Novel Clinically Validated Machine Learning Model for Early Pregnancy Loss in Recurrent Spontaneous Abortion:...

To explore the correlation between autoantibodies, ultrasonic endometrial receptivity parameters and early miscarriage in recurrent spontaneous abortion (RSA) patients during subsequent pregnancies, and to establish and validate a predictive model for early miscarriage.

A retrospective analysis was conducted on RSA patients who visited Xi'an People's Hospital from January 2019 to December 2024. Patients were randomly divided into a training set (70%, n=412) and a validation set (30%, n=177). Baseline data, serum autoantibodies (anti-β2-glycoprotein 1 antibody [aβ2-GP1], thyroglobulin antibody [TgAb], anti-sperm antibody [AsAb], anti-cardiolipin antibody [ACA]) and ultrasonic parameters (resistance index [RI], endometrial thickness, endometrial volume, vascularization index [VI], vascularization flow index [VFI]) were collected. Multiple machine learning models (logistic regression [LR], XGBoost, random forest, etc.) were developed. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, and other metrics. A nomogram was constructed based on the optimal model.

The abortion subgroup had significantly higher positive rates of aβ2-GP1, TgAb, ACA and RI, but lower endometrial thickness, endometrial volume, VI and VFI than the normal subgroup (all P<0.05). Eight variables (aβ2-GP1, TgAb, AsAb, RI, endometrial thickness, endometrial volume, VI, VFI) were identified as candidate predictors. The LR model was optimal, with AUC=0.94 and accuracy=0.93 in the training set, and AUC=0.92 and accuracy=0.90 in the validation set. The nomogram based on this model showed good alignment between predicted probabilities and actual outcomes.

A practical and accurate LR model for predicting early miscarriage in RSA patients was established using autoantibodies and ultrasonic parameters. It can assist in clinical risk stratification and individualized intervention. Future multicenter prospective studies with larger samples and more variables are needed to optimize the model.