Silawan Tassanee, Singhasivanon Pratap, Kaewkungwal Jaranit, Nimmanitya Suchitra, Suwonkerd Wanapa
Department of Tropical Hygiene, Faculty of Tropical Medicine, Mahidol University, Bangkok.
Southeast Asian J Trop Med Public Health. 2008 Jan;39(1):90-8.
This study aimed to determine temporal patterns and develop a forecasting model for dengue incidence in northeastern Thailand. Reported cases were obtained from the Thailand national surveillance system. The temporal patterns were displayed by plotting monthly rates, the seasonal-trend decomposition procedure based on loess (STL) was performed using R 2.2.1 software, and the trend was assessed using Poisson regression. The forecasting model for dengue incidence was performed in R 2.2.1 and Intercooled Stata 9.2 using the seasonal Autoregressive Integrated Moving Average (ARIMA) model. The model was evaluated by comparing predicted versus actual rates of dengue for 1996 to 2005 and used to forecast monthly rates during January to December 2006. The results reveal that epidemics occurred every two years, with approximately three years per epidemic, and that the next epidemic will take place in 2006 to 2008. It was found that if a month increased, the rate ratio for dengue infection decreased by a factor 0.9919 for overall region and 0.9776 to 0.9984 for individual provinces. The amplitude of the peak, which was evident in June or July, was 11.32 to 88.08 times greater than the rest of the year. The seasonal ARIMA (2, 1, 0) (0, 1, 1)12 model was model with the best fit for regionwide data of total dengue incidence whereas the models with the best fit varied by province. The forecasted regional monthly rates during January to December 2006 should range from 0.27 to 17.89 per 100,000 population. The peak for 2006 should be much higher than the peak for 2005. The highest peaks in 2006 should be in Loei, Buri Ram, Surin, Nakhon Phanom, and Ubon Ratchathani Provinces.
本研究旨在确定泰国东北部登革热发病率的时间模式并开发预测模型。报告病例来自泰国国家监测系统。通过绘制月发病率来展示时间模式，使用R 2.2.1软件执行基于局部加权回归散点平滑法（STL）的季节趋势分解程序，并使用泊松回归评估趋势。登革热发病率的预测模型在R 2.2.1和Intercooled Stata 9.2中使用季节性自回归积分滑动平均（ARIMA）模型进行。通过比较1996年至2005年登革热的预测发病率与实际发病率来评估该模型，并用于预测2006年1月至12月的月发病率。结果显示，疫情每两年发生一次，每次疫情约持续三年，下一次疫情将于2006年至2008年发生。研究发现，若月份增加，整个地区登革热感染率比降低0.9919倍，各省份则降低0.9776至0.9984倍。6月或7月出现的高峰幅度比一年中的其他时间高11.32至88.08倍。季节性ARIMA（2, 1, 0）（0, 1, 1）12模型最适合登革热总发病率的全区域数据，而最适合的模型因省份而异。2006年1月至12月预测的区域月发病率应为每10万人0.27至17.89例。2006年的高峰应远高于2005年的高峰。2006年最高峰应出现在黎逸府、武里喃府、素林府、廊开府和乌汶府。
