Welcome to Fertility and Sterility On Air, the podcast where you can stay current on the latest global research in the field of reproductive medicine. This podcast brings you an overview of this month's journal, in-depth discussions with authors, and other special features. F&S On Air is brought to you by the Fertility and Sterility family of journals, in conjunction with the American Society for Reproductive Medicine, and is hosted by Dr. Kurt Barnhart, Editor-in-Chief, Dr. Eve Feinberg, Editorial Editor, Dr. Micah Hill, Media Editor, Dr. Pietro Bortoletto, Interactive Associate-in-Chief, and Associate Editor, Dr. Kate Devine.
Welcome back to another episode of Fertility and Sterility On Air. We are in November 2026, volume 126, number 5. We're recording this just the week before ASRM, so by the time you hear this, we will have all seen you at ASRM, and I can't wait to see the three of you at ASRM. Good morning, Eve, how are you today? Good morning, Micah.
I'm excited to see everyone at ASRM. Good morning, Kate. Good morning, Kurt.
And it's a beautiful, crisp, fall Chicago day. Great to see everybody. We've got a lot to get through, so I'm going to go through the front matter pretty quickly.
We've got a great views and reviews from Editorial Editor, Anuja Dokres, talking about PCOS and the conversion of the name to PMOS. We briefly talked about that inkling in a prior month, but this is a full views and reviews that dives deeply into PMOS. I highly encourage you to read that.
We have a fertile battle from Editorial Editor, Michael Eisenberg, and are we ready for home sperm testing for all? And this is an interesting thing clinically as I've started to get more home sperm tests from guys, male in tests. I think sometimes they're helpful for are there sperm or they're not sperm, but sometimes not always helpful if I have to dive into clinical decisions like do I need ICSI, do I not? Do I need a microfluidics device, do I not? So it's a great debate on a fertile battle of sort of the state of the art and where we're at as at home testing continues to evolve and anything we can do to make point of care testing better for our patients is probably good. And finally, Editorial Editor, Dom DeZiegler, the time of men, can their involvement in the sphere revive live birth rates? Sort of looking at the global drop in live birth rate and how much of that is actual changes in fertility versus how much of that is social and what role can men play? So very timely article, thought provoking from Dom.
Kurt, before we dive into the content, the scientific content, we've got a acknowledgement of our reviewers of the year for fertility and sterility. Why is it important to talk about reviewers of the year? How do reviewers underpin the foundation of fertility and sterility? Well, I wish we could thank reviewers more often and more eloquently than we do, but every year we have to choose a couple of reviewers as reviewer of the years and it's my privilege to be able to do such. The reviewers are really the foundation of the journal.
There are thousands of reviews that are put together each year for this journal and they pick excellent papers which contributes to our growth, both in terms of our visibility and our impact score. So this year, we were able to choose based on the thoughts and the votes of all the associate editors and all the people that work for fertility and sterility. We chose Joshua Halpern and Heather Hipp as reviewers of the year for 25-26.
We asked them, and please read in the journal, why is it important to be a peer reviewer and how does it contribute? And I just want to highlight a couple of their comments. Joshua suggests that methodologic rigor comes first. To quote him, an ideal manuscript clearly explains what was done, how and why, while offering meaningful clinical impact and clearly writing tables and figures.
Couldn't have said it better myself. He also suggests that reviewing supports professional growth. It keeps physicians current, strengthens clinical knowledge, and lead to editorial appointments and other associate roles and editorial writing opportunities, which again is exactly correct.
Heather, who's been also a wonderful reviewer and for many years, suggests that a manuscript should fill a meaningful knowledge gap. The questions of method matter more than the ambitious study design. Studies in different populations and informative case reports can make a valuable contribution.
Sounds like I'm saying all of these things because I agree with everything that they're saying. It's really very, very true. And finally, she leaves us with the thought that reviewing provides fresh perspective and reciprocal learning.
Very, very nicely said. Reviewers improve manuscripts through an independent assessment while learning about study design, strengthening their own scholarship, and giving back to the field. Thank you, Heather, and thank you, Joshua, and thank you to all of those reviewers for telling us the reality.
And again, it's what makes this journal so wonderful. Well said. I loved reading both of those quotes from all of them.
They were fantastic answers, and it did sound like they were just saying in a very well said way things that you often say, Kurt. So that was great to hear reviewers thinking and feeling the same way that we do for the journal. Kurt, we're going to dive into the seminal contribution this year, which is looking at cumulative live birth over sequential euploid transfers.
Tell us what we learned this month from the seminal article. Yeah, this is a hot topic and probably will be for a long time. The article titled Cumulative Probability of Live Birth After Up to Six Sequential Single Euploid Embryo Transfers was the first author of Claudia Durbin and a senior author, Jamie Griffo, out of the team at New York University, as well as the medical school.
It's basically asking under the premise that patients have moved away from what's the chance that I'm going to get pregnant to how many embryos is it going to take for me to achieve a pregnancy? And therefore, we as a field have gotten to this concept of, well, how many do we need to give a cumulative chance that such will work? And there have been a number of papers that have suggested that three will give you a 90-plus cumulative pregnancy rate, three euploid embryos. Again, this is not a commercial for the use of PGTA and only transferring euploid embryos, but it gives you an idea of the prognosis with an embryo transfer. It also brings up not getting pregnant with a euploid embryo.
Somewhere along the line, there's got to be something wrong with me or the endometrium, and it gives this concept of recurrent implantation failure, which we'll dive into a second, and we have in papers in fertility and serality. But anyway, to get back to those topics, the authors hypothesize that the cumulative probability of at least one live birth would increase through six sequential embryo transfers, while the probability of success with each individual transfer would remain relatively stable. So that's the hypothesis of those that believe that it's really the embryo that's driving fertility success, and the other factors are perhaps more minimal.
So this is a single retrospective cohort study done at NYU covering 10 years, 2014 to 2024. It's patients that underwent a single euploid transfer and followed until their outcome, live birth or treatment discontinuation. The embryos could come from the same retrieval or additional retrievals.
It wasn't restricted to just one retrieval, and they looked at the primary outcome of live birth, but also the secondary outcomes we're used to looking at. They used Kaplan-Meier curves to estimate cumulative success, and they censored the patients when they stopped, meaning they didn't come back for a subsequent transfer or had no more embryos. Now, this assumes that patients who did stop would have had outcomes similar to those if they had continued, and I'll get back to that assumption and what that means as we go.
And it used mixed effects models, accounted for repeated transfers, and adjusted per transfer comparisons for age. So that's the technical aspect of it. So what they studied was almost 6,000 patients with 6,500 retrieved cycles and more than, well, actually 7,763 euploid embryos.
This was a relatively young group of people. The mean age was 36 at retrieval and 37 at first age of transfer. And approximately 88% of the embryos originated from one retrieval, but as I mentioned, they could combine embryos from other retrievals as well.
