Today, we're fortunate to be joined by three radiologists who are really leading the way and bringing the charge to bring forth diversity, equity, and inclusion into their practice. I'm so pleased to introduce today's speakers. We are joined by Dr. Yoshimi Anzai, Dr. Ryan Lee, and Dr. Janelle Scott. Now, first, we will hear from Dr. Janelle Scott. Dr. Scott will talk about systemic factors that perpetuate healthcare disparities and real practical solutions to bridge those gaps in radiology. Dr. Scott serves as the board liaison for RSNA's Health Equity Committee. She is Associate Director of Clinical Radiology at State University of New York Downstate Health Sciences University. She serves as Chief Quality Officer for New York Health and Hospitals, King County. Thank you, Dr. Scott. Thank you so much. And thank you for everyone who's participating in today's webinar. So today we're going to talk about closing the disparities gap and improving outcomes through radiology. We do have a part to play. So I have no financial disclosures and the two main objectives for today are to describe the patient-provider systemic factors that perpetuate disparities, and then to describe some interventions that we can do as radiologists that could potentially decrease disparities in outcomes. So before we get talking about interventions, let's talk about some of the etiologies that lead to poor outcomes in some of our minoritized groups. So when we're looking at poor outcomes, you can really separate the etiology into three factors or three levels, the patient level, the provider level, and the systems level. So system level would be, you know, entities such as resources, diagnostic algorithms, technology diffusion is a huge one. On the patient side, we can talk about language barriers, health illiteracy, medical mistrust, and of course, differences in quality of insurance or no insurance at all. For providers, implicit bias does play a large role at this level. We can also talk about ordering practices and also the availability of expertise and knowledge of advanced imaging. But it's very important for us to remember that this is all occurring in the milieu of racism that we experience in this nation. So Dr. Willard Edwards from the AMA was quoted as saying about two years ago that the AMA recognizes that racism negatively impacts and exacerbates health inequities among historically marginalized communities. He goes on to say, or she goes on to say, without systemic and structural level change, health inequities will continue to exist and the health of the nation will suffer. I like this quote from Dr. David Williams, who is a renowned author and scientific researcher, who says, specifically for health disparities, who says that racism is insidious and its structure and ideology can persist in governmental and institutional policies in the absence of individual actors who are racially prejudiced. And this is very important to keep in mind because a lot of people sort of internalize and get very defensive when we talk about some of these things, but the point is that it persists and it exists, even though you, yourself, or the individual, his or herself may not be racially prejudiced. So perhaps in no other area in radiology has been studies as much as breast cancer and breast imaging when it comes to disparities. So we have a graph here that's looking at deaths from breast cancer in women, for about over the period of 40 years, from 1970 to about 2020. And we can see early in when they started tracking that the mortality rates between black women in red and white women in blue were about equal. And then starting about in the 80s, the mid 80s or so, we see a gap starting to develop and it just gets wider through the decades to the point where now black women are 40% more likely to die from breast cancer than white women. So why is that? What happened? What's driving this disparity that we're seeing here? And when you look at it and when you kind of examine what was going on in the 1980s, that's when use of sputum mammography became widespread and especially as well, adjuvant systemic chemotherapy. That started in the 1980s and sort of revolutionized how breast cancer was treated and diagnosed. And overall, the mortality rate decreased, but you can see here that black women did not necessarily realize the full benefit of these advances in technology and advances in therapy as the white women did. And so now we have this large disparity that has been studied a lot in the literature. So I like to look, you know, I'm a quality person. I do quality and patient safety. So I like to think about things and I love this model. It's called the Reason Switch Cheese Model. And in healthcare, we use this model to explain the occurrences of systemic failures that reach patients that cause harm. So according to this model, in a complex system such as healthcare, patients are prevented from causing human losses by a series of barriers. However, each of these barriers can have unintended weaknesses or faults. And when these faults align, harm could actually get to the patient. So the question becomes, how are we as providers and as radiologists, what are we doing to close some of these holes up so that the harm do not get to all patients or patients? So let's look at the patient level. Medical literacy and screening awareness can play a huge part in whether or not patients actually have or experience poor outcomes. So when we look specifically at breast cancer, and I would ask the question to everyone on the line today, is your department involved with community outreach? And not just for breast cancer, but lung cancer screening, poloretinal cancer. Is your department involved with community outreach and education? And if not, who do you think is responsible for doing this? Because we play a role in this and we can definitely play a role in this. I like to think about my friend, the neurologist, and she's the one that really opened my eyes to this. She started a stroke program here at Kings County. And she noticed that several patients were refusing alteplase, the thrombolytics, when they came in with symptoms of an acute stroke. And they were refusing it because they weren't aware, they were kind of scared about the fact that they can actually be in their brain. So a lot of her denial rate or patient refusal rate was quite high. So what she did was started to work in the community. She went to churches. She went wherever they would listen and she started talking and talking about the benefits of the thrombolytics and how essential it is to get to the hospital on time. And that metric has completely turned