External validation of prognostic multivariable risk models for surgical site infections after open lower extremity...

Surgical site infections are common after open lower extremity revascularization for peripheral arterial disease, increasing morbidity, limb loss, mortality, and costs. Although universal bundles reduce surgical site infection rates, some prophylactic measures carry risks and costs, making targeted use in high-risk patients essential. Several multivariable prognostic models for surgical site infections after lower extremity revascularization exist, but few have undergone external validation.

We assessed 10 published surgical site infection risk models using 2 national registries: National Surgical Quality Improvement Program and Vascular Quality Initiative. Discrimination was evaluated with area under the receiver operating characteristic curve, and calibration with visual plots and expected calibration error, applying 200-fold bootstrapping for 95% confidence intervals. Sensitivity analyses examined severe surgical site infections and subgroup calibration.

Ten models published between 2012 and 2024 were identified: 4 from single institutions, 5 from national registries, and 1 from an international prospective study. In the National Surgical Quality Improvement Program, 8.5% of 61,586 patients developed surgical site infections; in the Vascular Quality Initiative, 11.2% of 75,193 patients developed surgical site infections. Across all models and datasets, the area under the receiver operating characteristic curve ranged from 0.50 to 0.60, indicating poor discrimination. Wiseman, Eslami, and Gwilym models performed best for discrimination, whereas Wiseman, Xu, and Gwilym showed relatively better calibration by expected calibration error. Visual inspection revealed poor calibration overall, with most models underestimating risk in low-risk patients and overestimating risk in high-risk patients.

Existing surgical site infection risk models demonstrate inadequate discrimination and calibration on external validation, limiting clinical utility. Poor performance likely reflects differences in cohorts, outcome definitions, and missing variables. Transparent, interpretable reporting and rigorous, independent external validation are essential before clinical implementation.