Predictive models of surgical site infections after coronary surgery: insights from a validation study on 7090 consecutive patients

BACKGROUND: The role of specific scoring systems in predicting risk of surgical site infections (SSIs) after coronary artery bypass grafting (CABG) has not been established. AIM: To validate the most relevant predictive systems for SSIs after CABG. METHODS: Five predictive systems (eight models) for SSIs after CABG were evaluated retrospectively in 7090 consecutive patients undergoing isolated (73.9%) or combined (26.1%) CABG. For each model, accuracy of prediction, calibration, and predictive power were assessed with area under receiver-operating characteristic curve (aROC), the Hosmer-Lemeshow test, and the Goodman-Kruskal γ-coefficient, respectively. Six predictive scoring systems for 30-day in-hospital mortality after cardiac operations were evaluated as to prediction of SSIs. The models were compared one-to-one using the Hanley-McNeil method. FINDINGS: There were 724 (10.2%) SSIs. Whereas all models showed satisfactory calibration (P = 0.176-0.656), accuracy of prediction was low (aROC: 0.609-0.650). Predictive power was moderate (γ: 0.315-0.386) for every model but one (γ: 0.272). When compared one-to-one, the Northern New England Cardiovascular Disease Study Group mediastinitis score had a higher discriminatory power both in overall series (aROC: 0.634) and combined CABG patients (aROC: 0.648); in isolated CABG patients, both models of the Fowler score showed a higher discriminatory power (aROC: 0.651 and 0.660). Accuracy of prediction for SSIs was low (aROC: 0.564-0.636) even for six scoring systems devised to predict mortality after cardiac surgery. CONCLUSION: In this validation study, current predictive models for SSIs after CABG showed low accuracy of prediction despite satisfactory calibration and moderate predictive power.