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. FNS 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 to Fertility and Sterility On Air. I'm Micah Hill, the Media Editor, and we are in July 2026, volume 126, number one. I'm joined by co-host Eve Feinberg.
Good morning, Eve. Good morning, Micah. And Kate Devine.
Good morning, Kate. How are you today? Good morning, Micah. I'm great.
How are you? Good. I am excited to talk about these articles today. There's a lot of really interesting things.
I'm curious to learn from both of you and hear your perspectives. We've got some good front matter this month from two of our urologic editorial editors. First is a views and reviews from Peter Schlegel on varicocele.
Takes a lot of really interesting takes on different aspects of a varicocele, including an article on low testosterone, chronic orchialgia, men with non-obstructive asospermia, and then what do you do with an adolescent who has a varicocele, a surgical conundrum. On a different line, Michael Eisenberg, another of our editorial editors, has an inkling on does teratozoospermia equal normozoospermia. He's essentially arguing that we shouldn't even really look at morphology in our semen analysis because clinically it rarely changes our decision-making or our algorithms, and the vast majority of the time those men do fine with their reproduction with our various options we have.
In the ASRM pages, we have one on protocol deviations and witnessing in the IVF and andrology lab, a very important one, especially for medical directors and lab directors out there, and then one on ICSI for non-male factor indications from ASRM. I'll just say that not much has changed on that. If you're using ICSI for things other than male factor infertility, for the most part, you're probably over-utilizing ICSI, as the evidence currently states.
So, great front matter in the journal this month. Eve, I'm really excited to hear what you're going to teach us on this seminal contribution about using AI to find ultra-rare sperm. This article fascinated me.
Yeah, agree. The title of this article is High-Throughput Imaging Cytometry Paired with Artificial Intelligence Identifies Ultra-Rare Sperm in Men Previously Diagnosed with Clinical Azospermia, and that's from Blair Stacks and others, with senior author Larry Lipschultz from Baylor. So, as we all know, microtessi is used for men with non-obstructive azospermia, and reported sperm retrieval rates approach about 50%.
But there are tons of limitations. So, current limitations in management include the inability to reliably predict preoperative microtessi success, and then difficulty identifying ultra-rare sperm in the ejaculate or in surgical specimens for men who have azospermia. Current methods can require up to eight hours of manual embryology effort, with success rates quoted to be or recovery rates quoted to be somewhere between 40% and 44%.
So, AI-based approaches, specifically deep learning with convoluted neural network models, have shown promise for rare sperm detection in microtessi samples, but their application to ejaculated samples in azospermic men is limited at this time. And for those of you that have not read it, there was a recent case report published in The Lancet by the Columbia University Group reporting the first successful ongoing pregnancy using AI-guided sperm identification from the ejaculate in a patient who was diagnosed with azospermia who had failed of micro-testicular sperm extraction, and this is a wow case report. This novel study introduces a new diagnostic platform with high throughput imaging cytometry paired with AI-assisted CNN-guided image analysis to identify these ultra-rare sperm in ejaculate samples that were deemed clinically azospermic.
So, what they did, and I know this is a lot, but it was a retrospective cohort review of remnant biobanked ejaculatory samples and microtessi samples from a single high-volume male infertility clinic at Baylor. Samples that were deemed clinically azospermic or cryptozospermic were identified from the biobank, and then they used control samples that were known to contain sperm. There were 148 total samples that were analyzed, 83 that were azospermic, 34 that were cryptozospermic, and the rest were higher volume sperm containing.
So, they prepared these samples for imaging, they fixed and they permeabilized the samples, and they stained them with three different markers to identify sperm-specific proteins for sperm DNA, the acrosome, the tail, and then a fourth marker to identify white blood cells in the ejaculate, and then they applied secondary fluorescent antibodies. They then did validation of antibody staining, they used high throughput cytometric image acquisition, and they did that using a ThermoFisher Attune CytPix Acoustic Focusing Cytometer. That's a mouthful.
They ran six batch runs, and the images were exported and analyzed, and then they used a pre-trained CONVNEXT CNN model, and I'm going to talk a little bit about that as the base model for image classification. So, this CONVNEXT CNN model is a convolutional neural network architecture that was introduced by Meta-AI Research in 2022. So, older CNNs work by sliding small windows over an image, and they look for simple patterns first, and then they combine these into more complex patterns, eventually recognizing whole objects.
A few years ago, a newer approach called Vision Transformers came along, and those started beating CNNs at their own game using a different strategy that they borrowed from language AI. So, CONVNEXT was researchers' way of asking, can we take the old reliable CNN, can we tweak it to design to match what the new transformer models can do? So, it takes those same building blocks as before, but it rearranges and resizes some of those design elements to form these newer models, and it performs just as well while staying simpler and more efficient. And so, they trained the model using this to identify the sperm.
They used a binary classification task where they identified sperm versus debris, and again, remember, they used those white blood cell markers to mark white blood cells to distinguish that from sperm, and those antibodies to identify the different parts of sperm. So, they took 80% of the data, they were using 80% to develop a training set, and then the remaining 20% was used as an external validation set. All the images that were classified as sperm by the model were then manually verified as sperm or debris by three independent reviewers.
So, positive predictions in samples from cryptospermic and severely oligospermic men were validated by a single reviewer until 100 sperm had been verified or all images in the sample had been reviewed. Performance of the model during cross-validation and test set validation was assessed using an F1 score as well as sensitivity and specificity, and an F1 score is a single number that combines both precision and recall into one balanced measure of how good a classification model is. Inter-rater agreement for manual review of positive model predictions were assessed using Cohen's kappa, and then all model training validation and testing were performed using Python 3.13. So, the initial training data set was comprised of 5,400 images encompassing normospermic controls from patients across each azospermia oligospermia domain.
