Tomorrow's Medicine, Ep. 7: Emerging Health Trends & AI - Dr. William B. Weeks

With Dr. William B. Weeks, Microsoft AI for Good Lab. Hosted by Arseniy Arsentyev. Published August 31, 2025.

https://aarsentyev.com/podcast/ep7-william-weeks

Machine-transcribed and lightly edited: filler words and false starts are removed, and the wording is otherwise as spoken. Where the two differ, the recording is the record. Times are minutes into the recording. Not medical advice.

[0:00] Arseniy Arsentyev: Hey, it's the seventh episode of Tomorrow's Medicine. My name is Arseniy and today I'm talking to Dr. William Weeks. Dr. Weeks is a physician economist leading the AI for Health effort within Microsoft's philanthropic AI for Good research lab. Dr. Weeks has an MD, MBA and PhD. He's a well-renowned scientist who published over 250 peer-reviewed papers and two books. The conversation with Dr. Weeks was a fascinating dive into the present and future of AI. Also, we discussed current healthcare challenges such as costs and administrative issues.

[0:45] Dr. William B. Weeks: My pleasure. Thanks for having me.

[0:47] Arseniy Arsentyev: Yeah, and my first question is, everyone, when he or she hears Microsoft, thinks about Microsoft Office, Microsoft PowerPoint, maybe Windows, but I think not many people know that there is a philanthropic branch to the Microsoft and especially doing something related to healthcare. Could you please describe what are you doing at Microsoft and what is this Microsoft philanthropic branch?

[1:16] Dr. William B. Weeks: Sure. So, Microsoft is a huge organization. I think it has around 220,000 people across the world and it's got particular divisions within it. I work within the division called CELA, which stands for Corporate External and Legal Affairs, C-E-L-A. And within that resides Microsoft Philanthropies. And there are several different types of philanthropies. Ours is actually a research philanthropy. So, what we do within the so-called AI for Good lab is we partner with not-for-profit organizations to see whether artificial intelligence can be used to solve some of the world's biggest problems.

[2:00] It's philanthropic. It's free to the partners. We don't have any expectations of sales or anything like that within our lab. We are just there so that we can explore this question. Can AI be useful for these massive problems that the world faces?

[2:21] Arseniy Arsentyev: Thank you for your answer. And also, as I know you're a writer, you wrote two books, two very good books. And I want to talk about one of them. It is AI for Good. Could you please outline the main idea of the book for our audience and describe what motivated you to write such a book?

[2:42] Dr. William B. Weeks: Sure. So, the book is called AI for Good. I think it's Applications in Sustainability, Humanitarian Action, and Health. And really, the idea of the book is to kind of consolidate the work that the AI for Good lab, where I work, has been doing for the past five or six years. So, the lab was established by Brad Smith, who is the president and vice chair of the board of Microsoft, as a philanthropic effort. Again, it was established, I think, in 2019. And so, it had been going on for about five years or so. And we just saw an opportunity to consolidate all the work that the lab has done into a book that would be more focused on lay readers.

[3:35] As part of the lab, what we do is we publish a lot of academic papers, right? So, when we partner with organizations or we even do internal work, we will have technical reports and academic papers published in medical journals and computer science journals and so forth. What we wanted to do is make that work more accessible to a lay public, particularly, people who might be interested in what Microsoft is doing in this area and wondering whether AI can be useful in those particular areas of sustainability, humanitarian action, and health. So, we just kind of decided to develop this book. And so, I worked very closely with members of the lab and obviously Juan Lavista, who is the director of the lab, to develop kind of an introduction.

[4:24] Brad Smith wrote the foreword to the book and to then structure chapters in such a way that they can be independently read so you don't have to have read the earlier chapter to understand the later chapter. And they kind of articulate the main findings and the key processes and methods that we use to get to those findings through different research projects in those different areas. So, again, the idea is make it accessible to a lay person, avoid a lot of the technological challenges that someone who isn't, doesn't have a PhD in computer science or something would need to have to understand the method. So, make it accessible and then also format it in a way that it's easy to use as an independent, if you're interested in this particular topic, you can just read it.

[5:20] You don't need to have a lot of background or follow-up to understand it. And the structure of the chapters is similar so that you're familiar. If you look at these three or four, you'll kind of understand, even across the different disciplines, the kind of structure and what we're trying to get at.

[5:38] Arseniy Arsentyev: All right. And let me at this point back up a little bit and just double click the concept for my audience and also check my understanding. So, you're working in a branch that helps other non-profit academic research organizations try to find the place for AI to improve the healthcare. Is it correct?

[6:00] Dr. William B. Weeks: Yeah. Again, the three different areas, but I particularly am in charge of the health aspect. So, I run the health portfolio in essence and kind of help, identify the questions and do the research and then write up the reports and the papers for the health-related stuff. But we do, we partner with exclusively not-for-profit healthcare organizations to do the analytics. The challenge, people often think that Microsoft has all sorts of data. We really don't. We are a platform company. The mission is to empower others. So, we don't just have a bunch of stuff and then want to use those data to do analytics. We need to have a partner who has data, who has a research question, who thinks that AI might be helpful in answering that research question. And then we partner with them.

[6:57] They keep the data on there, in their environment. They grant us access to data after we've, gone through any kind of regulatory or compliance issues like having an institutional review board review of the protocol or whatever is required. And then they always control the data. We never ingest the data. We simply do the analytics on the data and then write up the paper for publication.

[7:26] Arseniy Arsentyev: Okay. And that raises a question. How do you measure the success or long-term impact of deploying AI solutions in such a context? Do you have a certain branch on a certain board that after you did some impact that analyzes the outcomes? Because right now, AI is everywhere. Anyone is integrating AI to anything and I'm not sure that it works 100% of the time.

[7:54] Dr. William B. Weeks: Yeah, I agree with you. I think, so for us, the impact is really getting the word out that this is doable. Our projects are invariably proof of concept projects. So they are not designed around creating some kind of a medical product or something like that. They're simply, this, for this particular type of a problem in healthcare, AI can provide a solution with this kind of accuracy and sensitivity and specificity, for instance. So we can provide that, again, with the partner. Then the way we structure these agreements is that the partner has the intellectual property for any kind of algorithm or coding that is developed in that process.

