THE BUSINESS OF MEDICINE
Is predictive modeling in health care a house of cards?
Akhil Vaid, MBBS
This is exactly the point Akhil Vaid, MBBS, Assistant Professor in the Windreich Department of Artificial Intelligence and Human Health Medicine and the Department of Medicine at Icahn School of Medicine at Mount Sinai in NYC, recently made under the same heading in a Nature commentary article (1).
This has also been an issue investigators at the Center for Human Reproduction (CHR) have been keenly interested in for several years, as they've shared the same concerns. While medical journals likely experienced a peak in modeling studies during the COVID-19 pandemic—when they suddenly appeared everywhere and in every medical specialty—things have somewhat calmed down since. What computers allow you to do from home these days is truly amazing!
Therefore, it wouldn't be surprising if progress with machine learning reignites this discussion. In his article, Vaid indeed noted that "uncontrolled deployment of machine learning in medicine can distort patient information and sacrifice long-term data reliability for short-term benefits." He specifically highlights the many difficulties that may arise from how different models influence each other, or how new models built upon older ones may become misdirected. It's a very interesting article!
Reference
1. Vaid A. Nature 2025;642:864-866
Is Uterine transplantation cost-effective?
Some uterine transplant surgeons are genuinely claiming this, and—forgive us here at CHR—but we simply cannot accept this notion. If you consider for a moment what uterine transplantation involves (aside from IVF and other infertility treatments)—two major surgeries for the patient (transplantation and, at a later point, removal of the transplanted uterus), one major surgery for the uterine donor, the need for anti-rejection drugs, and the inherent failure rate of organ transplants—then as huge a scientific, technical, and momentary clinical accomplishment as every successful uterine transplant represents, it simultaneously also represents, at least as we see it, an abuse of basic logic and some of the most fundamental rules of medicine, including "first, do no harm."
The recent July issue of Fertility and Sterility, roughly at the 10-year anniversary of the first live birth after a uterine transplant in Sweden, almost appeared to be a special promotional issue for uterine transplantation. It contained several papers on the subject, all more or less asserting how wonderful a treatment uterine transplantation has become.
One "Reflection" article among these (for unclear reasons, F&S differentiates "Reflections" from "Inklings") was indeed titled, "From milestone to mainstream: the case for uterine transplantation." It included this rather astonishing comment: "Once viewed as a highly experimental procedure, uterine transplantation has evolved into a reproducible, increasingly safe (and when will it be really safe, we might ask?), and clinically meaningful option (we, of course, would also love to get a better understanding of what constitutes a clinically meaningful treatment option) for women with 'absolute uterine factor infertility' (one also wonders which talented marketer coined this obviously brilliant term for what medicine for decades called an 'absent uterus')" (1).
The author is, of course, one of the uterine transplant surgeons at the Cleveland Clinic, which has been one of the principal purveyors of uterine transplantation surgeries in the U.S. and, naturally, has no self-interest in promoting uterine transplants. This non-self-promotion was reportedly aided by ChatGPT-4.0 (OpenAI) in "rephrasing sentences and refining language for clarity and flow."
And then there were, lo and behold, two alleged cost-effectiveness analyses regarding uterine transplantations in the F&S issue. Both compared the costs of uterine transplants to the recruitment of a gestational carrier (GC), and both were the products of several academic centers, including economists.
The first study concluded that a singleton birth using a GC cost $97,712.90, which strikes us as a bizarrely low number. It's very difficult to believe that the same birth after uterine transplant would supposedly cost only $116,137.20 (seriously, they reported the cents), also in our opinion a completely unrealistic number in the U.S. Quality-adjusted life years were also higher with GC (0.93) than with transplant, as were live birth rates (94% vs. 77%). A so-called Monte Carlo simulation (i.e., a model) demonstrated that a transplant had a 37% chance of being cost-effective. The cost advantage flipped toward transplant with two births, but the overall rate of live birth remained higher with GC deliveries (86% vs. 66%). In short, the study suggested that for a single birth, GC was preferable, while for two children, a transplant was more cost-effective (2).
The second study, also by a U.S. consortium, offered what appeared to be much more realistic numbers. Using another modeling product (a decision-tree mathematical model), they demonstrated that a uterine transplant was $1.4 million more expensive than a GC, with, in addition, lower utility by 23.74 quality-adjusted life years using the same number of children born per every two FET cycles. After 10,000 simulations, the GC arm had two children born 42% of the time, compared with only 17% with the transplant arm. No children were born in 56% of transplants and in only 16% of GC arms. The authors concluded that their model currently marks GC use as the clearly more cost-effective approach (3).
And then there was another "Reflections" article which, by all rules, really should have been an "Inkling" article, since it was the adnexal commentary to the two papers comparing GC pregnancies to transplant pregnancies (4). Interestingly, it was written by an infectious disease expert from the CDC (probably the only financial model expert the editors could locate) who basically said nothing of interest except for regurgitating the two papers she was supposed to comment on. And considering her lack of expertise in reproductive medicine, one also cannot blame her for a concluding sentence full of cliché: "Uterus transplant offers the existing possibility for women with absolute uterine factor (she obviously learned the lingo quickly) to bear their own children."
Really? We coincidentally, in the discussion of the preceding paper, noted our dislike for modeling. Every model is, of course, fully dependent on the assumptions it is fed. And we know how far off from reality so many of our assumptions frequently are in medicine. We, therefore, have very serious doubts about all the papers discussed here regarding uterine transplants, and indeed, even more severe doubts about the basic rationale of uterine transplantation—and not only because of cost concerns, but also because of safety concerns for the mother, offspring (nine months under maternal immune suppression), and donors of the required uteri. As it is, too many women already have hysterectomies!
And, yes, we almost forgot, in addition to the four papers we just discussed on uterine transplantation, the July issue of F&S also contained three video presentations (and abstracts), offering detailed descriptions for "standardized" robot-assisted living donor hysterectomy (5), recipient surgery (6), and back-table technique for uterus transplantation (7).
References
1. Richards EG. Fertil Steril 2025;124(1):60-61
2. Walter et al., Fertil Steril 2025;124(1):121-132
3. Combs et al., Fertil Steril 2025;124(1):134-143
4. Honeycutt AA. Fertil Steril 2025;124(1):54
5. Tamate et al., Fertil Steril 2025l124(1):161-163
6. Tamate et al., Fertil Steril 2025;124(1):164-165
7. Tamate et al., Fertil Steril 2025;124(1):167-168