Read the Full Video Transcript

Jason Hafron: Our next speaker, Dr. Inderbir Gill, represents the GOAT of US urology. His entire career, he has developed foundational innovations and discoveries in advanced robotic urologic oncology. Currently, Dr. Gill is a distinguished professor and chairman, Catherine and Joseph Aresty Department of Urology, executive director of USC Institute of Urology and the Shirley and Donald Skinner chair in urologic cancer surgery at the Keck School of Medicine, University of Southern California, Los Angeles. Prior to this, he was at the Cleveland Clinic for many years. Dr. Gill has published over 870 scientific papers.

He has edited and co-edited 10 textbooks and has been on the editorial boards of nine urologic journals. He has been invited for over 450 visiting professorships, invited lectures, and live surgery demonstrations worldwide. He has received various honors, which are too far or too long to list, but I think his most impressive award is the Dr. B. C. Roy National Award for Eminent Medical Person, awarded by the president of India in 2005, which is equivalent to the Medal of Honor in the United States. His primary academic focus has been in advanced robotic urologic oncology surgery for cancers of the kidney, prostate, and bladder.

He has been involved in over 15,000 cases and those who have seen him operate know he's truly the Michael Jordan of robotic surgery. More recently, his interest has expanded to focal targeted therapy for prostate cancer and he and his team are now exploring artificial intelligence or AI applications in urology. In 2021, under his leadership, USC urology established the nation's first foundation-funded dedicated urology AI center in a urology department. It's with great pleasure I introduce the GOAT of US urology, Dr. Inderbir Gill.

Inderbir Gill: Thank you, Jason. I love you too. It's such a pleasure to see Dr. Hafron again after such a long time. We were hanging out at Cleveland for quite a few years and I just want to thank the LUGPA group. I've watched with admiration from afar how LUGPA has grown and what a service it gives to urologists in the country. Thank you, Dr. Spiess. Thank you to the LUGPA overall organization. Generative AI in healthcare, this is something that is a more recent interest we have been trying to learn. Dr. Maranchie is right on. This is very, very early days and if as a specialty in urology, if we do not really embrace this, or at least critically explore it, we are going to be the losers. AI is going to be all-pervasive in our lives and also in medicine and healthcare. This is a big topic, GAI. I am certainly no expert in GAI. What our interest is.

What are the practical aspects of artificial intelligence that can really help us today in our daily work to decrease work, to increase the rewards and make us smarter? My disclosures are listed at the bottom of this slide and I also want to recognize Dr. Kachi Amani who is in this room, who's my colleague, and we've been working on this together for the past three years or so. I'm not going to go through AI, machine learning, artificial neural networks, and deep learning. They have already been covered. So let's just get to it. Practical AI. We want to talk a little bit about how it can impact academic publishing and clinical practice. So starting with academic efforts, we created, as Dr. Hafron mentioned about two, three years ago, a multidisciplinary AI center in our department and multiple members, urologists, radiologists, pathologists, and from the school of engineering, software engineers, data scientists, et cetera, starting with a half a million dollar gift.

We then subsequently generated a significant amount of NIH funding and have multiple ongoing active projects. Dr. Kachi Amani has put together this Prisma AI reporting guidelines. I mean this is the level of interest in this arena, just putting together a guidelines committee to be looking at systematic reviews and meta-analyses in AI got us a publication in Nature Medicine. Taking it one step further, even hotter, ChatGPT. It's all over the place. This has the potential to do significant harm and significant fake publications, et cetera. So the guidelines for reporting this are essential and we have put together a strong steering committee. Look at the list of the journal editors, Nature, Nature Medicine, Lancet, JAMA, eLIFE, PNAS, et cetera, et cetera, are all part of the committee we have put together in order to come up with guidelines as to how GPT should or should not be used in academic reporting and this was just published in Nature recently.

In November, next month, our editorial is on the cover of the Journal of Urology and just in two months has generated over 1200 downloads again attesting to the significant interest in the community in this regard. I have never tweeted in my life, but guess what? I don't need to learn how to tweet anymore. We have put together an engine that will automate the tweets for you. So social media content generation for not only Twitter but also Facebook, Instagram, blah blah blah, multiple languages. And we looked at these three top journals, New England Journal, JAMA, Lancet. These are the original tweets that were put out by these journals for these 30 papers. And these are the tweets that our engine created. These are the tweets our engine created automatically without any human input. Each tweet took 13 seconds to generate automatically.

