and outside of radiology. And there are many interchanges that are happening. In an earlier analysis of communication errors in our department, we found that 30% of errors occurred while patients were being taken care of in radiology during the performance of the examination, and 48% were noted during result communication and in the radiology report. Errors at the time of ordering or scheduling the examination or in post-procedural care were less common. However, the severity of impact between errors that occurred at result communication and those at other steps are the same, and we need to pay attention to all of them. As radiologists, we are understandably very focused on result communication. When do errors occur in terms of who speaks to whom? This is where things get complex, to say the least. Rabaul et al. described in 2011 that handovers are the most vulnerable times of communication, which is also true in radiology. Most communication errors occurred between radiology staff and staff from other departments, about 78%. With communications between the radiologist and the referring physician being the most frequent vulnerable one at 35% total. This was followed by communication between technologists and transport staff in 25%, which that's probably more of a local issue for us, but then also nursing staff to the floor, the technologists and nursing staff, and 22% occurred within the radiology department and the communication between the radiologist and the technologist is the most important one. Awareness of the importance of these communications is needed when we are trying to avoid them. Now what are the root causes of communication errors? A classification system for communication errors was developed by Lindgaard et al. for the operating room, and this divided communication errors into those that are due to speaking to the wrong audience, content errors that are failure in the information being transmitted, such as missing or inaccurate information and unclear information with multiple meanings, errors in timing of the communication, and errors where the purpose of the communication is not fulfilled when the transfer of information is incomplete. Transfer errors were the most common type of errors followed by content errors. Looking at the sub-type of errors, the four most frequently encountered errors, causing 80% of errors overall, were missing information, lack of closed-loop communication individuals or teams, lack of contacting key individuals, and inaccurate information. Lastly, a quick thought on one possible way of instituting countermeasures, which is human factors engineering tools. They have been shown to improve performance in many aspects of healthcare and have been used to address communication errors, which we may use in the setting, such as checklists for missing information or forcing functions for inaccurate information, where these techniques have been proven useful, communication protocols and standardization for loop closure, and automated alert or forcing functions for the contacting of key individuals. So in summary, communication errors in radiology are harmful and costly in many ways. They occur most commonly when the patient is undergoing the examination during interaction between the radiologist and the referring physician or the radiologist and the technologist. They are related to content and transfer of information, such as missing information, lack of closed-loop communication, lack of contacting key individuals, and human factor engineering tools may help decrease these errors. Thank you very much for your attention. And now I'd like to invite my colleague Atul Chingari to the podium. Dr. Chingari is Associate Professor at Harvard Medical School and Vice Chair of Clinical Operations at DMAS General Brigham. Thank you. Thank you. Good morning, everyone. Thanks for joining us for this session. So every diagnostic test has some degree of uncertainty associated with it. And this is a non-exhaustive list, but some of it can be test-related, such as limitation of the modality or variable protocols, or it could be related to report content, such as report structure, missing details, or report language, terminology that's used in the report, or expression of diagnostic confidence, or how follow-up recommendations are communicated. And we have to understand that we cannot completely eliminate the uncertainty, but we can try to minimize it and express it in a way that the referrers and patients understand it. So that's where report standardization plays a role, because it helps us minimize this uncertainty and express it correctly, and also helps improve report quality and actionability. And standardization is the key to actionable reports. And we can do that by standardizing the scanning protocols, which I'm not going to talk about, or standardizing the report content and structure, or report language. And that's going to be the focus of this talk. So we'll talk about, very briefly, talk about structured reports, and a subset of those can be disease-specific structured reports. Then we'll talk about standardized terminology, diagnostic certainty, use of the word normal, which is often dreaded in imaging especially, and follow-up recommendations. So over time, the radiology report has evolved from something that looks like a wall of text, which was very easy to dictate, very flexible, but has marked variability in terms of length, content, completeness, to now most of the reports have some degree of structure, and increasingly disease-specific structured reports, which are more consistent, more specific, complete. Some people complain that they can be pretty complex and time-consuming, but on the other hand we have also noticed that once you get used to it, they can be very efficient. And so structured reports, and to some extent disease-specific structured reports, are the feature of radiology. They are more complete, they improve adherence to guidelines, they are associated with less uncertainty and improved communication, and overall are more actionable, they lead to higher provider satisfaction, and add value to patient care. So we should use structured reports to the extent possible, but then why care about report language? Well, let's see an example. So this is an email from one of the oncologists that I saw a patient's name in the clinic today, she has Mendelsohn's sarcoma, name of the radiologist, gave very detailed CT report with a plethora of small findings, but I cannot understand what the findings mean or