around for where most patients now, almost all patients, 100% of patients are getting the alteplase and some are even coming in asking for it. And that's the power of community outreach, community engagement, and patient education. We take it for granted sometimes because we're physicians, we kind of like, things just seem very self-explanatory to us, but our patients don't really know and we have to develop those relationships, those trusting relationships that will allow us to tell them about new technologies, new developments, new medications that they should be trying. I was reading the ACR Bulletin last week and I thought there was a great example there from Vanderbilt University Medical Center where I'm gonna just quote it, where it says, helping communities at large requires the work of many hands. And to that end, one of the strategies, the main strategies at Vanderbilt is to empower individuals throughout the geographic area to develop partnerships and collaborate with others working to improve healthcare. And I will quote our very own Dr. Lucy Spoluto, who is a co-chair of the Health Equity Committee, and when she says, to drive change, we have to get out into our communities and understand their needs. We can't stay in the bubble of academic medicine. We got into this field, most of us anyway, to help people. We have to meet people where they are. So what are some other barriers or some other holes in our defense against patients receiving experiencing poor outcomes? When we look at the systems level, long waits and missed appointments, what can we do to help patients not experience this barrier? And I would argue, and I would recommend, again, from my quality perspective, that we leverage performance improvement in this space. So we can actually decrease the risk of missing appointments by utilizing a more patient-centered approach. Ask yourself, look at your department and look at what you provide. Are there enough appointment slots during the off hours and weekends? Are you using appointment reminders? Can you bundle appointments? And this becomes really important, particularly in the rural populations where people travel great distances to receive care. To have to just come back to do a mammo today, lung cancer screening two weeks from now, colorectal cancer screening three, it's very inconvenient and patients may make some choices that ultimately will be detrimental to their health. So what can we do from a patient-centered approach, from a human-centered approach, to help patients not have to make these kinds of trade-offs and these decisions? And then another question is, what are your wait times like? So the patient actually is now educated because you've done all this education, they're coming in and they know they're there. And how long are your patients having to wait to get seen? And a study done by Ray and colleagues show that clinic times are significantly longer for racial and ethnic minorities. And why is there this difference? A lot of it is due to waiting. I like to call it the weight of waiting. They're there waiting and they're waiting and they have to pick up their children. They're there waiting and they have to get to their jobs. What are we doing to really look at this? And we need to be cognizant of this. We don't want patients who maybe cannot afford to miss a day of work because they're wage earners. And patients may say, I feel fine right now, there's nothing wrong with me. And if they have to choose between a health maintenance visit or health screening, which is going to work, they may more than often choose to go to work. And after doing that for several years, that may become a problem. So we have to help our patients to make the right decision. And I will say another issue, another hole, another area of weakness at the system level is poor quality measures. And again, racial and ethnic minorities tend to work in hospitals that unfortunately have poor quality measures for many different reasons. So having a strong quality assurance program is important. And I will say this, that this is not something that's going away. Health disparities is not going away. And recently the joint commission has sort of enshrined it through the CMS as they have enshrined it as one of their requirements now. Reducing healthcare disparities is here to stay as a joint commission requirement. And this is very important. Why is this important? Because initially when it was announced, it was going to be in what we call the leadership chapter of the book, of the manual, of the set of requirements. And they announced a month ago that they're actually taking it out of the leadership chapter and making it a national patient safety goal effective July 1st, 2023. That is going to be a national patient safety goal. And this is a huge, huge, huge deal. So what does that mean? That means if when they come to your hospital and they review your hospital, if they see anything that does not promote equitable healthcare, they can actually cite it as a high level finding in the red zone. And that's obviously not a good thing. So these are the list of EPs or elements of performance that they will be looking for in your hospital. And I just highlighted something that I think radiology can really play a role in. So for the first one, they want you to designate a leader. And I would like to also, again, reference Vanderbilt Center Medical, Vanderbilt, because they have a vice chair of health equity who's our own, Dr. Lucy Spoluto. So they want, so just the thing we have, the joint commission requires that you have a leader of the institution who's sort of like spearheading these efforts. You can have someone in your department look specifically at health equity. Identify healthcare disparities, element of performance number three. Now, the only way you're going to do this is if you're looking at your data and you have to disaggregate your data and stratify it by socio-demographic characteristics. And I'll give you a practical example of this. I was looking at my institutions for the past three years or four years. I looked at pre-COVID and post-COVID. Who was getting diagnosed with breast cancer? What's our incidence of breast cancer? And what are the characteristics of patients? And I saw that 30% of our patients were being diagnosed with breast cancer below the age of 50. Now, why is that important? That's important because our ambulatory colleagues, they follow the United States Preventative Services Task Force that sort of requires or recommends that patients get screened at 50 and every other year. So I was able to show them like, listen, 30% of our patients, 30% of