Each image, again, was classified binary as sperm versus debris. The CONVNEX model achieved an F1 score of 0.963 with a standard deviation of 0.006 with a sensitivity of 0.971 and a specificity of 0.955, and that's really good. The goal is to have a perfect model which is 1.0. When they evaluated this using the external test set validation of 6,117 images, the model achieved an F1 score of 0.874 with sensitivity of 0.881 and specificity of 0.970. So, the model classified at least one image of sperm in all azospermic patients.
Then when they did manual validation, they found sperm in 40 of 83 of these samples, so 48% of samples. In the control, sperm were correctly identified in all of the sample. I think probably the most impressive part was that the CNN inference speed was 265 images per second, and the full 30,000 image sample was processed in 113 seconds.
So, with their microtessie samples, so that was ejaculated sperm, and then with their microtessie findings, they had one azospermic patient who had sperm identified in the ejaculate by the CNN model, and that was manually confirmed. That same individual had previously undergone bilateral microtessie, but no sperm were identified by the surgeon or embryologist. Then they looked at the remnant microtessie tissue from both the right and left sides, and they were analyzed in triplicate, and the CNN predicted sperm in the right, two out of three replicates, and left, one out of three replicates, and that was confirmed by visual morphology and fluorescent expression of antibody staining.
So, pretty amazing findings. So, overall, when we summarize this, of 67 unique clinically azospermic specimens, these investigators found that 37 actually had sperm identified by the model and were manually validated as having sperm. 30 patients did not.
The underlying etiology of infertility in these patients did not affect whether the CNN predicted that sperm would be present or absent, and then when they took that high-throughput imaging cytometry and they paired it with AI-assisted image analysis, they identified ultra-rare sperm in ejaculatory samples that were deemed clinically azospermic by standard WHO-based semen analyses. So, this paper, I think, is very groundbreaking, and it highlights the potential for AI tools to exceed trained human embryologists, both in performance and in speed. So, 31 times faster image acquisition, 265 images per classification, and I think this doesn't only have tremendous potential for our azospermic patients, but I think it also has tremendous workforce implications for andrology laboratories.
And for ejaculated sperm, a 48% detection rate was achieved in samples previously classified as azospermic, and that's a clinically meaningful reclassification rate. And so, again, I want to highlight that this is an ejaculated sperm for men who are called azospermic. So, the authors propose a new sperm concentration category, which they call microzospermia, and that's defined as a sperm concentration so low that only advanced imaging-based search technologies can reliably identify ultra-rare sperm.
And the idea is that patients who are classified as microzospermic may potentially be able to avoid microtessy altogether if this platform can be applied clinically. So, before we open this up for discussion, there's two limitations that I want to highlight. One is that this model described doesn't actually have cell sorting or recovery capability.
They're staining the sperm, which renders them incapable for use clinically, so that's a huge limitation so far. It's easy to identify sperm when you label the DNA or you label the antibodies to the head or the tail. How well is something like this going to work clinically where you want to capture those sperm? So, it doesn't have the sorting or recovery capability, but once it's developed, it will need head-to-head comparisons with other platforms, such as the STAR system, which is what was described in the Lancet article, or another platform called SpermSearch AI.
And then the final limitation that I want to highlight is it doesn't distinguish between modal and immodal sperm. And so, the reproductive capacity of those identified sperm cannot be assessed. But I really think that it's tremendous work, and I want to congratulate the authors for publishing it.
And I'm thrilled it's in fertility and sterility. So, I want to open it up for discussion, Micah and Kate. There's a lot to talk about here.
I agree. This is tremendous work and really timely also with the recent paper by Zeb Williams and the team at Columbia that has also hit the popular press. So, I think many of us have been even asked about this by some of our patients.
I think it's going to be a really interesting story with this type of technology using AI for sperm, finding ultra-rare sperm, sperm sorting, and also there's parallel technology that's been discussed and is potentially coming down the pike for finding oocytes as well. To me, I think that the regulatory aspects of this will be really, really fascinating to follow. Obviously, this study was done under an IRB protocol, and any type of technology that a laboratory develops internally and uses only internally for their own patients does not require any type of FDA clearance.
However, as entrepreneurs decide that they're interested in commercializing these technologies, many would insist that they achieve FDA clearance. And this has been a really challenging thing in our field. Of course, there are hundreds of AI-approved technologies in the fields of radiology, et cetera, others in medicine, but in our field, there have been very few.
Some have tried and failed. Some have tried to bypass regulatory approvals altogether. And then there are other technologies that incorporate AI sperm selection into, for example, a larger robotic automated laboratory system.
Does that need FDA approval? So, to me, it digresses a little bit from the methods of this paper, but I think in terms of the practical application of these technologies, it's really an interesting time in our field, and to some extent, open to interpretation by clinicians in terms of what do they feel comfortable with from a regulatory perspective. It's actually going to ask you along those lines, since you're a chief medical officer and you kind of deal with these things all the time, why is our field behind radiology? Is there a different bar that we're held to because it's reproductive medicine, or have we just not taken the steps that the radiology or other areas have? What's your gestalt on why that difference exists? Yeah. I mean, I think some founders have just kind of taken the approach that our field will just use things without approval.
So, and honestly, not totally unfounded in the setting of some of the technologies that some of us have adopted. So there's that piece of it. And then there's also, it's very challenging to prove, especially things like oocyte or embryo selection to prove, quote unquote, prove safety.
And the FDA has just decided to have a really high bar in terms of, is there actually a benefit for these technologies and are they safe? And it's hard to show even with an RCT. Steve, what did you think about the new diagnosis that they proposed, microzoospermia for these sorts of things? I feel like if you can't pronounce it easily, then maybe we should question the name. I don't know that we can pronounce any of the sperm diagnosis easily.