what I need to do next. Patient is low risk, but worried sick. Now here the complaint is not about missing findings. It's about how the findings were communicated, because report language impacts the certainty and actionability of radiology report. And that's why we are going to focus on these four topics. This patient had ovarian cancer at an outside hospital. It was read as no peritoneal abnormality, mesenteric nodules. Now personally I don't agree with it, because mesentery is part of peritoneum, but that's how it was reported. Patient had second opinion at our institution, where one of the radiologists read peritoneal nodules representing carcinomatosis. And then patient got really worried, thinking there's a new peritoneal disease. Now oncologist was easily able to talk to patient, sort it out, but this was a confusion created solely by radiology, and we should not be doing that. And for that, we should use standardized terminology. We have many different RADs and lexicons available for this. I'm more interested in GU, GYN, so these are some examples in that space. But pretty much every area of radiology has this. And that's how we can improve the consistency and clarity of our reports, make them less uncertain, and increase the generalizability of the reports. So if someone showed you this picture, are you going to say this is suspicious for a puppy, but a cat or a small tiger cannot be excluded? No, that's ridiculous, right? So similarly, if someone showed you this picture, all you have to say is clear cell RCC, that's it. You don't need any differentials. The point is, whenever possible, avoid unnecessary hedging and differentials. Yes, of course, you cannot give a single diagnosis in every single case. You have to give differentials sometimes. There is going to be some degree of uncertainty, but to the extent possible, like limited number of differentials, do not list every diagnosis known to mankind just to be right 100% of the times, because then you're not being very helpful. And that's what leads to concept of diagnostic certainty, or radiologist's subjective confidence in the finding. It's a key component of any actionable radiology report. It helps referrers make important management decisions. And we use numerous phrases to convey this diagnostic certainty. And this is common in every area of radiology. For example, in abdominal CTMR reports, 86% of them have at least one diagnostic certainty term. And if you're using terms like probably, highly suggestive, worrisome, concerning, not inconsistent with, people can get very imaginative around there. So do all these words mean the same thing to us in terms of our subjective confidence? Again, do they mean the same thing to our referrers and patients? And more importantly, what you mean, the degree of confidence, is that the same thing that the referrers interpret? There are many studies that show that there is a significant disconnect amongst radiologists and between radiologists and referrers in terms of meaning of these many words. And we did a study at our institution just to understand how radiologists use these words. And we found a very significant variability. People use many different words to express similar degree of diagnostic confidence. And some words like probably, some people used it to express a very low confidence. Some people used it to express very high confidence and everything in between. So if you're a referrer or a patient, you see a word like probably or suspicious, you have no idea what the radiologist means. And that has important implications on the discussion with the patient. So to avoid this kind of confusion, we created this diagnostic certainty scale to standardize the expression of diagnostic confidence. And now many other, there is a lot of interest in this. Many institutions have their own efforts along these lines. And this way, we can standardize the communication of uncertainty or certainty in our report. And at our institution, we found a significant improvement. And now our referrers kind of expect radiologists to use this. But of course, there are some barriers. People like to have individual expression in the report. But when we educate the radiologists, once they understand the problem with it, most people are willing to change. But that requires practice change, which is always hard. We have to support them through this change. And then there is this concern, medical-legal concern. What if I say it's a high likelihood and I'm wrong? Well, that's possible. But the good thing is, any time you're expressing probability, inherently you're saying that this is not 100%. So our quality team and risk management team feel this is a safer approach, actually. And having some kind of standardized system is better than letting somebody else interpret your words. Now, another area is these sort of negative terms. Within normal limits, no definitive abnormality, unremarkable. Wherever possible, we should use the word normal. And yes, many radiologists are not comfortable with it. And we certainly cannot use it in every single case. But if you see that's a liver, normal on CT or MRI, just say it's normal. But one question that comes up is, how do I know it's really normal? Well, if you go to your PCP, have a physical examination, it's normal. How do you know it's normal? Any time you get a diagnostic test, you're operating within the confines of its limits. So if you see an organ normal, just say it's normal, it's fine. Also, some may ask, why should I use this word? Does it improve patient outcomes? Well, we don't know that. Unfortunately, I don't think that has been formally studied. But we do know that use of word normal reduces the confusion, improves the communication. Another thing that often comes up is follow-up recommendations. Sometimes they can be vague, indecisive, confusing. And that leads to more uncertainty. So any follow-up recommendation should have basic elements, which body part, which modality. Sometimes you may give one or two options and rank them first or second. And timing. And it also depends on whether you're making recommendations to a specialized referer or a primary care practitioner. For PCPs, you need to be even more concrete and very clear about communication. So for example, instead of saying something like, further evaluation with MRI can be considered if clinically appropriate, you see all kinds of vague language. Instead of that, just say, recommendation, liver MRI, four to six weeks. It's very clear, not confusing at all. So