DISCLAIMER: Today’s Advertiser helped make the production and delivery of this episode possible, for free, to you! But the themes expressed by the guests do not necessarily reflect the views of Inside Reproductive Health, nor of the Advertiser. The Advertiser does not have editorial control over the content of this episode, and the guest’s appearance is not an endorsement of the Advertiser.
Quality Disclaimer:
Despite our best efforts, technical issues can occasionally arise. Please excuse the audio in the following episode as it doesn’t reflect our usual quality standard.
The innovative potential of AI is an increasingly common topic in IVF and beyond. But on this week’s episode of Inside Reproductive Health we bring back Dr. David Sable and Abigail Sirus to ask a different question.
What’s preventing AI from completely taking over the fertility space?
Dr. Sable and Ms. Sirus discuss the seven big areas of AI investment in IVF and the obstacles standing in the way of full fledge adoption.
Tune in to hear:
The 7 categories of AI Investment (And their criteria)
Their visual for how they categorize AI in the fertility field (Corresponding to their seven categories)
Current developments in AI across the IVF space (Including the sticking points)
What’s preventing the inflection point of AI completely sweeping fertility treatment (And making their four principles the standard)
Abigail Sirus
LinkedIn
Twitter
David Sable
LinkedIn
Twitter
Transcript
[00:00:00] Dr. David Sable:
But the big part of that is doing the work is merging the software with the hardware so that you're getting reliable data so that the information they give you is based on the hardware and software shaking hands. in a, um, in a valid way and not giving you, it's not like, it's not like with chat GPT when you ask it a question and it makes mistakes and it makes things up.
The data we're getting now is not made up. It's really truly reflective of what the hardware is finding. Taking the next steps of plugging that into real decision making is going to be difficult.
[00:00:38] Sponsor:
This episode was brought to you by LEVY Health. Seeing more patients for a first consultation may actually decrease IVF revenue by 30 to 40 percent.
To see why, download the numbers for free at levy.health/conversion. That's levy.health/conversion. Today's advertiser helped make the production and delivery of this episode possible for free to you. But the themes expressed by the guests do not necessarily reflect the views of Inside Reproductive Health, nor of the advertiser.
The advertiser does not have editorial control over the content of this episode, and the guest's appearance is not an endorsement of the advertiser.
Despite our best efforts, technical issues can occasionally arise. Please excuse the audio in the following episode as it doesn't reflect our usual quality standard.
[00:01:34] Griffin Jones:
Why? Why? Why? If any of you are gracious enough to think that I'm good at my job as an interviewer, it's simply because I'm a child that wants to know more specifics.
I want to know why, and I'm not satisfied with the answer of fertility clinic workflows are just too complicated. Why is this tipping point that almost all of us can see we're so close to having a I move over this inflection point and totally dominate how fertility treatment is delivered? Why are we still not at that inflection point?
When we're so close, what's holding us back? So I bring back two very popular guests. They're Dr. David Sable, who you know very well, a former practicing REI turned investor, and his colleague Abigail Sirus, a venture capitalist who worked with IBM for a number of years. Last time Abigail and David were both on the show, they went into their four guiding principles for democratizing IVF.
That's have a la carte options for IVF services, go around incumbents if you can to incentivize them. Set the standard of today's highest pregnancy rates as the absolute basement for outcomes for the future and pay for outcomes, not cycles. In today's conversation, I asked David and Abigail, what's preventing the inflection point from AI completely sweeping fertility treatment and making these four principles the standard of the day.
David and Abigail have . visual of how they categorize artificial intelligence in the fertility space, corresponds into seven general categories, and we go into each of those categories. David and Abigail share some developments of what's happening in each of them, and they detail the sticking points for each.
That's oocyte assessment, embryo assessment, sperm selection, hormonal stimulation, non invasive PGT. clinic decision support and workflow. And in their other category, which may overlap with obstetrics and other areas of healthcare, you're talking about follicles, preterm birth, reproductive immunology, and ovulation.
If you're having trouble picturing this visual, we'll link to it in the show notes and follow along as I try to get the specifics of why we're so close to AI domination, each of these areas and what specifically is standing in the way of each. Enjoy this conversation with two of your favorite recurring cast members here on the Inside Reproductive Health Show, Dr. David Sable and Abigail Sirus. Ms., Sirus, Abigail, Dr. Sable, David, welcome back to both of you on to the Inside Reproductive Health podcast.
Always a pleasure, Griffin.
You're both troopers. You're both gonna have to go to the chiropractor after this episode because like our last episode, your backs are gonna break from carrying that episode.
Last time, it was after Thanksgiving and I... It was the first thing on the Monday morning after Thanksgiving. And I know business owners are supposed to say, I love Mondays. I love mornings. I don't, I don't love Mondays or mornings. And I really, really love Thanksgiving. I remember being in a funk and you both carried that interview.
Now you're going to carry it more today because not only did I ask to talk about a visual concept on a show where the audience is 90, 95 percent audio only, I also. Send one image where I sent you one image and said let's talk about this and then was thinking of going in a different direction so you have so many different visuals for investment areas of the IVF space so I will have you back on for yet another episode to talk about the.
Other map that you've used to, to wireframe the, the, the whole IVF process. But today let's zoom into AI and the visual that we're talking about today, Abigail and David, is this going to be something that we can either share or that we'll be able to direct people to, to your website or previous article that they can reference themselves?
[00:05:28] Abigail Sirus:
Yeah, so I think that we we've published it on David's blog, David's medium site, so it's what you're not seeing to everyone who's on on audio only is it's basically just a chart of the different areas of innovation and AI in the IVF industry, so we're Talking about things like an oocyte assessment, embryo assessment, sperm selection as well.
So just the different versions, uh, different flavors of companies we're seeing innovate in AI and IVF.
[00:05:58] Griffin Jones:
In this realm of AI, you break it into oocyte assessment, embryo assessment, sperm selection, hormonal stimulation management, non invasive PGT, clinical decision support and workflow. And then you've got an other box.