our patients who are diagnosed with breast cancer for the past four years have been under the age of 40. So if we're starting screening at 50 and every other year, we're potentially missing a large cohort of patients. And so that is problematic. So I was able to bring that to their attention. And of course, there's a lot of work there still to be done. So the next thing that we probably would have to do is to develop an action plan. This is something, again, once you realize that you have an issue, by looking at your outcomes measures, then you can develop an action plan and of course make improvements. And that's an iterative process, obviously, that can take a long time. There's no magic bullet. It's going to take a long time, but there is definitely room for us in this space to leverage performance improvement and quality assurance. So another, which I alluded to before, on the provider level, unclear screening recommendations. American College of Radiology, most radiology societies, most cancer societies recommend that women start screening at the age of 40 at every year. And if you have a post-degree relative with breast cancer, even sooner. Of course, the United States Preventative Services Task Force does not recommend that. And that's a problem. So providers are kind of caught in the middle, depending on who they listen to. So there are some studies that are showing that, or are suggesting that underuse of screening mammography among black women contributes to racial disparities in outcomes and mortality. Ahmed and colleagues demonstrated that black women utilize screening mammography at lower rates. And Omali and colleagues demonstrated that that was partly due to lack of physician recommendations. But clearly, if there's no agreement on current recommendations, that can create some ambivalence and that can create some confusion. And unfortunately, patients can suffer because of that. Another provider level issue that we should look at is generalists and a lack of specialty, subspecialty trained imagers. So unfortunately, again, low resource institutions that minorities visit, they tend to rely more on generalists who read screening, sorry, mammograms, and their readings are less sensitive. And so it contributes to a higher false negative rate as compared to some higher resource institutions. So Elmo and colleagues found that fellowship training is the only variable, the only variable associated with improved accuracy in mammographic interpretation. So therefore, as radiology departments, we should really prioritize hiring fellowship trained radiologists. And I would also include highly trained and skilled technologists. So another issue that patients may face is no direct communication or inadequate follow-up once a finding has been made. Some authors actually suggest that mammography and best imaging is not the rates are comparable between both groups of women, but lack of adequate follow-up after abnormal findings was sort of driving the outcomes that we're seeing. So there is research and colleagues found that for minority patients and for lower resource institutions where minority patients tend to frequent, there is a longer time interval between a suspicious finding and a biopsy compared to other well-resourced institutions that more white patients tend to frequent. And unfortunately, this longer follow-up time is associated with a higher likelihood, as you can imagine, because time is going, of diagnosis of advanced breast cancer. So what can we do to counteract this? Well, one thing I would say is to have a communication strategy. Nguyen and colleagues found that direct telephone communication with patients who had abnormal mammographic findings increased the percentage of patients presenting for diagnostic follow-up within 60 days. The study that they looked at, all racial groups demonstrated increased follow-up, and particularly Black women, their follow-up increased from 69% before the intervention to 85.9% after. So what is your communication strategy like? Do you have one-to-one, not one-to-one, sorry, do you have direct phone calls to patients to remind them of their appointments to follow-up? Let us, we have to be a little bit more hands-on to make sure that patients are following up with their abnormal findings. So hopefully you will have some ideas to take back to your colleagues to discuss how we can leverage our space in radiology and the imaging cycle to decrease disparities. There's several opportunities at every level in the imaging cycle, definitely in the pre-procedure, where we can talk about patient education, community engagement, even working with our physicians to educate them on best practices. During the procedure, what is our wait time like? What is the burden on our patients with paperwork and all these non-value-added steps? Can we bundle appointments? Can we offer more weekend hours, more after-work hours? Post-procedure, how are we communicating abnormal findings and diagnosis? Are we prioritizing hiring fellowship-trained images? And of course, clinical action, again, making sure abnormal results are communicated to the patient and providers in a timely fashion. So I hope I was able to provide you guys just a taste of some opportunities that we can have, and that there is room for everyone in radiology to take part in our efforts to decrease disparities in our communities. Thank you so much for your attention. Thank you so much, Dr. Scott, for that enlightening talk. I want to remind our audience that you may submit questions for our speakers in the chat box, and we will do our best to direct those to the appropriate person to answer. Next, I have the pleasure of introducing our next speaker, Dr. Yoshimi Anzai, who will discuss how radiologists can work to transform value-based health care by implementing diversity, equity, and inclusion in their caseload. Dr. Anzai is the Professor of Radiology and Director of Quality and Safety at the University of Utah. Thank you so much, Dr. Anzai. Great. Thank you so much for the invitation to speak. This is a fantastic webinar. I had a little challenging topics, but we can go through this. So nothing to disclose. The learning objectives are to understand why value-based health care transformation is necessary, and to understand the barrier to health care through EDI lenses, and also learn how radiologists can contribute to this system. Now, we all know that US health care spending reached to $4 trillion in 2020, thanks to COVID pandemics. When you're looking at what drive the health care cost in this nation, the top three items is advanced technology, such as robotics or imaging, administrative