[8:41] The partner, then, we would want to kind of scale any solution and pursue, if they want to pursue FDA approval as a medical device or whatever, they can do that. It's not us doing that. We're just kind of helping to say, is this doable? I think the challenge of research, which is what I am, and we are a research lab, right? There's kind of an old joke that if we knew what we were doing, they wouldn't call it research, right? And research fails. It's the Thomas Edison thing that he didn't find one way to create a light bulb. He found a thousand ways not to create a light bulb, right? So research is an iterative process. Sometimes there are failures.

[9:23] Sometimes you think this is a great idea and it turns out to be very, very much more complex than you had anticipated. And sometimes you find out that it's just, we're not quite there yet. It's too complex of a thing and maybe in a few years we'll be able to get to the solution. So it's always this kind of risk of failure with research that we have to kind of embrace and say, but that's how you stay on the cutting edge really of application of AI in these particular circumstances.

[9:52] Arseniy Arsentyev: And in addition to these risks, of course, Microsoft, as any other company, have limited resources to support such a project. And how do you prioritize which projects to support, to fund, given this limited resources? What is the criteria and things that just, when you see them, you think we need to support that?

[10:15] Dr. William B. Weeks: So, yeah, the prioritization scheme is primarily around potential for impact. So, if I tell people, potential partners often, that if this has a big impact on some patient population or health status of the world, then we're going to be more interested than if it doesn't have much of an impact. So, as an example, I'll give you, we did a study on retinopathy of prematurity. So, retinopathy of prematurity happens when babies who are born prematurely have this disease, their eyes don't develop normally, and it's the biggest cause of blindness in children in the world. It's actually been fairly long-standing. Stevie Wonder, the musician, has, who's 70 years old, he's blind because of retinopathy of prematurity.

[11:15] So, it's been long-standing, but it's becoming more common because more children are surviving prematurity, right? It used to be that not a lot of kids survive being born very early. Now, because of medical advances, that is the case that you're just kind of, kids are surviving more and more. The challenge with retinopathy of prematurity, which is treatable, you can give laser therapy and medications to actually treat it to prevent blindness, but you have to screen for it, and you can't just screen once, you have to screen repeatedly for retinopathy of prematurity.

[11:53] There are very limited resources for such screening. The people who tend to screen for these, for this disease are pediatric ophthalmologists, and they're very rare as a profession, even in the United States, but much more so in developing countries. So, what we developed was an algorithm that wherein you could use just a cell phone, right, a smartphone, and a magnifier, and you could take a video of the retina, right, the back of the eye of an infant, and the nice thing is you take this video, the video is automatically processed by AI to identify frames in the video that are good pictures of the retina, right, because kids are moving around, they're babies, right, they're moving around and they aren't cooperative and stuff.

[12:44] So, you can say, okay, either there are good frames in this video or there are not good frames. If they're not good frames, it goes immediately back so while the kid who may be somewhat sedated or may have his eyes dilated, probably has their eyes dilated, you can immediately redo the video until you get the good frames, right? So, then once you have these good frames that show, let's say, the 10 best pictures of the retina, there's an algorithm that says this patient is likely to have retinopathy of prematurity or is not likely to have retinopathy of prematurity.

[13:16] That data is then uploaded to a pediatric ophthalmologist who is using this process and that pediatric ophthalmologist then can verify that, yes, this looks like retinopathy of prematurity to me or no. Either way, it goes back into the system to improve the algorithm, right, so we're better and more consistent in the next time that we do this. But the nice thing is that that pediatric ophthalmologist is having all of this done for that person and not having to do it themselves, right?

[13:47] So if you think I'm an economist as well as a physician, right, so if you think kind of from a labor economics perspective, if you have this very scarce and expensive resource called a pediatric ophthalmologist and currently they're spending 80% of their time screening patients and 20% of their time treating patients, if you could flip that, right, so they're spending 20% of their time screening, most of the screening is being done for them, that means they can spend 80% of their time treating patients. So you've effectively quadrupled this very rare and scarce resource like overnight, which is kind of tremendous.

[14:31] So if we think about impact, right, on the world population, on children not going blind and being blind forever, right, it's not something you can recover from later, we think about how the impact of something like this, what the impact is pretty substantial, that here we have a relatively simple tool that took several months to develop that can help, expand a very scarce workforce, help patients who otherwise might go blind get access to treatment in a very time-efficient way and actually not go blind. And that's kind of an impactful and useful thing for the world, right, for not only for, from just kind of humanitarian help and people not having a disability for the entire rest of their lives from age zero to whenever they die.

[15:30] So it's a nice thing for them, but also, from a government standpoint, it may kind of relieve some longer term cost pressures in providing disability services or additional services that such people might need in the future. If we can cure this disease early on, identify it and cure it early on, there may not be the need for additional services around blindness for those folks. So, again, it's an impactful thing. It kind of does, it helps patients, it helps them be productive members of society in an easier way and contribute to the workforce in an easier way and it may avoid costs for ministries of health and governments overall into the future. So, that's a fairly impactful kind of a thing.

[16:25] Arseniy Arsentyev: Such a great case. You reminded me of my high school years because I was working on a similar project. We were doing a melanoma detector for the skin and at that time I was amazed how much time we can save for a specialist using such AI automation. Because the melanoma is some divergence from the normal of a nevus and you can literally design an ML algorithm and then predict it with the higher accuracy than the average specialist and put right into production.

[17:03] And, at this point, I want to ask you have an MD, PhD and MBA. So, what do you feel is the most useful qualification and what is the best experience that will help aspiring young men or women as I am to really affect the healthcare sector? Because, at one hand, you can work as a doctor, you probably need to get an MD, but if you want to work more on a policy side or just provide an impact on a more scientific way, not the way of the doctors, what would you recommend?

[17:53] Dr. William B. Weeks: Yeah, so for me, it's really the interplay of those degrees that makes it interesting. I frequently will tell people that I think it's a really good career move to work where Venn diagrams overlap, right, in that intersection. So you have IT in health, or you have economics in health, or you have whatever it is, there are two, because what I've found in my very long career is that people with particular disciplines don't necessarily talk well to one another. So you actually act as a bit of a translator.