And then in that other box, you've got follicle preterm birth. REI and ovulation. So, or excuse me, that's not, that isn't REI, that's reproductive immunology. That's, that's in your other box. Why did some things make it, uh, it, why did the things that are in your map make it to the central part that it is?
And others, Not appear here.
[00:06:37] Abigail Sirus:
Yeah. So the way we think about it, Griffin, we're venture capital investors and fertility is, is really looking at the market as a whole. So we map out a universe of all the companies. Find IVF and right now that number is around 280 and what you're looking at is our map of AI Subsection of that larger one, which is specifically the 20 plus companies It seems that the number is is growing more and more with every week that passes That are specifically focused on AI or are using AI as part of their processes and the way we map it out We're just looking at two buckets The first is companies that are looking to optimize IVF, the procedure itself, using AI.
And so that might be via oocyte selection tools, embryo grading tools, or things like, um, hormonal stimulation tools. And then the second bucket are the companies that are kind of adjacent to IVF itself, that are looking at the processes and procedures around delivering. IVF care and using AI to optimize those.
So we're talking there about, for example, an AI enabled chatbot that can help answer patient questions perhaps more quickly. Or, um, there are a number of companies that are focused on kind of the iOS for fertility or the operating system across a clinic or clinic network and using AI, AI to optimize things from billing to staffing, etc.
[00:08:03] Griffin Jones:
As the players become more fluid on either side, as there's more vertical integration, do you see the map changing? And that one bucket is the company's optimizing IVF itself, and there's others working on processing procedures. But as these players Start to overlap with each other and with what they provide.
Will we see this change over time? Or do you see these two buckets as a long term way of looking at this?
[00:08:30] Abigail Sirus:
I think it will change over time as we're early stage investors. So startups can, can. Pivot from time to time. But really, we're seeing a mixture of startups that are going after either point solution.
So we're really focused on making the best AI enabled salute or embryo assessment or those that are taking a more comprehensive approach. So might be doing The solution I just mentioned, while also looking across the clinic at ways to opt. So it really varies across the ecosystem in terms of whether people are focused on point or comprehensive solutions, but I think it's only going to continue to evolve.
It feels like there's been this kind of tidal wave of interest in AI ever since. Chat, GPT came to the fore and had, I think it was a hundred million users in the first few months, which far passes any product launch. But in reality, a lot of these companies that that we're mentioning in IVF have been working on it for a number of years, but it still feels early days.
[00:09:27] Griffin Jones:
The visual that you have is broken into these categories, and then as the offshoot, there's a, a blurred out section. Are you mapping companies that are in each of these different areas are working on each of these different areas? Right now? I know because you all are in venture capital, there's regulation about you not being able to talk about specific.
Companies. And so is that what's blurred out? Are you tracking who's doing who's involved in each of these sections?
[00:09:57] Abigail Sirus:
Exactly. So you can imagine that even behind this, I'm a data and spreadsheet person. So we have kind of our database that, that in those blurred section takes all of the 20 plus companies, which are what's represented as kind of the options that you were mentioning.
And we analyze them each across a number of dimensions. So one of them is. What I mentioned before, whether it's a point solution or whether it's comprehensive, another would be with something like A. I. Data is so critical to how we evaluate these. So it's understanding what are the data sources that each of these startups are using?
Are they proprietary? Is that data source going to grow? And we've been talking a lot, at least in the ecosystem about quantity of data. But from our perspective, it's really about quality. So it's about the signal that the data is releasing because IVF is is still growing. And in many ways, when you compare it to healthcare in general, it's a niche industry, even though we believe it's going to grow exponentially over the next coming years.
And so a lot of the companies that we talked to are using similar data sets. So what we try to understand is how are they differentiated because when you're building an AI algorithm and you're using all the same inputs, we're, we're curious to see how it shakes out in the coming years among these companies of how the outputs are really different and how that makes an impact on patient care.
[00:11:15] Griffin Jones:
How is it shaking out right now? Are you seeing meaningful differences? Are they all coming to the same conclusions and starting to build very similar solutions? Or are they coming up with radically different solutions using very similar data sets?
[00:11:33] Abigail Sirus:
I'd say it's still too early to tell in terms of which, which are going to be the winners of the pack and so on.
What we are seeing is kind of a convalescence or convergence around the same use cases, which are kind of laid out on that diagram that you mentioned before, embryo grading and selection. Hormone stimulation using AI, but what we are noticing as well is what's what's potentially differentiate companies or what clinics are they partnering with and going beyond that software or an AI powered algorithm is only as powerful as it can be actually applied in the clinic.
So what we're also looking for are companies that are focused on integrating their solution with Hardware and kind of bridging the gap between the digital and the physical worlds, because what we've seen is that companies who come into the space and might be really excited about this small part of the universe that they're innovating on, if you don't think more broadly about how it would actually be Impacting an REI's daily workflow or an embryologists and how you can make that part of their experience in a seamless way, solutions can have a hard, hard time taking off and being adopted when we're not thinking not only about the product itself, which is driven by AI or powered by AI, but also how it will be distributed and be made part of the IDF ecosystem as it is today, or the IDF ecosystem as it will be defined over the coming years.
[00:12:52] Griffin Jones:
Because hardware integration is so important, are you seeing multiple, in this blurred out section, are there multiple players that are appearing in different, in the different core categories, these eight boxes, as you have them laid out, or do you try to, do you try to identify which of the eight boxes they are most Best describes them and keep them in that singular category.
[00:13:19] Abigail Sirus:
It's a great question. And it's one that, that David and I talk about a lot because we like to be precise, although with things changing as quickly as they are, it's hard to do. So it really depends on the company and what we've learned and kind of. How far along they are. So yes, there are some companies that are tackling several of these areas that might be going after both oocyte selection and sperm selection.
