EPISODE 57

New Challenges and Opportunities in Pathology - Dr. Greg Osmond, Chief Medical Officer of Pathology Watch

09-23-2020

Despite being integrally involved in making diagnoses and treatment plans, pathologists remain fairly invisible to most patients. According to pathologist Greg Osmond, some of his colleagues don't mind staying behind the scenes, but out of concern that the profession in undervalued and at risk for automation, he's sees an opening for greater relevance in having pathologists provide a coherent picture to the wider care team of the many diagnostic and prognostic test results any given patient may have. In addition to considering that new role, the profession is also facing a deluge of digital tools and techniques that are coming online. Osmond, despite co-founding a digital pathology company, shares with host Dr. Rishi Desai that doctors really need to understand the limits of AI and other emerging modalities that are sure to change the practice of pathology in the coming decade.

Transcript

DR. RISHI DESAI: Hi, I'm Dr. Rishi Desai. Today, on Raise the Line, I'm going to be joined by Greg Osmond, Chief Medical Officer at PathologyWatch, which provides digital products and services for dermatology practices, including integration with electronic medical record systems. I'm looking forward to talking to him about how he sees pathologists' role on the care team and how pathology is changing, including the growing use of artificial intelligence in the field. Thank you so much for being with us here today, Greg. 

DR. GREG OSMOND: My pleasure. Thanks for having me.

DR. RISHI DESAI: Maybe we can just dive right in. Can you start by telling us about your background, and what got you interested in medicine?

DR. GREG OSMOND: Sure. I am originally from Utah. I grew up in Utah and Missouri for a bit. I've always been drawn into the sciences. I've been interested in science for a long time. I bailed on college for a couple of years and went on a mission to Honduras. One day, I was down in the middle of nowhere in a place called Choluteca, Honduras. We got home that night, and under our door, there was this letter from some general of the U.S. Air Force saying, “Your country needs you. Come to this address tomorrow.” We had no idea what that was, so we went, obviously.

The U.S Air Force was doing cataract surgery for these people down in Honduras. They spoke Spanish, but couldn't communicate with the Hondurans because it's just a little bit of an interesting dialect and we were really in the middle of nowhere, so we ended up interpreting for them, for their cataract patients for a couple of days and saw that was incredible going from blind to being able to see in a matter of minutes basically. That's what got me into medicine. When I came back, I was on a career in healthcare from there.

DR. RISHI DESAI: That's a pretty compelling story. That's amazing. From there, why did you choose dermatology and then eventually moved on into dermatopathology?

DR. GREG OSMOND: I did pathology rather than dermatology, but basically, I was always interested in the diagnostic specialties and made my way through...you know, as a medical student, you rotate through everything. I ended up fixating on dermatology. I went to Duke for medical school, and we got an extra year to basically pursue whatever we wanted. I did medical research in melanoma with dermatologists and oncologists and ended up spending more time at the microscope with the dermatopathologists and then in the clinic. 

I ended up liking that piece of it better. It’s just kind of the instant gratification of immediate diagnosis rather than having to wait for labs or anything, you know for a lot of it. I was really drawn to that, and at Duke, dermpath is in the pathology department, not in the dermatology department, so I ended up going into pathology at Duke to pursue dermpath. I knew that I wanted to do this subspecialty before I wanted to do pathology in general, but I really liked pathology, and the clinical pathologic correlation of it is what I liked the most and coming to something definitive faster is always very nice.

DR. RISHI DESAI: I can totally relate. My first passion was in dermpath as well, to be honest with you, specifically around psoriasis, so for all the same reasons, I totally get that. Tell us a little bit about PathologyWatch, how it got started, what it's all about.

