Is It Too Late? Machine Learning for Predicting Orchiectomy Versus Testicular Preservation in Acute Torsion.

Objectives

Testicular torsion is a time-critical pediatric urological emergency in which delayed presentation may increase the likelihood of orchiectomy. However, preoperative estimation of the immediate intraoperative outcome remains challenging. This study aimed to develop and internally validate machine-learning models integrating ischemic timing variables and complete blood count-derived inflammatory indices to predict orchiectomy versus testicular preservation at index surgical exploration in pediatric and adolescent testicular torsion.

Methods

We retrospectively analyzed 165 children and adolescents who underwent surgical exploration for confirmed testicular torsion. The primary endpoint was orchiectomy at index exploration versus testicular preservation through detorsion and/or orchiopexy. Clinical timing variables and complete blood count-derived indices, including neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, lymphocyte-to-monocyte ratio, white blood cell-to-monocyte ratio, monocyte-to-mean platelet volume ratio, hemoglobin-to-monocyte ratio, systemic inflammatory response index, and aggregate index of systemic inflammation, were evaluated. Five supervised machine-learning algorithms-logistic regression, random forest, XGBoost, LightGBM, and support vector machine-were assessed using nested stratified cross-validation with an outer 5-fold loop and inner 3-fold hyperparameter selection. Model performance was estimated from out-of-fold predictions.

Results

Orchiectomy was performed in 37 patients (22.4%), whereas testicular preservation through detorsion and/or orchiopexy was performed in 128 patients (77.6%). Symptom duration was significantly longer in the orchiectomy group than in the preservation group (48.00 [30.00-72.00] vs. 6.00 [2.00-24.00] h, p < 0.001). Monocyte count was also higher in the orchiectomy group (0.92 [0.68-1.23] vs. 0.65 [0.50-0.93] × 109/L, p < 0.001). Among the combined models, XGBoost achieved the highest numerical discrimination, with a ROC-AUC of 0.926 ± 0.066 and a bootstrap 95% confidence interval of 0.856-0.966. Feature-block and interpretability analyses identified symptom duration as the dominant predictor, followed by emergency department waiting time and selected monocyte-centered inflammatory indices.

Conclusions

Immediate intraoperative orchiectomy in pediatric and adolescent testicular torsion was most strongly associated with prolonged symptom duration, while selected complete blood count-derived inflammatory indices provided complementary risk information. The combined XGBoost model showed strong internal discrimination and clinically interpretable predictor patterns. However, the model was internally validated only and requires external validation before clinical implementation.