costs, and also drug pricing. Those three are the top major spending items. When you're looking at the imaging volume in radiology, this continues to grow, as we all feel every day in a clinical practice. This is the prediction in 2020, how much imaging volume will grow. And they expected maybe 2.5% to 3% of a growth of imaging volume by 2023. However, the recent study shows diagnostic imaging service grows at the CAGR, which is a compounding annual growth rate, 5.7% from 2021 to 2022. So double of what the prediction said. So we always feel the imaging volume continue to grow. So why the demand for the imaging continue to increase? Well, first of all, one, the medical imaging is accurate and efficient. Radiology has a tremendously efficient system to provide accurate diagnosis in a very efficient manner. Number two, we have an aging population in our country. When you're looking at the 10 top causes of death in the United States in 2020, heart disease, cancer, post-COVID-19 is number three now. Trauma, stroke, Alzheimer's disease, all of that, imaging plays a critical role to make a diagnosis for those conditions. So the demand is continue to grow. And number three, imaging is somewhat used as a triage tool, particularly for emergency room. And that's something we feel every day is practice. So the demand will never, ever going down. This is an interesting data by Kaiser Family Foundation. The level of complexity in the emergency room. When you're looking at the level one through five, five being most complex, the high level of complexity continue to increase. Now, half of the patient emergency room has a level four and five code. What does that mean? So let me take the example of a headache patient. So this is a 71% of the headache patient came through emergency rooms acclaimed at the level four and five. Now, one of the requirement for level four complexity is patient's necessity to get an advanced imaging, such as CT and MRI. So those neuroradiologists folks in the country that why we have so many CT ahead of neck for a headache patient, you can imagine when patient get CT or MRI, that qualified level four complexity, meaning higher the level of complexity, higher the payment, and higher the charges. So that may be one of the reason that we're getting too many CT ahead of neck for headache patient. So the challenges that we radiologists face is the demand for imaging continue to grow. And there's a pressure, increasing pressure to be more efficient, more productive, more effective. On the top of it, we have our workforce shortages, not only radiologists, but technologists and nurses. So all of that led to increasing burnout. The study showed that 54%, more than half of the radiologists are burnout in 2020. And that number increased 5% from 2021. So that is why we need AI machine learning to help us, our job, to make it a little more easier. But let's just put the whole perspective from patient side. Because of the health care is so expensive, employers is shifting the cost of health care to patient or employees through high deductible health plan. So this is out-of-pocket expenses over the years. Now we have an extremely high contribution of out-of-pocket expenses that reach to $433 billion. And that's 10% increase from 2020 to 2021. So it means that patient have to pay $2,000, $3,000, $6,000 out-of-pocket expenses in order to, before kicking up the health care benefit. So what does that mean? So the shocking data, again, from Kaiser Family Foundation, is 43% of patient either avoid or delayed necessary care due to cost concerns. And this is more common in women, younger patients, and low-income families. When you're looking at all this racial description, it actually affects all races, Black, Hispanic, white. But it's most affected in the Hispanic population. Now, 23 Americans have a medical debt in total, $195 billion. So that's not good things, because the patient cannot afford health care. So we need to shift focus or to change the dialogue from health care spending control to we have to increase affordability, increase access. And also, we have to focus our attention to social determinant of health. Now, what account for health in population? What makes people healthy and well? Well, it turned out that medical care only contributes 20% of modifiable contribution to population health. The rest of the 80% come from something called social determinant of health, which include educations or economic stability and neighborhood safety or social community context, which we don't really spend any effort or increasing resources to improve social determinant of health. But this is a big topic, so I'm not going to go into too much detail. So really, the big question is, are we going to continue to provide the latest and greatest advanced care only to those patients who can afford and ignore the rest of people who need access but can't afford the health care? But those patients, remember, will come back to emergency when things are getting out of control. You can provide an inexpensive pill for hypertension, but instead, those patients come back with ruptured abdominal aneurysm, need a surgery. Or we can provide a diabetic care, but instead, a patient became a chronic renal failure and need an unscheduled dialysis or macular degeneration and became blind or below the amputations. Or the patient can get a screening breast cancer, but instead, a cancer grow out of the skin at a distant metastasis that required not only surgery but chemotherapy and radiation therapy. So really, the big question is, are we going to keep the status quo, or are we going to try to innovate a path to provide better care for broader population of patients? And if so, what radiologists as we can do? So when you're looking at the oldest health care barrier through the DEI lenses, there's a four pillar that I would think, and some of them are kind of overlap, as Dr. Scott talked about. Access is the big deal. Not only access, meaning insurance status, but transportation. The two language barrier is huge. Health literacy is very limited. Those people speak English, but on the top of it, the situation is much worse for people who had a limited English proficiency. Cost affordability is a big deal, but most of all, the trust or lack thereof, trust of a health system, this is still one of the biggest barrier to get a health care through the DEI lenses. So we're going to show some of the example of some of the effort through the radiology community. Now, I found that state of Utah, the percent of eligible woman receiving breast cancer screening is below national average of 63%, as opposed to the