They did exclude patients with, as you might expect, uncorrected uterine abnormalities, congenital anomalies, donor oocytes, carriers, and non-euploid transfers. So let's go through the numbers, and I'll tell you, and then we can go back comparatively. So obviously, on the first transfer, they had more than 5,000 patients.
The second transfer, a little over about 1,500. The third, around 400. The fourth, around 120.
The fifth, down to 33 only, and the sixth, seven, and only seven patients. Now I'm telling you that just so you can understand the groups when I give you the success rates, that clearly the groups are getting smaller as we go and keep that in mind. So after the first transfer, the live birth rate was 63%, which is great, which means the cumulative live birth is 63% on the first transfer.
And on the second transfer, the live birth rate was apparently lower and statistically lowered around 54%. Now from the second, third, fourth, and fifth transfer, it didn't seem to change. The success rate was, again, 54%, 53%, 50%, and 57%.
And when you add those all up, you get a cumulative pregnancy rate after three transfers of 92%. Remember that number because compared to other studies, but 96% after four transfers, 98% after five transfers, and they get 98.8% after six transfers. So impressive data.
So the statistical answers is that the birth rate did significantly decrease between the first and second transfers, but showed no statistically detectable differences all the way through the fifth. Now that's impressive. And let's talk about that for a second, because I do think it's more than just a euploidy, that when you pick your best embryo, that obviously, and the first embryo makes sense to me that it gives you the first chance.
And then when you start going down into your pool, you might have embryos that are not as good quality, or there might be other factors. So that first finding didn't surprise me. But the one that's getting the publicity, so to speak, is that is the 98.8% cumulative pregnancy rate after six transfers.
So this is where I want to spend a little bit of time. So this is, it's not a modeled study. They didn't model data and model success rates.
This is actual data. But it did assume that as you would continue and had embryos would have a higher success rate. The actual number of people who actually did get pregnant after six transfers was only 81% because of the dropout.
So dropout and how you use that in numbers is going to be a problem. Now, let's go a little bit of just details, and then we can have a discussion about this. Yes, the patients who discontinued were older.
We need to understand that the patients that are dropping out are potentially lower prognosis. And yes, there is the issue of, did you change the treatment cycle? And they do comment that at least a quarter of the patients would change their transfer protocol along the way. And I don't know how much we could speculate how much that makes a difference.
They also want to comment a little bit about this idea of recurrent implantation failure. There's a paper in Fertility and Sterility a year or two ago that suggests that the definition should be at least three euploid embryos before you call somebody of having recurrent implantation failure. That's hard in our field because people are always asking, why didn't I get pregnant after the first or the second? And it's hard to say, do the same thing all the way up to the third.
But from a research perspective, we really do need a codified definition that everybody uses. The question is, are we making up a definition by artificial statistics or is this a real definition? So this paper tested that. If you claimed that recurrent implantation failure was three failures after three euploid embryos, they did have 184 patients that didn't achieve a pregnancy in that period of time.
And that was 3%. However, when you then move forward and go forward, another 80 of those patients or 43% of those that had quote unquote recurrent implantation failure got pregnant in the fourth attempt and another 59 got pregnant on the transfer after that. So they're basically commenting that we could still use an arbitrary cutoff of three that captures somewhere between three and 5% of the population as a definition, but recognizing that's a statistical definition.
But they're also giving a lot of evidence that doesn't mean you should stop or change things because there's a good percentage of those patients that are getting pregnant on the fourth and the fifth and even some on the sixth. So how robust is this evidence? The evidence is reasonably convincing that another euploid embryo transfer can succeed after previous failures. But I would be careful of the precise claim of 99% success rate after six.
That's where you have to be a little careful. So if you're reading or hearing people give the high level summary of this paper, that's where the cautionary tale is. So the principal concerns are, again, from having me geek out a little bit, is that you have something called informative dropout, that the people dropping out are of worse prognosis of the group.
They're older. You're getting down to embryos that are lower quality. You're getting down to people that don't have any embryos left and have to go another cycle.
And it's an important point. This is saying that if you had six embryos, you might get up to 99% pregnancy, but it doesn't say the probability that you can get six embryos. So some people are never going to get six euploid embryos, and they're going to look at this and consider themselves a lower prognosis or a failure.
The other obvious statistical issue is that you're talking about very small numbers in the latter transfers, only 33 patients in the fifth transfer, and only seven patients in the sixth transfer. Again, statistical significance doesn't demonstrate equivalent success rates, so we can't say that all of the embryo transfers have the same success rate after the second one, even though they do appear to look pretty good. So with that in mind, let's dig into the literature a little bit.
So our own Paul Pertea has what we often call a landmark paper where he suggested that the success rate was 92.6% after three transfers with a sustained inflation of 95. So this closely matches this Durbin study at 92. So I think that that reasonable replicates what very good large programs are getting.
There's also a paper at Yale in 2024 that estimates live birth was 98.1 after five transfers. So again, we're replicating some of the data, although I'm the first to admit that people tend to publish their best work, and this is not going to be replicated at all centers. And there's also an Al-Mohadadi paper in 2024 that estimates cumulative live birth was only 68% after three transfers, again showing not all programs are the same, and not everyone has the same success rate.
So this is a nice study. It's a large cohort. It's got successful follow-up.
It's got reported dropouts and conservative sensitivity analysis to make the uncertainty visible in the paper. You can read into the paper really closely if you'd like to, but it is a retrospective cohort at one study. The laboratory has changed over time.
There's very few people out in the tail end of this, so we have to be careful. So the commentary, which was terrific and that you should also read by Alexandra Herwick and Leah Bernardi, I have to give you the title because it's also cute, Chasing Certainty, Quantifying Hope, Cumulative Outcomes After Six Euclid Embryos, their message is clear. The reassurance from this paper is very valuable, but approximating 99% success rates should not be a promise or a target.
It's just not going to happen in all programs. So the main point of their commentary is that another transfer can offer meaningful hope, and I agree with that, but the dropout really does change your interpretation. Now, persistence is required for this kind of discussion with your patients.
It does give you some good ammunition to say, hang in there, everything else looks good. It's not telling you you should change nothing or that you shouldn't re-evaluate your patients, which we all do at all steps to find out what we could do differently or what could be improved. But it does, in my mind, encourage persistence, which is a good thing.
Certainly, the age and embryo availability must be guided by counsel. As I mentioned, not everybody's going to be able to get to six euclid embryos, let alone transfer all six and have the persistence to do that, and it shouldn't become a quote-unquote target that you need to bank up to six embryos before we start transferring. That also, I think, would be a mistake.