DR. GREG OSMOND:  Sure. PathologyWatch officially started about three years ago. Digital pathology has been around for a long time, but during residency, it became more prominent, at least to me, so I ended up doing a concentration program, at the GME office, in health policy and I focused on digital pathology and business models around digital pathology, And then through fellowship, I went to the Harvard system for dermatopathology. At the time, we would rotate through the various different institutions affiliated with the Harvard Medical School, I was already interested in digital pathology looking for models, and at the time, I met one of my main business partners, He has a background in machine learning and was the IBM executive track, worked with Watson, and has done other successful start-ups in the healthcare space.

After I'd been in practice for about three years, I stayed in touch with him looking at various things, and that's when the FDA approved the first digital pathology platform for primary diagnosis in the United States, which was the Phillips IntelliSite platform. Once that happened, it basically green-lighted that this was going to happen sooner rather than later in the United States, and that's when we started PathologyWatch. 

So the reason we’re an AI company...you know, digital technology is great for various reasons, but it's very expensive, and it cuts out no cost. It adds in a ton of expense. You're not making any more money. You're making less money. Without AI being part of the digital pathology rollout or systems to make it more efficient, you really can't justify it, and a lot of systems have been candid about that. They're betting that at some point, this becomes valuable, but it's really an expense. It's not helping them with their bottom line any.

DR. RISHI DESAI: I know that the pathology community has been trying to raise awareness in recent years about the role they play on the care team, and it's specifically pushing for a more direct relationship with patients, I guess. What do you think is behind that? From your perspective, what are the benefits of pathologists being more visible to patients in general?

DR. GREG OSMOND:  It's interesting, if you ask different pathologists, you'll get very different answers on that. Some of the key thought leaders inside of pathology have pushed this for a long time. This is the narrative that I've heard from them that basically says “We need to be visible, so people see what we're doing.” It's not a matter of trying to do more things. It's a matter of letting people know like what it actually is because it's highly involved and highly technical and when you put something in and get an answer out, there's a host of things that happen inside of that black box that you need highly qualified and highly trained people to do. That's one of the main drivers. 

There are plenty of other reasons, in my opinion, as to why it's important for pathologists to be visible. A lot of it has to do with patient care. Especially now, with all of the new diagnostic tests that have come out over -- not diagnostic, even like prognostic tests, just the other modalities that have come online -- pathologists in my opinion are probably the best positioned to manage all of that. That space is not well-defined inside of healthcare, who's going to control that.  Across the board, I think that pathologists can improve in filling that role. 

I've also seen some other people advocating that, “Oh, we should be more directly involved with the patient, actually talk to the patient,” social media, things like that. I think it's fantastic and there is value. It's just not compensated. Unless you're doing an FNA or a bone marrow biopsy, you're not getting paid. It's charity work to go out and do that. Whereas overseeing the laboratory and the diagnostic testing and interacting with clinicians, making sure the patient has the best diagnosis, there is, at least depending on how you're structured, the pathologist has an incentive to take that over and then they will be compensated either directly or indirectly for a lot of that work. 

Exactly what people should be doing, I don't know, but for the specialty overall, there is a win for you to be visible and engaged both for the patients to win and for the specialty to win, and for you personally.  If the pathologist steps out and fills this role for a hospital or an organization that recognizes there are these eight things in whatever diagnostic area that you're looking at, that  all need to be integrated together in a comprehensive report to give the best care. Who's going to do that? I really do think pathologists are best suited. They just may not want to do it. Some of them just want to sit in their office and just look at the glass, and they don't want to do the rest of the work. That’s how I see it evolving.

DR. RISHI DESAI: What you're talking about is fascinating, the incentive structure, shaping the scope of practice. I'm curious to get your thoughts on how the practice of pathology has changed in recent years.

DR. GREG OSMOND: A ton. It's changed so much. A lot of the new technology, a lot of the investment in the pharmaceutical like venture capital markets and even some private equity have focused on pathology and diagnostics, and it has changed a lot. There are still pathologists that are practicing that never trained on immunostains. If you want to say how it’s changed...immunostaining was a huge change. You look at all of the changes in hematolymphoid diseases or neoplasms over the last 15 years, and it's amazing how much it's modified and changed the different testing modalities that go into it with, like flow and FISH, PCR testing, and next-gen sequencing.