And so they would be listed in that blurred out area that you mentioned before in both sections. But for others who might be in the early days of exploring one use case, but farther along, much farther along a different one, it would only list them in one section.
[00:13:55] Griffin Jones:
You mentioned that they use similar data sets very often.
Is that equally true if they're in very different categories? Meaning you might expect the folks that are working on O site assessment that they are using similar data sources, but are. Does it matter, are the people that are working on PGT, non invasive PGT, or the people that are working on clinic workflow, and the people that are working on sperm selection, is there still great overlap in data sources for them?
How much did data sources vary depending on if they're in entirely different categories?
[00:14:35] Abigail Sirus:
I think it's, Definitely different. If you're focused on oocyte selection, you're not going to be looking at the sperm data, unless maybe that's going to be part of an algorithm and your future roadmap. So in that case, what I would be talking about is, so in oocyte selection specifically, I'm looking at the map on my computer now, but that's not blurred out.
We have five plus companies just focused on oocyte selection. So they would be like using the same or similar data sources.
[00:15:03] Dr. David Sable:
One of the problems, Griffin, that we do see is that you see a difference in data sets, not so much the data being collected to make the decisions, but in some of the categorizations.
Something as simple as a diagnosis. One clinic's unexplained infertility will be another's ovulatory dysfunction. Depending on what the algorithms pull out, the connections they make, you hear that one of the great The beauty of AI is that these are unbiased mechanism agnostic algorithms. So if we use these things kind of differently from one clinic to the next, or maybe one operator within a clinic to the next operator, that's going to sort of infect the data.
And we're going to see these kind of artifactual differences. And since an IVF, we're limited by the fact that the data sets are pretty small, relatively. These are not big. When you talk about big data, when we're only doing a few million cycles worldwide every year and capturing only a tiny percentage of those into these data sets, we're handicapped from the start.
You start throwing some, I don't want to call it sloppy data, data collection, but let's just say inconsistent data collection. On top of it, and that makes our job that much more difficult, and it makes the algorithms have to work harder to avoid presenting things to us that are not really real.
[00:16:37] Griffin Jones:
Is this part of your advisory role for your portfolio companies?
Do you advise on the data sources that people are using? Oh, we, we've seen this before. We know that the data that you're showing us here is inconsistent and you might consider this other source.
[00:16:56] Dr. David Sable:
I think the, I think the data people, the data scientists know that going in, and they're, unfortunately, they're limited by who's going to, who's going to work with them, who's going to send their data to them.
Let's say, as you can imagine, and rationally, rightfully so, clinics are very protective of their data. So it's a, it's a bit of a trench warfare, trying to accrue enough things to feed into the systems to start building the training sets and building the algorithms themselves.
[00:17:27] Griffin Jones:
What would allow for more data sources to become available?
So if there's, if, if there's. So a lot of redundancy and overlap in different companies using similar data sets, what would allow for more data sets and more data sources to become available?
[00:17:48] Abigail Sirus: I think even before tackling that question, there's a level of kind of data uniformity or unification. So in a previous life, I was, uh, focused very much on data, uh, while I was working at IBM.
And we would build out a, you know, garbage, garbage out. And what we've heard in the industry is sometimes even using a popular EMR system for one clinic, they might use a specific data field or mapping that's, that's different than another clinic. And so there's this high level of customization across the industry today.
In terms of how they're just thinking about how they talk about data, and that's going to create challenges of bringing the disparate sources together in a way that makes sense and unifying them so that you could even be run an AI algorithm on top of them.
[00:18:39] Sponsor:
Seeing more patients may actually decrease IVF revenue by 30 to 40 percent.
Many REIs see patients that haven't had any testing and are not ready or suitable for treatment. Patients that don't need IVF often feel they're being sold treatment. And patients that do need IVF are choosing other centers because the new patient wait list is too long. What's worse is patients often feel unengaged by the clinic because they have little to nothing to do between scheduling their appointment and the visit.
LEVY Health uses precision diagnostics including lab testing, cycle data, symptoms, and medical history to triage fertility patients and prepare them for IVF. The first consult can consequently be used to plan and discuss treatment. Increasing conversion to IVF. LEVY Health's clinical decision support software for fertility and endocrine disorders handles patient prescreening, diagnostics, and triaging for your fertility clinic to save your team valuable time so you can treat more patients.
LEVY Health has estimates for improved clinic KPIs. Conversion to appointment, no shows, conversion to follow up, conversion to treatment, and IVF treatment start. Download these numbers for free at Levy.health/conversion. That's Levy.health/conversion.
[00:20:05] Griffin Jones:
I wanna ask, uh, you, you can't talk about particular players, but I do wanna go through each of these categories and give the audience an update on, on what's going on in development.
And I understand that in some things you're gonna be limited and I'll try to press. for specifics where I can, but tell us what's happening in the development in an O site assessment right now. What's going on with that grading and with regard to AI.
[00:20:32] Dr. David Sable:
In a large sense, the big challenge is demonstrating the value proposition.
The, uh, we, we know that the use of. Uh, advanced imaging systems and vast data handling can give us legitimate, uh, insights into which oocytes are more likely to turn into good blastocysts and good pregnancies. Uh, they can rationally rank the embryos as to which should go back First. Problem is how, when, you know, you and I and and Abigail have talked before about our criteria of dollars to baby time to baby and life disruption to baby.
And if we can do it little bit better by offering a point solution to put back an embryo in February instead of, instead of in March, so the pregnancy occurs quicker, how much is that worth in the grand scheme of things? By itself and how much better are we doing some of the business plans that we've got early in the AI journey were beautifully engineered solutions to answer a question and absolutely terrible business plans with the expectation that the price points that you could achieve.
Sometimes per cycle, sometimes hundreds of dollars per embryo assessed. We're just completely unrealistic in terms of getting the patient to where she wants to be. And we kind of, you know, then you may ask about advising the companies. Well, you sort of start advising them and say, Listen, you got to go back and put this into the grand scheme of What the patients themselves in the clinics are trying to solve for.