So the main point of this paper is reassurance that you can continue with euclid embryo transfers, but because of the dropout and over-interpretation, be careful of the headline. One qualification, again, as I mentioned, is that 81% of people that achieved a pregnancy is different from the 99% that they actually estimate could get a pregnancy at 99%. That's a big difference, and that's because of real-world patients and their journey.
It's not necessarily just the laboratory. So the combined message here, I think, is another euclid embryo can retain a substantial chance of success after previous failures, but achieving this opportunity in itself is a bit of an uncertainty. These data support reassurance and individual decisions.
It doesn't guarantee success, and it doesn't require the banking of six embryos. So I still think this is an excellent paper that is a good seminal contribution because of the discussion that's going to evoke. I hope that was a clear presentation, but I'd really love to hear your thoughts on this paper as well.
It's not necessarily new, but it really does confirm what we have, and that kind of confirmatory knowledge can be very powerful. Eve, Kate, Micah, what do you think? Yeah, I thought it was great. I caution with the, you know, quote, dropout, how much of that is patient dropout due to psychological distress or other factors.
How much of that is physicians saying, your lining is really thin. I don't want to do another transfer. I think you need to move forward with a gestational carrier.
And I worry a little bit that if this paper is misinterpreted, that in clinical settings that weren't moving on to alternative strategies that patients may push and push and say, but this paper said that if I do six transfers, I have a 99% chance of achieving pregnancy. And I think that that's a good message, but I think it's the wrong message. I think you have to look at that pessimistic contribution of those patients who didn't complete six transfers, that what I would call a pessimistic contribution that those patients who didn't complete it didn't get pregnant.
And that's, you highlighted it, and it's buried a little bit in the paper. I would have really liked to see the Kaplan-Meier curve show not just the top end, the optimistic prediction, but the pessimistic prediction of that final 82% of all patients. Because I think that there is a point in time where you look at a patient where it's not working, their lining's thin, they have Asherman, they have other morbidities, and you encourage them to move to an alternative strategy.
And so I don't take this paper to say, I'm quitting too soon. I should be doing six transfers before moving on to a gestational carrier. And I'm curious if you guys look at this the same way.
Yeah, I do look at it the same way. That said, this is something that all of our patients think about all the time. And it really is a very meaningful contribution just in terms of the size of the data set required to even have a real crack at answering this question well.
I think one way to look at it is, what cohort of patients does this analysis apply to? And so it applies to patients that can get a lot of embryos, right? And so we don't want to be indiscriminately applying rainbows and bunnies hopping through fields to every clinical scenario based on this paper. We want to be thinking critically about how can we paint what is a realistic but useful picture and appropriate optimism to guide our patients in their decision making. So I love this paper.
It's a question that's near and dear to me. We have been working on a very similar analysis. And I'm a little bummed that they beat us to it, actually.
But I extend my heartfelt congrats. I think it helps a lot. And I think the reflection, it also helps a lot, because it's really important that clinicians don't overextend the findings.
I have a quick diagnosis question. When you're defining a disease or a syndrome or an entity, and in this case, we're using a percentile. And the ASRM practice committee came out and said it's the 95th percentile, whatever that is for your embryos.
Because some people globally use cleavage. Some use untested blastocysts. Some use euploid embryos, which is what this study did.
It sort of bothered me that we were coming down on just a statistical construct for a definition. But the more I've thought about it, I don't know how we get away from that. I mean, that's how thyroid disease is defined.
It's how other diseases are defined. Dr. Justin Marchegiani How infertility is defined. Dr. Tim Jackson Yeah, exactly.
How infertility is defined. And if we do find a new test that we can use for recurrent implantation failure, it's gonna move into the pre-IVF diagnosis, right? So let's say the ERA worked, a new ERA works. Then it's gonna become pre, like pre-done, but for everyone before you do IVF.
And so it's no longer gonna be in the RIF just like we do saline ultrasounds to rule out other uterine factors before anyone goes to IVF. So do you think we're ever gonna get away from a statistical construct to define implantation failure, recurrent implantation failure? I guess that's my question for you. Dr. Justin Marchegiani This is our question of our generation.
You know, it goes back to is it the embryo or the uterus? I think that this kind of paper suggests that's the uterus that's driving the boat, but it never drives the boat fully. There are other things to worry about as well. But so until we really understand the implantation better, molecularly, scientifically, I think we have to go with what we know, which is that, you know, if you keep putting back a good embryo in most majority of patients, you're gonna be able to achieve a pregnancy.
But it doesn't take away from the optimism of this paper. I just wanna end this conversation with again, this is a really optimistic scenario. 36-year-old patients that make a ton of embryos.
You can't apply this to everybody. I would love to have this many patients that are 35 and 36-year-old that are going through PGTA. I might argue they might not even need PGTA.
But the point is, this is the best circumstances. And when we learn more about implantation, we do–obviously, it's got something to do with this. It's just a question of how we define it and how we quantitate it.
And we just don't know the answers yet. Wonderful paper, great seminal contribution. And thank you for leading that discussion.
I thought about new things as you led that, Kurt. That was great. Kate, we've got a couple papers.
If we have a theme this month, artificial intelligence would be one of the themes. And we're starting with you on looking at AI for embryo ranking. I know this is something you've thought a lot about in your research role and medical officer role.
Yeah. Thanks, Micah. This paper comes from Danilo Cimadomo and colleagues at IVRMA in Rome and is entitled Artificial Intelligence-Based Embryo Banking Can Match or Improve Traditional Assessments but Cannot Predict an Aneuploid Chromosomal Constitution.
So we'll get back to that title a little bit later. But the authors ask whether AI can put a euploid blastocyst first when a patient has several embryos. So getting back to these patients with several embryos, also a theme for today.
Most AI validation studies assess euploidy prediction embryo by embryo. But the clinical task is relative is what these authors argue. They're stating that really what we're asked to do when we're selecting an embryo is to rank sibling embryos within one patient's cohort.
So they tried to design a study that would answer whether AI can do that. If found to be true would be that maybe PGTA one day might be avoidable. So this is a retrospective single center study starting out with 786 PGTA cycles that had at least one available blastocyst.
So that actual starting point in their study flow diagram is actually a little bit arbitrary. Why do they start with only the cycles that had one PGTA euploid as the top and not their actual group that they analyzed? It's not a central point to the question. They excluded patients with fewer than three biopsied embryos, leaving 363 cycles.
And then they also excluded cycles where every embryo was euploid or every embryo was aneuploid. So that left only cycles that had at least three tested blastocysts of which at least one was euploid and at least one was aneuploid. And that left 35% of what they considered their top line, again, which was already kind of a couple of attrition steps in, at 279 cycles.
And those cycles contributed 1,260 blastocysts. And embryologists graded the embryos blinded to PGTA results, of course, as we always do. They compared five different ways of picking the top embryo.
So static Gardner grading, which some of us might consider to be traditional assessment. Gardner grading plus morphodynamics. So they looked at embryoscope data and added to Gardner grading timing of blastocyst expansion and expanded blastocyst area.
