MACHINE LEARNING model has demonstrated potential to stratify miscarriage risk in women with polycystic ovary syndrome (PCOS) who achieve clinical pregnancy following fresh in vitro fertilisation (IVF), although researchers have cautioned against using it to guide treatment decisions.
Women with PCOS who become pregnant after fresh IVF or intracytoplasmic sperm injection (ICSI) face an elevated risk of miscarriage. However, predicting individual risk after pregnancy confirmation, but before the outcome is known, remains challenging.
The multicentre study developed and validated a machine learning model using data from 8,911 clinical pregnancy IVF/ICSI cycles. The findings suggest that routinely collected clinical variables could help identify women at relatively higher risk and support more individualised counselling.
Researchers compared 13 machine learning and traditional statistical algorithms using data from 7,241 clinical pregnancy cycles, including 973 miscarriages. The model was then tested unchanged in an external validation cohort of 1,170 cycles and a prospective temporal validation cohort of 500 cycles.
During internal cross-validation, the model achieved an area under the receiver operating characteristic curve (AUC) of 0.885, indicating strong discrimination between pregnancies that ended in miscarriage and those that did not.
However, the model’s performance declined in the validation cohorts. Its AUC was 0.787 in external validation and 0.763 in prospective temporal validation, indicating reduced ability to distinguish between outcomes when applied to new populations and later clinical data.
Shapley additive explanations analysis identified anti-Müllerian hormone (AMH), endometrial thickness on the day of human chorionic gonadotropin administration, and basal luteinising hormone as the leading predictors of miscarriage risk.
Miscarriage rates were broadly consistent across the development, external validation, and prospective temporal validation cohorts, at 13.4%, 12.5%, and 12.4%, respectively.
Risk-quartile stratification using thresholds established in the development cohort consistently enriched for miscarriage events across both validation groups. However, positive predictive values remained modest, meaning the model could not reliably identify which individual pregnancies would end in miscarriage.
The researchers concluded that the locked model could support miscarriage risk stratification and counselling for women with PCOS following fresh IVF/ICSI.
However, calibration drift and reduced discrimination across cohorts limit its immediate clinical application. The model should therefore be used as a risk-ranking tool rather than for diagnosis or treatment selection.
The authors emphasised that local calibration and ongoing performance monitoring would be essential before real-world deployment. Further validation will be important to establish whether the approach can provide reliable, clinically useful risk estimates across different fertility care settings.
Reference
Tong J et al. Miscarriage risk prediction for women with polycystic ovary syndrome who achieve clinical pregnancy after fresh embryo transfer: multicenter development, geographic external validation, and prospective temporal validation study. J Med Internet Res. 2026;28:e101829.
Featured Image: Inna Doctor on Adobe stock
Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.
