(499) Predicting Sperm Retrieval Success in Secondary (Salvage) Testicular Sperm Extraction: A Machine Learning Perspective

Introduction Reported sperm retrieval rates (SRRs) following secondary TESE range from 12.5% to 74.7%. Given that secondary surgery may impose substantial economic and psychological burdens on couples desiring fertility, careful consideration is warranted when establishing the indication for salvage TESE. Additionally, knowledge of factors that increase or decrease the likelihood of a positive outcome following TESE may help physicians and patients determine the optimal timing for TESE. Objective This study aims to evaluate the predictive factors for the success of secondary testicular sperm extraction (TESE) by utilizing multi-institutional patient data and applying the advanced predictive capabilities of machine learning models. Methods A total of 503 men from couples suffering with infertility with a history of one unsuccessful TESE were retrospectively investigated in 2 groups regarding their surgical sperm retrieval status during the second TESE. Patient characteristics and preoperative laboratory findings were evaluated using machine learning (ML) models designed to predict the success of secondary TESE procedures. Ten different algorithms, such as Xtreme Gradient Boosting (XGB), Random Forest (RF), Gradient Boosting (GB), Decision Tree (DT), AdaBoost (AB), Logistic Regression (LR), Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), K-Nearest Neighbors (k-NN), and Naive Bayes (NB) were applied for model development. Overall accuracy, sensitivity, specificity, negative predictive value, positive predictive value, F1 score, Youden Index and the area under the ROC curve identified the best-performing model for predicting secondary TESE success. Results Sperm retrieval was successful in 211 patients (41.9%), while 292 patients (58.1%) experienced unsuccessful retrieval. Among the evaluated algorithms, XGBoost consistently demonstrated superior performance across all metrics (Figure 1). The most influential predictors of secondary TESE outcomes were body mass index prior to surgery, TESE location, preoperative LH level, and spermiogram volume. Testicular volumes (left and right) and duration of infertility also had a substantial impact on model predictions, whereas preoperative serum testosterone and FSH levels contributed moderately (Figure 2). Conclusions Integration of diverse patient characteristics into predictive machine learning models may improve patient counseling, guide surgical decision-making, and reduce unnecessary procedures in men undergoing secondary TESE. Disclosure No