We estimated the number of AIDS cases in Brazil from official statistics, after correcting for delays in reporting, by the use of two statistical models. First, we fitted a Kaplan-Meyer survival model with right censoring to the distribution of reporting delays. The second approach is based on a Poisson regression model which was fitted to the contingency table having as one dimension the number of AIDS cases diagnosed per calendar time and the other dimension the amount of delay in reporting. Before the latter approach, we explored, by the use of generalized additive models (GAM), the search for convenient smoothing functions that fit the observed functional relationships in question. The estimates we obtained were then compared to the cases actually reported in the second semester of 1996 and in the first semester of 1997. Comparing the observed number of cases and the predicted figures by Poisson regression, a new reporting pattern seems to start to take place. The official reported number of AIDS cases is very close to the actual epidemic, probably due to government policies of free distribution of medication. Survival models that do not account for calendar time changes in the number of diagnosed AIDS cases in each semester seem to underestimate the behavior of the disease in exposure categories where transmission is increasing.
Models, statistical; Regression analysis; Acquired Immunodeficiency Syndrome; Brazil