So we haven't really seen data telling us how much better we can do on specific, specific items. Embryo assessment, the same embryos are going to be sitting in the lab anyway. So it's a matter of when an embryo goes back versus other embryos that might not turn into a pregnant. Oocyte assessment is a wonderful, uh, science project.
I think it's, it's fabulous knowing which oocytes to use. But frankly, what we do now is we let nature tell us which oocytes to use because Some of them will fertilize, some of them won't. Some of the ones that fertilize will develop well, some of them won't. And then we've got other means of assessing the subsequent embryos.
So, up front, Determining which oocytes to use really is valuable, I guess, only if you have very, very limited numbers of sperm to use and you're triaging those sperm. Our work in AI for sperm selection is brilliant work being done right now. But it may be work that's being done looking at the wrong things.
For ever since we put sperm under a microscope. Hundreds of years ago, we've used this, the number of sperm, the way the sperm look, and the way they swim, as surrogate reasons for choosing one sperm versus another to, to fertilize the eggs. When we do, certainly when we're doing ICSI, in old fashioned IVF, we just kind of squirt the sperm into the dish.
Nature does that choice. Now, we don't know yet Whether the computers have been smart enough to find new things to look at versus what we look at. Because the important things are really the quality of the DNA and whether the DNA is normal within the sperm and the sperm's ability to kind of direct early regulation of first early cell division.
We have no way of measuring that. So while we have trained the algorithms and in some cases the algorithm and hardware combined to do things in a more systematic fashion. And some of the things that we as human beings just don't have the capacity to do, like we can look into a dish under a microscope of sperm and how many sperm can we track versus a brilliantly designed surveillance system with computers that can track all the sperm and then choose them based on certain criteria.
The algorithms haven't been trained back on the pregnancies yet that result from those tests because there just haven't been enough experiments done. Again, getting back to the limited amount of data that we have. So it's a very incremental process by which we can put in the, put the playbook into the various steps and try to extricate better chance of having a baby.
Now the stuff that we can judge earlier are some of the efficiency steps. We do something quicker. Can we do it? More error free. Can we free up people so to let machines do work that people do that takes longer or it's subject to fatigue? Or frankly, we just can't do as well as a well trained machine.
And those are things we're getting data back on. Now, I don't think we're in a position yet where we can judge company A does it better than company B, because these are all iterative processes and AI and IVF is such a new thing that it's exciting to watch things get better with the expectation that we're going to get some real, real decision making quality data. We're just not there yet.
[00:26:20] Griffin Jones:
The answer to the question of what's going on in these two categories, is the answer not enough?
[00:26:24] Dr. David Sable:
Well, they're, they're, they're in spring training.
[00:26:27] Griffin Jones:
Is that gonna be the, is that gonna be the answer for each of the categories? So before I go through each of the categories and, and just find out that everybody's in, in spring training, maybe instead of going through each category one by one, where would you say of, of these eight categories, oside assessment, embryo assessment, sperm selection, or hormonal stimulation, non-invasive, PGT.
clinical support and workflow, and then other, where would you see, where would you say we are, who's beyond spring training, or which of these categories is seeing players that are beyond spring training?
[00:27:03] Dr. David Sable:
Well, certainly clinical support and workflow. That's pretty straightforward. In embryo assessment, there is data showing there's validity to the ranking systems.
Again, we're challenged in proving that we have a viable value proposition, but certainly the rankings, some very good publications on the ability to rank embryos in ways that improve the efficiency of selection there. OSI selection, sperm selection, we have, a big part of that is, Doing the work is merging the software with the hardware so that you're getting reliable data so that the information they give you is based on the hardware and the software are shaking hands.
in a, uh, in a valid way and not giving you, it's not like, it's not like with chat GPT when you ask it a question and it makes mistakes and it makes things up. You know, the data we're getting now is not made up. It's really truly reflective of what the hardware is finding. Taking the next steps of plugging that into real decision making is going to be difficult.
The clinical data support, having the algorithms choose. What stimulation should be done having them make the decisions along the way for how the stimulation should be should be run That needs a lot more cycles to chew on and to have those cycles and have them collect connected to the outcomes So and there we're going to stratify the value in two ways The first will be that when it shows that we can relieve doctors from having to look at dozens or hundreds of data points per day for each cycle they're monitoring, and the computer can do it for 98 percent of them, just kicking the outliers out, we'll know that before we learn that the computers make better decisions than the doctors did.
So this can be a two step process there. So it's, yeah, I think that given that AI infertility was pretty much non existent, just a few years ago. We've made some terrific progress, but it's kind of like, like in biotech, what we say, we're still preclinical. We've got, we've identified molecules that can make a difference in the body.
Now we've got to stick them in the body and see what happens.
[00:29:20] Griffin Jones:
You mentioned that some of that is down the road. What's happening? What are these players that are in the support, the clinical support and workflow category right now doing that's being implemented right now that maybe wasn't even happening a year or so ago?
[00:29:36] Dr. David Sable:
There's, there's two areas. One is they've got to plug them into, into existing clinics and have them adopt them into their workflow. The other end of it, some of these companies are actually starting their own clinics. And they're running clinic prototype clinics with the AI systems in from the start as foundational elements.
And it's going to be really interesting to see in those two settings, what kind of difference we make in just the efficiency in which we can run an IVF program and those efficiencies will flow to both the clinic operators themselves later. do what they do cheaper, and hopefully to the patients themselves at a lower cost point, and in down the road, a faster time to getting pregnant.
[00:30:28] Griffin Jones:
Those that aren't starting their own prototype clinics and those that are still selling into clinics and being implemented by clinics, are you seeing a different rate in, uh, adoption than was happening a year or two ago? Have we passed a threshold where. They are starting to be implemented, or is that still the beginning of a mountain yet to be climbed where most clinics are not implementing these solutions?
