# Development of machine learning prediction models for postoperative outcomes in adult male circumcision

## TL;DR

This study uses machine learning to predict post-surgery complications in adult male circumcision, aiming to improve preoperative planning and patient outcomes.

## Contribution

The novel contribution is the development and validation of interpretable machine learning models to predict complications in adult male circumcision.

## Key findings

- The SVM model achieved the best performance with an AUC ROC of 0.907 and strong sensitivity and precision.
- Intraoperative blood loss and surgical technique were identified as the strongest predictors of complications.
- The models showed potential for guiding individualized preoperative decisions in clinical settings.

## Abstract

Male circumcision is among the most commonly performed and clinically endorsed surgical procedures globally, deeply rooted in medical, cultural, and religious traditions. While circumcision confers well-documented health benefits such as reduced infection and inflammation, adult patients often experience variable outcomes related to anatomical variations and comorbidities, emphasizing the importance of optimizing procedural planning.

The objective of this study was to develop and internally validate prediction models for short-term postoperative complications following adult male circumcision. This retrospective study evaluated the ability of supervised machine learning models (logistic regression [LR], random forest [RF], and support vector machines [SVM]) to predict short-term postoperative complications following adult male circumcision, using procedural and intraoperative variables, including surgical modality (scalpel- and clamp-based (traditional) vs. laser-based), intraoperative blood loss, and operative technique. Data from 194 adult male patients (≥ 18 years) who underwent circumcision between 2023 and 2024 at a single clinical center in Milan, Italy, were analyzed. Models were trained using standardized preprocessing pipelines and evaluated via stratified 10-fold cross-validation using classification metrics, calibration curves, and SHapley Additive exPlanations (SHAP)-based interpretability analysis.

The SVM model demonstrated superior predictive performance, achieving the highest area under the curve of the receiver operating characteristic (AUC ROC) of 0.907, sensitivity of 0.862, average precision of 0.832, and the lowest Brier score of 0.105. SHAP analysis identified intraoperative blood loss and surgical technique as the strongest predictors of postoperative complications.

These findings support the clinical utility of interpretable machine learning models for individualized risk prediction in adult circumcision, guiding tailored preoperative decisions, particularly in high-risk or resource-limited clinical settings. Study strengths include rigorous evaluation and interpretability, while limitations encompass single-center data and the absence of external validation. Therefore, future research should assess generalizability across more diverse surgical populations and healthcare environments.

The online version contains supplementary material available at 10.1186/s12894-026-02072-x.