I mean next-gen sequencing is still coming out in some institutions, and it's still questionable who's going to oversee it. Some pathologists do molecular training, and they're highly trained to do that, but you don't need to be a pathologist to be in that space. And those are all kind of like ancillary testing. Some pathologists specialize in clinical pathology where it's more like that type of testing or blood testing or chemistry use or whatever, but most pathologists tend to be in the anatomic pathology space where they're looking at glass slides and diagnosing and then they'll order these extra tests to help them, but that's kind of the bread and butter of the pathologist that hasn't changed for hundreds of years. 

That is what has been a dramatic shift -- I guess it's shifting under us at the moment -- which is the digitization of pathology and the technology associated with it. That's the space that Pathology Watch is in, but it's not just us. There are a host of start-ups that are targeting different parts of anatomic pathology in terms of AI as well as digital pathology, vendors that are coming out with new systems that are smaller, cheaper, more efficient. It's rapidly changing. For most pathologists, nothing's really changed because they're not digital. 

They're like, “We'll see how this really plays out.” I don't think that they recognize the investment that's in it right now. It is massive, the amount of money that's coming into digital pathology and AI right now, and that has over the last couple of years. There's no way that this isn't going to change healthcare within the next five years. At some point within maybe 10 years, I would say that at least a portion of anatomic pathology, the standard of care will be to digitize it and to use some type of algorithm development. I just can't see that not happening in my career, at least.

DR. RISHI DESAI: A topic that I think that dovetails with what you're saying is specifically artificial intelligence. I'm curious how you've seen some common uses of AI in pathology.

DR. GREG OSMOND: In terms of common uses, first of all, what I would say is that the FDA has not approved any diagnostic algorithms in the United States or even diagnostic assists that I'm aware of. It’s all experimental, or people are using it for non-clinical reasons, for research reasons only. How it's actually being used in the United States by different organizations, I'm not sure, but I know what different companies are doing and the uses that they’ve seen.  A lot of the initial use cases that I've seen at least marketed which are non-clinical has to do with case sorting where AI will sort cases based on what it thinks a potential diagnosis could be to try and optimize things that way. 

Sometimes, they'll do notifications. I know radiology has some FDA approved applications that'll do use cases like that, kind of case triage with the preliminary possible and notification that something could potentially be a certain diagnosis. That's one application that I've seen people working on. 

Predictive analysis is another one that some companies, at least early on, focused on that exclusively. So if you can predict a mismatch repair syndrome or Lynch syndrome based on the H&E alone, using an algorithm versus having to do all of the immunostains or additional testing, that'd be one example of predictive type testing. Drug companies have been looking at AI to predict drug responses. Based on different features that an algorithm can determine, you can predict how well one drug is going to work for you versus another one for your tumor type. 

Diagnosis is another one, and some people are using case sorting. Diagnosis could mean a bunch of things, and it does mean a bunch of things to different companies. So if you're doing case sorting, any type of case sorting, you're relying on a preliminary type diagnosis in order to sort it versus some companies are going to diagnose it, subtype it, call the margin status, generate the report --  basically try and automate the pathologist’s job. Those are two ends of the spectrum of what diagnosis could mean. Some people are looking at diagnostic algorithms solely for things like QA/QC, so after the fact pathologists have already diagnosed it, and then it'll run through some type of QA/QC in the background to try and flag a potential miss or something like that. 

Those are the main use cases that I see. There are a lot of other use cases like virtual stains. If you digitize it, you can use the features of the digitization to auto-generate a different type of staining that doesn't look like an H&E. If you're trying to highlight elastin or other types of factors in there, you can do it digitally, and digitally enhance it. Other algorithms are bypassing a bunch of the upfront TC process completely, or they'll just take an actual tissue, or unstained tissue and then use some type of algorithm to generate a pseudo H&E and move on to other types of staining patterns to diagnose things. 