Both of you are doing a great job. Yeah. I mean, it's, I don't know, it's another term to me that was not the highlight of the paper.
They seem to focus a little bit on that. But I think just the idea that maybe we don't need to do as many micro testes as we're doing is that's the part that really struck me. And with, you know, with the Columbia paper as well, I mean, that was a patient that had undergone I think it was her 10th or 19th IVF cycle.
I don't remember, but the female partner had severe diminished ovarian reserve. The male had undergone several micro testes and they used ejaculated sperm for IVF or for ICSI in that situation. And so I think like, whoa, proof of concept, we have the ability to do this.
So this model is, this was more of the pre-work in terms of the identification proof of concept. This is what it takes. This is how we train our model, which I thought was really interesting.
And for those of you that are interested in learning more about these models, I was just on Claude yesterday asking Claude, what is a convoluted neural network? Explain it to me like I'm 10 years old. And actually the descriptions in Claude were quite helpful in the language. They use similar models to how they interpret language.
It's built on one word and then it's built on a sentence and the image processing is built on a curvature and then it builds to a shape and then it builds to an entire object. And so they layer it to train the model in that way. And so I think Mike and I were texting yesterday.
I'm like, I'm in a rabbit hole learning this. It's fascinating. And I learned a lot by reading this paper.
I thought the method section was excellent and I tried to simplify it. One other point that I didn't get into is this idea of spillover where you have to really clean the machine between use. And they found that when they didn't clean the machine optimally that they were actually getting sperm from a previous sample.
So that to me is very scary with these ultra rare sperm. You have to be incredibly cognizant. There's got to be a way to sterilize or be a hundred percent certain that when you're running samples through these machinery that you're completely cleansing it of the sperm sample of the patient used previously.
So they talk about that in their methods as well. I just didn't. I could have gotten a lot more granular in how I described it.
I tried to stick to the high level points of how they did this. I had to Google spillover also. So I had to learn what they meant by that.
To me, I actually, Kate knows this about me, but I get more excited about AI in this sort of application than I do about the sort of low hanging clinical places. Like how do we dose people? What data trigger people? I think the actual ability of AI to meaningfully move those sorts of decisions is relatively small. Here it's massive.
These could change. You don't have a biologic kid until you could. And as you both mentioned, they're finding 10 to 20% more eggs in some of these.
They're finding sperm. It's just the reality of the human eye. Our embryologists are the most amazing scientists under the microscope on the planet, but there's a limit to the human eye.
And that's definitely a place where AI can meaningfully move the needle. Well, not only that, but how many times have I had to tell a patient, like, I'm so sorry, our embryologist spent six hours tearing apart your sample and didn't find anything. And here in 113 seconds, the AI gets through it.
So it's not only better than the human eye, but it's faster. And obviously you still need a human to verify. But I think I agree with you.
I'm far more excited about the technology in this particular application than I am about, is it a 12 or an 11 day trigger? And when do we start STEM? Do we adjust the dose? Do we trigger on this day? Those are all interesting applications of AI, but they don't really move the needle in the field. Whereas this is one application of AI that I think really moves the needle and really has the potential to change clinical care. Final word on this paper? It's exciting.
And I think more work is needed. Obviously, we need to be able to use this clinically and not destroy the sperm in the process. But I think that it's getting there.
And develop a single use cartridge disposable to prevent the spillover. But yeah, I couldn't agree more. Incredibly fascinating and really promising work.
Awesome. Well, we spent a lot of time on that one, but it's because it's cool. When you think of Kurt's Venn diagram, this one hits the novelty, which is why we talked about it so much.
Hopefully, clinically down the road, we'll see things that we can use from this. We're moving on to ART. I have the article, euploidy rates are comparable between progestin protocol ovarian stimulation and gonadotropin releasing hormone antagonist protocol ovarian stimulation cycles.
This is from authors Boris Atta and Heumann Fatemi and the rest of the team from the United Arab Emirates and a global collaboration. And so, as the title said, they're comparing euploidy rates between the progestin protocol and the antagonist protocol. This is a retrospective cohort study, and they're using digergestrone at a dose of 10 milligrams TID.
And they basically say in their introduction, which I appreciated, that there's no biologic plausibility for the progestin protocol to have an adverse effect on euploidy, which is exactly my thought when I read it. I was like, do we really need another article on this? But they cited a couple papers that state that maybe there is an increased risk, and this particular progestin has not been studied yet. Others have.
There's actually quite a bit of data from other progestins, including clinical trials that don't show an increase in aneuploidy. So, I like their statistics. They use propensity score matching.
They looked at patient age, BMI, AMH, AFC, and stimulation medicine. But then they also did some multivariable and mixed effects approach to then control for those same models afterwards. And so, for the fellows and the learners out there, this actually makes a lot of sense.
With the propensity score matching, you're bucketing patients so that they're similar to each other before you even start. But for example, their AMH was a bucket of 1.1 to 3.5. And you could just imagine that one group averages 1.5. The other three, those end up being two different groups, even though you've tried to propensity score match them just by random chance. So, then you additionally control for those variables on the back end.
So, they're essentially controlling for them twice to try to get the precision of their estimate to be better. There was a difference between the two groups before they did the propensity score matching. Actually, there was a difference in several of the baseline characteristics.
But after propensity score matching, they were mostly aligned. The main outcome, what do we care about? The euploidy rate, 43% with antagonist, 42%. They were off by 0.6% between the two protocols.
So, essentially, almost exactly the same numerically and obviously no statistical difference. They also did an interaction test to see if this changed based upon age. This is based on one single paper retrospective that's out there that says as you get older, the progestin protocol is more likely to give you increase in aneuploidy.
This interaction test tests to see if there is an interaction between the age and the protocol type with the euploidy rate. They did not find a difference in that. So, overall, I think this is a nice paper.
