Análise de dados de vigilância epidemiológica por meio de diferentes tipos de modelos de séries temporais

Abstract

The analysis of time series obtained in the databases of public health plays an important role in processes of health surveillance. However, implementation of methodologies for time series has not yet become a routine in the midst of healthcare practitioners. The objective of this study is to present a theoretical review about time series analysis used for epidemiological surveillance data and practical application of statistical methods for the estimation of three models for notifiable diseases: the Box and Jenkins methodological in the presence and absence of exogenous variable (ARIMAX and ARIMA) and vector autoregression (VAR) model. For this, we perfomed a cross-sectional study using secondary data from SINAN (Information System for Notifiable Diseases) consisting of cases of hepatitis A and leptospirosis recorded in Rio Grande do Sul, in the period January 2008 to December 2012. The models were analyzed and discussed through comparison of performance measures. The ARIMA models presented the best properties for the prediction of new cases of the diseases studied. The one-way causality between the diseases was also established.