national average of 71%. So this is the University of Utah Health in conjunction with the Cancer Huntsman Institute created a cancer screening and educational program. So bringing a breast cancer screening to vulnerable population through MAML, I think I would say MAML bus instead of MAML van, because it's so huge, inside is gorgeous, beautiful, top of the notch technology. And we drive 200 miles across the state of Utah, not just a couple of miles, 20, 30 miles. And one of the story that we received from those breast cancer team is that it's not just that you drive to some places and hey, people coming over for screening mammogram, but rather you really need a lot of prepping. And then what it require is community partnership. Dr. Skoll emphasized that as well. To facilitate the cancer screening, we need to have a community partnership and build a trust and a collaboration. We need to build a long lasting relationship with a community team such as primary care, family medicine, social workers, and all of the people so that we can work together rather than I coming over, you correct a patient, we're gonna screen breast cancer. And we also have to be very sensitive about cultural nuance and the language barrier as well. So it takes a village to get community outreach program, but it's much, much worse end of it. It's an interesting things of COVID pandemics broad is that telemedicine. Now we know how to use technology for medicine. Telemedicine is here to stay, or I would say even accelerated further. And this is a great pilot project by Canadian team. So the instructor show how to use an autism probe using this kind of video. And then that image is projected through augmented reality to the students or even MA or somebody who don't have a skillset, but being able to scan a patient. And this could be a huge impact for the rural health or global health where the health professional are scarce resources. We can't have enough radiologists or technologists to cover the entire population. But when you use technology, we might be able to get certain level of care for those people who don't have right now. Now we'll talk about transportation. This is a big barrier that we live in a city don't understand. 3.6 million American don't obtain the Medicare kit because of the lack of transportation. Those people may have insurance, they may have money, but they don't get that care because of lack of transportations. And those are common in a woman and those people who are poor, elderly, and less educated and a minority patient. It's an interesting project in a H-Health in Minnesota. They use software, the link with the EMR to ride share lift. So when patients schedule for colonoscopy, breast cancer, blood test, the EKG, I think Dr. Scott talk about bundling. So the bundling of the appointment, the lift will pick you up from the home, bring it to the hospital. And then after all the exam is done, bring it back home. And this is a really great way for people who live in a far away to get access to healthcare. Now, patient-provider racial language concordance is a very interesting topics. When you're looking at the Kaiser Permanente data for 131,000 diabetic patients, adherence to cardiovascular medication improve when provider-patient racial concordance is present for African-American patient and also Spanish speaking patient. So that is also the emphasis on diversifying a healthcare professional. Another patient, much more smaller sample, 723 hypertensive patient was African-American or white patient in primary care. They didn't find that adherence is different depending on the racial concordance or lack of concordance. However, the trust to the provider measured by the survey of the patient was associated with better blood pressure control. So again, the trust is very important part of the health outcome. If the patient don't trust the provider, they don't obey or comply with the medication. Now, healthcare finance is the hardest part. And I'm sure that many of the institution like University of the Health provide a various financial assistance program. We also provide a huge discount for the self-pay and ensure the patient who need access to the healthcare. It's important to provide some educational material about balancing billing or surprise bill protection assistance so that patient don't become a victim of surprise bill. I'm gonna touch a little bit about 340B drug pricing because this is so important for safety net hospital. The hospital or clinic that treat a low income or uninsured patient, the certain level, I can't remember, I think UTI is like 11.5% or whatever, are qualified to get a discount outpatient drug pricing by 25 to 50%. This is a pretty huge benefit. So we wanted to keep the 340B drug pricing so that we can continue to provide a charity care for those patients who can't get access to healthcare. But this is a debated and a pharmaceutical company wanted to reduce the benefit or completely kill the 340B. So I cannot emphasize enough that a 340B drug pricing is so critical to provide health equity in a system. So in summary, we'll talk about access, transportation, language, cost and trust, a four important key element to improve the value-based healthcare through the DEI lenses. And I think we all be part of this whole journey so we can work together with through RSNA or other organizations. Take one point, healthcare is too expensive to afford for many patients resulting in a care avoidance and delay. So we have to change the dialogue from healthcare expenses to affordability and increasing access. And barrier to healthcare, including a transportation, languages, cost, but most importantly, the trust of the health system. We have to work well together to gain that trust. And again, I cannot emphasize enough to how important it is to have a community partnership. With that, I'd like to thank you for your attention. Thank you so much, Dr. Anzai. That gave us so much to think about as we're working towards health equity. I'd like to now introduce our next speaker. We will hear from Dr. Ryan Lee. Dr. Lee is Chair of the Department of Radiology at Albert Einstein Medical Center. He has recently been promoted to Professor of Radiology in the Sydney Kimmel College at Thomas Jefferson University. Congratulations, Dr. Lee. Today, Dr. Lee will discuss the role of digital tools and solutions to increase healthcare access, address unmet needs, and personalize care for patients in underserved communities. Thank you, Dr. Lee. Thank you for that introduction. And thank you for RSNA for the invitation today to speak about radiology's role beyond the reading room, advancing