The first thing I loved, PPOS, they kept the acronym, but they used the estrodiagnosis. They removed priming because there's nothing about priming in this. I still think progestin protocol is fine.
I don't know that we need ovarian stimulation because the progestin isn't stimulating anything. Antag protocol, long luteal protocol, progestin protocol, but whatever. I'll take it.
It's a win. We've been harping on this for like five years on this podcast. So, I love that.
The other thing I just noted is that there's a danger in publishing false positive data. So, if you're trying to publish a study, don't try to find something statistically significant so that you can get it published because then other researchers have to come along with bigger studies that are more rigorous and show that there isn't an effect. I feel like half the studies we publish with our fellows end up being along those lines.
Null studies are fine. Fertility and sterility loves null findings. Most studies should be null findings.
That's helpful for us as a field. My final comment is I don't need to see another study on progestins and aneuploidy. I'm very convinced.
I don't need to see another study on outcomes. I'm convinced. They work.
We don't have to go through all of the dozens of progestins that are available worldwide and publish all of them. I think they needed to do this. I'm glad it's in FNS, but those are my take-homes.
Kate, Eve, what did you guys think? Amen. I couldn't agree more. It's well-studied now.
I just hope that we get more uptake in the U.S. population. It's happening. It's happening slowly, but it's good for patients.
So, great protocol and great paper. Agree. I thought it was really nicely done.
I think it's the nail in the coffin to say we've split this and spliced these data 10 times over and looked at it in every possible way, and it's no different. So, let's use more of it. Our patients love it.
It's one less injection. It's so much less expensive. I think that the ovulation breakthrough rates, I actually have not yet seen anyone have breakthrough ovulation when using the progestin protocol.
I really also smiled when I saw the primed was out of the title. I was so happy to see that. So, I think it's great, and I agree.
We don't need more studies of this nature. That said, and Micah and I agree on this, if there's a fellow out there looking for a thesis to do that has a big enough population to draw upon in terms of participants, we could still use a dose-finding study because if we can prevent ovulation at a high rate as we do with a high dose progestin, and I think most of us are using 10 milligrams of medroxyprogesterone in the U.S., there might be even less suppression. Maybe we do better than we would do with an antagonist protocol or at least just minimize the exposure to the patient.
So, I still don't know what the optimal dose and regimen for progestin is, but overall, not worried about aneuploidy, that's for sure. I love that comment, Kate. It's a sidebar off the Ganarelix dose-finding study that's like 20-plus years old now, and the lower the dose, the better the yields, but the higher the dose, the less the breakthrough ovulation.
So, it's finding that sweet spot between those two compromises. We don't yet have that with the progestin protocol. So, great paper.
Encourage you to read that. I'm sticking in ART with the next article, Association Between Serum Estradiol Level Decline in the Days Preceding Ovulatory Trigger and ART Outcomes. This is by Peter Linder, one of our past fellows, and Philip Romanski, and some other authors from RMA New York and Shady Grove.
This study, obviously, is looking to see what happens if you have a drop in your estradiol in the one to four days prior to your trigger. This was a retrospective cohort study. They limited the population only to ANTAG cycles and only to ICSI.
I'm not exactly sure why only to ICSI, except they're probably just trying to keep their population tight up front rather than control for these things on the back. They excluded things that would mess with estrogen levels. So, if you're on letrozole or clomid or something that's intentionally trying to suppress estradiol levels.
The primary outcome was the number of usable blastocysts, which were defined as SART good or fair, which would be Gardner score A's and B's with expansions three through five. They used generalized estimating equations, and they only adjusted for female age and AMH, which is very simple, but I actually like that. Having done a lot of modeling on data, those two things, when you're looking from the female perspective, almost always drive 98% of outcomes.
Other things drive a little bit, but they're in the margins. So, I actually like that they were simplistic on what they adjusted for. I think sometimes, and I've been guilty of this, we tend to throw the kitchen sink at our adjustments and it's too much, and especially if your adjustments lie between the exposure and the outcome.
So, they're along that causal pathway between those two things. It can give you some funky results that aren't real. So, I appreciated that they did that.
There were no differences between the groups and their baseline outcomes for the most part, except for some key things, and I just want to describe the groups real quick. So, consistent rise was the estrogen went up every day. There was never a decline.
That was almost 6,000 of the 7,000 cycles they had. A single drop was 1,000 patients, and that makes sense. I think most of us, if we're at the end of the stem and the patient drops, we're triggering that day.
We think of it as a sign that the granulosa cells are sort of giving all they've got, and maybe those eggs, it's time for them to come out. Two to four drops, they actually had 200 patients that had two to four drops. Now, that's interesting to me because I rarely let a patient drop that much, unless it's intentional.
Those of us who are older remember coasting, and we were intentionally dropping people, and we do lose eggs when we coast, but we could safely get them to retrieval. So, that group of 200 patients is different. Their AMH was higher.
Their estrogen was higher when they started the antagonist. So, these are good responders, but then they end up using less gonadotropin. So, again, good responders, and their E2 is lower on trigger.
So, this tells me that those 200 patients, the physicians are intentionally slowing them down, but they're probably good prognosis patients. So, that group maybe is a little bit different than someone who we're letting crash for two to four days, just because we're not paying attention. So, what did they find? Their primary outcome, blastocyst.
Five, if you had a consistent rise. 5.2, if you had one drop. 5.9, if you had two to four drops.
So, again, I think that group's probably a special population that we're bringing along, but they're a good responder. That's 16 comparisons, no statistical difference. They then decided to break down AMH patients less than one, AMH patients greater than three, to see if maybe that effect of their ovarian reserve and response was what was driving this.