[00:30:58] Abigail Sirus:
Yeah, and so when I used to do software development, we would describe it in three phases. There's proof of concept, when you're just getting started, testing things out. The next is pilot, when you're maybe working with a couple clinics or a few clinics or a handful at a time. Having them initially adopt the software, testing it out with them with maybe some real data, some simulated.
And then the third stage is production, when you're fully live, maybe across a handful, a number of clinics or clinic chain. And so I think for clinical decision support and workflow, we're seeing a mixture between still in proof of concepts phase, but also some that are doing pilots with clinics, with some live and simulated data.
I wouldn't say that any. Solutions that I've come across are quite production scale yet. It's still early days there, but I will say that what has been interesting for me to see is the difference. It's the difference in how incumbents, so existing clinics are integrating AI solutions and the new startups that are coming to the fore with kind of AI as part of their, their backbone or, um, their core foundation.
It's, it's kind of like with other platform shifts we've seen with. The Internet coming to the fore, for example, there was this general assumption that a lot of the advertising companies that already existed would just simply port everything that they already did onto the Internet or the World Wide Web and would continue to maintain their market leadership.
And then there were these new upstarts like Google's and others of the world at that time who were originally written off who came to the fore with the being on the Internet, and ended up being able to kind of come after the advertising industry and really flip it on its head. So, we're still in the early days of understanding how the adoption is going to be spread.
And these, these clinics are powerhouses for a reason. They are innovative and thinking towards the future. And they also control... pretty much all of the data that upstarts would need in order to have meaningful algorithms that actually make a difference in patient care. So it's going to be something that we're monitoring closely.
[00:33:08] Griffin Jones:
So a lot of the programs in the, the players in the clinic decision support and workflow category are still in pilot. Mode. What are they working on specifically? Is this that is using smart technology so that when supplies are low, they're automatically reordered. Is it that when a certain prognosis is given that, or a certain diagnosis or prognosis, it automatically schedules tasks, tell us about what specifically is happening.
[00:33:37] Dr. David Sable:
Yeah, it really depends on clinic to clinic. And one of the things we can't ignore too, is there's still the incredible amount of consolidation going on with larger and larger networks being formed and there they've got to homogenize their processes before they can, they can, before they can even think about adding something new in terms of the technology and what they're doing on the ground.
A lot of it depends on. what specific problem they're solving. Some of this, like doing order procurement or deciding which test to order relative to a single diagnosis, these are not exactly sophisticated decision, decision treats. Some of the things that we've been presented with are, yeah, I used to say it's kind of like making the Instructions, instructions to the babysitter.
It's like the baby wakes up and cries. If it's this time, you do one thing. It's not a heck of a lot of choices, not a lot of decision points. And applying, quote unquote, AI to it is, you know, kind of glorifying a little. And really, the benefits that we're going to see are in the much more complex decision making, where you just have a tremendous amount of data that's all being aggregated that needs to be looked at.
We've been making connections that we haven't been able to make yet. We've got 40 years of great human artificial intelligence based on the work that the embryologists and scientists have been doing in the lab before. They're just iterating and iterating and iterating slowly because that's what humans do.
If we're adding this extra layer on top of it, the greatest amount of benefit we're going to get are the toughest things to instill. And realistically, the more complex Problem you go after the more constituents you have to. Get behind it. When you run the lab, you've got to get the doctors behind, you've got to get the embryologists.
The scientific director needs to get each embryologist to sign off on it because, you know, you get one person in your workflow who wants to slow walk the implementation of a new system. And as, and where are we going to get the best information? We're going to get them from the largest clinics that do the same things, but they're also the ones that are consolidating most.
So it's a, you asked a perfectly good question. And here I am doing my best to dodge the answer because the reality of the IVF industry right now makes it a lot, it makes it kind of tough to get to that so what kind of thing where we say, oh yeah, we absolutely need this and we can define precisely what the benefit is.
So we know how much we should be paying for it. Once we get to there, we're going to see really rapid adoption. Now, you see the, a lot of the entrepreneurs, the founders will come to us and say they'll approach a clinic. The clinic will say, all right, well, you got to make it effortless to do it. We don't want to pay for it.
And we're going to give you nothing for what you're going to do. The, the founders themselves want to go to the clinics and say, all right, we're going to do this for you. Here's what we want to charge you for, and we want access to all of your data, so we can advance what we're going to do moving down the line.
Those are not easy negotiations to have. So in some cases, they're really left at the, all right, so let's make a little micro step along the way. But that microstep is not particularly clinically meaningful because they're being asked to optimize something that's easy to optimize. Frankly, this could have been done by systems that don't have the AI name on them, but are really just some combination of arithmetic or math or basic computing.
So it's a kind of a multi tiered answer, uh, a long winded non answer to your very good question.
[00:37:40] Griffin Jones:
Well, let's maybe get answers in a different category because even a general answer would be more than I've covered on the show before. I never have really delved into the category of hormonal stimulation management and the solutions that are coming in that category.
This is the fourth category that you have in your visual and so is Am I understanding it correctly by thinking of it, this is how AI is going to impact pharma and dosage and, and, and med protocols that talk to us about this category.
[00:38:17] Dr. David Sable:
Let's, let's go to the do it yourself IVF cycle and let's, let's fast forward to when every, every one of these systems works perfectly.
So there the patient. Does her own diagnostics because there's a list online of all the things you can order from Amazon You need a lot of testing done very easily If you need some type of invasive test it can be done the way a colonoscopy is done You just make an appointment in a place that does it you never get to know the doctor gets it's done So you line up all your basic testing?
You have, you've disaggregated stimulation from the big box, big tent IVF program, and there are OBGYNs that do it, or maybe freestanding IVF stimulators that do it, run by whatever combination of medical professionals. They take the information that you've put together from your checklist of pre IVF testing.