And then Dr. Kachi Amani recruited 1300 Amazon Turk users, 1300 Amazon Turk users, guys out there. And we said, "Here you go, blinded. There are two tweets. Which one do you like more? Which one do you feel gives you more information, et cetera, et cetera." Uniformly the auto-generated tweet by our engine outscored the original posts from these journals. And so at least in this day and age, publication is one thing, but its dissemination is another and people focus on the dissemination quite a bit. And so we feel that this, and we have submitted this paper to the New England Journal AI. Let's see what happens. And you can see in this plot that the GPT-generated tweets outperformed the actual journal tweets. And these are not any ordinary journals. These are the best in the business yet it outperformed it. And as regards the data from this, the title of the article was correct a hundred percent of the time, 97% of the time, findings, hashtags, et cetera.

I don't even know what that means. Emojis, again, I don't know why emojis have any place in life period, but apparently in medicine emojis need to be improved. But bottom line, these things can be automated. We don't need to waste our time working on this. But this has real practical implications. NEJM has said, given us the paper back for revision. So it looks like they're going to be looking at it seriously. We also have been working on automating the layperson summary that many journals do. And the lay summary generated by our technology outperforms the original paper abstract and also outperforms the original layperson summary. Actually, this last paper that I just showed you showed up on the cover of Urology Practice. One step further, it actually was cited in Time Magazine last month. Going on to clinical practice. So just want to... Burnout is all around us. Okay, and we went to medical school to take care of patients not to service the computer, which is what we do for a large part of our daily work.

And this is because we are doing multiple daily mundane tasks, mundane, repetitive, scribe-like work that just drives us into the ground and takes away the joy of medicine. So there is a good potential that AI can do this mundane work for us, leaving us for higher-order tasks that are at the top of our license. Clinical co-piloting, for example. Now this MedPALM2 is available wherein you just put up a chest x-ray and ask the thing to, "Hey, can you write me a report for this thing?" And boom, it will be able to automatically do it for us without any prior training. So this is available today, apparently. I have not used it, but I'm told that this is available today. Physician burnout is what I was talking about. I've already listed my disclosures. This is real. Everybody in this room knows about it. I don't have to belabor the point. This is in JAMA Surgery. Surgical residents, they're still in their residency yet 75% are already burnt out. They haven't even finished training. 75% are like, "We are done." JAMA Health Forum, this got worse in COVID.

How much time do we spend worshiping and servicing the electronic health record in the outpatient clinic? 74% of our day in the outpatient clinic is just hanging out with the EMR. We'll get to the patient whenever we are ready, but first we need to take care of this. Otherwise, I can't bill, I can't get out of the clinic and I'm going to be getting suspension notices from my administration. I really thought urology was amazing. I still believe it's an amazing specialty. Then why the heck are 54% of us burnt out? And I was being shown today that amongst all jobs in the United States, urology has the highest burnout rate, which is surprising to me, stunning to me. But those are the data. And this has been recognized by Vivek Murthy, the Surgeon General, saying that time for incremental change has passed. We need bold fundamental change that gets to the root of this burnout crisis and basically he has made this a national priority.

The problem is the operational inefficiencies that we just basically have to power through. 50% of our time is spent documenting care. Many visits are not even billed and there's significant lost revenue. Three-fourths of US patients wish their healthcare experiences were more personalized and two-thirds feel that they would be able to come to the physicians more often if they felt they had a more personalized experience. The scope of the problem is significant. 1 million physicians in the US today do 1 billion outpatient clinical consultations. And as discussed, 50% of the time is spent in EMR work, 27% with the patients. And when we are with the patient, 85%, we are spending again looking at the computer and no real conversation with the patient. And this is all because of the daily repetitive scribe-like mundane tasks. The time to create a new clinic note, on average, is 20 minutes.