[18:29] So one nice thing I can do is I can take the work of a computer scientist who may not know healthcare all that well, right, but they know computer science, and I can translate that into something that people in the healthcare arena can understand and appreciate, right? So it's not just a technical thing, but it actually becomes a clinical application for them. And I think for me, I learned early on after my, I did a residency and stuff, I learned pretty quickly that I thought a lot of healthcare in the U.S. was driven by monetary decisions, right? So I got an MBA first and then later a PhD in economics because I like this interplay of healthcare and finance.

[19:16] And there too, I think a really important lesson for me early on in my career was that the clinicians, when you go to medical school, you do a whole bunch of science and stuff, right? You do organic chemistry and biochemistry and things like that, but you don't do, you never do accounting. You never can read a balance, right? It's not part of your curriculum unless you build it in. And most people, there's so many requirements for pre-med that most of it is science, right? And on the other hand, MBAs don't really learn about healthcare. They learn about management and the numbers and stuff like that, but they don't really learn about healthcare.

[19:57] So again, what I found was being able to kind of bring those two worlds together, was actually very helpful for improving healthcare delivery because if you just have people clashing and they just don't understand one another, they're never going to get anywhere. But if you can find some common ground that they both align with, and most people think, yeah, I want people to be healthy, that's a good thing.

[20:22] So if you tell the MBAs, this is how things can be healthy, the population can be healthier, and this is why it's important to a clinician, and you can also tell the clinicians, this is why it's important to the MBAs and the finance guys, then maybe they can have a little bit more flexibility in how they approach things, and they can kind of come together in some kind of a common ground to kind of achieve the overall aims. So kind of having those two different perspectives, or even three with a PhD in economics, I think is really what makes you kind of see the world a little differently. Like this retinopathy of prematurity example, right?

[21:02] Clinicians were all for it, and that's wonderful and good, but as an economist who might be more appealing to a government, right, who's got limited resources and is trying to figure out how to what's the best method for having a provider workforce to serve their population, right? If I can tell them, hey, you do this and you can quadruple your workforce overnight, you know how long it takes to get a pediatric ophthalmologist? It takes four years of undergrad, four years of medical school, five years of ophthalmology, and then another specialty of a year or so, right? So if I start tomorrow doing this, it's going to take me 10 or 15 years to get some kind of a workforce, right?

[21:45] And meanwhile, all of these kids with retinopathy of prematurity are going blind because they don't have access to care. So if I can do something that is technologically savvy to immediately quadruple their workforce, that's a pretty helpful thing for them. And they see that from an economic perspective. Sure, of course they care about the health of their population, but sometimes I think for the folks in government, they've limited resources and the money's got to go so far and how do they manage it is a really important question. For clinicians, they just want to do good and probably, knowing how doctors think, pediatric ophthalmologists would rather be treating patients than screening them.

[22:26] They get, it's more fun for them to do the laser treatments and to see people get better than to just kind of check and see if people have these diseases. So it's kind of a win-win-win if you put it like that, but you have to know these different kind of perspectives to understand how to make that win-win-win work.

[22:43] Arseniy Arsentyev: Yes, and healthcare is such a conservative field as we know and it is sometimes slow to adapt. My next question is, I'm curious type of question, whether you met anyone without advanced degrees, so with no MD or PhD, having a bachelor's degree or maybe even not having a bachelor's degree, who's the leader and transforming healthcare right now.

[23:12] Dr. William B. Weeks: Yeah, so Bill Gates. Bill Gates dropped out of college, I think, after his first year, right? But he is a very thoughtful man who reads very deeply. So he understands things in part, I think, because he goes into such depth about them independently. And he reads, a whole bunch of books and a whole bunch of different disciplines. So he kind of does what I had to do. I'm not smart enough to do it on my own. I had to have people show me how to do it. But he does what I've done and I think he's very able to look at different perspectives and understand things.

[23:56] So he gets, I think a lot of his work in the Gates Foundation in supporting, trying to address infectious diseases which are much more impactful in low and middle income countries than they are in developed countries. He's trying to do that with a knowledge, a really very deep knowledge of medicine as well as of the kind of the business aspects and the economic aspects of developing a stable long-term workforce for these countries so that they can develop better into the future.

[24:35] Arseniy Arsentyev: I'm asking you because I have many friends in computer science who work in the industry without having advanced degrees and I know literally no one who's working in the healthcare sector without having an advanced degree. So Bill Gates is of course an outlier in such context. And do you think it would change? Do you feel that healthcare would have more young people and people not with extensive experience but with extensive ambition to change the field?

[25:12] Dr. William B. Weeks: Yeah, I think it's possible. I think like any discipline, a lot of what you learn in medical school is a different language and a different approach. So you learn that with economics, you learn that with whatever it is you're studying. It's really important to understand that language. One of the challenges that I have found is if an organization is pitching a technology to a healthcare organization, they need to speak that language, which again, that's the beauty of that Venn diagram overlap. You know both languages and you can kind of translate.

[25:50] If they fail to speak to the language that the CEO or the chief medical officer or the chief nursing officer understands whose priorities are around patient care, if they fail to kind of articulate the value of technology to that goal, then they're not going to be very successful. So I think any of these folks you're talking about, as long as they understand the language, and what I tell people is again, it's an advanced degree, but you can usually get it in one year, is a Masters of Public Health, just because that gives you a really broad exposure to general language.

[26:33] So you don't have to do all the clinical work and things like that, but you kind of understand the importance of clinical work to population health, you understand that integration between the two disciplines, and you can understand the language that then can be helpful in thinking about how IT or computer science or whatever other discipline the person might be in, can again express themselves in a way that the person to whom they're pitching appreciates it and sees the value.

[27:05] Arseniy Arsentyev: All right. And you're an adjunct professor at NYU School of Global Health right now, is correct?

[27:12] Dr. William B. Weeks: Global Public Health, yeah.