It gets fed into the computer. The computer says, all right, here's the optimal. stimulation regimen. Frankly, there's not that many regimens. We, again, we have a version of that written out on a yellow legal pad as instructions to the babysitter now. So that stimulation starts, the patient gets her medication from lots of different ways of sending medication to someone.
And she's monitored the monitoring. We hope to move to the home. Urine testing instead of blood testing, ultrasound only when it's needed. We do over, over scan people now and maybe we'll invent a really good cost effective home ultrasound, kind of like putting a pro, you know, patient. Places the probe herself saves hours of going to the clinic each time data that's collected the hormonal levels Whether it's from urine or saliva or whatever and the images from the probe go to the cloud the cloud sends them to a Processing system that in a big data AI way Uses those inputs to make the decision as to what the medication should be changed to or kept the same and when the next monitoring should be.
Now, on the ground, having done thousands and thousands of IVF cycles over the years, personally, and as a field, we've done millions of them, we know that most of these decisions are pretty routine, so that the computer will do maybe 98 percent of them, and kick out the 1 or 2 percent that are outliers, and that will go to the reproductive endocrinologist, who may be in a consulting role.
We've talked to you, I don't... Griffin, we talked earlier about moving the reproductive endocrinologists from doing a couple hundred cycles a year to overseeing thousands of cycles a year. This will be part of that. So that the AI system has chosen the stimulation, the AI system does the monitoring, and in conjunction with the overseeing RE, decides when the trigger for retrieval should be.
At that point, the AI system can take a break for a couple days. We go to retrieval. The oocytes are retrieved. The AI system is part of the microscopy. It talks to the microscope, sees the eggs as they come out. If there's a need to rank the eggs to be fertilized, because there's very, very few sperm, Or, if we get smarter about oocyte culture, and maybe different eggs need to be treated differently depending on things that the AI system may be able to see that we can't, the AI system will kick in there, and now the AI system is working in concert with the embryologists.
So it helps us choose the eggs to fertilize if there's minimal sperm, or stratifies the type of handling of the oocytes themselves. And then the fertilization will occur. This is where, hopefully, we'll be in a system where the AI system is really good at choosing the sperm that should be, maybe based on some type of marker that it sees that we as humans can't, that correlates really, really well with the genetics of the sperm.
Something we can't tell now, like we look at a sperm, the way it swims, the way it looks now, many sperm. It's like trying to figure out what's in the trunk of the car by looking at the license plate. So here we've got the AI system can look inside the trunk of the sperm and know what's inside. So it tells us which sperm to use and, and for which eggs then mechanically.
Let's say we're doing ICSI, there is an optimal angle that the needle should be at. There's an optimal speed that the needle should go through the zone of pellucida. There's an interplay between the elasticity of the shell of the egg. And the speed button, the speed and the sharpness of the angle needle itself.
This is a Toyota assembly line type optimization. May make a big difference or may not make any difference at all, but as AI gets smarter and smarter and smarter and smarter, it's going to turn ICSI from a procedure that maybe hurts a certain percentage of the eggs, maybe doesn't hurt to one that doesn't hurt any of them.
Then we go to the development of the embryos and the culturing. Right now, it's sort of a one size fits all type thing, where the embryos are treated all alike. They may be all in their own little wells with a probe inside, monitors the vital signs of the embryo. How much fuel is this embryo eating?
What's the pace by which the cells divide? Maybe we should hit the gas a little bit or hit the brake a little bit on the specific embryos themselves. And then ultimately we'll reach a time when we need to do the choosing. So there's tons of things that a really great hardware software hybrid that measures everything in ways that we as humans can't.
And over time, if we implement systems that are efficient enough and, very important, it's cheap enough to gather these data, then it's going to start telling us things that we had no idea we were doing wrong. All of which, hopefully, will result in being able to do the procedure cheaper and better, getting better yields at every step, higher fertilization percentages, higher number of blasts, higher numbers of percentage that develop well because we change what we do during culturing, and better decisions.
So that'll result in higher pregnancy rates and cheaper implementation of the cycle itself. A real virtuous process. Problem is, there's so many things that we could work to optimize, that it's just to figure out which ones make a difference first. So what we've been doing is we've been choosing the stuff that's easiest to do.
Like, okay, we got 12 embryos in the dish. Let's train the system on the embryos and start matching up which ones get pregnant based on what they look. And so the solutions that are being find now may not be the ones that important, but there's the ones that in this early stage, I hesitate to say this embryonic stage of AI infertility.
It's sort of that really, really early auto assembly line in the 1920s. Let's say, okay, there are some things that we can just do easily. It may not make our outcome that much better, but let's just start checking them off. So, uh, that's, that's sort of where, where we are in terms of the, you know, what's being looked at now and where it can go.
[00:46:24] Griffin Jones:
How do these changes, particularly those in hormonal stimulation management, impact the pharmaceutical, the, the pharmaceutical manufacturers, the, the Drug volumes other than just ordering more of them because it ostensibly if you have a I doing the monitoring and they are doing 98 percent of what the area I used to and they can scale that volume that there be an increase in the use of pharmaceuticals, but are there other input?
Other implications that these changes will have on the pharmaceutical side?
[00:47:02] Dr. David Sable:
Well, AI and drug development is a huge thing now. And we're trying to figure out what these same huge data crunchers, these mechanistic huge data crunchers, can tell drug developers about how molecules should be different. There may be a modification to the drug itself, or the drug's delivery itself.
Or, something that it picks up in the dynamics of when a dose changes, that can take that information, take it back to the drug developers, and they could do something different to their drugs to make them more effective. Or it may turn out that a combination of hormones that's been used rarely makes a big difference and We can package the drugs in a way that takes advantage of that.
So it's certainly, you're right, the most likely is great, cheaper, easier IVF, more drugs. Terrific for the pharmaceutical industry. But in so far as they're always looking for better versions of what they do or novel versions. All these data that we collect may make connections that just never occurred to us or never dawned on us.