This is national data and our goal is to shave seven to 10 minutes off this note, thereby returning back to the physician one to two hours per clinic and decreasing this mundane work. So issues that need to be fixed are the redundant scribe-like repetitive tasks, let's automate it. Clinic workflow, let's automate it and thereby elevate our game so that we work to the top of our license. What we went to med school for, what the PAs and NPs went to their APP school for. They didn't go to school to follow us around. They went there to again take care of patients and MAs and scribes also elevate their games such that they can help us better take care of patients. So the technology options available today are... Let's look at the EMRs and then some apps. So EMR, here are the top four or five EMRs that is their market share within the United States.

As regards the apps are concerned, I have divided them into physician-generated and patient-generated. Physician-generated are the AI scribing platforms wherein the clinical note... So what are we trying to do? We are trying to facilitate the creation of the outpatient clinic note. So physician-generated, if the physician is creating the note, AI scribing platforms are now beginning to be available. If it is patient-generated note, intake platforms are beginning to be available. So the AI scribing platforms are listed here. All these scribing platforms work during the consult. Voice recognition. It is speech-to-text technology that will take away the typing that we have to do. And so let's look at the EMR apps for the physician-generated. This was in the New England Journal recently, GPT-4 as an AI chatbot. So here is a clinician-patient interaction that occurred.

So the clinician is talking to the patient, he's put the little device on the table that goes to the cloud and this thing comes back. Here is the note clinician-patient and using the chatbot then this was created by GPT, a brief summary that, remarkably, is able to focus in only on the important stuff and get rid of the fluff. Two points I'd like to make. Number one, this entire thing, if the clinician did not ask the patient all these questions, nothing would go up to the cloud, nothing would come back. And if you are asking, you're seeing five prostate cancer patients or 10 BPH patients in one day, you're asking the same question over and over and over and over. I mean just shoot me. And instead, why do I have to ask the patient these questions? Why can't these be asked for me in advance of the clinic? And I walk in there, all those data, every question that I want to ask technology should be able to ask for me, individualized to the patient.

So one issue here is with this technology is that you actually have to think the question, then ask the question, only then you get the output. Secondly, the note comes back to you after the consultation is finished. You still have to do work in the evening. Saturdays, you can kiss goodbye to four hours a Saturday. That's what you're doing. So it would be wonderful if we were able to have all this done without physician input. That brings us to patient-generated intake platforms, which are pre-consult. So you will see over here, all these are pre-consult, all these were during consult. All the patient-generated occur before the consult, and these are during the consult. And in here there are various technologies that are beginning to be available. The first two lack clinical detail in depth and also AI extraction of outside records. So I want to talk a little bit about the last one, the novel AI platform, which is automating the outpatient clinic note. Two components to this.

Number one, as soon as the patient calls your place to set up an appointment, a link is sent to the patient and there are patient and specialty-specific adaptive questionnaires that mirror real-time physician logic. And then is the AI engine to which you upload outside records, PSA, MRI, etc., and it'll go extract only the relevant information and put it all together on the backend invisible to the physician, all happening in the backend. And the editable note automatically shows up in the EMR before the patient has been seen by the physician. So again, first is the AI takes the subjective history. It is an intelligent simulation of an expert physician. I want to highlight the word expert. Expert physician stepwise history taking. Then AI extracts data from the CAT scan, the bone scan, the PET PSMA. You can see that little prostate little thing over there. That's where the biopsy, the pathology report, what percent core positive with that horizontal line. The red in there, 80%, 40%, GG2, right side, left side, etc., all put together for us automatically.

And then all this being integrated upfront in advance for the patient visit. So the workflow of our clinic could be in the near future, it could be that the patient calls your clinic to schedule an appointment. Automatically the appointment is given and the patient receives the link. At home, the patient does the AI pre-consult at home with the family. And you could say, well no one does this stuff. The data would suggest 88 to 92% will do it. It takes about 10 to 15 minutes for them to figure it out. Their son, daughter, granddaughter, family, somebody is going to help them put it together because they've been told that Dr. Hafron would really like for you to do this, it'll help you and it'll help him as well. So as soon as he hits submit instantly the engine creates the note. This is the highest quality note, that beats the notes that we have currently.