[27:14] Arseniy Arsentyev: Global Public Health. Yeah, I remember I met some people from this department, such a great place to work, so congratulations for you. And I want to have some details about this master's program on global health. Can you describe what is the difference between pursuing any other field related to healthcare and medicine and pursuing global health studies? So what masters are studying? Can you provide sample courses for our audience to understand it better?

[27:51] Dr. William B. Weeks: Sure. So there's kind of just fundamentals of public health, how population health and public health work, the differences between clinical health, clinical processes and public health goals and their commonalities. And then there's some things like statistics and epidemiology and kind of research skills that are necessary, too. A lot of public health graduates go on to be epidemiologists and work in public health, arenas in governments, right, whether it's for WHO or for local governments or something like that. So that's kind of the goal is to produce people who can take on those roles, but also to maybe do research around epidemiological research and understanding of social determinants of health, things that impact the overall health of the population that are outside of the health care, the clinical health care realm.

[28:50] So those are the kinds of things that they study. That's the kind of work that they do. I think, it's important, it's kind of a common thing that they teach in public health is that while clinical care kind of impacts or is associated with 20 to 30% of overall health outcomes, non-clinical stuff like social determinants of health and genomics and health behaviors account for the rest, the other 70 to 80%. So population health can actually have a kind of a bigger impact than clinical health on the population's overall health status, which is kind of motivating for people. The challenge in the U.S. is that not a lot of money is spent on public health.

[29:42] If you look at other OECD countries, they split the public health and the clinical health kind of dollar much more equivalently than in the U.S. where we have something like 95% of the health care dollar goes into clinical health and 5% goes into public health. And many people think that that's a lot of the reason that our outcomes, despite the fact that we spend more than any other country, like almost double any country per capita on health care delivery, but our health care outcomes are not so great, is probably because of an underinvestment in public health in the U.S.

[30:23] Arseniy Arsentyev: Got it. So let's talk for a little about today's health care and then we'll proceed to the main topic of my podcast, which is tomorrow's medicine. And tomorrow's health care. So today's health care. You mentioned many times throughout this discussion, when you were talking about Bill Gates, when you were talking about global public health, these monetary issues that are constantly chasing today's health care. What can you say about the ways in which today's health care is working from the perspective of business and from the financial side? So please outline the broad context that we're living right now.

[31:19] Dr. William B. Weeks: So I'm going to talk really mostly about the U.S., right? Because that's what I know.

[31:23] Arseniy Arsentyev: Yeah, for sure.

[31:25] Dr. William B. Weeks: So the U.S. is very financially driven as far as health care is concerned. And that's part of the reason, like I said, that I got into, got an MBA because I kind of figured that out fairly early in my clinical career that a lot of these decisions are financially made. And we have a private health, we have an interesting situation in the U.S. We have private health insurers, right, that take care of most commercially insured folks who work for, during their working age, right? So while you're employed, you have employed-sponsored health insurance that is provided by private insurers. And you simply, your company on behalf of you purchases health insurance. You get access to care that way.

[32:11] There's also government-sponsored health insurance, which is about, I think, 50% or more of the total health insurance pie. And that includes things like Medicaid and Medicare, with Medicaid being for poor and younger folks, generally, and older folks. So Medicaid covers people who are under a particular income threshold, and it tends to be people who are poor. About half or more of deliveries, babies, are born through the Medicaid process. But it also supports, Medicaid also supports nursing home placement for older patients. So when, if patients don't have, someone needs a nursing home and they don't have, the funds to support it themselves, they can access Medicaid if they're eligible to get access to a nursing home so that they can have that kind of care.

[33:04] Medicare is for people who are disabled, so they have some level of disability that prevents them from entering the workforce or staying in the workforce, or they have given, they've worked for 40 quarters, so 10 years overall, not necessarily in a row, but 40 quarters they've contributed to Medicaid, Medicare, the Medicare trust fund, and that makes them eligible when they're 65 for Medicare, which is a government-sponsored and subsidized process of care. The challenge for if you're a healthcare deliverer, so if you're a healthcare system like a hospital with physician practices and things like that, is each of those different groups pays differently for the identical service. And generally, it follows that Medicaid is the, pays the least for a particular service, then Medicare, and then commercial insurer.

[34:00] So, to be able to survive and to pay their employees, right, and to pay for the facilities and to pay the doctors and stuff like that, the healthcare system has to have a mix of patients that is balanced, balances the Medicaid population with the commercial population. So, in essence, if I'm commercially insured and I get some service, I pay more for that than the cost of the service, but the Medicaid person who comes in after me to get the identical services pays less. So, in combination, the healthcare service is covered with someone paying more than the cost to deliver and someone paying less. And that's kind of the way it is in the U.S. In almost every other country, healthcare is very heavily subsidized and provided by the federal government.

[35:06] It costs very little to get access to care from a copayment perspective. The entire population is covered and so everyone pretty much has access to care. With the complaint sometimes being that they have to wait for a longer period to get elective surgeries like knee replacements or back surgeries or hip replacements. In the U.S., it's very different. There's somewhere between 11 and 15 percent of the population is uninsured because they cannot qualify for Medicare or Medicaid or any of the other kind of government. There's the Indian Health Services and the VA Service and the FQHCs, all of which are federally funded, but nonetheless, they can't qualify for those.

[35:55] So they are so-called self-insured. They have to buy insurance through the Obama marketplace, the ACA marketplace, but they can now opt out of that. So in that case, the insurance is fairly expensive. It will become more expensive in the next couple of years and people will increasingly opt out of purchasing that. That means that they will have to pay out of pocket and they have to kind of pay the full price, which is pretty expensive. It's higher than the price that I get as a commercially insured person.

[36:33] And that, and as people have pointed out in the U.S., you know, medical debt is one of the biggest causes of bankruptcy in the U.S. So our system is quite different in the U.S. than most others. It's challenged because we leave a certain population, like 10 or 15 percent of the population, out. They're kind of on their own. They may then do things like not seek care if they need care. Or alternatively, they may wait until they're very, very sick, in which case they simply default on their medical debt. They can't pay it, so they never pay it. And so those costs have to be absorbed by other people, which makes the costs of my insurance as a commercially insured person go up.

