Accounting for incomplete postdischarge follow-up during surveillance of surgical site infection by use of the National Nosocomial Infections Surveillance system's risk index.

Objective: We examined the usefulness of a simple method to account for incomplete postdischarge follow-up during surveillance of surgical site infection (SSI) by use of the National Nosocomial Infections Surveillance (NNIS) system's risk index.

Design: Retrospective cohort study that used data prospectively collected from 1993 through 2006.

Setting: Five private, nonuniversity healthcare facilities in Belo Horizonte, Brazil.

Patients: Consecutive patients undergoing the following NNIS operative procedures: 20,981 operations on the genitourinary system, 11,930 abdominal hysterectomies, 7,696 herniorraphies, 6,002 cholecystectomies, and 6,892 laparotomies.

Methods: For each operative procedure category, 2 SSI risk models were specified. First, a model based on the NNIS system's risk index variables was specified (hereafter referred to as the NNIS-based model). Second, a modified model (hereafter referred to as the modified NNIS-based model), which was also based on the NNIS system's risk index, was specified with a postdischarge surveillance indicator, which was assigned the value of 1 if the patient could be reached during follow-up and a value of 0 if the patient could not be reached. A formal comparison of the capabilities of the 2 models to assess the risk of SSI was conducted using measures of calibration (by use of the Pearson goodness-of-fit test) and discrimination (by use of receiver operating characteristic curves). Goodman-Kruskal correlations (G) were also calculated.

Results: The rate of incomplete postdischarge follow-up varied between 29.8% for abdominal hysterectomies and 50.5% for cholecystectomies. The modified NNIS-based model for laparotomy did not show any significant benefit over the NNIS-based model in any measure. For all other operative procedures, the modified NNIS-based model showed a significantly improved discriminatory ability and higher G statistics, compared with the NNIS-based model, with no significant impairment in calibration, except if used to assess the risk of SSI after operations on the genitourinary system or after a cholecystectomy.

Conclusions: Compared with the NNIS-based model, the modified NNIS-based model added potentially useful clinical information regarding most of the operative procedures. Further work is warranted to evaluate this method for accounting for incomplete postdischarge follow-up during surveillance of SSI.