Machine learning prediction of clinical pregnancy in endometriosis patients following fresh IVF/ICSI-ET

Comprehensive Summary

Dr. Wan and friends developed and compared machine learning models to predict clinical pregnancy in endometriosis patients after IVF/ICSI-ET treatment. In a cohort of 1752 EM patients from 2014 to 2024, 1226 were first used to train six machine learning models on 24 clinical and embryonic predictor features. To identify independent features associated with pregnancy, researchers performed a multivariable logistic regression analysis with backward stepwise elimination. Notably, the XGBoost model demonstrated the best overall performance, specifically ranking first in accuracy, recall, F1 score, and AUC. SHAP analysis was performed to quantify each feature’s contribution to ML model predictions. Of the factors influencing clinical pregnancy, male age, number of normal fertilizations, and number of transferred embryos were identified as the three most significant independent predictors. This study provides a stepping stone to using ML algorithms to model nonlinear relationships representative of biological complexity. Still, there are mentions of potential selection bias, as this is a retrospective study.

Outcomes and Implications

Endometriosis (EM) is a chronic condition in which endometrium-like tissue grows outside the uterine cavity; this is often accompanied by severe pelvic pain. Endometriosis can cause many changes in the reproductive system, leading to infertility in 40-50% of EM patients. Assisted reproductive technology (ART) like IVF and ICSI can be extremely demanding and costly for the patient. This research provides a timely strategy to exploit the therapeutic window before fertility is compromised, in addition to providing realistic expectations and personalized treatment strategies. Machine learning is hypothesized to outperform current logistic regression models in both accuracy and personalization of treatment planning.