Big picture, they looked at 32 outcomes, so now we have 48 total outcomes. Three were marginally different. The big take-home is that AMH didn't really affect their response to a drop.
And then, what we all care about, live birth. Live birth was not different in any of these patient groups. So, overall, I think this is a great paper.
I trained Peter through medical school, residency, and fellowship, as did Kate, and when he mentioned that he wanted to do this, I said, yeah, that's great, because we definitely perseverate over this when we see those drops, and this shows us that maybe it's not quite as big of a clinical difference as we might expect. My one caveat is that two- to four-day drop group I really do think is different, and I don't think we can extrapolate that to a sort of normal responder or low responder and just let them keep crashing. I do think from the coasting data, it's very clear you'll lose eggs and embryos if you do that.
So, that's my two cents on that article. Kate, what do you think? Yeah, the other clinical scenario that this calls to mind for me, in addition to coasting, and I couldn't agree more that that's a very different population than the other two groups that were evaluated, is the patient who starts to plateau or drop who has an isolated lead follicle and a growing secondary cohort, that we start to see that one follicle that was making probably the most estrogen of all start to peter out, and we let that happen to salvage what is overall the greatest good for the greatest number. So, I think while those patients couldn't be specifically teased out in this particular analysis, I think probably that also explains some of what they describe here and overall just clinically reassuring for taking that kind of a strategy, especially now in the era of lots of freeze-only cycles, not so worried about rising progesterone and so on.
Yeah, I guess what I was a little bit curious about was how many of these drops were due to dose reduction versus constant dose followed by drop in estradiol, and I didn't see that in the paper, and I think that those are two different populations. I never worry when we drop a dose or we start antagonist and the estrogen plateaus or drops a little bit. Where I do start to worry is in those older poor responders where they're dropping, and typically, and I know that they excluded all the cycles that were canceled before retrieval, but I think a lot of times when we do see those large drops, we're canceling those patients, and so this study has a little bit of bias in that these were the ones that continued to retrieval, so probably their drops were not quite as significant as those who were canceled.
You're spot on, especially in that two- to four-day drop. I think to me the more interesting was the one-day drop, and they did good. So, I think that we can be reassuring to patients that, yeah, it plateaued this day, it's time to trigger, but your outcomes are going to be just fine.
So, overall, I find it a good reassuring null study. Again, I guess theme of the day is null study. Kate, we're moving on to you.
The outcome of this study in infertility actually surprised me. I expected that they would find something different, so tell us what you learned that's sort of an epidemiologic or research design issue. Yeah, I would say this also sort of fits in peripherally with our theme of null study.
So, this paper was entitled Association Between Trial Registration and Reporting Large Treatment Effects in Randomized Controlled Trials of Infertility Interventions, a Meta-Epidemiological Study. So, this was first authored by Feng, along with an international collaboration of authors, including senior author Ben Moll, really nice paper. They highlight the importance of trial registration, and they look at it specifically within the confines of the question of whether trials that are not prospectively registered or trials that are not registered at all, and whether those trials should be included in systematic reviews and meta-analyses.
And so, you know, they highlight that the purposes of trial registration are mostly transparency to avoid selective reporting and to be able to compare what the pre-specified study design was relative to what the authors ultimately go on to report. And so, to get a sense of whether there is really intrinsic bias differences between those studies that were prospectively registered, those trials that were prospectively registered, versus those that were not, they look specifically at the outcome of a large statistically significant relative risk. And they defined that as a relative risk less than 0.8 or greater than 1.25 for any pregnancy or live birth outcome.
So, that includes clinical pregnancy, biochemical pregnancy, or live birth. So, if the study reported a statistically significant large relative risk for any of those outcomes, it would be considered to have to be in the positive for the outcome. Whereas if it did not, if either it was not statistically significant or it was smaller than what was just described, then that would be in the negative in terms of this binary outcome that they assessed.
In terms of the actual results that they found, so they looked at 1,369 RCTs that met their inclusion that were published between 2012 and 2023. Importantly, they excluded conference abstracts, secondary analyses, and RCTs where the registration timing was unclear. I think it's important to note that this particular study, the study that we're discussing right now, was registered with PROSPERO.
Somewhat ironically, it was a secondary analysis of a larger study evaluating characteristics. That's awesome. I totally missed that.
That's great. That is a very good paper, by the way, looking sort of at lots of characteristics of trials in our field. The original study is a great paper that's published in Human Reproduction Open, and it was published in 2025.
So, in terms of getting on to their actual results, of the 1,369 RCTs evaluated, 20% were prospectively registered, 30% were retrospectively registered, and approximately 50% were unregistered. So, big numbers still unregistered over this time period. Reassuringly, prospective registration increased significantly over the study period.
China had the highest number of RCTs in our field, followed by the EU, and with Italy having the highest number in the EU. Just of note, the United States was sixth in the number of RCTs meeting the criteria. Retrospective and unregistered studies were less likely to be multicenter, had smaller sample sizes, and were published in lower-impact journals.
All of this unsurprising to me. They were also less likely to receive government funds. In terms of the analysis that they did, they adjusted for known confounders, so those that had been reported in the literature as well as those found to be significant on univariate analysis.
So, those included, as we just said, sample size, funding source, single versus multicenter, and developed versus developing countries. In terms of their primary outcome, 21.6% reported a statistically significant large treatment effect, of which 14.4% did so specifically for the outcome of live birth. There was no difference in the reporting of large statistically significant differences on univariate or multivariable analysis based on registration studies.
So, this is a negative study, basically telling us, to some extent, that unregistered trials are not less likely to be negative studies. Sensitivity analyses and subset analyses were undertaken by publication year, pre versus post-2018, and by sample size. The only comparisons that yielded statistically significant differences was that among small trials, meaning those less than or equal to 200 participants, and trials published prior to 2018, retrospectively registered trials were less likely to report a large treatment effect.