this kind of change management requires some strategies. Radiologist education goes a long way. Especially trainee education is very important. Multidisciplinary consensus to secure buy-in. That way we create some expectations from radiologists as well. Giving radiologists feedback using AI and LP tools. Including this in your QI and incentive programs. And strong leadership engagement, both in terms of authority and influence, goes a long way. So in summary, uncertainty is inherent to every imaging test. It cannot be eliminated, but we can try to minimize it and appropriately communicate it. Standardized reports reduce variability and uncertainty, and they are more actionable. And disease-specific structured reports, or at least structured reports, is the future. We should minimize the uncertainty related to report language by using standardized terminology, clear expression of diagnostic certainty, using the word normal wherever possible, and clear communication of follow-up recommendations, because that's what leads to improved patient care. Thank you. Thank you very much for this excellent presentation. Our next speaker is Dr. Hannah Safar. Dr. Safar is a social professor of radiology at University of Pennsylvania and vice chair for quality and safety. Welcome. Good morning, everyone. Thank you. So in the next few minutes, I'm going to try to cover a little bit of the what, how, why, and how of e-consults. Talk about some practical barriers for people who may be contemplating this at your institutions and also some future challenges. So what are e-consults? Well, it's basically a provider-to-provider consultation, which we've all experienced many times in our lifetime. Here though, you leverage an electronic platform in order to do that consultation, which could be either live or synchronous, as you see here, or it can be asynchronous, meaning it goes into a queue and somebody would look at it at a later point in time when it's convenient for them. Overall, the goal of e-consults, very clear. We're trying to provide efficient and improved access typically to specialty expertise, right? There's a question a provider has that's outside their bandwidth, outside their ball of wax, so to speak, and they want to get help with this, and so oftentimes this will involve primary care physicians. Asynchronous e-consults have been used at some institutions, and I would say this is really similar to the traditional reading room consultation, right? There's all kinds of interruptions here, the phone rings, you have to stop reading out the trainee, you have to put down what you're doing in your work, and then eventually you get back to the case you were reading, and as we all know, this is associated with a higher rate of errors for radiologists because you may forget exactly where you were, like which organ you were looking at, where in your search pattern, et cetera. This type of asynchronous e-consult has been used at some institutions, NYU in specific, and here you can see that a provider will enter a request into an electronic platform and sit and wait for the radiologist, a radiologist, to acknowledge their consult, who will then open up the case and go over it with them. In contrast, you can have an asynchronous workflow, and that's really less disruptive to the radiologist, so more conducive to us, the way that we operate. It's going to minimize our interruptions and our error rates. Really more suited, I think, to the outpatient setting, obviously, and in this case, a consult is placed, and then at your leisure, the radiologist, either between cases, after hours, or you may be a dedicated radiologist assigned to the service, you would open up those cases and answer those questions. So again, the choice between whether you'd want to do synchronous or asynchronous, it's really going to depend on the care setting that you're trying to address here, whether your providers are emergency physicians, if you set expectations that you could even send an urgent question, or are you going to make this really more outpatient-friendly, where you say, I'll get back to you within 72 hours? So that's just something that has to be contemplated at your institution. So why? Why do radiology consults in the first place? A couple of reasons. We all are familiar with rising imaging volumes. Surprisingly or unsurprisingly, both providers and radiologists prefer this method of consultation, and the literature does show that it's associated with decreased costs. So here we have a graph showing that imaging rates in U.S. and Canada increased between 2000 to 2016 for all imaging modalities, with the exception of nuclear medicine, and you'll see that that growth was really occurring among adults, particularly among older adults, where you see the steepest curves. Data has shown that, and I want to clarify here, asynchronous e-consults are preferred both by referring providers and radiologists. So to clarify, e-consults are preferred by all referring providers, whether they're synchronous or asynchronous, with rates somewhere in the 70 to 90 percent. This particular article showed that radiologists also had a high likelihood or preference for e-consults. I just want to clarify, these were radiologists who were consulted asynchronously, and you can do the math and understand why they have higher rates of preference than those who are consulted synchronously. There's also data that shows that e-consults are associated with decreased costs. So this comes out of a health system in Connecticut, and what they did is they focused specifically on Medicaid patients, and they just let the providers choose whether they wanted to go the traditional route of sending a patient for a specialist referral or if they wanted to use the e-consult. And what they found is across these four different specialties, and there were variabilities, that they saw decreased costs, again, ranging depending on which clinic was using the e-consult, but overall a saving of $84. So that's $84 per patient per month, which annualized at their health system to almost $600,000 of savings just among their Medicaid patients. So you can extrapolate to say what that might be for other payers as well. I think one of the most important and compelling reasons for us to document consultation is that we all know that most institutions have policies around wet reads, right? So at an older time, we would do curbside consults, wet reads. Well, the problem is that there's no documentation in the electronic medical record, and if that