An analysis of 4,300 euploid embryo-transfer cycles identified important variables including age, AMH, endometrial thickness and previous treatment cycles.
Machine-learning models achieved an AUC of up to 0.778, suggesting a possible role in more individualized fertility care.
What deserves the closest attention when estimating the chances of frozen embryo transfer: a woman’s age, the thickness of her uterine lining, or the condition of the embryo after thawing? These are familiar questions for people preparing for IVF.
Pregnancy does not depend on any one condition. Treatment can fail despite younger age and an adequately thick lining, while pregnancy can occur under less favorable circumstances. Even transferring an embryo with a normal chromosome count does not guarantee pregnancy.
A recent study used artificial intelligence to predict pregnancy after frozen embryo transfer. By considering clinical information together—including age, ovarian reserve, endometrial condition and embryo quality—the researchers identified seven important variables associated with pregnancy.
The study, conducted by researchers at Istanbul Sisli Memorial Hospital and Mimar Sinan Fine Arts University in Türkiye, appeared in the October 2026 issue of Reproductive BioMedicine Online.
Seven important variables in 4,300 frozen embryo-transfer cycles
The researchers retrospectively analyzed 4,300 frozen embryo-transfer cycles performed at Sisli Memorial Hospital between October 2011 and February 2023. Each involved the transfer of one embryo confirmed by preimplantation genetic testing to have a normal chromosome count.
Restricting the study to euploid embryos was an important feature. It allowed the team to examine how patients’ clinical circumstances related to pregnancy while reducing the influence of embryo chromosome abnormalities.
The team applied five machine-learning algorithms to 26 clinical variables. Seven emerged as important for predicting clinical pregnancy: the number of previous IVF cycles, anti-Müllerian hormone (AMH), endometrial thickness, embryo quality after thawing, female age, the number of frozen embryos and the method used to prepare the endometrium.
These variables are useful as information describing the overall reproductive and treatment context, rather than as independent determinants of success.
Why embryo quality after thawing matters even with euploid embryos
One notable finding was the inclusion of post-thaw embryo quality among the important predictors.
Chromosome abnormalities are an important cause of implantation failure and miscarriage in IVF. However, selecting euploid embryos through preimplantation genetic testing does not ensure pregnancy.
The findings indicate that the embryo’s appearance after thawing may provide useful predictive information alongside chromosome status.
The number of previous IVF cycles was also important. More previous treatment was associated with a lower probability of clinical pregnancy. This does not mean that previous failures themselves reduce the chance of pregnancy; patients undergoing repeated treatment may also have other unfavorable clinical characteristics.
AMH and endometrial thickness cannot determine success on their own
AMH is a commonly used marker of ovarian reserve. It helps estimate response to ovarian stimulation and the number of eggs that might be retrieved, but cannot by itself determine embryo quality or pregnancy success.
In this study, AMH contributed to prediction alongside other clinical variables. The finding should not be interpreted to mean that higher AMH necessarily produces a higher chance of pregnancy.
Endometrial thickness was another important variable. The lining provides the environment for implantation, but a thicker lining is not automatically more favorable. Its condition, the hormonal environment and the embryo’s developmental stage all matter.
The method of preparing the endometrium also contributed to prediction. Frozen embryo transfers may use natural cycles or hormone-prepared cycles, depending on ovulation and the treatment plan.
The study did not establish that one preparation method is superior. It identified predictive variables; treatment selection still requires individual assessment and clinical judgment.
An AUC of 0.778 is not a 77.8% pregnancy rate
The five models achieved AUC values of 0.760–0.778. XGBoost produced the highest value, 0.778.
AUC measures how well a model distinguishes cases with and without the outcome—in this case clinical pregnancy. A value of 0.5 corresponds to chance-level discrimination, while values closer to 1 indicate better discrimination.
An AUC of 0.778 must therefore not be read as a 77.8% pregnancy rate. Classification accuracy for models using the seven selected variables was approximately 70.2–72.6%.
The models showed some ability to predict outcomes, but could not determine with certainty whether an individual patient would become pregnant.
The researchers also noted the limitation of using retrospective data from a single center. Additional validation is needed to establish whether performance is similar in other hospitals and patient populations.
AI as a possible aid to individualized fertility care
The study illustrates one potential use of AI in fertility treatment.
Clinicians already consider age, ovarian function, the uterine lining, embryo quality and previous treatment when planning a transfer. AI may help analyze these interrelated details together and identify patterns that are difficult to detect in existing records.
It may be particularly useful for understanding why outcomes differ even after transfer of euploid embryos, linking outcomes that chromosome status alone cannot explain with patients’ wider clinical circumstances.
Predicting pregnancy and improving pregnancy rates are different tasks, however. Further validation and prospective research are needed to determine whether using a model in clinical practice improves outcomes.
The outcome predicted here was clinical pregnancy, not live birth. The study does not show that the models accurately predict subsequent miscarriage or birth outcomes.
The central message is that the chances of frozen embryo transfer cannot be explained by one factor such as age, chromosome status or lining thickness.
If adequately validated before clinical implementation, AI models could become useful supporting tools for assessing patients more precisely and helping clinicians develop individual treatment strategies.
This article is based on a machine-learning study of factors associated with clinical pregnancy in 4,300 frozen-thawed single euploid embryo-transfer cycles, by researchers at Istanbul Sisli Memorial Hospital and Mimar Sinan Fine Arts University, Türkiye. It was published online on July 4, 2026, and in the October issue of Reproductive BioMedicine Online, volume 53, issue 4, article 105864. DOI: 10.1016/j.rbmo.2026.105864. PMID: 42566879.
The accompanying image was created using generative AI (ChatGPT, OpenAI); any people depicted are fictional.
Related guide: Understanding implantation failure after euploid embryo transfer
Copyright © 2026 Fertility Information TokTok Fertility News | THE FERTILITY NOTES. All rights reserved. Unauthorized reproduction, copying, redistribution, scraping, AI training, and other data use of any articles, photographs, images, or other content are strictly prohibited. Violations may result in legal liability.