It’s interesting. The technology is amazing. It's much farther along than people probably realize. It'll be interesting to see what use cases really get used like what really ends up being practical and efficient for pathologists and clinicians in terms of all the technology because there's a bunch of different things being developed, and I think that some of it is not really going to do much and other stuff I think would be highly useful. It just depends. We'll see what the market decides 10 years from now. 

DR. RISHI DESAI: Given those use cases you've illustrated, if you were to sum it up, what would you say are some of the key advantages of using AI in pathology?

DR. GREG OSMOND: I still think they're kind of theoretical. They haven't well established the benefits of AI yet, and it might not work well right now. That's what the investment community is betting on. When they pick a company, they're betting on who's going to actually build something that's useful.

Right now, it is quite a big question mark, in terms of what's going to be adopted and truly be useful. I think it is theoretical, but theoretically, AI could make it significantly more efficient which allows the price points to drop internationally especially, where there are these dramatic needs and coverage for pathologists people can afford. You just can't afford diagnostic tests in certain countries. It’s just the reality. Taking an algorithm and how well the algorithm works will come into play, but if your algorithm works really well and you can exclude high-risk diseases with 100% sensitivity, then you should be able to automate a certain subset of diagnosis. 

When the option is no diagnosis versus a diagnosis that's right 99.9% of the time, even if it's automated, that's significantly better care for that community. That's not in the United States. An international implementation could look something like that. I think it's theoretical, but at the end of the day, in my opinion, AI needs to make your life better. It's got to make the physician more efficient somehow.  I'm not sure that case triage really does it. I know that it's an early use case in the United States that people pitch because there's nothing diagnostic with it. It's just case triage. 

And going down that point, depending on what you want out of your AI, will dictate what you choose to implement inside of your system. When you're looking at a platform, you say you're trying to go digital, and you're like, “Well, who should I go digital with?”, there's a bunch of different vendors. There are a ton of software companies that will try and plugin, and it really depends on what you want AI to do. If you want AI to do triage cases, versus if you want AI to diagnose your case, set up your report, and truly make you efficient in the long run, those are two very different propositions. The AI itself is architected very differently. One is significantly more expensive than the other one to roll out. Some of the investment community, as well as plenty of the physicians, they hear AI and they're like, “Oh, I want to do AI. It's going to be able to do X, Y, and Z.”

The reality is that AI is not going to be able to do X, Y, and Z. It is going to be able to do X, or it's going to be able to do Z. What do you want on your platform? That's going to dictate what you choose to do. There are some prominent companies that have published different use cases already, and some of them are prominent publications, but the way that they're architected makes it impossible to do certain use cases with at least that portion of their algorithm. You're basically betting on a horse to pick a platform right now without seeing how things come out, and there are so many factors on the pre-analytic space, so before the algorithm runs, you've got a file type, you've got a scanner type, you've got a stain type, you got a processing type and all of those impact how well the AI performs. You can make your AI perform decently on all the different variables, or you can control those variables and build something that works amazingly, but then it's a more brittle algorithm. It works amazing, but you have to control certain factors upfront. You can't just roll it out like, “Hey, this software has AI. It's going to be great.” That's not how it works right now. Maybe in the future, we'll get there, but there are more factors involved in rolling out an AI to make it practical than I think people realize, at least in the current state.

DR. RISHI DESAI: Flipping it around then, what do you see as some of the risks to applying AI in pathology, and for the risks, are there any mitigants that you're aware of?