consultation alters patient management, you need to have a record of that. The referring provider wants it, the patient wants it. And in our case, it's also important because it's a record of the radiologist's time that was spent providing that consultation. And if you think about making this scalable and sustainable, and you want to get financial reimbursement for providing the e-consult, you need to have that track, so to speak. So it's important for that reason too. So what is the ideal radiology e-consult design? Well, it's going to have a couple of things. You want to embed it in the EMR so that you can easily access the patient's medical record and understand the question a little bit more. It should be bi-directional, right? You want to continue that conversation until the referring provider has received their answer. And usually that's just going to be one communication, but there are times, and I'm sure we're all familiar with that, where you may require a slightly longer, you know, several chats back and forth before the provider feels comfortable and clear as to what you're saying. And again, as I've pointed out, asynchronous is really going to be ideal for most radiology practices. This is data that comes out of the Partners Health System in Boston. They did a study where they looked at the rollout of e-consults across five different clinics that you see here. And overall, there were about 6,500 e-consults. Interestingly, what they did is the first thing is they took a subset, about 11% of these consults, and they did a manual review to determine if the consults were appropriate, meaning was the urgency and the complexity of the question and the type of question appropriate for sending on an e-consult. And they found that overall, 70% of these consults were appropriate. And you can see that there's some variability. So psychiatry had relatively high rates of appropriateness compared to some of the other clinics. Not surprisingly, those clinics that had higher rates of appropriate e-consults were more likely to reduce unnecessary specialist referrals. And that's important, right? So if you're using the tool correctly, it's more likely that you're going to avoid a visit. And the way that they defined avoided visits is they looked for a visit by that patient within that specialty clinic up to 120 days after the e-consult was placed. So a pretty long time interval. Similar data, this time coming out of Canada, where they looked at a very large network. This is pre-COVID data, interestingly. And what they found is that primary care providers received advice on new or additional course of action in half of e-consult. So in half of e-consults, they learned something that they didn't already know or they were not already suspecting simply through this electronic consultation. And you can see that the likelihood that it would change action, which is showed here in, excuse me, the burgundy is going to vary again by the specialty. So some specialties are more conducive to being effectively used with e-consults than others. That would be the way to look at this. They also found that nearly three quarters of specialist referrals were avoided completely through the use of the e-consult. So again, these are primary care providers who may not be familiar with all of the nuances of care and specialty guidelines and follow-up recommendations as Atul was saying. So again, that's going to have a higher impact in this audience. So what about radiology? That's going to be the actual next question. We are a bunch, mostly radiologists sitting in this room, I assume. So it turns out radiology consultations are a much smaller piece of the pie. This is actually the same group. So the same group in Canada, taking a look at their data, there's a longer time interval. So now we've gone up to almost 21,000 e-consults across the entire, I think it was the state of Ontario. They included all of the clinics there. So only 1.5% of those consultations, those e-consults were for radiology. And what a punch, man. I mean, even though it was only 1.5% of the consults, they found that those e-consults altered patient management in 55% of cases. You might want to stash that number away in your mind. 55% of cases. So very high impact. And they avoided unnecessary referrals and tests in about a quarter of cases. So 1.5% of consults overall, very, very high yield return on investment. You might also ask, well, what about the time that it takes to answer these? Well, the providers spend typically about less than 10 minutes answering these consults. And in this particular case, they found that about a third of radiologist responses were accomplished in less than 10 minutes of time. That's consistent with our experience as well. We typically find that the overwhelming majority of these are answered within less than 15 minutes. So what about our health system experience? Well, we have a large health system with multiple hospitals. Many of you do as well. Just in 2022, we had about 7 million outpatient visits. So we decided that for us, we really wanted to focus on the outpatient setting when we embarked on this e-consult journey. In addition, around the time that we were contemplating this, we had an expanding network of primary care providers and they really needed access to all kinds of expertise, including radiology expertise. So what we did in our case is that we were able to actually leverage an existing platform in order to add our radiology e-consult. So here we are down here at the bottom, and I just wanna show you that we have e-consults available for multiple different specialties. So we worked with our counterparts and other departments in order to take advantage of this tool. So anytime a consult is placed, there are several questions that are gonna be asked and I put them here just to make it easier to look at. And this is something really important as well. If you're gonna make the investment in going into e-consults, you wanna think about what metrics you want to extract on the backend in order to show who's using it, why are they using it, and what trends can you pick up in order to finesse it on the backend when you move to your version 2.0, so to speak. So one of the basic questions, should you have referred this patient to a referral if the e-consult was not available? That's a very low-hanging fruit. And if you read the literature, a lot of authors have taken the time to do that. As a radiologist, when we pick up this cue, it actually just shows up right over here