Explainable machine learning for perioperative surgical site infection risk enrichment after operative treatment of closed pilon fractures: a multicenter retrospective study with external validation - Report - MDSpire
Advertisement
Explainable machine learning for perioperative surgical site infection risk enrichment after operative treatment of closed pilon fractures: a multicenter retrospective study with external validation
Clinical Report: Utilizing Explainable Machine Learning for SSI Risk Assessment
Overview
This study developed and validated machine learning models to predict surgical site infection (SSI) risk in patients undergoing surgery for closed pilon fractures. The random forest model demonstrated high specificity and limited sensitivity.
Background
Surgical site infections (SSIs) are significant complications following surgical interventions for closed pilon fractures, leading to increased healthcare costs and prolonged recovery. Current prediction models for SSI risk are limited in external validation and clinical interpretability.
Data Highlights
Cohort
SSI Events
ROC-AUC
Sensitivity
Specificity
Internal
74 (3.9%)
0.899
0.294
0.987
External
11 (3.1%)
0.902
0.636
0.974
Key Findings
The random forest (RF) model achieved ROC-AUCs of 0.899 and 0.902 in internal and external cohorts, respectively.
RF specificity was high at 0.987 internally and 0.974 externally.
RF sensitivity was 0.294 internally and 0.636 externally.
The preoperative-only RF model had ROC-AUCs of 0.884 and 0.905 in the internal and external cohorts.
Decision-curve analysis indicated positive net benefit for the RF model across various threshold probabilities.
Clinical Implications
The machine learning framework developed in this study provides a tool for identifying patients at risk for SSIs after closed pilon fracture surgery.
Conclusion
The study presents a machine learning approach for SSI risk assessment.