So, to some extent, a little bit surprising. You might think that, actually, those that are prospectively registered are more likely to report their data, whether it was negative or positive. I think this is a really good paper.
I think it's reassuring. I 100% agree with their conclusions that, especially when we have so few high-quality trials from which to draw conclusions, that these trials should not be excluded from systematic reviews and meta-analyses. That said, I'm not sure we can totally close the book on whether there's bias based on the status of registration based on this study alone.
The authors mentioned astutely that they chose this outcome as opposed to whether it was just a negative or a positive study, meaning did they find a statistically significant difference or not, because of the fact that that outcome would be heavily influenced by power, sample size, et cetera, which may also very much be associated with registration status. I would have liked to see that analysis as well. That's not an analysis that was included at all.
It's not one that I can do myself, looking at the raw data provided here. I think it's also very, very important in terms of assessing this. Of note, only 0.8% of the trials reported had a small statistically significant treatment effect.
Also not surprising, given that means that they had a really, really big sample. But 73% reported no statistically significant differences. I'd like to understand how that breaks down based on registration status.
Again, I think a great paper. Congratulations to the authors. Reassuring overall.
Again, please do register your trials. For the record, fertility and sterility will not accept a trial if it's not registered. It has to be prospectively registered for it to be published.
Yeah, Kate, I actually expected the exact opposite finding, especially coming from Ben Mole and the group of world-leading people thinking about bias in our study. I was surprised by this. But as I think about it more, I think it's because they chose the large effect size as their decision-making, not a small one.
I think not registering a trial, if it's going to have small biases, isn't going to make it so big that you're going to get outside the 0.8 or the 1.25 of the point estimate. That's why it found the null. I think we can at least conclude that not prospectively registering doesn't have an influence that's that big.
But I 100% agree with you. I can't say that we can say from these data that it doesn't have a small influence. Unfortunately, I think most clinicians, most of us don't think about the effect size.
Was it statistically significant? I'm going to implement it if there's a benefit to it. Even if the number needed to treat might be 200, we don't think about that number needed to treat and what that cost is. We just hear the p-value and go with it.
I think your nuance to that finding is spot on and was my take-home as well. Yeah, I agree. I was actually surprised that the conclusion was to not exclude unregistered trials.
When I first read the title, I thought for sure they're going to be on a soapbox, that all trials need to be registered. I thought the exact same thing. Yeah, we see a lot of bias in these unregistered trials and we shouldn't be using them.
But I think there'll probably be a historical time in our field where trials were unregistered. My hope is that moving forward, more and more trials are going to be registered and that if this paper were to be redone in 10 years, they would have the opposite conclusions that we should be excluding trials that aren't registered. I'm always a little bit suspect of those that don't register.
Yeah, I think there's another analysis to be done in this same data set as well as to the trials that were registered, whether it's prospectively or retrospectively, what's the proportion of trials that had a deviation between their a priori analysis plan and trial design and that they ultimately reported. To me, that's fascinating, having done multiple trials. I always say, you don't do the trials, the trials do you.
Sometimes changes are called for and it's really a judgment call as to whether it's going to affect the validity of your results and when it's appropriate. Obviously, there's a lot in the literature about this, but I think they did a great job amassing a very good data set and that would be another interesting way to look at it. I am disappointed with the amount of detail that clinical trials registry require because it does allow people to sort of play in the margins.
We wrote a letter to the editor to HR a few years ago. The trial was looking at oocyte maturity and their intervention actually had worse oocyte maturity and that was what they said their outcome was, but what they reported was the percentage of patients that had more than 60 mature oocytes and that was better with their intervention, which is statistically really weird. It means overall it was worse, but there had to be a bunch of patients that were just above and just below that margin of 60 percent and they reported it as a positive trial and that to me just felt like endpoint switching, but because clinical trials doesn't require you to be very precise with what you put in there, it allows authors to play in the margin and again gets back to the theme, null is okay.
You don't have to force it to fit a box that it's positive to get it published. I would actually highly recommend that we don't do that. The null is good.
Kate, we're staying with you. We're moving on to research letters now. One of our favorite areas of the journal.
Thanks, Micah. Yeah, this was a study that I was very excited to read as a hypothesis generating study, perfect for a research letter. It is entitled Atosaban as a Potential Treatment for Endometriosis-Related Pain.
Results from the Endoban pilot study and this is by first author Mark E. Shupp and senior author Theodore E. Nibor and colleagues. This is a really interesting study. Atosaban is a competitive antagonist of both oxytocin receptor and vasopressin receptor.
It decreases intracellular calcium, thereby reducing uterine contractility. It has been looked at for a number of different indications. It actually has EMA, European Medical Agency approval in Europe as a tocolytic, so as a medication hopefully to treat preterm labor, but it has no regulatory clearance in the United States.
This group has also looked at it to evaluate it for improving embryo transfer outcomes based on the hypothesis that if there were to be any type of cramping during embryo transfer, that might reduce the chances of success. Here though, they are looking at a population of patients who have either surgically or imaging proved, which I think reasonable people can have a whole long discussion about what does it mean to have imaging proved, but endometriosis, aged 18 to 45 years old, who also report a pain score of 5 to 10 for their dysmenorrhea. These patients were administered IV Atosaban on what they reported was the most painful day of their period over a time period of six hours.
There is no control group here. It is a pilot study. There are 11 patients.
What they did was they looked at subjective endpoints with the primary endpoint being what was their numeric pain score at time zero and then going out to time six, six hours after the initiation of the infusion of the Atosaban. This is a very small study. They did find a statistically significant reduction in pain scores, but it is obviously, as I said, hypothesis generating in such a small group and kind of hard to pin down such a subjective outcome in terms of statistical significance, I would say.
