Time series analysis of dengue surveillance data in two Brazilian cities.

12 Mar 2018
Cortes F, Turchi Martelli CM, Arraes de Alencar Ximenes R, Montarroyos UR, Siqueira Junior JB, Gonçalves Cruz O, Alexander N, Vieira de Souza W

The aim of the study was to evaluate the temporal patterns of dengue incidence from 2001 to 2014 and forecast for 2015 in two Brazilian cities. We analysed dengue surveillance data (SINAN) from Recife, 1.6 million population, and Goiania, 1.4 million population. We used Auto-Regressive Integrated Moving Average (ARIMA) modelling of monthly notified dengue incidence (2001-2014). Forecasting models (95% prediction interval) were developed to predict numbers of dengue cases for 2015. During the study period, 73,479 dengue cases were reported in Recife varying from 11 cases/100,000 inhab (2004) to 2418 cases/100,000 inhab (2002). In Goiania, 253,008 dengue cases were reported and the yearly incidence varied from 293 cases/100,000 inhab (2004) to 3927 cases/100,000 inhab (2013). Trend was the most important component for Recife, while seasonality was the most important one in Goiania. For Recife, the best fitted model was ARIMA (1,1,3)12 and for Goiania Seasonal ARIMA (1,0,2) (1,1,2)12. The model predicted 4254 dengue cases for Recife in 2015; SINAN registered 35,724 cases. For Goiania the model predicted 33,757 cases for 2015; the reported number of cases by SINAN was 74,095, within the 95% prediction interval. The difference between notified and forecasted dengue cases in Recife can be explained by the co-circulation of dengue and Zika virus in 2015. In this year, all cases with rash were notified as "dengue-like" illness. The ARIMA models may be considered a baseline for the time series analysis of dengue incidence before the Zika epidemic.

America region; Arima; Brazil; Dengue; Forecasting; Time series analysis