[37:19] So it's a very complex arrangement of multiple insurers, some government, some private, some just individuals that creates kind of a challenge in the financing of health care in the U.S.

[37:37] Arseniy Arsentyev: Could you also outline how decisions are made? So we have healthcare economists, we have private companies, we have this big pharma, and we have also the government, the HHS, and so on. Could you tell me how new treatments are put into the system, how policies are changed, and other, committee meetings to sign and to accept certain policy?

[38:14] Dr. William B. Weeks: Yeah. So if you have, say, new pharmaceutical, right, it has to go through a certain process. And so generally the new pharmaceutical, you identify a compound, a particular chemical compound, and then the pharmaceutical company would patent that compound. And that patent lasts, I believe, for 17 years. And then it goes off patent and it's available. So within that 17-year period, what the company has to do is they have to do phase one, two, and three trials, right? So they start with just seeing if it's dangerous to use the compound and then how does it do in mice or in apes or some other animal model? And then how does it do in humans for a particular disease?

[39:05] Then they have to do these human trials, which then have to demonstrate that the medication is at least as good as or better than an existing treatment, right? So you can't really, compare against just placebo if something else is already existing and seems to work. So it's got to be, either a superiority or a non-inferiority is not worse than the current treatment regimen. And then if it's able to demonstrate that through medical research, which costs a lot of money to get these, you usually have to pay patients, you have to recruit them, you have doctors who are being paid to evaluate the patients and provide the care, you have the, all the investment in R&D by the pharmaceutical and so forth.

[39:55] You do all that and then it goes to FDA approval. So the Food and Drug Administration has a committee that then oversees this, they look at the science and they say, yes, this works. And then the company then is allowed to market that medication for a particular indication, say whether it's depression or arthritis or whatever or cancer, for a particular thing. They market it for that and then they try to promote it. So they have a sales force that goes out and tells doctors about it. They put on conferences, they talk about, the use and the advantages of that medication over another medication or another regimen and then they sell it.

[40:40] And then that, so during all this time period, right, that 17 years is being eaten up. And for a large scale trial of a thousand patients, it may take two or three years to get the patients enrolled to do the follow-up because the follow-up isn't like the next day. People might have to be on the medication for six or eight months or something like that and have a follow-up, right? So a lot of that is being chewed up. And so then they have a limited time period really to generate revenues that will kind of pay them back for that initial investment of maybe eight or ten years. And the pharmaceutical companies will also point out that those are just the ones that are successful.

[41:27] In the meantime, they also have a whole, just like research, like I was saying before, right, it can fail, right? So they have a whole lot of compounds that didn't make it to even phase one trials because they failed, right? So they have all this investment and so they have to kind of, they have to pay for that somehow. And so that's how they pay for it is that during that time period where the patent is available to them, they can then kind of recoup their investment in not only that compound but in other compounds. So that's kind of the process for a medication that one would go through to get FDA approval.

[42:01] In the EU, for instance, they have a different approach where they have an organization called NICE, and I can't really recall what NICE stands for, but they have an overall, the EU kind of looks at the cost effectiveness of a medication. So they look for not only is it effective clinically, but is it cost effective? Does it cost less to do this? Does it generate cost savings when compared to another medication? So the FDA doesn't do that part of it. It just says, is this effective or not? In other countries, they determine whether it makes sense from a cost perspective, from a government cost perspective.

[42:44] And then I think from a pharmaceutical perspective, the big difference between the US, actually US and New Zealand, I believe, are the only countries wherein pharmaceuticals can direct to consumer advertise. So they can provide information directly to consumers, to potential users of their medication through TV or any kind of media that might then stimulate demand for that pharmaceutical, that particular pharmaceutical agent, where a patient might come into a doctor and say, here's an advertisement I saw in a magazine. I have this condition. What do you think about this medication? That's not legal in, I believe, every other country but the US and New Zealand.

[43:31] Arseniy Arsentyev: That sounds like it. And we're now in this debate between fee-for-service and value-based payment models. And I just want to hear your comments and your opinion about this famous example of a drug for spinal muscular atrophy that costs $2 million. For a single person. And how this example fits into all the things you described. And I'm curious, if my understanding is correct, would the price drop in the near future after such a monopoly for 17 years or something as you described will officially end? So yeah.

[44:19] Dr. William B. Weeks: Yeah. So after the patent ends, the medication is then available for generic companies to develop, right? And so the prices for generics are almost invariably cheaper than the prices for the brand labeled thing. The reason, the rationale that a pharmaceutical company might provide for the $2 million thing is that there may be no other treatment for that. And that's not a big population, right? So if you're treating hypertension in the U.S., like half of the population, half the adult population has hypertension.

[44:54] So a $1 difference is a lot of money, right? Because people are on it repeatedly over time for years for the rest of their lives usually. But if you only have 10 patients, right, how do they recoup all of that investment that I talked about and the patent development and the, they have a limited time, how do they recoup that? It costs, their argument is it costs a lot of money. They need to be reimbursed a lot for access to the medication. So that's kind of the rationale for that.

[45:27] Yeah, it does become cheaper as generics come in and maybe even generics compete with one another to drive down the price. Because just like if you have a monopoly on, essentially a patent provides you a monopoly on anything that you got the patent for some limited time period, you're less price sensitive than a generic might be. As far as like fee-for-service versus, it's really kind of fee-for-service, value-based care is still part of the fee-for-service in that fee-for-service just means you pay for the service, right? So if I go see a primary care doctor, I pay for that visit. If that primary care doctor says you need these labs and this x-ray, I pay for each of those. And if I need some kind of follow-up, I pay again.

[46:19] Value-based care might be something that I'm being treated for back pain or for, I'm getting a knee replacement. So now I don't pay for each of those services separately. I combine them into one kind of a service and I might have, which might be a bundle, right? And in the U.S., these happened a long time ago originally with something called DRGs, diagnosis-related groups for hospitalization. So it used to be that if you got, if you went to the hospital before 1981, you went to the hospital, you get, 3,000 pages of every time a nurse gave you an aspirin, you would be charged for the aspirin. You'd be, so there'd be all these charges. And they said, no, this is ridiculous. We're going to combine this.