They also looked at secondary endpoints in terms of subjective rates of bleeding, as well as their pain over the course of their entire menstrual cycle that followed. I think it is a really interesting study. I think, obviously, this is not a practical drug to have to administer as an IV infusion over the period of six hours.
An alternative delivery mechanism would need to be created in order to make this useful for our patients. That said, we are really still at a terrible loss to help patients with the horrible pain that comes with endometriosis and adenomyosis. I am, again, really happy to see this paper in Fertility and Sterility.
It is not an ART paper, but it is looking at something that is really central to our work in reproductive endocrinology and the treatment goals of our patients, which is to improve their quality of life and to help with this really horrific pathophysiology. I do think there is a little bit to be desired here in terms of the biologic plausibility of this having an across-the-menstrual-cycle positive benefit. Looking at the pharmacodynamics of this drug, it has an 18-minute half-life and no effect beyond 12 hours.
That really could be even hypothesized based on what we know about it. I think that bit is a little bit of a stretch. I think investigators who are looking for novel ways to reduce pain among patients with endometriosis are to be commended.
I think this one has a long way to go before clinical applicability, but I found it to be a really interesting study. I agree. I was also struck by the half-life of it being so short.
Pharmacodynamically, half-life times four is when it is exiting your system. Yes, it may provide some pain relief in the immediate time points surrounding the infusion, but I agree with you. The only thing that I can think about is the old saying, pain begets pain, that maybe you start this pain spiral and that activates some neurotransmitters and you get into more of a pain circuit.
If you stop that in the beginning, then it may subsequently decrease pain signaling down the road. I am certainly no expert in that arena, but I was trying to stretch my brain to understand how could an infusion on day two of the menstrual cycle help a week or two later with periopulatory pain or luteal phase pain. I wish they had carried on their analysis beyond just the six hours of the infusion.
Even if you check them the next day for a week and just have them write down their scores, that would have been incredibly informative to the point that both of you are getting at. Yes, or to have a control group. With 11 patients, just to have somebody get saline, 11 patients get saline, that would have really helped a lot in terms of our ability to interpret these data.
I agree. To that point, most phase two trials do have a control group and most have one to 300 patients. This had 12.
Interesting data. I agree with you, Kate. I'm glad it's in FNS.
I 100% think it's just hypothesis generating and confirming that they can do more research. The delivery mechanism, prolonging the half-life, can we make it oral, all of those things that the pharmaceutical community works on, I think will be huge to see if this actually has benefit. As you said, probably only during certain parts of the menstrual cycle as far as benefit, but could be huge.
Eve, we're staying on research letters. Tell us on the last one we have today. Yeah, this is a nice one.
The title of this is, Age May Be a More Important Determinant of Oocyte Yield Than AMH. This came out of the MGH group with first author Evelyn Minnis and senior author Shruthi Mahalankaya. As everyone knows, egg freezing or oocyte cryopreservation has grown substantially since ASRM removed the experimental label in 2012.
But recently, there have been reports of patient regret following elective oocyte cryopreservation. I think that's driven primarily by the high cost of the procedure and then perceived inadequate counseling regarding expected outcomes. So often, I think patients have these unrealistic outcomes that they are going to get all of the eggs that they need from a single cycle.
Mike and Kate, I think there was a great paper that came out of the Shady Grove group, looking at the actual number of cycles that patients needed to get to their 15 to 20 eggs, and most patients couldn't do it in a single cycle. So currently, we all use AMH as a primary marker of ovarian reserve, but oocyte yield does not always match AMH predicted response. Prior data from oocyte donation cycles suggests that age modifies this relationship between AMH and oocyte yield.
So donors who are age 30 to 35 have been found to have lower yields compared to younger donors with the same AMH level. So it's really both age and AMH. And so the objective of this study was to assess the effective AMH on oocyte yield across different age groups of patients undergoing planned oocyte cycles.
This was a retrospective single center cohort study from MGH, and it was over a five-year time period, and they had 519 patients undergoing planned oocyte cycles. They stratified their patients into five groups at the time of retrieval, less than 30, 30 to less than 35, 35 to less than 38, 38 to less than 40, and then greater than or equal to 40. And in their cycles, about 65% used antagonist, and they stated that gonadotropins were dosed conservatively.
Their mean peak estradiol levels were 2,300 to 2,600 across all groups. So primary outcomes were the total number of eggs retrieved and then the number of mature eggs retrieved. And they used a beta coefficient to calculate the, to assess the effect of AMH on total and mature egg counts.
And a beta coefficient is a measure of how much a variable changes in response to a one unit change in a predictor value variable when you hold all of the other predictors constant. So what they found, the mean AMH level ranged from 2.3 nanograms per ml. And I'm going to give a shout out to our Australian colleagues who have asked us to use picomol per liter in our, in our podcast, since we have a large cohort of Australian listeners.
So for our colleagues, that's 16.4 picomol per liter. And then in the 38 to less than 40 age group, AMH was 3.6 nanograms per ml, or 25.7 picomols per liter in this group. And then the mean number of mature eggs, not surprisingly, decreased for each advancing age group.
So in those who are less than 30, there was a mean number of 14 eggs retrieved, 10.6 were mature. And then in those greater than or equal to 40, there was a mean 10.7 eggs retrieved and 7.8 who were mature. So their analysis did not show a correlation between AMH and the total number of eggs retrieved in the under 30 population.
But they did see an association that was positive in the 30 to less than 35 age group with an increasing beta coefficient. And then also in the 35 to less than 38 age group and the 38 to less than 40 age group, they saw a strong correlation between AMH and the total number of eggs retrieved. And then again, in the oldest age group, the total number of eggs retrieved increased with higher AMH, but that result was attenuated.