We'll go back not to the way the cycle is managed, but to the way the drugs themselves are designed or manufactured, which would be enormous for the pharmaceutical companies.
[00:48:26] Griffin Jones:
Are there specific features that we might expect to see because it like other than press release around Esri time that we're not that close to oral FSH, is there features that we should expect to see?
We have a An article that will probably come out before well before this episode airs just about some things happening in the pipeline, but what's of note Abigail?
[00:48:52] Abigail Sirus:
Yeah, but maybe before we get to that topic, I just wanted to mention that there are there's a couple companies that are focused on the hormone stimulation and one.
Release paper last year that showed that they could potentially decrease the amount of drugs that were needed for a cycle. So you could maybe decrease the cost. And we know the average IVF cycle is expensive and out of reach for, for most patients today. So being able to decrease that cost could be a part of it.
And then it would be that kind of cost decrease, which would be obviously less sales or fewer sales for the Pharmaco, would hopefully be offset by more cycles being able to be done over time as the industry expands.
[00:49:30] Griffin Jones:
Are we starting to see any features that might be, that we might see in drug development in the next year or so, or are any of them close enough to call?
[00:49:41] Abigail Sirus:
No. No. I think there are some exciting developments happening in drug development and IVF in general, but I haven't come across data to suggest that they were driven by any kind of age.
[00:49:56] Griffin Jones:
Yeah. Okay. So let's, well, let's get into PGT because I have been in this field as a non clinician for nine years. And as a non clinician, it seems to me like the debate is still the same, Dr.
Gleicher's camp talking about PGT being overused and then other. Other folks saying that we might not be using PGT enough. Are we, is AI being used to break this stalemate yet? And if not, will and how it'd be.
[00:50:33] Dr. David Sable:
Two areas. One is. You're right. AI at PGT has done too much and it's done not enough because we really haven't figured out who we should be doing it for.
That is a great AI challenge and we need a ton of data for the AI to tell us, to answer that question for us. So that's going to be, that is an ongoing issue. The other is making PGT better and the obvious thing there is using a mass data Processing AI system to help us figure out just to what extent we can do non invasive genetic assessment and other means of embryo assessment.
There are other things we can do without biopsy. It's, it's, it's got some encouraging data sets, but they're way too small to be anywhere near conclusive. AI should be able to answer both of those questions once the once enough data has been fed in. And once the AI here, really, particularly in non-invasive assessment, it's gotta be able to look at things that we don't.
So here we're doing a lot of mixing and matching of the data handling capabilities with things like new visualization tech. All these systems were based on light microscopy. And some rudimentary staining. And now they're based on more sequencing technology. So which also has its limitations. So we've got sequencing and we've got visual visualization of embryo characteristics.
So we let the AI systems digest all of that and tell them to tell us the stuff that we've missed. We've been probably been pretty good. about optimizing within the context of the limitations of the systems we have now. Then the AI systems are And you didn't notice the connection between this and that.
And if we start throwing all that stuff in, that's where the AI system is smarter than we are. And it's going to turn around and say, Okay, you can get all the information you want, the problems that you want to solve in terms of detecting implantability, ability to turn into a good, a good term pregnancy and genetics and disease prevention in ways that we're just not smart enough to do yet.
So the AI can do all that, but again, you're going to hate me for saying this, but we're way too early.
[00:53:01] Griffin Jones:
For it to be clinically meaningful. I'm trying to salvage this with some, somebody that's, that's kicking butt. If I can, if I can think of an area where it might be happening in, in your other categories of your seven categories in other, you've got follicle, preterm birth, reproductive immunology, ovulation.
And is, is part of the reason that this is an other category is because that's where you have a lot of overlap. So in this category, you've got overlap with obstetrics and, and genetics and, and, and broader areas of women's health. And, and so are there things that are happening in this other category developments that are happening fast that we might expect to.
Be adopted in the fertility field fast because they are mature. They're in other areas and now they're starting to to take. on like wildfire in the fertility field. Is that happening at all in this other category? If not, tell me the damn reason why. And if, and if it is who, what's happening?
[00:54:02] Abigail Sirus:
Let me give you, first of all, it's, it's unfortunate about that.
Not that exciting. This other category in terms of why. And broken out separately. It's just that in this area, there are, um, typically just one or two companies working on each of these segments. So that's why I just kind of grouped them together just because they're not necessarily haven't reached the point where they need to be broken out on their own, like embryo assessment, which has the most.
On that topic. So that's not an exciting answer. However, what I will say is that there is a company working on, they're adding AI to a, that they're using a software system to look at follicle development, which is already being deployed in clinics. So I would actually put them at kind of a mature pilot stage.
So that is exciting. And they are, they are maybe farther along than I am. Uh, majority of other names on this, on this image specifically. So hopefully that's, that satiates you, Griffin, and we salvage it a little bit there.
[00:54:58] Griffin Jones:
What are they doing?
[00:55:01] Abigail Sirus:
They have a software solution that you could use while you're looking at follicles, obviously.
And they are using AI on top of it to help identify development and to make the process faster and more efficient along the way.
[00:55:16] Dr. David Sable:
Griffin, for the part two of your question, the why has been so few things that reached the kind of clinical so what stage and in the other areas, women's health is really difficult that way, both fertility and women's health in general.
It goes back to the information we use medically. It used to be all pattern recognition, analog, it's like it's syndromes and yeah, diseases were defined globally or by organ. Now we're at a molecular and cellular level. The more molecular and cellular you are, the more usable data you can, you can plug into systems that will look at them digitally because the data is much more homogeneous.
Women's health is back with psychiatry and neurology in areas that are really difficult. to get reproducible quantitative data. You can't just stick a probe into the uterus during early pregnancy. And figure out what's going on and measure lots of stuff. And that filters all through women's health. Even the diagnoses we have.
A lot of them have the word syndrome attached to them. Syndrome is almost like the Latin word for we have no idea what's really going on. Just a bunch of observations and we make logical things that we try to do. But we really don't know. So, that shows up in a lot of the stuff that we try to develop.
