A machine learning–based prediction model for early pregnancy loss after IVF/ICSI: a retrospective cohort study

Abstract

We aimed to develop machine learning (ML) models using routine clinical features to predict early pregnancy loss (EPL) after IVF/ICSI and implement an online risk calculator. This retrospective study included 1,201 couples undergoing their first IVF/ICSI cycle resulted in ultrasound-confirmed intrauterine clinical pregnancy. Data were randomly split into a training cohort (70%) and an internal validation cohort (30%). Feature selection integrated univariable, Boruta, and LASSO regressions. Eight ML algorithms were compared, utilizing SHapley Additive exPlanations (SHAP) to interpret the optimal model. Outcomes included 979 live births and 222 EPLs. Five core predictors emerged: female age, body mass index (BMI), basal follicle-stimulating hormone (FSH), platelet count (PLT), and thyroid-stimulating hormone (TSH). Among the evaluated models, XGBoost showed the highest AUC and a relatively low Brier score. SHAP analysis ranked FSH as the largest contributor, followed by BMI, TSH, age, and PLT. A web-based calculator was developed to facilitate implementation. An XGBoost model based on five routinely available indicators (age, BMI, FSH, PLT, and TSH) may assist in estimating EPL risk after IVF/ICSI. The model is intended to support risk stratification and counseling and requires external validation before broader use. Not applicable.