And so overall, the authors conclude that AMH was not associated with egg yield in patients under 30. And this suggests that for younger patients, ovarian reserve is measured by AMH is not the key driver of response. So it's kind of like a dose response relationship.
And then in patients age 30 to 40, AMH was significantly associated with oocyte yield with this effect peaking in that 38 to 40 age group. And then again, in patients over 40, AMH oocyte relationship was significant, but was attenuated compared to that 38 to 40 cohort. So age itself, and you guys know I love to say age is queen, age had a significant independent inverse effect on oocyte yield across the cohort.
And the authors conclude that AMH alone is insufficient for counseling patients about expected oocyte yield, that you have to look at age alongside AMH. And I think overall, it's a nice research letter. It is somewhat limited as the outcomes are just oocyte yield with no data on fertilization, blastocyst formation, euploid rates or live birth rates, but I think it's a perfect research letter.
It's a well-done study. It's a nice nugget of information and very useful for counseling, especially in the older population. Even if your AMH is the same as a 28-year-old, your response to stimulation will not be.
So I really liked it. And I think it just adds to the literature there. Our group actually has some data that we're presenting at ESHRA that takes it one step further.
We looked at live birth rates and we looked at AMH actually as a predictor of live birth rate. And in the older population, we did see that AMH can be a predictor for live birth rate, but it does not tell you that you should not be stimulating. So it can be positive that there was a positive relationship, but you can't use AMH to exclude people from stimulation.
I 100% agree with that last point. I actually don't believe in the biologic plausibility that AMH is not predictive in women over 30. I would like someone to explain to me biologically how AMH does not predict how many eggs you have.
It's a measure of granulosa cell mass, which is a marker of how many follicles you have, which is a marker of how many eggs you're going to get. They had 26 patients in that group. And I just think including from 26 patients that there's not that interaction between AMH and how many eggs you're going to get, I just don't buy it.
You mean in the younger population? I think it's sort of that you can do anything in a younger population and they're going to respond well. So an AMH of 10 and an AMH of 1 is going to be the same in a 25-year-old? You believe that? That's what they're saying. Well, I think with the caveat that if you dose conservatively, that's what's going on here.
And they don't give us the dosing in the table. I think if they gave us the dosing in the table and they didn't adjust for the dosing, so if you don't tell us what the dosing was and you're dosing conservatively in young women and you don't adjust for it, sure, this is what you would find. But I don't believe this tells us AMH in women under 30 is not predictive of egg yield.
Yeah, I don't. And our data that we're presenting shows a linear relationship with AMH and with live birth, showing that especially in the older population over 40, that your AMH is actually predictive of live birth because you can get eggs. And so if you can't get eggs from somebody, you're not going to have a baby.
But again, there's no lower limit where below which we didn't see live birth. Sure. I 100% agree with that, but that also makes biologic sense.
In a young woman, 75% of her embryos are going to be euploid. So whether you stem hard and get 15 or stem harder and get 25, you're still going to have a euploid embryo on your first transfer. If you're 42 and you're going to get 10 embryos versus one embryo, then it makes a huge difference in their chance because only 25% will be euploid.
So to me, that completely makes biologic sense what your data shows. But it's interesting because most papers say that AMH does not predict the likelihood of live birth. And so I actually disagree with that, which is why we looked in that.
Yeah, I mean, the devil's in the details in all of those studies, I would say. But to me, the most interesting thing about this paper is that it calls into question, what is the right dose to get as many eggs as we possibly get from any particular patient? And that's something that we still don't know. That's super hard to answer.
What's the ROI per unit of gonadotropin? I know it's something we've talked about on this podcast before, but I fall asleep sometimes at night wondering about the perfect study design for that one. Yeah, I mean, I've spent a lot of time looking at whether or not AI can help us to predict that. And no, because it's dependent on operator scale.
Are you getting every egg? What's the BMI of the patient? There are a lot of variables that we can't predict. But I do think this idea of at both ends of the dose response curve, you're going to see flattening of it where, and this is where the idea of like mini stim comes in for patients who have incredibly low ovarian reserve, that a higher dose of gonadotropins is not going to yield a better response than a lower dose of gonadotropins. And it's probably true at the low ends of ovarian reserve, you're not going to get a lot of eggs no matter what you do.
I do agree though, at the higher end of ovarian reserve, if you really blast those women with gonadotropins, those are the patients where you can get 50, 60 eggs from those patients. Whether or not that's a good strategy is a whole other debate, but we certainly are getting upwards of 30 eggs in many of our young patients doing planned OC cycles with a higher starting dose of gonadotropins. So I think it was a nicely done study, definitely appropriate as a research letter.
And I think it just reminds us that age is queen, as I always say. Age is absolutely queen. I don't think there's any doubt about that.
Yeah, it drives over 90% of all prediction models I've seen. It's definitely the biggest thing. All right.
Well, shout out to Paul Linn. You say you want us to be a little more argumentative on it. So I tried to bring a different perspective to what he was saying, but I think at the end we agree.
It's just, I think AMH does matter. I don't know a biologic plausibility of why it wouldn't. So as always, we have a bunch of other letters and articles in the journal.
Just so people know, what we try to do is have someone who's not on the podcast assign the articles, pick them, and choose who presents it. So I presented one of my own past fellows articles, but I didn't choose that article or choose to present it. So we try to reduce bias in which articles we talk about by doing it that way.
So I encourage you to read the entire journal, even Kate, as always. It was great seeing you. Shout out, we will be at ESHRA, all of us, as well as Kurt, in about two to three weeks.
So look for two or three podcasts from the best of what is being presented at ESHRA. Even Kate, I'm excited to see you there. Excited to see you, too, and all our listeners.
And don't forget to get your UK visa. We'll see you there. Yeah, we'll see you in London.
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.