And then the other methods that they included were two versions of a commercial AI score. So this is IDA score from embryoscope. And they included version one and version two and compared those ranking systems along with the others I've mentioned, as well as random selection as a chance baseline.
So again, the outcome was how often an embryo, a euploid embryo, was ranked first. The results showed that a euploid embryo ranked first in 58% of cohorts with static grading, 64% with morphodynamics added to static grading, 62% with AIV1, and 68% with AIV2. And random selection gave 44%.
So not surprisingly, more or less correlated with the percentage of blastocysts that were euploid in the cohort. Only AIV2 was significantly better than static grading in their analysis with a p-value of 0.022. And they reported that it was not better, though, than static grading plus morphodynamics. There was no statistically significant difference there.
Thankfully to me, random selection was statistically worse than static grading. The supplement stratifies by maternal age groups that with approximate SART age categories, not exactly the same, but almost, they found that the probability of choosing euploid fell with age as expected for nearly every method, including random selection. And of course, that's because euploid fraction falls with age.
Only AIV2 showed no significant decline in this outcome with increasing age. So all the other models got worse at choosing an embryo that would be euploid as the age of the cohort increased. For patients under 35, static grading and morphodynamics numerically outperformed both AI versions.
In the 40 to 41 year-old cohort, both AI versions were ahead of embryologists. And at 42, AIV2 was the highest, but that group had only 18 cycles. And results in the middle cohort, 35 to 39, were similar across all the five methods, really four methods in randomness.
And so I think appropriately, no between-method tests were shown within the age strata since the strata sample sizes were small. And then a second analysis was done. So the authors calculated coefficient of variation for each of these methods of ranking embryos.
And that was calculated as the standard deviation of the scores, the embryo scores within the cohort, rather, divided by the mean within each cohort as a measure of how different the siblings within a cohort looked from one another in the view of that ranking method. The median CV, so that's the coefficient of variation, was higher for embryologists than the AI models. And then they looked at, in an age-adjusted model, whether the degree of difference or coefficient of variation was associated with a higher likelihood that embryologists would rank an embryo first.
And they found that that was the case for the embryologists, that the more different the cohorts looked from one another, the more likely the model was to select an embryo that was euploid. But they did not find a difference for either of the AI versions. And so the take-home from this paper is that even the best method of embryo assessment, if we're assessing it as defined by these authors, the probability of choosing a euploid, was at least 32% likely to choose an aneuploid embryo.
And so the authors conclude from this that ranking tools beat chance, but they can't replicate or replace PGTA. So this is a clear, actually refreshingly clear for an AI paper, in my opinion, and clinically relevant evaluation. And the authors deserve credit for reporting what is essentially a negative result on replacing PGTA.
Again, I think a credit to this group that does work on AI modalities to say, you know, it really can't do this well. I do have some misgivings about this paper, though. So in terms of the primary analysis, because every method ranked the same cohorts, confounding between models isn't a concern.
So they didn't need to do any modeling. It's a very straightforward analysis. That said, the five methods imply, and they report them, most of them, up to 10 pairwise comparisons, and the paper doesn't mention any multiplicity correction or say whether paired tests were used.
So a p-value of 0.022 for the version two, which was the only statistically significant difference, and that was versus static embryology assessment, it wouldn't survive a Bonferroni correction of even four comparisons. So I kind of question that claim that V2 outperformed the embryologist. I'm not so sure that this analysis can say that.
I certainly am interested to hear what Micah and Kurt and Eve have to say about that, though. And so in terms of the coefficient of variation analysis, they appropriately adjusted for age, but adjusting for age alone, I think, can leave some residual confounding. So if a cohort has a few clearly poor embryos, the CV is going to rise, and the best embryo looks easier to pick.
So I'd want a model that includes not just age, but also includes the euploid fraction and the total number of blastocysts, maybe some other variables as well. More fundamentally, I just struggle to see the clinical relevance of the CV analysis. So I also worry a little bit about generalizability.
This group in the center, GORD embryos all the way down to CC grade A7s for the ICM and trophectoderm, and many labs wouldn't biopsy these. So it would be interesting to see these results in a system that biopsies only fair and good grade embryos. And now getting back to that title.
So again, this was AI-based embryo ranking can match or improve traditional assessments, but can't predict an aneuploid chromosomal constitution. So we already covered why superiority over embryologists isn't clearly shown. So I think the first clause is an overstatement.
But in terms of the second clause, the paper also reports no area under the curve or other predictive accuracy metrics. So it can't really conclude that AI isn't predictive. And the authors even cite data to the contrary, including a meta-analysis with a pooled AUC of 0.8 of AI tools for euploidy prediction.
So then finally, and somewhat philosophically, as we're framing success as picking a euploid embryo, it kind of also raises the central question of what PGTA is for. And I'll put a plug in also for a conversation we're going to be having at ASRM. I hope you'll all come and join us about what the utility of PGTA and validation of PGTA platforms.
But from a high level, reasonable people can disagree about whether the main goal of PGTA should be to deselect embryos that are unlikely to succeed and lead potentially to abnormalities, or should the goal of PGTA be also sort of treated something like a grading system for finding the embryo that is most likely to succeed. So given how contested that question is, I would really hate to see the future of AI in our field defined by the ability of AI to predict PGTA. My question for this group is, if euploidy is an imperfect target, what endpoints should embryo ranking be judged on? And then again, what in general did you guys think about this paper? Exactly to the heart of my biggest question I was going to ask back to this group at the end is, why is PGTA the standard euploid that a test should be having for AI? Shouldn't it be live birth that patients come to us for? Unless you're saying it's embryo banking, and we're getting back to the discussion that Kurt kind of got to for family building and how many embryos do I need to bank to build a family size, then I could see that being of value.
But I think what patients care about is a baby. What's my chance of a baby from this embryo? Yeah, and I feel like I say this a lot on the podcast, but we know that in nature, abnormal embryos implant and abnormal embryos make it to live birth. So the idea, the biological plausibility that you can detect aneuploidy based on physical characteristics, like I really struggle with that.
Where I think that AI might help is if you have three euploid embryos, can AI help to select the best of those euploids for the first embryo transfer? But again, we all know that PGTA doesn't change the ultimate probability of live birth. And we reviewed some data last month that showed that it doesn't even improve the time to live birth, especially in donor egg cycles. So I think at the heart of this question is, what are we trying to study? And I feel like, Mikey, you say this again and again, we have been looking at the use of AI for embryo selection for over a decade, and it seems to be a technique really without an indication.
