Toward the rational use of standardized infection ratios to benchmark surgical site infections

Abstract

The National Healthcare Safety Network transitioned from surgical site infection (SSI) rates to the standardized infection ratio (SIR) calculated by statistical models that included perioperative factors (surgical approach and surgery duration). Rationally, however, only patient-related variables should be included in the SIR model. Logistic regression was performed to predict expected SSI rate in 2 models that included or excluded perioperative factors. Observed and expected SSI rates were used to calculate the SIR for each participating hospital. The difference of SIR in each model was then evaluated. Surveillance data were collected from a total of 1,530 colon surgery patients and 185 SSIs. C-index in the model with perioperative factors was statistically greater than that in the model including patient-related factors only (0.701 vs 0.621, respectively, P < .001). At one particular hospital, for which the percentage of open surgery was lowest (33.2%), SIR estimates changed considerably from 0.92 (95% confidence interval: 0.84-1.00) for the model with perioperative variables to 0.79 (0.75-0.85) for the model without perioperative variables. In another hospital with a high percentage of open surgery (88.6%), the estimate of SIR was decreased by 12.1% in the model without perioperative variables. Because surgical approach and duration of surgery each serve as a partial proxy of the operative process or the competence of surgical teams, these factors should not be considered predictive variables.