Epidemiologic surveillance of foodborne diseases: demonstration of the analytic and predictive capacity of Box-Jenkins models

In order to demonstrate the feasibility of using the Box-jenkins models in the analysis and prediction of data coming from the surveillance of foodborne diseases, two discrete time series were defined corresponding to the monthly number of notified cases of typhoid-paratyphoid fever and viral hepatitis, during the period 1977-1989. The fitting procedure to adjust a Box-Jenkins model to each of the observed time series was performed in three iterative stages: identification, estimation and diagnosis. The adjusted models were used to predict the values of the series for 1990 and to compare them with the observed values during that year. The results pointed out a decreasing tendency in the number of typhoid-paratyphoid, cases, confirmed by the autocorrelation function of the series that decreases slowly; a seasonality, with a maximum number of cases during the warmest months, proved by the autocorrelation function that shows peaks every twelve months; a high autocorrelation between adjacent values of the series demonstrated by the regular autoregressive terms; and a persistence of a random effect among observations separated by a period of twelve months, indicated by the seasonal moving average component. In relation to the viral hepatitis series, the analysis of the results indicated an increasing tendency proved by the estimated autocorrelation function descending slowly; a seasonal variation with many cases during the colder months, demonstrated by seasonal peaks in the autocorrelation function; and a persistence of a random effect between adjacent observations and among observations separated by a period of twelve months, confirmed by the regular and seasonal moving average component, respectively. It was possible to conclude that Box-Jenkins models showed an adecuate analytic and predictive capacity and can be used in the surveillance of foodbome diseases