Multivariate time series analysis in nosocomial infection surveillance:: a case study

Background The present study describes the use of time series analysis in the evaluation of the incidence of nosocomial infection. The main hypothesis analysed was that monthly occurrence of nosocomial infection in a hospital may be related to work-related factors such as the control and training of personnel imposed by a surveillance system, strikes supported by medical personnel and movement of personnel. Time series analysis was used to quantify, model and statistically evaluate these interventions. Methods The data employed (March 1982-December 1990) were supplied by the nosocomial infection surveillance system of a primary-care general hospital. The monthly time series incidence of nosocomial infections (measured as percentage cumulative incidence) was analysed by curve fitting, autoregressive, integrated and moving average (ARIMA) modelling (Box-Jenkins) and intervention and dynamic regression analysis. Results The imposed control and training of personnel by the surveillance system was associated with a 3.63% decrease in the accumulated monthly incidence of nosocomial infection from 7.82% to a baseline level of 4.19%. There was a strong indication that an increase of infection incidence of 4.34% corresponded to a medical strike. This increase was maintained over the following months raising the baseline level to 4.84%. An increase of 0.18% was associated with each new nursing contract. Evidence was obtained for the possible relationship between incidence of nosocomial infection and vacation periods. Conclusions The results suggest the need for strict control of the activities of hospital personnel and for the adoption of certain preventative measures during vacation periods to avoid an undesirable increase in the incidence of nosocomial infections.