Like, and we've talked earlier about in IVF, there's only so many things we know how to measure. There's a few hormone levels, there's a number of eggs, number of sperm, percent fertilization, percent to blast, genetics of the embryo, and whether you get pregnant or not. Now, when you try to engineer the system, and you try to intervene at different points along the way, whether it's selecting an embryo, whether it's...
The angle by which your ixy needle enters the egg, which eggs that you choose to fertilize and when, the change that you make on the sixth day of stimulation, you try to figure out what difference does that make in getting pregnant down the line, each of these things gets drowned by what we call confounders.
All these other things that are happening in an uncontrolled way along the cycle and it makes observing, making meaningful observations very, very difficult. And the data scientists tell me, it's like, yeah, you just need a data set big enough. to plot all that noise. Problem is in IVF, the data set is maximally, if we had every cycle in the world being analyzed maximally, still only 3 million, which isn't a lot.
If it's something in pregnancy, and you're trying to figure out, well, what should we be doing in that 11th week of pregnancy to avoid the chances you're going to have hypertensive disease in pregnancy in the third trimester, you're going to deliver early or something of that sort. Well, you've got... 130 million pregnancies worldwide, but you've got nine months of observations to lose the validity of that one intervention.
And women's health is just, it's one of the reasons, in addition to women being kind of discriminated against and women not being put into clinical trials, it's one of the reasons that the amount of investment into women's health. has been relatively low because it's damn hard to do. So when you asked before, say, well, why, why haven't we gotten there?
It's, I have no doubt we will get there and we need. Technology to take us that next, next mile on the backs of all the really great human artificial intelligence that we put together. But that's why in 2023, when I first saw my first AI company in IVF in 2018, we're still in that kind of early, it's like, it's like when the genome was elucidated around the turn of the century and three or four years later.
Who cares? Like what's come out of it? Well, now it's incredible what's come out of it, but we're in that kind of foundation building stage, like spring training, if you will, where we haven't gotten to the so what test and maybe two years from now, five years from now, seven years from now, it's going to be dizzying, all the benefits we're getting from all this, it's kind of frustrating and I can see from your point of view, you want some headline stuff for us, we want to be able to give it to you, but it's kind of all, it's all inside baseball.
[00:59:45] Griffin Jones:
But it still is a fitting sequel, even though I prepared for the wrong sequel to our previous conversation. It still is a successful sequel because the last time when we were discussing the four guiding principles for democratizing IVF, I wanted to know why, what, what's blocking us from this inflection point that we're Almost at you mentioned you started investing in AI in this space in 2018 2023 and the next two to seven years are going to be dizzying, similar to how no one had heard of chat GPT.
And then that became dizzying all overnight is barely hyperbole in terms of its. Release and recognition to the public. And I, I, I can see that we are so close to that point. I needed to go in specifically into each of these categories and find out what's preventing us from being that at that inflection point that we're almost certainly going to be at very soon.
And we did that today in detail. And I want to give each of you the opportunity to conclude of what we might expect to see as we march toward this inflection point in the next year or so.
[01:00:54] Abigail Sirus:
I think that it feels public sentiment talking about AI is at a fever pitch. It's almost a subsection of every news site you look at now, but we are still only in the early days.
And I know that that's frustrating, but for me personally, thinking about the future of IVF enabled by AI, We have one in six people are struggling with infertility. Only 2 percent of that population is actually getting the care they need. We have a massive... Massive match between supply and demand. And it's only going to be technology like AI and bringing those into the clinic, optimizing existing processes, making it more efficient that we can close to serving the number of patients who are struggling with this.
And so I'm really excited to see more of these, these proof of concepts emerge as pilots and these pilots start to gain traction. And we start to see results that are actually making an impact, whether it's on the time to baby, the cost to baby. or on cycle outcomes as well. So still early days, but definitely lots to be excited.
[01:02:00] Dr. David Sable:
Griffin, what I'm most excited about is AI is going to catalyze the trying of new delivery methods, cutting up the cycle and disaggregating the cycle away from the big box, enormous lab, trying to find ways to pull in that 90 percent of people that aren't even in the arena yet that want to be, whether it's by Cost efficiencies or just setting up prototype programs that treat people for much less expensively and discover things that result in operational efficiencies.
I think it's going to be a little ways until we start seeing specific techniques that result in. higher pregnancy rates, only because pregnancy rate, by the time you get to the pregnancy rate, again, the data of teasing out the influence of one thing. But I think that it's going to show up first in the ability to deliver IVF cycles of one type or another to a lot more people than we do now.
And I think that's in a way that just as exciting as getting higher pregnancy rates, which will virtuously happen faster. The more people we get into these systems. So, I think let's look for operational efficiencies, let's look for people opening clinics, whether it's people outside the field or whether it's the big networks saying, look, there's other ways to do this and there's lots more people we can help.
Let's start going after that as well. I think it's a great facilitator for lots of areas of one degree of separation, uh, from the pure tech part innovation.
[01:03:39] Griffin Jones:
Abigail Sirus and David Sable, you're like the Star Wars or the Marvel franchises to my sequels in that we'll just keep making them forever and people will keep eating it up.
I look forward to having you both back on.
[01:03:51] Dr. David Sable:
Looking forward to coming back. Thank you.
[01:03:54] Sponsor:
This episode was brought to you by LEVY Health. Seeing more patients for a first consultation may actually decrease IVF revenue by 30 to 40%. To see why download the numbers for free at levy.health/conversion. That's levy.health/conversion.
Announcer:
Today's advertiser helped make the production and delivery of this episode possible for free to you, but the themes expressed by the guests do not necessarily reflect the views of inside reproductive health, nor of the advertiser. The advertiser does not have editorial control over the content of this episode and the guest's appearance is not an endorsement of the advertiser.
Thank you for listening to Inside Reproductive Health.