[47:01] So now you get a bundle of services that are provided by the hospital, separate services provided by the physician, but nonetheless for the hospital, a bundle of services. And that then becomes a bundle. So a payment for a compilation of services. The value bit gets to be in, we're going to withhold some of that money unless the outcome is a good outcome, right? So here we're trying to see, can we drive the cost down and improve outcomes by withholding some of the money and not paying it until they demonstrate that they've met certain quality metrics or outcome metrics. And then the other way, which most, again, most governments around the world pay for healthcare is they kind of do it through a capitated system.

[47:52] And here, if I'm a healthcare system in the U.S. I might, be taking care of 10,000 patients or something like that. And I would just get paid a fixed amount for each patient, right? For a year. So I get a certain, it's a per member per month. It's a per member per year, kind of a payment. And it covers everything that they have, or maybe it's specified what's covered and that's fine. You can see what the incentives do, right? If I'm getting paid for every service, right? I'm incentivized to provide a lot more services, right? Cause I get paid every single time I provide them. So I'm going to do more x-rays, more lab tests, more pharmacy, more things like that.

[48:35] If I've, if I'm given just a single amount of payment for a whole year for a patient, I'm now incentivized to do fewer services, right? To provide less care. The challenge there is people then worry that, oh, I'm not going to get the care that I need. Right. So you have to balance that by saying, no, no, no, but you have to provide these kinds of services and you have to bring out and we're going to watch it at how many times those kinds of services are provided. So you, there's always a balance because people will, there's this old adage. When people, if you provide an incentive, people will do that. And so you have to worry about unintended consequences because they will do that to an extreme.

[49:19] The old joke is right. That it's a true thing. I think it was the cobra. So in India, there were cobras and I believe it was India. I may have this wrong, but there were cobras and people were given, they were, incentivized to provide dead cobras, right? Because they wanted to kill off this cobra population. So they did, they killed off, they were killing off this cobra population, but the people who were incentivized to do this were also breeding cobras, right? So they had, they were making more and more cobras because every cobra they killed, they got money for. So they didn't want to just kill them off. They wanted to have more and more and more money, right?

[50:02] So the challenge with any kind of incentive is there's always these perverse outcomes sometimes where people will behave in such a way to maximize their own profit from the incentive. So you always have to kind of be very careful about what you ask for because you will get that. And you have to then balance with some, countervailing, metrics to make sure that that is the system is not being abused, right?

[50:25] Arseniy Arsentyev: Yeah. That's probably what an MBA should teach you.

[50:28] Dr. William B. Weeks: Exactly.

[50:30] Arseniy Arsentyev: All right. And the last question on today's healthcare, are there any emerging global health trends? I heard somewhere that depression rates are growing. Compared to the past. And do you notice any statistical or anything else, some changes in the health of populations today?

[50:51] Dr. William B. Weeks: Yeah. So a couple of things, the entire world's population is aging, and people are simply living longer than they used to. So since chronic disease is associated with aging, just by virtue of the longer you live, the more likely you are to have a disease. If you die when you're five years old, you're very unlikely to have a chronic disease because you're, it takes time to develop diabetes or hypertension or kidney disease or arthritis, any of these other common chronic diseases. So the longer you live, the more likely you're going to have chronic diseases. And we know the world is just aging because they're not starving to death.

[51:31] They're not having, it used to be that infectious diseases were killing off a lot of people at a young age and now they can live through that. So that's not happening. And just, so nutrition, even though maybe it's not the best nutrition in the world, but it's at least calories, people are not starving. They're surviving these, infectious disease outbreaks and they're just living longer. The conditions in which they live are better for longevity. So we will experience more and more diseases of aging, which include multiple chronic diseases and things like neurodegenerative diseases like Alzheimer's, right? Where you just kind of outlive your brain, in a way. So that's happening, depression and the kind of the pressures of society, is growing as well, across all kind of population demographics.

[52:25] Because of the fact that nutrition that we don't have to go out and hunt and kill something to eat it is available at a vending machine next door. Right. We have the obesity crisis, and obesity, which is happening even in low and middle income countries that pick up the nutritional habits of the West, is growing considerably. And so that has its own set of chronic diseases associated with it, including diabetes and hypertension, and metabolic syndrome, which increases deaths from heart disease and things like that. So people will live longer with multiple chronic diseases and more and more chronic diseases will develop by virtue of both the nutritional impact and the fact that people are just kind of living longer. And so they will just have more opportunity to develop those diseases.

[53:16] I think, those are probably the biggest ones. I think because outside of Africa, the world population will be declining. The shape of the population curve will change. So there will be fewer, younger people and more older people. This is already happening in the U.S. and that puts a stress on healthcare systems because it's the older people that consume most of the healthcare resources, right? Because of these very things I just talked about, they have more chronic diseases, they live longer, they run into more healthcare challenges than younger people who tend to be healthy.

[53:53] So if your healthcare financial structure is dependent on money coming from those who are currently employed, who are younger, to subsidize those who are older and retired, which is the case in the U.S. and all developed countries, then that's going to be a real challenge going forward because you're going to have fewer and fewer of those people who are in the working age group and more and more relative, of the population that's in the older age group. So there'll be a healthcare strain, increasing healthcare needs, but probably decreasing, healthcare resources and financing of healthcare to deal with those needs going into the future.

[54:39] Arseniy Arsentyev: Wow. So such a deep insight, cause I also thought about this demographic problems, but I haven't considered, the things you said that this percentage of older population relative to the young and people who pay the taxes, who finance healthcare research and actual clinical healthcare stuff, is changing. So yeah, that's a really great insight. I think I'll also do a clip of this episode to showcase it. And it's a great moment to transition into tomorrow's medicine and tomorrow's healthcare. As I know you're a professional in things related to AI and what do you feel would change in 10 years? Let's say 10 years, how this AI hype or AI technology would influence our medicine in 10 years.