What are we using it for? And are we really getting any closer to embryo selection through AI than we were a decade ago? And I'm just not that convinced. I think that this is complicated by a couple of factors that we don't have time to get into, but would make wonderful discussions and you all listening should think about it. One concept is, what is AI? In other words, is it embryo development with an embryoscope? Is it a static image? Is it just replicating what we think is good, which is embryology? And what is PGTA? And again, is it really a definitive test or not? And the final concept is, once you work out that specificity, is we have trouble with the concept of prediction.
Prediction doesn't mean you're right 100% of the time. So the fact that AI might predict an embryo, predict as the best embryo, one that's aneuploid, shouldn't surprise anybody. Our goal shouldn't be never make an error and never pick an embryo that doesn't result in a live birth.
So there's a lot of concepts going around that we haven't really defined very well to clarify this muddy area. Obviously, the goal is to pick the best embryo that's going to result in a live birth with the least expense, invasiveness, and difficulty. The question is, do we have those techniques yet? And I would argue we're still working on it.
Good discussion. Kate, we have one more AI article. It's at the end of our table of contents, but do you want to stick with the theme and cover that now? Sounds good, Micah.
So, yeah, I was really excited to get to put my AI hat on and leave it on in reviewing these articles for this November issue. So, again, keeping my AI hat on. Next, I'll review a research letter from the Tel Aviv Swarovski Group.
This paper had first author Rosa Berkovitz-Sperling, and it was entitled Artificial Intelligence-Driven Trigger Timing in Planned Oocyte Cryopreservation, Learning versus Non-Learning Models. It appears alongside another really good reflections piece, this one from Tanya Carreon and Marcos Messiger, and I'll draw on their really excellent discussion as well. So, patients increasingly are asking chatbots health questions, and so the authors compared a general chatbot-style learning model with a model that was trained on the clinic's own data, and they used the physician's actual day of trigger as a reference.
So, the model is looking at whether, or rather the paper is looking at whether two different models can accurately track the physician's selection of trigger day. It's a retrospective cohort of 373 first elective fertility preservation cycles in women 31 to 41 with normal ovarian reserve, so it's a good prognosis group. They looked at two GPT-5-based models that received identical inputs, and they deemed them the non-learning model, which they described as reasoning from physiology alone.
So, another way of thinking about this is just general knowledge. So, it looked at the cycle, and it looked at what is known and would make a prediction as to when to trigger, and then also a learning model, which combined the GPT-5, so what I just described, with a random forest algorithm trained on prior physician decisions, so not outcomes. So, again, the models were trained on physician decisions, and then more trying to predict physician decisions.
The primary outcome, again, was agreement with the physician-selected trigger day. So, the learning model, in terms of results, agreed within two days, so that was their primary outcome, percent agreement within two days, in 89.3 percent of cycles, whereas the non-learning model did worse and agreed with the physician in about 64.9 percent of cycles. When they narrowed the window of agreement, it was 71.6 percent agreement within…rather, for the learning model versus 36.7 for the non-learning model.
The non-learning model, interestingly, also recommended earlier trigger in 82.5 percent of cycles, with 1.6 days earlier recommended relative to the physician, and the learning model did not have a bidirectional bias. They also commented that the learning model achieved 84 percent agreement with the physician after only 50 training cycles, so it was a fast learner. The letter refers throughout to, quote, physician override, meaning, you know, leading us to believe, more or less, that when the model disagreed, the physician could decide whether it was going to take its advice or not.
But actually, these models were run retrospectively, so every trigger analyzed, because they secondarily look at the outcomes of these retrievals, was actually selected by a live physician. So, in the 43 cycles where the physician and the learning models differed by more than two days, the unadjusted yield was higher than in the cycles when the model and the physician agreed. So, you know, it's a little bit of a complex extrapolation or deduction, but essentially suggesting that the physician might be better than the models.
In the 43 cycles where the physician and the learning model differed by more than two days, the unadjusted yield was higher when they did not agree, right, than in the agreement cycles. So, with total eggs having a median of 19 for disagreement cycles versus 15 with agreement cycles, and then for mature eggs, 16 for non-agreement cycles versus 12 for agreement cycles. And they provide supplementary data that showed lower agreement for both models, learning and non-learning, in the 78 cycles with four or more monitoring visits, which may reflect complexity of these cases and, again, extrapolating further, may reflect that a physician's judgment is needed when cases are more complex.
So, the reflections piece really makes several good points. A couple of points are that learning clinical practice is not the same as learning the optimal decision, and a high concordance between the model and the physician just essentially suggests that the model can reproduce local behavior, but not that the decisions maximize yield or even safety. So, whether, you know, the chatbot triggering early, for example, by, you know, one and a half days would have reduced the yield is plausible, but this paper does not test that.
The cohort is normal reserve patients, freezing eggs. So, these results shouldn't be extended to poor responders, PMOS, conventional IVF patients, really any infertile patients. They also note astutely that AMH was a model input and an inclusion criterion, but it was only available for a subset of the patients, and it's unclear how the authors managed this missingness.
And this is, to me, one of the most important methodologic things to consider is that the authors don't at all describe, and of course, it's a research letter, so maybe just not enough room, but how the training and evaluation sets were split, really important for evaluating validation. And so, they have, again, astutely note that the next step should be external validation and prospective studies looking at how this model would perform in different datasets and populations. And they also suggest that disagreement or model uncertainty, another really important point, could flag cycles for review.
And so, I'd add a couple additional thoughts, which is that I would have really liked to see them build out the comparison to also include a no random forest-only arm, so basically a simple rule-based algorithm and also a random forest-only arm. So, those, I think, would just provide two additional useful controls. You know, the learning model reached 84 percent agreement with 50 cycles, and so a simple rule may actually capture much of that signal.
Another point is that all but one cycle in the entire dataset of 373 cycles were antagonist or PPP, or progestin-priming stimulation protocols, as we so lovingly call it, and 99 percent had a GnRH agonist trigger. So, really can't generalize these data to agonist protocols or HCG triggers. So, to me, the most interesting thing that would be missing from this analysis, though, is the characterization of the patients with discordance, which, again, the reflection touches upon, because if we're going to be integrating AI tools into care, patient selection will be key, right? And so, understanding which patients are not well characterized or not well served by relying upon models, AI models, is going to be really helpful if we're going to try to actually implement these tools in practice.
So, Micah, Eve, Kurt, how are you using such tools to manage stimulations? How do you think they should be used? You know, what would convince you that we should be using them? I'd like to see a tool, not that mimics what I do, but I'd like to see a tool that improves outcomes, and we're not there yet. I think we're inching closer towards it, but I think, you know, basically, different physicians think about this differently, and I think when you are using one data set to train your model, you're training your model in the way that your clinic thinks about trigger decisions. And so, I think having a larger data set, having it be more robust, having it be from multiple centers, and I think part of the challenge is that none of us can really, I think everyone thinks they do it the best way, but what is the real best timing for trigger? I would love AI to help us to better determine that, not to be trained to mimic what we're already doing.
