(123) Predictive Model for Adverse Pregnancy Outcomes in Patients With Recurrent Miscarriage Based on Back Propagation Neural Network

First Affiliated Hospital (Army Medical University)

Abstract Text:

Objective  To explore the related factors for the failure of recurrent pregnancy in patients with unexplained recurrent spontaneous abortion (URSA), and to construct a prediction model based on the back propagation neural network.

Methods  Patients with URSA admitted to our hospital from January 2021 to December 2023 were selected as the research objects. All URSA patients were followed up until the termination of pregnancy and divided into the successful pregnancy group and the failed pregnancy group according to the pregnancy outcome. Univariate and multivariate logistic regression analyses were performed on the baseline data of this population to screen out the risk factors that might lead to the failure of recurrent pregnancy in URSA patients, and a prediction model was constructed based on the back propagation neural network.

Results  Among all URSA patients, those with pregnancy failure accounted for approximately 67.73%. Multivariate analysis showed that advanced age (OR = 1.297, 95%CI: 1.139 - 1.455, P = 0.013), multiple abortion times (OR = 5.954, 95%CI: 3.339 - 7.855, P < 0.001), abnormal autoimmune factors (OR = 3.154, 95%CI: 1.339 - 6.155, P = 0.023), and impaired fasting glucose (OR = 3.954, 95%CI: 2.639 - 7.155, P = 0.031) were independent predictors for the failure of recurrent pregnancy in URSA patients. A back propagation neural network model was constructed based on these factors, with an area under the curve (AUC) of 0.871 (95%CI: 0.817 - 0.896), and the sensitivity, specificity and accuracy were 77.8%, 87.2% and 83.6%, respectively.

 Conclusion  The machine learning model constructed based on the back propagation neural network has certain value in predicting the adverse outcomes of recurrent pregnancy in patients with recurrent spontaneous abortion.