E.S. Singh1, H. Shafique1, A. Karpurapu1, C. Cui2, Y. Kim2, L. Mureebe2, A. Johnson2 1Duke University Medical Center, School Of Medicine, Durham, NC, USA 2Duke University Medical Center, Department Of Vascular Surgery, Durham, NC, USA
Introduction: Surgical site infections (SSI) are a persistent problem in open lower extremity revascularizations (LER) performed for peripheral arterial disease, leading to increased morbidity, limb loss, mortality, and health care costs. Interventions such as incisional wound vacs, prophylactic sartorius flaps, alternative autologous conduits, and prolonged antibiotics carry their own risk and costs, but may be of value in particularly high-risk patients. This study aims to validate published risk models of SSI in LER in the NSQIP procedure targeted data sets.
Methods: The 2013-2020 NSQIP Participant Use Files were queried for open aortoiliac and lower extremity revascularization CPT codes that involved open groin incisions. The variables were mapped for four published models (Gwilym, Bennett, Leekha and Wiseman) to the available NSQIP variables and performance was validated with receiver operating characteristic (ROC) curves using R and the tidyverse, tidymodels, and finalfit packages.
Results: A total of 4578 (8.4%) of the 54,601 patients in our validation data set developed SSI after LER. Ten variables were mapped for Gwilym, with an AUC of 0.57; three variables were mapped for Leekha, with an AUC of 0.54; six variables were mapped for Bennett, with an AUC of 0.56; and 17 variables were mapped for Wiseman, with an AUC of 0.63 (Figure 1).
Conclusion: Model performance was poor for all previously published SSI models for LER, calling to question the external validity of these models. Poor performance may also be due to difficulty mapping published risk factors to variables collected in NSQIP. Without capturing previously published LER SSI risk factors, NSQIP-derived models and benchmarks may not accurately reflect patient risk. Additional efforts to align risk factor collection and model development are necessary to improve the real-world risk prediction for LER SSIs.