And so, I sort of envision it like a little pop-up box. We're not yet using AI decision tools in our practice, but I envision it as a pop-up box that says, like, optimal timing for trigger detected, and then the physician can then interpret, like, do I agree with this? Do I want to override that decision? And using it in real time, I think, would be effective. But I think in order to get there, we have to have better trained models that will optimize what we do and not simply mimic it.
Dr. Michael S.: Eve, I agree with that, and I've been trying to eloquently say that for a long time. I think AI is just learning what we do and making the ceiling and the floor tighter, making everything more replicated so human error is minimized. But I don't think it's telling us that we're doing anything.
It's not telling us that we could be doing things better. It's not telling us that really you should be doing something else because all it can compare to is what we're doing. The other potential application is when centers are trying to just level out their number of retrievals.
So, will it ever make it better for getting the number of eggs? I don't know. Maybe, maybe not. I'm a little skeptical about that.
To me, the reassuring thing about both these papers, if you're a doctor or an embryologist, neither of these papers tell me that you're out of a job anytime soon. Like, AI is not coming to replace you at best. They're somewhere getting close to what you do.
But, you know, whether your limit for cycles is retrievals in a day is six or 20 or 30, we see spikes in that. And so, if maybe there's a tool that can help you say these are the three patients that maybe could trigger a day early or trigger a day later and push a day, that might just help you get through that and make life better for your embryologist. I think that's at least a clinical utility goal for our embryology labs.
I know we've been trying to do that and I personally am not convinced we're there with the tool we've tried in our clinic. Kate may have a different view than what I do. But I think that's the second goal beyond just will it be better than us at predicting when to trigger.
Yeah, I totally agree with that. It's kind of the holy grail. If we're limited by predicting just what we do anyway, based on the data sets that we're feeding, that we're allowing it to learn from, having it help us logistically would actually be a big win.
But there are challenges there as well. I mean, it has to be able to take the data up lightning fast to be useful. And frankly, the technology is still pretty limiting there.
So, you know, hopefully with the real, you know, massive shifts we've seen in the quality of AI models recently, we'll get there soon. Great, great discussions in AI. There are some very big podcasters and non-physician, non-embryologist voice in our field that have been asking physicians on podcasts, what are you going to do when AI does all your work for you? And asking embryologists, what are you going to do when AI does all your work for you? I will just say the more and more I read about this and stay up on the technology, it's barely catching up to being equivalent to what we do.
And some of that's based upon as being trained on what we do. But even once we get past that, it's not going to be close to the ability to make clinical decisions. So I agree with the people that say AI is not going to replace your job, but the physicians that know how to use AI might.
The embryologists that know how to use AI might. So the lesson to the young people out there, learn what AI is doing, become the masters of it, because then you're going to be driving our field in the future. All right.
We are going to move on to assisted reproduction. I have an article titled Effective Blast Assist Stage versus Cleavage Stage Strategy on Cumulative, that's the key, Cumulative Live Birth and ART, Systematic Review and Meta-analysis. Senior Arthur Busnelli and Levi Setti.
This is from Italy, France and Spain. The title says exactly what it is. Cleavage versus Blast, Cumulative Live Birth, Meta-analysis, Systematic Review.
They included both randomized clinical trials of which they had four and some observational cohort studies. So I'm going to go through most of the results pretty quick because I think they're intuitive. But for our listeners, I just want you to think at this point, why do you use if you use blast transfer? What is the purpose? Is it to improve cumulative live birth or is it as a selection tool to get to maybe different embryos at a different time in the embryo transfer order? So I want to spend just a brief second on the methods because this is interesting, especially for our young learners, our fellows.
They used the I-squared test to look at heterogeneity. So what is heterogeneity in a meta-analysis? You just think about your forest plot. Every single study is going to have a dot that's going to be where its effect size is and it's going to have a bar, which is the 95 percent confidence interval around that effect size.
The bigger the study, the tighter that bar will be, the smaller the study, the larger that bar. The bars are going to differ from study to study. Why is that? Well, the four of us, Kate, Eve, Kurt, and I could all do the same RCT.
We could say over the next five years, let's give us each five million in funding. Let's do the same RCT. We're going to design it a little bit different.
We're going to define cumulative live birth a little different. We're going to have different patient populations. So there's two reasons why the bars in our four individual studies would vary.
One is just sampling randomness, right? If we each do a study of 500 patients, there's going to be sampling error. There's going to be variation just by random statistical chance. So a fixed effects model assumes that the difference between the bars in the study is purely sampling error, but we know that's rarely true.
Studies are different. In the example that I just gave, even the four of us who think similarly and have worked together on studies would design different studies, have slightly different patients, have slightly different definitions. So the random effects model assumes that there's also heterogeneity between the studies.
And so that's why in general, most people will use random effects models. Now, a random effect model is always going to give you a bigger confidence interval. That's just the nature of it.
And so in general, when I approach a systematic review, I never do fixed effects models unless I can show that the studies are exactly the same. So they mostly use random effects models, which I agree with. They then went a step further, something called the Kapp-Hartung adjustment, which further adjusts and further widens your confidence intervals if you have only a few studies.
And in this case, they had four clinical trials. So this is a very conservative way, which I would agree with doing in that analysis, to say, let's be really careful to limit our type one error and not overemphasize what we find. So it's essentially two adjustments from a fixed effects to a random effects that widens your confidence interval, and then to this further adjustment that widens it even more and makes it very hard to reach the bar of statistical significance unless the signal is very clear.
So that's great. So what did they find? Overall, they found that there was an improvement in live birth in the four RCTs with blastocyst transfer. It was 1.1 on the relative risk with the p-value just under 0.5. But when they apply that adjustment, again, remember, if you apply the adjustment, it has to make the confidence interval wider.
The estimate stays exactly the same, 1.1, but the confidence interval widens, and it's now p of 0.13. So it's not statistically significant, but that's the statistical construct they did. It could only go in that direction, applying that construct. As we go through all of these sub-analyses, and this is why I asked you to think at the start about why do you use blast transfer if you do, the signal is kind of what you would expect it to be.
The more embryos you had, the higher your AMH was, the younger you were, the better off it was to go to blastocyst. That's what the study showed. The fewer transfers you do, the better off blastocyst is, right? So if you do just one transfer, blast is going to get you to a pregnancy quicker.
If you do two transfers, blast is going to get you to a live birth quicker. If you do more than that, if you add in all of the embryos over time, they even out. The two arms become more equivalent.
And that finding persisted in all of their subgroup analyses. So essentially, the authors concluded that there might be an advantage to doing cleavage transfer, but it's not clear. But overall, both strategies remain reasonable.
I think that's probably true. Before I get to my final comment, I'll just say that it's always interesting when people add qualifiers to their statistical findings. When the p-value was statistically significant, but just under p.05, they said it's barely statistically significant or marginally.