in our, oops, sorry, right over here in our electronic medical records so that we can access it when we want to take a look at it. The most common categories that we find for questions are the next best test to order, follow-up guidelines, going back again to what Atul was talking about, clarification of an impression. So again, just as Atul was saying, they read the impression, the patient and provider, and they don't understand what it means and what to do next. And then occasionally there's also some contrast in protocol questions. Similar to the group in Canada, we found that the overwhelming majority of these consults are for body and neuroimaging. So we make sure that we have a body and neuroradiologists available in order to answer these questions. And many of them, in fact, are quite simple. And so even though they may deal with a specialty outside of your own, you would be surprised. You will probably have the expertise to handle them. At least that's what our group has found. The overwhelming majority of these consults are placed by PCPs, even though e-consults are available for all specialties. And we've seen some interesting changes in utilization over time. Overall, I would say there's been an upward trajectory, and we can all agree on that. There have been some handoffs between when radiologists come and go, which I've shown with these two orange arrows. This last arrow, I wanted to pause on for a moment because this is a common refrain for all, I think all IT projects, which go through growing pains, which is that we changed our IT mechanism in anticipation of having financial reimbursement. And that created a disruption, if you will, within our system so that previously, any of our radiologists, we have six radiologists in the pool, they could go in, they could click on a question, and if they didn't feel quite comfortable answering it, they could back out of the queue and let somebody else come in, and it would still appear that the consult was unread. However, with this new mechanism, where apparently, I guess we could have done a better job of communicating with our IT staff, in addition to our providers, that is not the case. And so that has led to an increased drag. So whereas before, 75% of consults were answered within one day, now we're seeing much higher spikes coming. And again, it's all due to this change in workflow. Again, a normal growing pain, something we're gonna work through and we'll figure out how to get there. There are several barriers if you're considering implementing e-consults. As you may imagine, there is no standard reimbursement across payers. How does this affect compensation for radiologists and for specialist providers? And you obviously need to have a strong IT infrastructure in place. There is literature about how you can go about doing this. If again, you are considering this, there are systems in place about how you wanna measure your prevalence, track referrals, and then use that data in order to engage in meaningful QI work, which would obviously be the end result here. So when you're moving to that version 2.0 as your system matures, you can actually use some of the learning that you've gained from your first system in order to try to do some cute things where you could actually embed a frequently asked question, for example, in a dropdown menu, so that a referring provider or PCP can be helped or guided into figuring out whether it's an appropriate referral before ordering it. And I did wanna end on saying that there was a systematic review that came out in 2020 that looked at all of the different e-consult literature out there. And I think something that's really important is to have some humility about this, that there is only modest empirical evidence for effectiveness of e-consults on important outcomes. So we have a lot of work to do. We embrace the idea, providers like it, radiologists like it, and we need to have more robust data on how it actually affects patient outcomes. So thank you very much for your time. Thank you. Thank you very much. Our next speaker is Dr. Matt Davenport. He is professor of radiology and urology and service chief and vice chair of the department of radiology at the University of Michigan. Thank you. Thanks. Hello everybody. So I'd like to challenge us to think for a second about what our core business is. And I'm asking this question because this whole session is about communication. And a lot of people think about their core business is creating reports. And I don't think that's true. So here's a series of companies that have existed and for various reasons don't exist so well anymore. And I want you to think about what's common about them. These companies in part define themselves by their product or delivery method rather than the customer need that they served. They also had challenges in which they ignored technological change and they had some inertia. They were paralyzed by their own success and they failed to pivot when customer preferences changed. So think about that in the context of a radiology department. What's our product and delivery method? It's a report. And what's the customer need we're serving? It's to reduce uncertainty. So in that lens, you might ask yourself why is this whole session on communication? It's because our goal is not to make a report. Our goal is to reduce uncertainty because that is our core business. So here's an example value chain where we have an image. That's me, radiologist. We make reports, it gets delivered to people and then some clinical action is taken. And in this value chain, what is our value proposition? It's information. We're in the information business. We're information brokers. Our function is we extract information from imaging data and then we deliver that information and our intent is to reduce uncertainty and the purpose of it is to improve health. Now, a lot of radiology training is to take the information that's embedded in this test, this imaging test, and figure out ways to optimize our extraction of that information from the test and put it inside our brain. That's what we think of as making a diagnosis. And then also in radiology training, we spend a lot of time trying to make an accurate report. So take the information we extracted that's now in our brain and stick it into a report. We encode that data into words. Now, we assume that this accurate diagnosis extracts full value from the test but we're learning now that from opportunistic screening, for example, there's a lot of information in the