Model context: Trained on 194 adult males (ages 18-93) at a single center in Milan, Italy (2023-2024). Observed complication rate 29.9%. SVM (RBF kernel), AUC 0.907, Brier 0.105, operating threshold 0.238. Not externally validated.
Only comorbidity in the model. Others not represented.
Complications likely? -
Predicted probability of any complication: -%
Classification threshold for elevated risk: 23.8%
· Composite outcome: bleeding, edema, pain, or infection within 7 days
Laser subgroup note: In the training cohort, the laser group (n=62) had only 1 observed complication, which the model did not detect. Predictions within the laser subgroup have limited empirical validation and should be interpreted with caution.
Disclaimer:
Adult circumcision remains a safe and commonly performed procedure with many proven benefits,
including reduced rates of urinary tract infection, penile cancer, and certain sexually transmitted
infections. This tool is intended to help surgeons by providing an estimated probability of composite
complications based on intraoperative vitals and patient characteristics. It should be used as an
adjunct to, not a replacement for, clinical judgment and shared decision-making with the patient
or guardian.
We compared three supervised models: logistic regression, random forest, and support vector
machines to predict short-term complications after adult male circumcision. Data from 194
patients (≥18 years) at a Milan center (2023-2024), using age, BMI, blood loss, surgical
technique (traditional vs. laser), intraop vitals, and diabetes status.
Cohort
n = 194 adult males (ages 18-93, median 34)
Single center, Milan, Italy, 2023-2024
Complication rate: 29.9% (58/194)
Surgical technique: 68% traditional, 32% laser
Diabetes: n=30 (only comorbidity retained)
Resampling & Training
No resampling (SMOTE/ROS) in final model
Stratified 10-fold cross-validation
MinMax scaling; balanced class weights
Performance Metrics (SVM Best)
Metric
Point Estimate (95% CI)
AUC ROC
0.907 (0.855-0.950)
Precision / PPV
0.725 (0.623-0.826)
Average Precision
0.832 (0.735-0.913)
Sensitivity / Recall
0.862 (0.765-0.939)
Specificity
0.860 (0.797-0.915)
F1-Score
0.787 (0.706-0.857)
Brier Score
0.105 (0.077-0.134)
Classification Threshold
0.238
Top Predictors (SHAP)
Intraoperative blood loss and surgical technique were the strongest predictors, followed by age, diabetes status, and overweight BMI category.
Known Limitations
Single-center retrospective data; no external validation
Laser subgroup had only 1 observed complication (model did not detect it); predictions within this subgroup have limited empirical validation
Only diabetes retained as comorbidity; others (hypertension, hypercholesterolemia, Parkinson's, etc.) not represented
Intraoperative predictors mean the tool is a hybrid intra/postoperative risk estimator, not a purely preoperative calculator
Probabilities from this deployed tool reflect the final model refit on all 194 training patients. They will differ slightly from the out-of-fold (OOF) predictions reported in the SHAP analysis, which were generated by fold-specific models during cross-validated evaluation. Both are legitimate quantities: OOF probabilities represent honest held-out performance for patients in the training set, while this tool returns the appropriate inference-time prediction for new patients.
