Development and validation of a clinical prediction model for poor ovarian response in assisted reproductive...

The aim of this study was to develop and validate a predictive model for poor ovarian response (POR) in patients undergoing Assisted Reproductive Technology (ART).

This retrospective cohort study included 1,789 patients who underwent IVF/ICSI with a GnRH antagonist protocol at the reproductive medicine center of Shenyang Women's and Children's Hospital between January 2020 and December 2023. Data from January 2020 to April 2023 were used for model development, and data from May 2023 to December 2023 was used for validation. Clinical and ovarian reserve markers were collected. Four prediction models were developed and compared: (1) logistic regression with stepwise selection, (2) full-variable logistic regression, (3) LASSO regression with the minimum lambda (λ_min) criterion, and (4) LASSO regression with the 1-standard error (1-SE) rule. The optimal model was selected based on discrimination (area under the receiver operating characteristic curve, AUC), calibration, and Brier score.

The final model, based on stepwise logistic regression, identified AMH, basal FSH, BMI, and antral follicle count as significant predictors of POR. The model exhibited high discriminatory ability and good calibration across the training, internal validation, and test datasets. The nomogram developed from this model provides an easy-to-use tool for individualized risk prediction, enabling clinicians to optimize treatment strategies for patients at risk of POR.

This study provides a comprehensive predictive model for POR in ART, incorporating a wide range of clinical and ovarian reserve markers. The stepwise logistic regression model offers superior predictive accuracy and clinical utility, making it a valuable tool for personalized ART management. The nomogram developed enhances decision-making in ART by offering an intuitive, evidence-based method for predicting POR risk.