test that we've so far not been pulling out. So there's penalties that happen, a loss of information fidelity at each step. And as information brokers, it doesn't really stop there because to maximize our value, we have to be sure that we don't lose information further downstream. So from the report, that report can be misinterpreted by the patient or by the treating provider. And when we put the things in our brain onto a report, we might say, well, I mean, we put down what we thought, what's the big deal? How many of you ever read an email or read a text message in which your conclusion was completely different from what was intended or you sent a text message or an email and the information in there was misinterpreted? This is very common because most of the information content that's delivered in a message between humans is nonverbal. And this has been studied extensively in psychology literature. And from there, it's still more porous because even if the information's received, it may fail to change the person's perspective and it may be that they're trying to do something but no action actually takes place because it was misunderstood. And then it could be that the action that's taking place is not actually effective. So if you think about value creation, what's our core business? We're trying to improve health. The only way we can do that's by changing behavior. The only way we can change behavior is if we communicate the information that we've extracted in a way that influences behavior change. The Freibach and Thornberry hierarchy, which was released in the 1990s, talks about how we don't create value unless there's a patient outcome difference. Simply creating reports doesn't change outcomes by itself. So in that lens, communication's critical because it's central to our business. We're information brokers. And without communication, we don't exist. So here's an example of how we can think about this rubric around the idea of closed-loop communication and why it might be helpful. There's this information that's embedded in the test and there's some information lost when I try to extract it from the test. And then there's more information lost when I try to encode that in words in a report and more information lost when the urologist tries to read what I said. Closed-loop communication, when we're talking on the phone and having a dialogue about it, the intent of that in this rubric is to bypass this encoding step, which prevents that information loss, and then mitigates this failure to understand what's written by my words. So it improves the fidelity of information transfer. Now the ACR guideline talks about closed-loop communication and there's three times in which it's recommended or to be done. And I think from a medical legal standpoint, we're sort of held to this. Immediate or urgent intervention is gonna be based on that result, or if it's discrepant with a prior interpretation, or if it's, quote, significant and unexpected. But I would like us to think beyond this. This is kind of when we're being held from a medical legal stand, but I think a better question then, when do I have to call? Which is, again, it's focused on us. From a value creation standpoint, a better question I think is, how can I optimize communication to maximize value creation? Because that's, again, that's our core business of what we're doing, is to deliver information and improve health. In 2015, the Committee on Diagnostic Error in Healthcare released this important report. It's over 400 pages long and it talks about improving diagnosis in healthcare. This is the same crew that came out with To Error is Human and Crossing the Quality Chasm. And if you've not taken a look at this, it's over 400 pages, so it's gonna take a bit to look at it. But take a look at it, because it's got a lot of compelling information. They focus on five key tenets of team-based healthcare. And the two I wanna anchor to is effective communication and mutual trust. Trust is team members earn each other's trust. And effective communication is the team prioritizes and continuously refines its communication skills. It has consistent channels for candid and complete communication, which are accessed and used by all team members across all settings. So what's the deal with this trust piece? I mean, I said what I said in my report. I mean, can't they just read it? Like, why do they have to trust me? Well, think of this thought experiment for a second. You have this omnipotent creature and it's releasing flawless reports. They're perfectly accurate and they send them to your institution. What's gonna happen? You're gonna ask them to do a reinterpretation. You're gonna do a second interpretation of that scan. Why? Because you don't trust the person. You don't necessarily know who they are. You don't have like a bank of information in your mind of who they are. So the trust isn't there. If there's not a trusting relationship, there's no clinical action taken based on the report. High quality communication improves fidelity and creates trust, increases your influence and generates material value for the business function that you're providing, which is to improve health. That's why this kind of meaningful communication is so helpful. There's all kinds of evidence that supports better communication. Here's just a handful of them. We know that better communication reduces hospital time and costs, reduces diagnostic error, results in better test selection, improves diagnostic performance and results in fewer diagnostic errors. So we wanna over communicate to improve our value because we are information brokers. And if that's our business as information brokers, how do we maximize our value? And this little article is something where me and one of my colleagues wrote about this. Fundamentally, we're translators. That's our core function. And this is a quote by Richard Barron. No matter how many great new technological developments are implemented by radiologists, meaningful clinical effect and outcomes will only come because a radiologist is an outstanding translator of the language of images. And that's an interesting construct. What does that mean? It means we translate and communicate imaging data to change provider behavior with a goal of improving health. This is a cool quote from George Bernard Shaw. The single biggest problem in communication is the illusion that's taken place. A lot of times we think we're saying something and what someone hears us saying, whether they're reading it or hearing