[55:46] Dr. William B. Weeks: Sure. So as we just discussed, right, we have across the whole world, this changing demographic setup where we have fewer people, providing resources for an increasingly needy older population, people who are going to live longer and have more healthcare needs. So the harsh reality is to not just go bankrupt. We need to do more with less. And I think AI can be quite helpful in doing that. I think, in a several ways, first, even though people will have multiple chronic diseases and as they age, they spend the vast majority of their time outside of the healthcare system.

[56:27] So I frequently, when I give a talk on this, I will put up a pie chart that shows that even if someone is quite sick and they spend two full weeks, days and nights in the hospital, and they see a provider for 30 minutes every single month, they are still spending 96% of their time outside of the healthcare system. And so what I think AI can do is can very much help, with self care management and providing accurate and up to date information on how best to manage one's own health. And as an example here, your readers could just do a Google search or a Bing search and just put in without any AI enhancement, how do I best manage my diabetes? And what they will generally get is a long list of advertisements, right?

[57:16] These are things that can, you can use for diabetes, but it doesn't tell you how to manage your diabetes. It's just selling you stuff, right? If you actually use a Copilot or a Llama or a GPT, and you ask the very same question, it will give you very good information on how to best manage your, diabetes. And you combine stuff like that with AI powered algorithms that are linked to the internet of medical things, kind of devices, Fitbits, things like that can be used to kind of monitor ongoing daily, health.

[57:52] One can then use these AI algorithms to intervene so that we can avoid care that might be avoidable, right? Unnecessary hospitalizations or downturns in health that we can anticipate several days in advance and avoid that downturn from happening. So if we think about that, both of those examples, right? Self-management, improved self-management and using AI to avoid, expensive care that could be reasonably avoided, we are doing that very thing we talked about. We're doing more, by which I mean we're taking better care of this population with healthcare needs, with less resources, with less need for intensive hospital stays, additional unnecessary labs, and additional visits that maybe aren't needed. So I think AI can help, quite a bit with that.

[58:42] I think AI can help, particularly in low and middle income countries where there might be, healthcare provider shortages. It can help provide some healthcare insights to places with healthcare shortages, through using kind of distance care technologies, to people who have healthcare needs and need those resolved quickly without spending a lot of money and effort in expanding a healthcare workforce. So similar to the example I gave you with the pediatric ophthalmologist to the degree we can use technology to kind of expand the healthcare workforce very quickly and very cheaply. That would be a good use of AI.

[59:33] I think the other thing, the other kind of reality of healthcare is that not everyone gets the same quality of healthcare. There's an old joke that you know, what do you call the person who graduates last in their medical school class? You call them doctor, right? So there's variety in the quality of care that's being provided by different providers. I think AI can help provide decision support to improve quality of care for providers who may be, providing less than optimal quality of care. And maybe just like someone like me who you know, went to med school quite a long time ago and maybe has difficulty keeping up with the medical literature and the newest and latest changes, it can integrate that and make that that kind of decision support available to everyone.

[1:00:23] So as to kind of raise all boats. One thing we don't want to do in healthcare is it costs a lot of money from a societal perspective to develop a provider, right? In med school and undergraduate school and stuff like that just takes a lot of time and money. So we don't want to just fire people who are not performing well. We want to help them perform better. And AI can help, do that. And so those are kind of the broadest ways: AI can more efficiently help patients manage care without such a reliance on healthcare systems. It can avoid, care that might be, expensive and avoidable. It can expand the physician or the provider workforce so that so that the current workforce is adequate, for the needs of the population.

[1:01:16] And it can be used to kind of optimize health for populations and improve access to healthcare for populations around the world. And not just access to any healthcare, but access to high quality care.

[1:01:31] Arseniy Arsentyev: All right. And as you know, there are differences between LLMs, this side of AI, generative AI, and computer vision, the other topics. Speaking about computer vision in particular, do you see the place, the gap in surgery, surgical robots, different medical imaging and medical videos, that will be overtaken by AI? And AI will do maybe even some parts of surgery, some processes. What are your thoughts on that regard?

[1:02:09] Dr. William B. Weeks: My sense is that we'll always have a human in the loop. But even now with robotics, the robotics are helping the surgeons do better and more precise care for things that are important, like prostate cancer. Right. So brain, of course, brain stuff. Right. But prostate, actually: every man over age 60 has prostatic hypertrophy, and pretty much every man who's over 80, if they do an autopsy on them, has some micro prostate cancer cells in them. If they do an autopsy on them has some, micro prostate cancer cells in them. Right. So, and again, as people age, there's just going to be more and more common. So there'll be maybe more need for TURPs, transurethral prostatectomy. The challenge with that is that if you make the cuts wrong, you can cause impotence and you cause incontinence.

[1:03:05] You can cause all these, bad side effects that people would rather not have. So the precision of robotic enhanced, surgery is that you can maybe avoid those things because you can identify the nerves or the connections that would prevent, the bad side effects from happening. So my guess is that I seriously doubt that surgeries will ever be done, entirely, by robots, without a human being right there in the room and managing the robot. Right. And overseeing it. But I do think that those kinds of techniques will be more and more helpful to, again, avoid, a surgical knife, another millimeter deep can cause maybe really bad problems.

[1:03:54] So it can look at what's the tissue that's a millimeter deep and identify that so as to avoid some kind of complications, which would be really good for patients, I think.

[1:04:06] Arseniy Arsentyev: Okay. And do you have certain projects right now that are related to this future use of AI? Can you maybe give us an insight? What are you doing right now at Microsoft?

[1:04:18] Dr. William B. Weeks: Of course. Yeah. So we're doing, we are doing computer vision, exercises with looking at prediction of pancreatic cysts. So the challenge sometimes with radiology is that as it gets better and better, it can identify smaller and smaller lesions. And the question is, are those lesions clinically relevant? So if you identify a small pancreatic cyst, of which supposedly 70% of the population has a pancreatic cyst at autopsy, you need to do something about it. Right now it becomes kind of this liability. If you don't know whether it's going to go on to become cancer, because pancreatic cyst can, some of them are very aggressive. They can become cancer. Some of them need to be watched and some you can just ignore because they're never going to be anything other than a cyst.