patient-centered tech equity. The learning objectives for my session is one, to describe factors that can impact deployment of AI equitably, to discuss examples of algorithms that can address disparities, and also identify some questions you may want to ask when you consider AI algorithms for your practice, especially from an equity perspective. So I have no disclosures for this talk. I want to just start out, and many of you already know so many of these numbers, but just as a recent case for COVID, 30% of COVID-19 cases in 2020 actually were Black Americans, yet Black Americans only represent 13% of the US population. That two times number represents the death rate from COVID-19 in the Black population compared to the White population. And I'll give one more here. The prevalence of diabetes for those on Medicare is 9%, yet the prevalence of diabetes for those on Medicaid, which may be a proxy for those of a lower socioeconomic class, is a staggering 76%. So I'll only give you just a few of these examples just to show you the, and I don't think this is news to anybody, of inequities in our healthcare and the way it's deployed, and also even the way the morbidity or mortality is spread across our population. So I'd like to start off with this, and this is a little cartoon I developed, which really talks about the life cycle of the patient when they touch radiology, all the way from ordering, scheduling, protocoling, acquisition, interpretation, reporting, communication, and peer review. And in fact, believe it or not, AI can touch each and every one of these aspects of the patient radiology life cycle. And I can give examples for all these, but for the purpose of today's discussion, we're going to talk about a few of these, particularly as they equate to the equity issue. So let's start off from an interpretation standpoint. And so I would say the largest amount of AI that we hear about most predominantly is about the pixel-based AI, those AI that make diagnoses. So it may be intracranial hemorrhage, it may be PE, maybe even be chest x-rays. This is an interesting paper that came out in 2021 by Said et al. And what they did was they actually looked at how a specific algorithm applied to chest radiographs could have different ramifications depending on the group that we're talking about. And here is just a graph from that paper. And what this is, is they defined a false positive rate as that in which a study was called falsely negative. So in other words, a study was actually positive, but was graded by the algorithm as negative. In other words, we were potentially denying the patient proper care because the algorithm called it incorrectly. And so what you're seeing here is the likelihood or the rate at which these incorrect decisions by an algorithm was made. And the interesting thing about this, and these are all statistically significant, and you can see starting at the left-hand corner, the difference between male and female, there was a slightly higher rate of having the algorithm miscall and call something negative when it in fact had a finding on there. And this is, you can see the difference between female and male. When you go over to that green graph, you can see the dramatic increase when you talk about a false positive rate, or in other words, the algorithm falsely calling something negative when it had a finding for the very young patients, so the zero to 20 compared to the population that was over 80. And as you move over to the red graphs, and you can see the differences between Black, Hispanic, Asian, Native Americans, and white, that there is statistically differences between Black and Hispanic patients compared to white or Native American patients. And you can see from, as a proxy of socioeconomic class, the much higher increase in falsely calling a study negative by the AI algorithm when there are Medicaid patients as opposed to other, and other represents private insurance, Medicare, apparently the data was not usable. So you can see that even with looking at these groups here, there can be a difference in how an algorithm performs. And they went further, they actually looked at the intersection of various different groups. So for example, on the left-hand side there, the female cross-reference between age, ethnicity, and also Medicare versus Medicaid insurance. And you can see that there is a difference even within that subgroup. And same thing you can see from the young patients, the differences between a Medicaid, which is a very high false negative rate or false positive rate in which it was falsely called negative, and also in the Black versus white population. And again, I'm not going to go through all of these, but you can see that the same algorithm on different subgroups and then on the intersection of subgroups can have very different results. And this is something to be cognizant of when evaluating these algorithms. Now let's talk about how these things could be mitigated. For example, let's look at from a protocol and acquisition perspective. Now this paper, actually with the good Dr. Flores as part of this paper, looked at possibly using a automated translation, web-based translation for those patients that could not speak English. And what they did was they provided the technologist with an iPad and basically for various common phrases that a technologist would have to explain to a patient, have ready-to-go translations that could be used to then communicate to the patient. And what you see there is one of the big outcomes of this paper was that although the times the patient spent in the department for these chest x-rays was about the same, whether you use this RAD Translate program or you didn't, the variability and how long a patient was staying from patient to patient was much more consistent. So you could see where this is going. This could lead to better scheduling. And ultimately, even though it's hard to tally, you may get better studies. And you can see how if this was applied to other more complex studies, perhaps the study itself is better. You have less motion artifact if we're able to communicate these things better. And these are things that are going to disproportionately affect the underserved population. I'd like to talk about something else in terms of communication and not in the traditional way we think about communication. So the traditional way we think about communication is we have a study that the radios reports, and then we have to communicate that result to the patient and to the referring physician. That's the traditional way we think of communication. But another way to think about communication is being proactive. When