OBJECTIVE: To assess the accuracy of nosocomial infections data reported on patients in the intensive-care unit by nine hospitals participating in the National Nosocomial Infections Surveillance (NNIS) System.
DESIGN: A pilot study was done in two phases to review the charts of selected intensive-care–unit patients who had nosocomial infections reported to the NNIS System. The charts of selected high- and low-risk patients in the same cohort who had no infections reported to the NNIS System also were included. In phase I, trained data collectors reviewed a sample of charts for nosocomial infections. Retrospectively detected infections that matched with previously reported infections were deemed to be true infections. In phase II, two Centers for Disease Control and Prevention (CDC) epidemiologists reexamined a sample of charts for which a discrepancy existed. Each sampled infection either was confirmed or disallowed by the epidemiologists. Confirmed infections also were deemed to be true infections. True infections from both phases were used to estimate the accuracy of reported NNIS data by calculating the predictive value positive, sensitivity, and specificity at each major infection site and the “other sites.”
RESULTS: The data collectors examined a total of 1,136 patients' charts in phase I. Among these charts were 611 infections that the study hospitals had reported to the CDC. The data collectors retrospectively matched 474 (78%) of the prospectively identified infections, but also detected 790 infections that were not reported prospectively. Phase II focused on the discrepant infections: the 137 infections that were identified prospectively and reported but not detected retrospectively, and the 790 infections that were detected retrospectively but not reported previously. The CDC epidemiologists examined a sample of 113 of the discrepant reported infections and 369 of the discrepant detected infections, and estimated that 37% of all discrepant reported infections and 43% of all discrepant detected infections were true infections. The predictive value positive for reported bloodstream infections, pneumonia, surgical-site infection, urinary tract infection, and other sites was 87%, 89%, 72%, 92%, and 80%, respectively; the sensitivity was 85%, 68%, 67%, 59%, and 30%, respectively; and the specificity was 98.3%, 97.8%, 97.7%, 98.7%, and 98.6%, respectively.
CONCLUSIONS: When the NNIS hospitals in the study reported a nosocomial infection, the infection most likely was a true infection, and they infrequently reported conditions that were not infections. The hospitals also identified and reported most of the nosocomial infections that occurred in the patients they monitored, but accuracy varied by infection site. Primary bloodstream infection was the most accurately identified and reported site. Measures that will be taken to improve the quality of the infection data reported to the NNIS System include reviewing the criteria for definitions of infections and other data fields, enhancing communication between the CDC and NNIS hospitals, and improving the training of surveillance personnel in NNIS hospitals