[1:05:10] So can we use AI algorithms to determine from radiology, and from maybe pathology specimens, whether or not this needs to be followed? Because if we can't, we simply can't afford, again, going forward to follow up. If 70% of the population has a pancreatic cyst, we can't afford to follow up annually with a CT scan to see if they continue, see to watch their cyst progress. Right. It would use all of the CT scan, capacity in the U.S. to do that. So you couldn't do any other CT scans. It's ludicrous, right? So you need to have these algorithms that can kind of help you classify into whether or not something is going to be, needs to be followed up or not. So as to best use, scarce medical resources and not, have waste.

[1:06:04] So that's one thing we're doing. One thing, another thing we're exploring right now is you know, identification of fraud, waste, and abuse. So the thought is that 30% of the healthcare dollar in the U.S. is wasted on unnecessary care. And how can we then use large data sets to identify such, care and identify and give feedback to providers who might be giving such care. So they will quit doing that or to the insurers who are paying for it and tell them to stop paying for it, to change their algorithms, to stop paying for it. Other things we're doing, right now is we're looking at sudden unexpected infant death, and looking at, it's a fairly rare disease, but it, causes pretty devastating outcomes for people who just have had a kid.

[1:06:57] And then the child dies for kind of an unknown cause. Might there be predictors that can identify risks that can then be mitigated so that is a less, an even less common outcome. So we've done work around that with Seattle Children's Hospital, for instance; the pancreatic cyst stuff we've done with Johns Hopkins University. We are doing work with the Novartis Foundation to identify, at a population level, what interventions and social determinants of health might have returns on investment to the health of that population. So as I already, suggested, and when we're talking about population health, something like 70 or 80% of health of an individual is accounted for by social determinants of health and their genetics and their health behaviors. Right.

[1:07:55] So you might, if you intervene with social determinants of health, like the you know, air quality or water quality or health behavior, improving, exercise or reducing obesity or something like that. Where might you get a return to investment in lower, or improved health outcomes, lower health demands that might then reduce the needs, the demands on the healthcare system. So if you're a politician, you might strategically then say, if we know for different areas, for different census tracts in the U.S., if we target this intervention in this, census tract, we might anticipate less heart disease in five years. And that will have a financial impact on us as well. Not only will people's health be better, but we will have less of a demand in the emergency room and the hospitals.

[1:08:51] So we will have to pay less for those services. So we can kind of fund this intervention that might be cheaper than those healthcare services, in anticipation of getting a return on those healthcare, in a return in the form of a reduced healthcare service demand. So those are kind of several areas that we're looking at across the broad areas of computer vision and social determinants of health or public health, models. I think with regards to large language models, we're getting into that more and more. I think there, the opportunities are around identification of better and more patient specific and precise, interventions around constellations of diseases. So we can apply large language models to very large, data sets, and identify the best care pathways and then try to steer patients into those pathways.

[1:09:53] And maybe even into the providers that provide those pathways most efficiently and most effectively. So as to improve population health by just steering patients to avoid this unnecessary care and this wasteful care to the optimal care for them.

[1:10:13] Arseniy Arsentyev: Dr. Weeks, last but not least, what is the medicine that you wish to see? Just forget about where we're heading and the current state: you as a person, as a patient, what kind of healthcare system would you love to see in the future?

[1:10:31] Dr. William B. Weeks: Yeah. So the kind of healthcare system I'd like to see in the future is one that provides the care that I want and need, and only that care, in an efficient and an effective manner. So by the care that I want and need: we will often in medicine just say, this is the process you need to follow. And that may not be consistent with my own values. In fact, this happened to my mother-in-law. She had hypertension, right? And they put her on a medication for hypertension, right? She's 95 years old. She had no other healthcare problems. One of the side effects of hypertension, of a hypertension drug can be, it can make you hypotensive, right? In which case you stand up quickly and you get dizzy, right?

[1:11:40] This happened to her at the top of stairs. She fell down her, the stairs, broke a couple of vertebrae, had a head injury because of it. And since then has been disabled essentially, and has been getting worse. She was doing pretty well at, for a 95 year old, she's doing very, very well. And they put her on a medication to treat a number and not a patient, right? Because she had just like, at 95, it's not hypertension that's going to kill her, right? It was a bad move to do that. And it caused then the sequelae, that were, bad for her and have really disrupted her life.

[1:12:18] So instead of just saying, here's a number, here's the thing you got to do. Let's involve the patients in the care. Let's get, make sure that their values are appreciated in that care. And let's understand what their healthcare trajectories are likely to be. And here, I think large language models can be very useful in aligning values with care, with the care that's being provided and offered to patients. So that's the care that I want and the care that I need. I don't need, if I'm 95 and this is the first time I've ever had hypertension in my life and it's mild hypertension, I don't need a treatment for it. That's not going to kill me.

[1:12:54] Where, if I was, if I'm 30 and I have a moderate hypertension, that will be helpful to me because I may avoid strokes in 20 years. But as a 95 year old, I'm not going to live for another 20 years. It's just not going to happen. So that's the care I want and need. So effective care is stuff that actually works. We know that there's stuff in healthcare that works and stuff that doesn't work. And it may work for certain populations and not others. I think where large language models and AI algorithms can help is to identify where, which patients might be most responsive to which medications or which medication and therapeutic combinations there are.

[1:13:33] So to align, given that I want the care and I need the care, to align the best care treatment options for me. So it actually is going to work. And then efficient care is care that is not bombarded by waste and time delays. Right. So here I want to get the care that's going to be provided to me fairly quickly, going to be effective the first time we try it, and not going to involve a lot of unnecessary tests and visits and stuff that consumes my time and my money in getting that care. So that's kind of how I would simply describe, the optimal healthcare system for myself.

[1:14:14] Arseniy Arsentyev: Thank you very much, Dr. Weeks. It was absolutely great to talk to you today. And thanks again for your insights.

[1:14:23] Dr. William B. Weeks: My pleasure. Thanks for having me.

[1:14:25] Arseniy Arsentyev: It was the seventh episode of Tomorrow's Medicine podcast. The best thing you can do is to subscribe to my channel, share this video with your friends, and of course, support me on my Patreon page. Goodbye, everyone.