Comparison of surgical site infection (SSI) rates across institutions has been an effective infection control measure, but success relies on the quality of risk adjustments. This study assessed desirable risk-adjustment methodologies for use in the Japan Nosocomial Infections Surveillance (JANIS) network. Patients who underwent 1 of 6 digestive system procedures (APPY, BILI, CHOL, COLN, GAST, or REC) were included. Logistic regression analysis was performed to predict the risk of developing SSI in the following two models: (1) selected variables that consist of an NNIS Risk Index, or (2) all variables that were collected at SSI surveillance. Model performances were assessed using the c-index. Two regression models were also developed that included or excluded factors regarding surgery duration as well as laparoscopic surgery. The difference in the standardized infection ratio (SIR) in each model was then evaluated. Surveillance data were collected from a total of 37,251 procedures from 37 institutions. Odds ratios regarding the development of SSI were generally different between procedures and risk factors. Except for APPY, the c-index was statistically greater in the model with all variables than in the model including risk index factors only (p < 0.001). The estimates of SIR were considerably different between models with adjustment of surgery duration and laparoscopic surgery versus models without these adjustments. The two models offered contradictory evidence regarding hospital performance. Multivariate logistic regression analyses that use all available variables from SSI surveillance were found to be superior to NNIS risk index methodology. When calculating SIR, we should consider the exclusion of surgery duration and laparoscopic surgery as risk-adjustment factors.
All Science Journal Classification (ASJC) codes