DR. GREG OSMOND: If a physician is going to double their case volume because all of a sudden, AI can help them in some form or fashion...there are definitely risks associated with that, one of which is the physician turning their brain off. I'd say that's the biggest risk. You can't trust AI, especially not right now. I think it's going to be a long time before you can fully trust it and automate anything. The mitigant to that, I would say. is that the physician needs to really understand what the AI is doing. You can't just put it into an actual black box and pump it out and trust what it says. The way that you architect your AI makes it possible to understand what it's doing, whereas other ways if you architect your AI in a different way, you can't really understand what it's doing. 

So in my professional opinion, you would want to understand what the AI is doing, how it's architected, and that will help you know how much you can trust it. Obviously, you need to do significant validation studies in your specific laboratory or specific instance with all of the exact same variables locked down, so it's a consistent environment, and then you'd want to test it extensively, and the physician needs to understand what the algorithm can miss and what is the probability of missing it is. 

For skin basal cell carcinoma is extremely common, the number one cancer in the world, it's a very easy diagnosis to make visually. If AI can diagnose that and make you more efficient in the basal cell carcinoma, you need to know exactly what it might call basal cell carcinoma -- like a false positive. If the physician isn't aware of the limitations of the AI and they don't understand how it works, then that would be a significant risk that would otherwise be mitigated with training, basically, and education.

Dr. RISHI DESAI: Our audience is composed of a lot of students, early-career health professionals. What's your advice to them about meeting the challenges of this moment and approaching their career in healthcare in general, given the direction you’ve seen in pathology and even more broadly in healthcare? 

DR. GREG OSMOND:  Exactly. This isn't just in pathology. Healthcare at large is changing rapidly, and the same technology is being plugged in everywhere that it can be right now. I would say that you need to do what you love and are interested in. If you're really interested in diagnostics and pathology, but you're worried that the market's changing and you don't know what it's going to be like, there's always going to be room in every specialty for people that are highly competent and passionate about what they're doing. It doesn't matter. There's always room for you. That's what I would do, first and foremost, is making sure that you pick something that you enjoy.  

I actually think that pathology is going to change. For people, as part of the change, there are going to be a lot of opportunities. Whenever there's change, there's just as much opportunity as there is risk to more established players. So choose what you love and what interests you, make sure you're doing something you're passionate about. I'd say that money is important to consider, at least if you have as much student debt as I had when I came out of training, but now that I've been out seven years, I would say money is a lot less important. I'm really glad I didn't pick a specialty that would pay me more than what I thought pathology would. 

Your happiness isn't tied to money. Once you get past a certain number and that number usually is across any specialty. In choosing a specialty, I'd recommend that you go talk to physicians in that specialty that are in the middle to the later stage of that specialty just because physician burnout is real, so if you know what kind of lifestyle you want to have, what kind of person you want to be, you should talk to somebody at the later stage of that specialty and see how happy they are and how they feel about their job because there can be a big difference between specialties and it depends on what you want out of life. It is the reality, so I would definitely do that. 

The other thing that I would say is there are a ton of pathways in healthcare that I didn't recognize at all, at least initially. There are so many opportunities outside of it, especially if you're sub-specialized, and you find your niche, whatever it is inside of your specialty. You've got two physicians right here that their primary job is not seeing patients in the clinic. In the business world, education, research, pharma, academics has all kinds of opportunities.

Digital tech right now is massive in healthcare across any subspecialty. If you're interested in Silicon Valley-type medical healthcare start-ups, there are tons of job openings in that space in basically every specialty that I've seen. So it's a great time to be in healthcare. I think healthcare is going to be a lot better in 10 years because of it. 

DR. RISHI DESAI: That's a very helpful synopsis, and I appreciate the optimistic outlook. I think a lot of us are looking for reasons for optimism right now, so I think that's a great way to end. Greg, thank you so much for being with us today.

DR. GREG OSMOND: Thanks for having me. It's been great. Nice to meet you.

DR. RISHI DESAI: I’m Dr. Rishi Desai. Thanks for checking out today's show. Remember to do your part to flatten the curve and raise the line. We're all in this together.