it by voice, they're not hearing what we're saying. How many people have gotten into an argument with someone because what you said was not what was heard? Probably everybody in the room. This is a study that we did at our institution looking at how in-person discussion about cases changes clinical decision-making. There were 100 patients who had already had their radiology reports written, they were already read, people already evaluated them, they had a diagnosis and a treatment plan, but then there was a conversation. And after that conversation, we recorded the diagnosis and the plan and measured discrepancies between those plans. So the function of talking, just having a conversation about the case resulted in a change in operative plan about one out of five times, which is a huge percentage just by having a conversation. Here's just a Venn diagram of cases that were discussed. And you can see in 43% of the cases there was a change in surgeon management. And there was actually, sometimes they would change their impression and not change the management. Sometimes the management stayed the same and they would not change their impression. When we discuss, have a conversation, it creates a shared mental model. So this is just one example where the person had a small bowel obstruction at two different time points. And the radiologist said, oh, by the way, these are the kind of the same luminal place for this transition point. And the surgeon said, I didn't know that. I didn't know it was the same transition point. And the radiologist said, well, I didn't know that was important to you. We didn't have that in the report. So having these dialogues is really helpful. We do a really good job of cloaking what we know in complicated words and jargon. Whenever you use jargon, you're reducing the information fidelity transfer. You're making it more confusing. So think about this kind of like hierarchy of words. We've got physics that we learned. Then we have appearance, anatomy, and physiology. And then we have our diagnosis. And we're always told, don't put impressions in the finding section. The finding section is where we see all the confusing things and we'll tell you what you need to know in the impression. But I would challenge that. So what is this structure that I'm describing? Most people probably don't know, because I wouldn't know. Okay, this is now appearance, anatomy, and physiology. Or we can give it a diagnosis word. And this kind of silly rubric here, you can take that and apply it to lots of the things that we do. The more we cloak it in complicated words, radiology, jargon, the harder it is for things to understand. Here's another one for you. There's an expanding carpet of red and orange and yellow photons emitting from energetic electrons falling to their base state at approximately 800 to 1200 degrees centigrade. I mean, that would be a ridiculous way to say that there's a fire. So most of the time when someone sees a fire, they're just gonna use a single word. And that's the same idea, they're using a diagnosis word. Here's an example from radiology. There's a unilocular cystic lesion in the pancreas A lot of words. All these things are kind of probably not necessary. So looking for ways to shrink that down and be a little bit more parsimonious, communicate more effectively. So everyone's gonna have their 10 cents about how to communicate. And I thought to myself, like if I was making recommendations about how to communicate, what would I recommend? So I made a list. Use diagnosis words, not radiology words. Don't bury the lead. Don't put the worsening metastatic disease in impression point five or the incident of lung cancer in impression point six. Explain why you think something, don't assume that they're gonna know. Be parsimonious, less is generally always more. And when you're translating, because that's one of your core functions, do so for your recipient because jargon's gonna kill that information transfer. And I like to describe this empathic reporting. Imagine you're them. If you read this and you were them, what would you wanna know next? And then tell them. Ask clarifying questions when you're on the phone so you can help improve the quality of your report and over-communicate, especially by voice. So fundamentally, we're information brokers and our core functions are translation and communication. We convert image data into words that influence behavior. So if we're putting all of our attention on the accuracy piece, we're really missing a lot of the influence piece and the behavior change piece and the value outcome. The customer need we serve is not report creation. It's reducing clinical uncertainties. The next time you think twice about calling, remind yourself what your core function is. And to generate maximum value in this context, we wanna maximize fidelity of information transfer at every step, not just in report creation. Create trust and be open to new modes of information extraction delivery because our business is not report creation. Thanks. Thank you very much for a great presentation. Our last speaker is Dr. Alex Tobin. Dr. Tobin is professor of radiology, chair of radiology informatics and associate chief of radiology and clinical operations and radiology informatics at Cincinnati Children's Hospital. Thank you. Thanks. I'm happy to be here. I'm gonna be talking about that cutting edge that Matt just said was not that valuable. He didn't really say that. I didn't really hear his message though. He didn't communicate effectively. We're gonna be talking about AI and informatics tools to improve communication. At the start of this lecture, I'll give my objectives. I'd like to describe six ways that communication can be improved within the imaging value chain and identify six artificial intelligence or informatics tools that can be used to improve communication. At the outset, I need to say that this is a bit of a pie in the sky type discussion. Many of the things I'm gonna talk about are things that are not yet reality, but many of them could and could be there pretty soon. I like to think of imaging as a value chain. And while it's often described as a linear pathway, I think of it more circular. Many times one thing leads to the next, there's feedback. And so that communication affects every different part of the imaging value chain. I will talk about it in a linear pathway though, so you won't get the weave. We start with an order, move from the order to some sort of protocoling step. Imaging happens, then