we communicate a unexpected finding for a study that's performed for another reason, the so-called concept of opportunistic imaging. So taking results from studied for one purpose, but deriving additional information and communicating that back to the patient. And one of the things that we're trying to do at my institution is to improve the follow-up rate of studies that were recommended by the radiologist. And so we did a pilot study with using natural language processing for impressions and figuring out based on the impressions, if a study was recommended by a radiologist to then send a notification to the patient. And that notification was done in our instance via text. And so what we did was we actually used the ACR recommended compliance range. This is still all in development, but we used a recommendation due date as the date that it would be due. And up to three weeks before, we would send up to two messages to these patients. And we would send up to two, so if it became scheduled or completed, then we would stop the messages. We would send a certain number of reminders up to three after the due date, based on the same concept. And we wanted to see if that would actually change the habits of follow-up and the rates of follow-up. And this has a profound effect, I think, potentially on the underserved population. So what we found, and we actually presented this at RS&A this past November, was that there was a significant increase. So previously we had a compliance rate of about 54%, I think relatively low. Our main population is an underserved population, but we did get that up to 70% with our intervention. Let's move on to communication. And communication in this sense is going to be also a little different because we're talking about communication from a population health setting. So I'll briefly describe our project. And for both of these things, for that previous project I talked about as well as this project, we'll see, I'll talk about how this actually can have bigger effects on an underserved population. So the gist of this project is you start off with a non-contrast chest CT, and we have an AI algorithm that can detect coronary artery calcifications. And normally on a regular chest CT, you would not typically be looking at that. And more and more, we're now seeing vendors that can have algorithms that can quantify coronary artery calcifications that are separate from the CT coronary artery calcification studies that you can obtain. Once we get that result, it is then looked at by the radiologist because the radiologist is already reading the chest CT. And once that has been vetted, we're then able to transmit that information to the nurse navigator. Right now it's manual, simply through a spreadsheet. And the nurse navigator will then go through the patient's history and other things within the EMR and determine, hey, is this something that we may want a cardiologist to consult on? And if the patient doesn't have a cardiologist, then she might recommend that. And what happens is she'll potentially reach out to the patient directly, and that's where it's going down to the bottom, or talk to the primary care physician. But ultimately, the goal is to get a consult with the cardiologist and then see what sort of changes might occur. And just to show you what it looks like in our system, it basically is automated in terms of it gives us a quantification, and you can see the quantification in the top right here, and make an assessment of whether something is low, whether, for example, it could not be scored. Is it, for example, in this case, this patient has atrial appendage clip, or has medium score, it's a little higher, or finally, a high score, which we would recommend that perhaps a patient, a possible analysis for whether a consultation with a cardiologist is warranted. In phase one, we did this this past year, just to see if it could even work. How many patients could we get to go see the cardiologist? In that short summary between December and March, we had 10 finally arrive. We've subsequently had a lot more data, and that's going to be for phase two. And the gist of this is to see how many patients we can identify, incidentally, that have moderate high coronary artery calcifications, and of those, how many might be eligible for consultation with cardiologists, and ultimately, how many had management changes. And we can break this down into various different management changes. These are just some examples. But ultimately, and our preliminary data shows that we're changing management in about 20% of these patients. Now remember, these are patients that would never ordinarily have had any management change, because that scan was done for something else. So even a 20% change could be quite significant. So the question is, for both of those projects that I just described, can this help an underserved population? Because clearly, this could, if it works out, could help all patients. And the answer is, I think yes. Because when you think about the underserved population, they're less likely to schedule routine visits, they're less likely to schedule problem visits, and they're also less likely to follow up. Whether it's access, whether it's time, because they're working, all those various factors for those social determinants of health that were discussed earlier. And potentially, using a combination of the algorithm, with radios vetting, a nurse navigator to go through the results to see who's eligible, and the cardiologist to manage, this leads to proactive healthcare, and we could potentially prevent more of these patients from having trips to the ED, inpatient visits, all those things. So yes, I think that algorithms applied in a population that has less access, that's less likely to have routine visits, could have an even disproportionate effect to help the underserved population. And you can see where this is going. There are other projects that we're exploring that have a very similar outlook to what I just described for coronary artery calcifications. Bone demineralization with endocrinology, can we prevent more fractures that could have been managed by the endocrinologist? Abdominal aortic aneurysms, can we get those patients the management they need from the primary care specialists? So you can see there's a lot of upside for similar appearing projects. And I want to leave you with just a few questions. When you're looking at algorithms such as those, what you should consider from an equity perspective. What was the background of the development team? Is it diverse? Similarly, and what population was the algorithm created? Does the population