Modeling diarrhea disease in children less than 5 years old.

PURPOSE Identification of the temporal pattern of diarrhea disease in children less than 5 years of age in Rio de Janeiro City (1995-1998) to provide support for decisions about prevention and control of the disease. METHODS The weekly counts of hospitalizations and deaths due to diarrhea disease were analyzed separately. An initial generalized linear model (GLM) was derived using variables related to weather and month. Displays of fitted generalized additive models (GAM) including a spline smoothed function of time suggested additional predictors that were used to obtain new models. RESULTS The initial models did not properly account for the observed cyclical pattern of the data. Graphical displays of the GAM model show a nonhomogeneous decline and annual cycles. Stepwise fitting of GLMs with two factors (cycle and season), and a time trend, showed that the full three-way interaction model was required. Plots of the residuals from the death model suggested a mixture of distributions while the residuals from the hospitalization model were approximately normal. CONCLUSIONS The same general pattern for both time series was found by graphical inspection and fitting of appropriate GLMs. This study provides some additional evidence that severe cases of diarrhea disease may be attributed to rotavirus.