Predicting pediatric surgical site infections using machine learning: clinical implications and limitations

Abstract

Dear Editor, Surgical site infections (SSIs) remain a significant source of postoperative morbidity in children, contributing to increased hospital stays, healthcare costs, and patient burden[1,2]. Despite being largely preventable, pediatric-specific predictive models for SSIs are limited, as most frameworks are derived from adult populations. The recent study analyzing over 1.1 million pediatric surgical cases from the NSQIP–Pediatrics database represents a landmark effort to provide individualized SSI risk estimation using machine learning[3]. The authors developed and compared five machine learning models with similar performance, selecting elastic net–regularized logistic regression for its balance of accuracy and interpretability. It includes elastic net-regularized logistic regression, random forest, gradient boosted trees, k-nearest neighbors, and neural networks. All models demonstrated comparable performance (Brier scores 0.023–0.024; c-statistics 0.72–0.77). Regularized logistic regression was selected for its optimal balance of predictive accuracy, computational efficiency, and feasibility for clinical implementation. Key predictors included procedural codes, comorbidities, perioperative diagnoses, acuity markers, laboratory results, and patient demographics. Notably, while the mean predicted SSI risk was 2.4%, a small subset of patients (3%) had a predicted risk ≥10%, emphasizing the potential for targeted preventive interventions[3,4]. These findings have immediate clinical relevance. Integration of validated predictive models into electronic health records could enable real-time, patient-specific SSI risk estimation, allowing surgical teams and infection prevention personnel to tailor interventions for high-risk children. Early identification of at-risk patients can guide prophylactic strategies, optimize perioperative planning, and improve patient safety outcomes. Moreover, using interpretable models like regularized logistic regression facilitates clinician understanding and adoption in complex clinical workflows[4,5]. Despite these advances, several limitations should be considered. The analysis relied on retrospective registry data, which may not capture granular clinical nuances, socioeconomic factors, or institutional variability. External validation and prospective implementation studies are essential to ensure the model’s applicability across diverse pediatric populations and healthcare systems. Importantly, the applicability of these findings in low-resource settings such as Rwanda requires careful consideration. Differences in healthcare infrastructure, data systems, and infection prevention capacity may limit direct implementation of such machine learning models. Additionally, critical real-world factors – including caregiver availability, financial constraints, and access to perioperative care – are often not captured in large registry datasets but significantly influence patient outcomes. These gaps highlight the need for context-specific validation and adaptation. Furthermore, the moderate predictive performance (c-statistics 0.72–0.77) suggests that current models may not yet be sufficient for high-stakes clinical decision-making without further refinement. In conclusion, machine learning-based prediction of pediatric SSI represents a transformative approach to personalized perioperative care. Regularized logistic regression provides a practical and accurate tool for individualized risk stratification, which can be integrated into clinical workflows to inform targeted prevention strategies. This study followed the transparency in the reporting of artificial intelligence (TITAN) guidelines 2025[6].