.png){kind=icon}
.png){kind=icon}
Return
Prediction of live birth using machine learning based on β-hCG and embryo quality in patients with unexplained recurrent spontaneous abortion undergoing PGT-A
.png){kind=icon}
OA
AI
J
X
Y
B
H
C
X
J
E
X
DOI：10.3389/fendo.2026.1898690.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
PurposeThis study aimed to identify key predictors of live birth and to develop a machine learning prediction model for women with unexplained recurrent spontaneous abortion (URSA) undergoing preimplantation genetic testing for aneuploidy (PGT-A).MethodsThis retrospective study analyzed 127 patients with URSA who underwent PGT-A. Patients underwent frozen-thawed embryo transfer (FET) following either a hormone replacement cycle (HRT) or a down-regulated HRT. Data were collected on clinical baseline characteristics; endometrial thickness; embryo quality; medication use; and pregnancy outcomes. Binary logistic regression was used to assess the correlation between these factors and live birth.ResultsAmong the 127 patients with URSA; 79 achieved live births and 48 did not. Through comparative and regression analyses; the study identified key predictive variables; providing a reference for evaluating pregnancy outcomes. There were no significant baseline differences between the live birth and non-live birth groups in terms of age; Anti-Müllerian Hormone (AMH); Body Mass Index (BMI); duration of infertility; thyroid-stimulating hormone (TSH); erythrocyte sedimentation rate (ESR); endometrial thickness (EMT); embryo quality; or post-transfer medication (all P>0.05). Univariate regression analysis showed that β-human chorionic gonadotropin (β-hCG) levels on day 14 post-transfer was significantly positively correlated with live birth (P<0.001). Multivariate analysis further identified the day 14 β-hCG level post-transfer and transferred embryo quality as key variables. The constructed random forest machine learning model demonstrated strong predictive performance [Area Under the Curve (AUC)=0.917; 95%CI 0.778-1.000]. The day 14 β-hCG level cut-off for predicting live birth was 457.75 mIU/mL; with an optimal range of 828.69-3571.77 mIU/mL. SHAP analysis showed that high-quality embryos increased the chances of live birth; while low-quality embryos decreased it. Different medications or luteal support regimens did not significantly affect live birth rates.ConclusionsFor URSA patients undergoing PGT-A with euploid embryo transfer; day 14 β-hCG levels and embryo quality were key predictors of live birth. Empirical interventions and luteal support type had no significant impact. The random forest model can accurately predict live births; aiding personalized clinical management and minimizing unnecessary treatments.
Keywords:
machine learning
prediction model
immunosuppressant
live birth
PGT-A
URSA
Journal
IF:
4.6
Papers:
1.9W
Citations:
6.4W
Organization
Cited Papers