Application of Exponential Smoothing for Nosocomial Infection Surveillance

Abstract

Detection of outbreaks of infection or increases in bacterial resistance to antimicrobial agents is an essential component of hospital infection control surveillance. The authors applied the method of exponential smoothing to microbiology data from 1987–1992 to investigate a suspected outbreak of gentamicin resistance among Pseudomonas aeruginosa bacteria at the Department of Veterans Affairs Medical Center, San Francisco, California, in 1991–1992. The years 1987–1990 were used to develop the baseline for the forecast model. Application of the model indicated that two observed prominent peaks in the annual cumulative incidence of gentamicin-resistant P. aeruginosa were within the upper bounds of their respective 95% confidence intervals as estimated by the forecast model–i.e., that no epidemic was in progress. This prediction was supported by investigations by the hospital's infection control team which indicated that the apparent increases were due to readmission of patients previously known to harbor these organisms. in contrast, application of a typically employed method that ignores the time series data structure indicated that there were 6 months in which incidence rates exceeded the upper bounds of their respective 95% confidence intervals, thereby erroneously suggesting that an epidemic was in progress. Recursive algorithms and some simplifying assumptions that do not affect the validity of inferences make the application of this method practical for nosocomial infection control programs.