When it was just over, they said it's clearly not statistically significant. So the authors are telling me what they view here. They view the null as being true, that both arms are reasonable.
Blast transfer, cleavage transfer is reasonable. You could have had someone else who clearly believes that blast transfer is right this the exact opposite way and say this is clearly statistically significant. There's clearly an advantage by doing a blast transfer.
So just a qualifier to people writing papers out there. I do it in my own papers accidentally. It shows my own biases.
I always encourage my fellows to remove your adjective in front of your p-value in front of how you're interpreting it because you're telling me what you think as the reviewer, as the editor, when you put that in there. So then I just come back to my final thought. Why do we do blast transfer? To me, it's kind of like PGT.
It's a selection process. By allowing the embryos to grow for a couple extra days, I'm going to allow them to self-select out as they undergo genomic activation and some embryos will arrest naturally. We get to better selection.
I would not expect it to improve cumulative live birth. But if you do first transfer, the Cochrane Review shows with over 30 studies, the relative risk is 1.4. So this was a relative risk of 1.1. 1.4 if you go to first transfer. So you might get to a first baby quicker.
Maybe that will reduce the dropout. Similar argument to PGT. It's more of a selection.
So my take-home, if you're a blast transfer person, you're going to take this data and you're going to say this is why I do blast transfer. If you are selective cleavage transfer or an all cleavage transfer lab, you're going to say this is why I still do it because it's not any different. You can take this paper and you can adjust or use it to support what you believe is true.
So I don't know that this is going to move the needle as much as it's going to make us all feel good about whatever our strategy is in our clinic. That's my take on it. Maybe I'm a little cynical, but that's my view.
It's an interesting study. I liked reading it. I enjoyed the thought process of working through it.
I agree with everything you say. Using blast transfer is basically letting the marathon time determine your marathon team as opposed to letting their 50-yard dash time determine it. That's the way that I describe it to patients.
But it's also going to be obviously dependent on the quality of the laboratory, which this study can't really assess. You have to have a lab that has excellent culture systems and excellent blastulation rates. Ultimately, it's about patient selection just like everything else.
What the paper also can't address is what about those patients where we could have known we have nothing to transfer and we move on as opposed to potentially exposing them to risk of a pregnancy loss that can take up valuable time. Sort of becoming age-old questions in our field, but I congratulate these authors on shedding light on some of the important issues here. I like that this was in the literature because we all have very strong opinions, and regardless of which adjectives you use to describe your p-value, which I love that you gave that comment.
It's very telling. It shows, at least in my mind, that the data is not really dramatically different. Therefore, as Kate said, we're making decisions based on other aspects of care because this is not a done question.
It's not obvious that blastosis is better or that cleavage is better. I think that's the take-home message. Again, be careful of your—or just not be careful of your opinion, but know why you have your opinion.
What is driving your opinion and what your rationale is? Blanket statements are often good for patient counseling, so we don't have to get into the weeds. We should know that we don't really know what's best, or at least what we're doing isn't making that big a difference. Yeah, I think you can justify what you do.
That's my take-home message from this, is you can justify what you do and why you do it, and I think that you can easily rationalize both approaches, and I also think that there's room for individualization patient by patient. All right, Eve, we are coming to you. One of my favorite topics in journal clubs, platelet-rich plasma.
What did we learn about PRP in the endometrium for IVF this month? Thanks, my God. This is a great paper, and I'm glad that it's published in Fertility and Sterility, and I think no doubt it will raise a lot of discussion about what PRP can do and what PRP maybe cannot do, and so the title of this paper is Autologous PRP on Endometrial Growth and Reproductive Outcomes in Women with Thin Endometrium, a Multi-Center Cohort Study, and this was also from the IVY group out of Valencia, and so I think most people will agree that thin endometrium, as we define as under seven millimeters, despite adequate estrogen exposure, affects a small proportion of ART patients, but it is linked to lower implantation, pregnancy, and live birth rates, and may also, I think, be linked to more obstetric complications. There is no standard treatment that's out there.
High-dose estrogen often fails in this scenario, and hysteroscopy carries a risk of adhesion formation or adhesion recurrence, and so PRP has been discussed as a possible help in this scenario, and what PRP does is it concentrates platelet growth factors, things like VEGF, TGF-beta, EGF, that drive angiogenesis and tissue repair. Some studies have reported thicker linings, but live birth data are limited, and the results are certainly heterogeneous, and so what this study is, it's a retrospective cohort study, so this group is routinely using PRP, and they retrospectively analyze their results between January of 2020 and January of 2024. They had 280 FET cycles, 94 with PRP and 186 with no PRP, and eligible participants were women with an endometrial thickness between 3 to 6.9 millimeters during the previous cycle, and all of these patients had experienced at least one negative cycle that they call implantation failure.
I'm now cautious after discussion of our first paper, but they had experienced at least one negative FET cycle, so all transfers were ICSI-derived, day 5 or 6 blastocysts, and there are multiple different ways of preparing autologous PRP and many different protocols out there, but they use the PRGF endo-RET closed system kit. They used 18 to 20 milliliters of blood. This was drawn and spun, and then between 0.5 to 1 ml of PRP were infused into the uterus with a soft intrauterine catheter, and then interestingly, in this study protocol, patients had up to three installations per patient.
Each was 48 to 72 hours apart, and they were performed between cycle days 7 to 12. Endometrial preparation in these patients was mostly a programmed cycle with estradiol valerate 6 milligrams per day. They increased up to 8 or 10 milligrams per day if the lining stayed under 7, and they used vaginal progesterone for luteal support.
I see Kate's horror on her face with that. I'm just kidding. A minor, a minority of these were natural or stimulated cycles.
The outcomes were evaluated, and they looked at endometrial thickness after treatment, biochemical, clinical pregnancy, ongoing pregnancy, and live birth, and then they did some multivariable regression analyses adjusting for age, embryo quality, oocyte source, PGTA use, number of embryos transferred, and baseline endometrial thickness. So overall, the two groups were quite similar in age. They were a little bit on the older side, 41.6 versus 42.2. They were similar in their BMIs, cycle type, trigger, and embryo day, day five versus day six.
And interestingly, in this population, most cycles use donor oocytes. The PRP group at baseline had a worse prognosis. They had a thinner lining.
They used more PGTA. They had more double embryo transfers, and they had more poor quality embryos, which again, I think is interesting coming from donor eggs. But women who receive PRP started with a thinner endometrium than controls, 5.7 versus 5.96, which actually for control or comparison group, I think that's actually still quite a thin endometrium.
What they found was that PRP significantly increased the endometrial thickness, although I want to temper that by saying that the final thickness was 6.65 millimeters versus 6.12 millimeters in the comparison group. So still, I think it's a significant difference, but you're comparing it to I think an artificially low comparison group. And so they had this mean increase of about 0.95 millimeters with PRP versus 0.16 millimeters without PRP, finding that they found to be statistically significantly different.