there's some sort of post-processing. I know post-processing doesn't typically happen today, but I'm talking about AI, so there has to be something with post-processing. Then a radiologist makes an interpretation, and finally it's communicated downstream somewhere else. So let's talk about the order. Clinical decision support has been thought of as the panacea for improving communication and guiding communication at the appropriateness of an order. Unfortunately, it has been a massive failure within our specialty, and that's because of all of these annoying pop-ups. In an ideal state, a physician would be writing his or her note within the electronic medical record. You can tell it's a physician note because you can't read it. And a large language model would be trolling all of the information within the electronic health record, recommending, identifying, and then recommending information within the study, or within the patient chart, and using that information in total to recommend an imaging study. Similarly, with protocoling, a radiologist working at her work list can see the studies coming in, and a large language model, or a small language model could say, here's all the information in the note. Here's the type of order that's performed. We're bringing the clinical history in. Let's perform an MRCP with intravenous contrast and secretin, that protocoling step happening automatically. Now, I know many different organizations do different things, and there can be some automation with that protocoling step. It does not need to be through large language models. It can be a simple relational database based on an order, or guided by specific pieces of history. Those large language models, or even a small language model may not be needed to do this protocoling step. However, something like this could happen more efficiently. Okay, talk about imaging. This is where we'll move away from large language models. Here, I think one of the keys is in that bracelet that patients wear. If those bracelets were embedded with a radiofrequency identifier, or RFID, all of the information related to the patient and the imaging study can be embedded within that piece of technology. So, as the patient walks into the room, they're welcomed into the room, the protocol is automatically applied to the scanner, and all their demographic information happens automatically. So, the technologist does not need to enter information. The technologist is doing less, and yet we're communicating more and more efficiently. Moving into the post-processing step, we'll talk a little bit about traditional artificial intelligence. Most artificial intelligence does one of three tasks. It detects disease, it makes a diagnosis of some disease, or it identifies that disease and takes it from a work list full of many different studies, and triages it. While each of those steps are cool, these are the types of things that you'll see on the floor today. So, many of these things are possible and can help us with communication, but it's only the first step of communication. The next step moves into the radiologist as they're reporting. This is something where we're talking about now the cutting edge of radiology reporting with common data elements, or CDEs. Hopefully, you'll see common data elements at RSNA this year. If you go to the Radiology Reimagined demonstration, you'll see how common data elements can drive radiologist workflow. So, let's explore them, but first, let's define them. A common data element can be thought of as a question, series of answers to that question, and instructions for how to answer the question more specifically. The RSNA and American College of Radiology have worked together to start publishing common data elements. This example is for pulmonary embolism with two common data elements in a set. The first element on the screen says if a pulmonary embolism is present or absent, and then if it's present, the second data element says where is it located. So, let's bring that with artificial intelligence. The artificial intelligence could identify the pulmonary embolism within the left main pulmonary artery, but instead of outputting that diagnosis, it could output a coded data element, or the coded common data element. That gets bundled up into an interface that allows for communication between systems. HL7 FHIR is the way that that happens, and that FHIR interface transmits the information directly into our reports, and then translates it into the English language, or a language that we're expecting to see directly in our reports. But in the background, that coding remains, so that same HL7 FHIR interface can take that information, help us communicate to downstream systems, such as things like the American College of Radiology Grid Registry, or it could go to other places, and trigger things like clinical decision support hooks, or CDS hooks. This is where clinical decision support can get cool. Clinical decision support hooks, or CDS hooks, do three things. They can present information to an end user. They can present an order to that user. So, in that first step, where we were talking about presenting radiology orders to the user based on notes, a CDS hook could do this, or it could launch a different application. So, taking that pulmonary embolism example, it could present information, such as this patient may be an ideal patient suited for a thrombectomy. Or it could say, this patient should have anticoagulation. Would you like to place the order? Here's the information. Or it could launch a pulmonary embolism burden calculator, helping you to identify the risk for that patient. So, at the outset, I said I would describe six ways that communication can be improved within the imaging value chain, and then identify six AI or informatics tools that can be used to improve that communication. So, within the imaging value chain, we start with the order, the protocol, imaging, post-processing, reporting, and then downstream communication. And the tools that can be used include large-language models or small-language models, RFIDs, directing information through identifying protocols, as well as DICOM modality worklists, common data elements, as well as artificial intelligence, making diagnoses, and then CDS hooks to drive workflows at the end. All of that with a touch of fire. Thank you.

RSNA 2023

radiology communication

patient care

communication errors

structured reports

e-consults

artificial intelligence

large language models

diagnostic outcomes

informatics