align with the institution's population? A lot of times the algorithms may have been trained on a certain set of data that may not reflect the population in your practice. And to that end, what is the generalizability? Or you could ask the opposite question. Everybody talks about making sure you have a diverse data set, which is definitely one way to go. But some also advocate perhaps using algorithms for a tailored data set. So perhaps algorithm that specialize in black population or specialize in Hispanic population. Perhaps that would have better output and would have less misses, such as I described earlier from that paper on the chest x-rays. Can this bias help or hurt those impact? And we've seen examples where it could do either or depending on the population we're talking about. And what ultimately is the impact? And I think this is very important to consider because we're such at the beginning of AI that it's hard to quantify what some of these benefits might be. And as we've seen, depending on the population you're at, some of these marginalized populations could be impacted negatively. And ultimately, how do we monitor these algorithms, their performance over time? And this is a really critical question because right now, most places, ours included, don't have a robust way of checking how well these things are working over time. And having a dashboard as to how these things can work, for example, comparing using NLP of the report versus the output of the algorithm, that's something we're actually working on and could at least give us a first approximation of whether something is working well or not. So with that, the last thing I'll leave you with is that ultimately, this is going to be a partnership between how AI functions and any of the tools we use in radiology with the radiologists. And I think together, it's always better whether we're talking about the underserved population. Thank you. Thank you for that incredible presentation, Dr. Lee, and to all the speakers, Dr. Scott, Dr. Einstein, and for sharing your expertise and really discussing high priority topics in terms of how to advance health equity in different areas. So, you know, we have a few minutes left and just want to start with some of the questions that came through the chat Q&A. The first one, and this is directed to Dr. Scott, given your discussion on the screening mammography component at your institution. The question that came through from the audience was the best age for screening mammogram and at what age can someone with positive family history of breast cancer can start screening mammography? So, you know, I'm not a breast imager, so I would just preface my answer by saying that. But I think, like I said, the American College of Radiology, most cancer societies recommend that women start at 40 and every year after that. The United States Preventative Services Task Force recommends 50 and every other year after that. So you can see that there's a huge difference in what the recommendations are. And, you know, the experiences and outcomes of patients can really, really hinge on ultimately what guideline an institution chooses to follow. So in my particular situation, I am really pushing to change what my system is following now, which is the United States Preventative Services Task Force, to following the ACR recommendations. Because, like I said, 30% of our patients with breast cancer are below the age of 50. So there is definitely room to change that. And I think in terms of if you have a family history of breast cancer, I think it's even earlier than 40. Like, I want to say maybe 10, I don't want to, I'm not going to quote, but I think it's even earlier than 40. But I would say, though, that sitting in a safety net institution, there are challenges because our volume is so high. And the question and the pushback that I get, which makes me, you know, it is a reality. Can you handle the volume that we will send your way if we do change these recommendations? So I think there is definitely work to be done to get to that point. But this is something that we as a system, I feel as if we have to do, because we have, I think, an obligation to make sure our patients are getting the care that they need and preventative services that they need. Thank you so much, Dr. Scott. We have another question, and this one is directed towards Dr. Anzai, given the discussion of AI. The question is, do you think that with great development of AI, the number of specialists in radiology will decrease in the future? I do not believe that specialists will decrease, but the interesting how AI is going to implement it and actually function as a whole system, not just a testing environment, a developing algorithm. For example, outside of the United States, brain radiographs are not read by radiologists. They are enough to just a handling for sectional imaging. So imagine outside of the United States, again, this is no U.S., X-ray can't be screened through AI and machine learning and then flag something abnormal. So that will go to radiologists. That is a totally near future. And then I think that the still AI algorithm can detect and diagnose pulmonary nodule, the large vessel occlusion for acute stroke patient. But radiologists have to verify the diagnosis so that we can exclude a false positive and false negative both way. So radiologists and specialists are still in high demand in the future moving forward. The other things that, you know, we don't go to primary care doctor to get that blood pressure measurement. We all had a blood pressure measuring devices at home, pulse ox, blood pressure measuring, and ultrasound may become one of those tools that every household may have. If they can scan ultrasound and upload it to the whatever the AI program, they may have a diagnosis. And then depending on what imaging study shows, AI will say, well, go to this. Well, this is whatever. They may guide the necessity for further management or diagnosis. So again, I think that care at home, that concept is going to be expanded in the future. Thank you so much, Dr. Anzai. I want to thank all of today's faculty for sharing their expertise. I want to thank everyone for attending today's webinar. Please make sure that you copy the link from the resources panel so that you can complete a brief survey and earn CME credit for attending today. Please also check the RS&A website for additional upcoming educational activities and resources. Thank you, everyone. Thank you.

radiology

diversity

equity

healthcare disparities

mammographic screening

value-based healthcare

AI in radiology

community partnerships

digital solutions