But I think that the major point here is that there were no statistically significant differences in biochemical pregnancy rate, clinical pregnancy rate, ongoing pregnancy rate, or live birth rate between groups. And with multivariable adjustment, PRP was not associated with improvements in any of these. So what they found is that PRP resulted in a modest but statistically significant increase in endometrial thickness when compared to their control group that had a very thin lining.
However, this thicker endometrium did not translate into better pregnancy or live birth outcomes. And so the authors conclude that embryo competence may have a greater impact on IVF success than modest improvements in endometrial thickness. And I think that goes along very nicely with the first study that we discussed on this morning's podcast.
And the authors also conclude that routine use of PRP to improve reproductive outcomes cannot be recommended based on these data. So I think, you know, in the strength section, this was a large contemporary real-world PRP cohort for thin endometrium. It has a multicenter design.
They did include live birth as an endpoint, which I like. And they performed adjusted insensitivity analyses looking at various subgroups. But I think it also has some important limitations.
It was retrospective. This is not a randomized study. There's clearly bias in who they selected for their PRP group.
And these PRP patients had poor baseline prognostic characteristics. They also admittedly only talked about endometrial receptivity as ultrasound thickness. Is ultrasound a real marker? Is it a surrogate marker? And there was a lot of variation in cycle preparation and a lot of variation in how they timed PRP.
And also, I think this older donor egg population may not be generalizable to younger patients using their own eggs. So I think if I were to drill it down to a few points, like my take-home messages, would be that PRP made the endometrium thicker, but it did not improve live birth rates. And I think that this study challenges the assumption that increasing endometrial thickness alone improves outcomes.
But the finding reinforces this concept that endometrial thickness may be a surrogate marker that doesn't actually reflect functional receptivity. And so I want to open it up for discussion. I thought it was a nice paper.
I'm glad it was done. I think there's a lot of PRP being used in a lot of different patients with a lot of different protocols. And I think we're just not there yet.
It's got interesting biologic plausibility, but does it actually work? I don't know. And I don't think that this paper is the definitive answer to that question. But I do commend the authors on publishing it and sharing their results and their protocols with us.
If I want to raise the possibility that it's even worse than the null based upon their findings. So they had 36% live birth in the non-PRP arm, 24% in the PRP arm. Now they say the P-value is .07, but that's their adjusted analysis.
I ran the chi-square on that. It is worse. So PRP, you had lower live birth than the unadjusted.
Now you mentioned that the PRP arm had a thinner endometrium, but it's by 0.2 millimeters. Yeah, it's thinner at baseline, but who, which of us uses 0.2 millimeters as a threshold of anything, one or two millimeters, maybe 0.2. The PRP group was younger. They were twice as likely to use PGT and they were four times more likely to have a double embryo transfer.
So there's more PGT. They were younger and they had a double embryo transfer, but yet their live birth was 12 percentage points less. Somehow when they did the adjustment that moved from statistically significant to not.
So I would love to know what the other variables were in their modeling. They tell me, but I would love to see that model because to me, there's a signal here that PRP may have actually done harm. They did 12 percentage.
So not only, and I think you've seen this, I think you've told us before, you've, you've seen people lose their ovaries from PRP injection into their ovaries from other programs with an infection. Like our goal as physicians is to first do no harm. So I actually see a potential signal for us to be very, very cautious with using PRP here.
Yes, it needs more research, but I see a signal at least on the red side that would say, Hey, let's be careful here as a field before we jump into doing this. Yeah. I mean, the only thing that I will push back on that, and this is just from our own centers, very limited experience with endometrial PRP is typically the patients that are selected.
It's like this last ditch effort of patients who, you know, really have done worse. And I couldn't tease that out from these data. I mean, they said at least one implantation failure, but they didn't really characterize exactly how many and what the quality of.
So they could be worse prognosis. And that's why I think that explains it is that it's a selection bias that the patients that they're selecting for PRP are so much worse. And especially in a non-randomized study, you're really opening it up to that selection bias.
But I agree like this paper, to me, doesn't answer the question of, should we be using it? But they should have characterized the differences in the groups in terms of their prior reproductive history a little bit better. If we want, we can't just assume that it was selection bias. And I think Micah's caution is, you know, extremely important to readers that it's when they're trying to get a patient with a better endometrial thickness, if they take these findings to be a reason to do that, they might actually be hurting their patients' chances of success.
Based on this paper, we don't know. And there also may be some bias on the part of the sonographer, right? Because you're looking at a patient who, you know, has had this really poor history in PRP and, you know, we all want to give patients good news and you might just eat that calculator up a little bit. So, yeah, I mean, I think it's a really interesting question, but I mean, don't go ordering your next batch of PRP just yet, folks.
This was a wonderful discussion you just heard, listeners. You know, why was this paper published if it had so many flaws? And the idea is that this paper would not have been published if it had had very grandiose conclusions or proposed, you know, this could help people in certain circumstances or whatever way people kind of construe the positive. The reason this paper was published was because it was tempered as conclusion.
And you just heard an outstanding review on how to really look at a paper about bias, about regression to the mean, and all the reasons that we might conclude the answer that's not true. So, that's why it's published because we want to be able to have this discussion and arm you with the ability to look at the data like this and recognize that it doesn't change care. Even though this paper doesn't change care, it is a contribution to literature.
This data does deserve to see the light of day. Just be careful on how you interpret it. I'm really pleased.
That was a great discussion. That is a great way to end this segment. As we often say, the idea that you have to force a positive finding into your research to get it published is not true with fertility and sterility.
In fact, it's the opposite. Beat up your hypothesis. A null study has a lot of value, and we just appreciate a rigorous approach to the scientific question.
What is the hypothesis? What is the null? Did you approach it rigorously? Is that question clinically relevant and methodologically robust? Fertility and sterility is likely the place for you to be published. So, we have some other studies we wanted to talk about. In the interest of time, we're going to save those for our ASRM podcast.
We have something new that can reduce intrauterine adhesions. They've agreed to do their RCT with us at ASRM, so we'll cover that topic there. But, as always, don't use this as a to read the journal.
Use this as a place to dive into the journal, to read fertility and sterility, and to, as Kurt would say, let the science flow over you. So, we are excited to see all of you at ASRM, and please submit your great papers to fertility and sterility. This concludes our episode of Fertility and Sterility On Air, brought to you by Fertility and Sterility in conjunction with the American Society for Reproductive Medicine.
This podcast is produced by Dr. Molly Kornfield, Dr. Adriana Wong, Dr. Elena HogenEsch, Dr. Selena Park, Dr. Carissa Pekny, and Dr. Nicholas Raja.
