Development and validation predictive models of sperm retrieval for azoospermic men undergoing testicular sperm aspiration: a multicentre, retrospective, cohort study.

Purpose: Testicular sperm aspiration (TESA) is widely used to retrieve sperm from testis. Diagnostic testicular biopsy should not be routinely performed for azoospermia. Therefore, a good predictive model is needed before TESA.

Methods: A total of 1972 azoospermia patients constituted the modelling set, and 260 azoospermia patients from two other centres constituted the validation set. An integrated predictive model was built using logistic regression. Receiver operating characteristic (ROC), calibration and decision curve analyses were performed to evaluate the performance of follicle-stimulating hormone (FSH), semen volume, testicular volume and the integrated model.

Results: The FSH level was the best univariate predictor for successful sperm retrieval (SSR) and was better than semen volume and testicular volume alone (p<0.001, threshold 6.17 IU/L, modelling set area under receiver operating characteristic curve (AUC) 0.80, accuracy 0.79; validation set AUC 0.87, accuracy 0.78). The integrated predictive model had excellent accuracy for predicting SSR (modelling set: AUC 0.93, accuracy 0.89; validation set: AUC 0.96, accuracy: 0.89). Calibration curve analysis indicated that the integrated model calibration was good and better than that of FSH, semen volume and testicular volume alone. Decision curve analysis indicated with a threshold probability between 0.05 and 0.98, the integrated model added more benefit than treating either all or no patients.

Conclusions: The integrated model has excellent discrimination and good calibration. It can help azoospermic men make better decisions before TESA. It should be noted that TESA is not the first-line treatment for non-obstructive azoospermia because of a low sperm retrieval rate.