And the AI engine then also extracts anything from the outside data that has been uploaded. These two are combined on the backend. A near-complete note lands automatically in the EMR days before that you're seeing the patient. And then during the actual consultation, the physician reconfirms findings. We always have to reconfirm findings. Doesn't matter who has created the note for you, your PA, your NP, your fellow, your resident, whoever. If it's not you who's created it, you have to reconfirm it. And then you do the physical exam, finalize the management plan and sign off. And the only work prior to the patient being seen is done by your team is sending the link.

So the sound is not going to play because of the thing here. But the bottom line is this is the patient-facing active sheet where from your phone you can just create a quick little video, personalize it for the patient. The patient has not even seen you already is aware of you, has heard your voice. And then they fill out this questionnaire. It is just click, click, click. All adaptive, all specialty-specific, highly customizable and mirroring our reasoning. Different for different patients depending on their presentation. And then this is the physician-facing thing where you can see that note on the right side of the screen. This is the automated note that shows up in the EMR without any work having been done on your part. Now your job is to check this out. You're not doing extractive questioning, you're doing confirmatory conversations and you can see the subjective history.

Everything is nicely there, and the objective data as well. And this saving seven to 10 minutes per patient translates into one to two hours saved per clinic. Two to four more patients seen per clinic, a reasonable amount of new revenue based on the patients that you see, and this is based on conservative estimates of having only two clinics per week. So the value add here is automated, highest quality documentation, working smarter, not harder, and providing detailed history from the comfort of their home, standardized across your practice with increased billings. And here are some summary clinical data from some of these AI engines. Excellent patient compliance. For a normal survey, if you send me a survey, I'm not filling it out, but if I'm a patient and your secretary says you got to do this, I will fill it out. 88 to 91% is the data.

Higher patient satisfaction you can see, provider satisfaction. In comparing the current manually created note with the automated AI-created note, a 59% improvement in just the sheer quality, academic quality of the note. Provided time savings about an hour or two and some increased billings potentially. Much other stuff can be automated. The pre-visit registration, consent forms, ID, insurance card, e-signature, etc. All this stuff is the simple stuff that can be done in advance. And then also after the visit. Avatars can be created that would help humanize some of this clinical support. So the final points I want to make are, I just want to compare the physician-generated AI scribing platforms with the patient-generated intake platforms. So AI scribe platforms are basically voice to text. AI intake platforms are AI, OCR, optical character recognition, logic-driven platforms. So in the AI scribe platforms, the technology works during the consult.

You are asking the questions. It's being scribed. In the AI intake platforms technology works prior to the consult. The note in the AI scribe platforms is created after the consult. In the AI intake platform, it is created before the consult. The note is finalized also after the consult in the AI scribe. It is finalized during the consult in the AI intake such that when you're done with talking to the patient, you walk out the door, you are done, done. Physician must engage in extractive questioning in the scribe platforms, in the intake platforms. They are confirmatory conversations. Pre and post-clinic work still remains if you're using the AI scribes. However, pre and post-clinic work by the MD is not going to be happening since all the work is happening during the clinic. And finally, external records have to be manually extracted by the physician/provider in the scribe platforms. But in the intake platform, the possibility exists that these records are automatically extracted by the technology.

So how could something like this benefit our practices? The goal being to decrease work and increase reward. Quality of note is a higher quality, standardized note compared to the manually constructed notes. I showed you some data. Patient access, decreases appointment wait times and more patient throughput, saving seven to 10 minutes per consult. Patient experience, good, happiness, satisfaction data. Physician joy and wellness. If you are working and creating my note for me, of course, I'm going to be having joy. Optimized coding can help from that perspective as well, and some new revenue. So again, reimbursements are decreasing. We are working harder, we are working longer hours. And I've heard from many folks in this conference today that there is no money really being made in clinical medicine per se, it is more the ancillary stuff that is generating the majority of the revenue.

So we have to figure out a way to work smarter, not harder and AI, I think, can really help us get there. So in conclusion then, I think that we must pay attention to these new technologies that are now pretty much all around us and we're going to be increasing. Physician burnout obviously is real and urologists must explore the potential of AI robustly. We are working on a little publication here and we are doing in our practical urology symposium a little plug for our course. We are going to be doing the artificial intelligence symposium on February 1. We would love to welcome you to USC and with that, thank you very much.