Molecular Medicine Unit, Faculty of Medicine, University of Kelaniya, Ragama, Sri Lanka.
Department of Statistics, Faculty of Science, University of Colombo, Colombo 03, Sri Lanka.
Parasit Vectors. 2018 Apr 24;11(1):262. doi: 10.1186/s13071-018-2828-2.
Dengue is one of the major health problems in Sri Lanka causing an enormous social and economic burden to the country. An accurate early warning system can enhance the efficiency of preventive measures. The aim of the study was to develop and validate a simple accurate forecasting model for the District of Gampaha, Sri Lanka. Three time-series regression models were developed using monthly rainfall, rainy days, temperature, humidity, wind speed and retrospective dengue incidences over the period January 2012 to November 2015 for the District of Gampaha, Sri Lanka. Various lag times were analyzed to identify optimum forecasting periods including interactions of multiple lags. The models were validated using epidemiological data from December 2015 to November 2017. Prepared models were compared based on Akaike's information criterion, Bayesian information criterion and residual analysis.
The selected model forecasted correctly with mean absolute errors of 0.07 and 0.22, and root mean squared errors of 0.09 and 0.28, for training and validation periods, respectively. There were no dengue epidemics observed in the district during the training period and nine outbreaks occurred during the forecasting period. The proposed model captured five outbreaks and correctly rejected 14 within the testing period of 24 months. The Pierce skill score of the model was 0.49, with a receiver operating characteristic of 86% and 92% sensitivity.
The developed weather based forecasting model allows warnings of impending dengue outbreaks and epidemics in advance of one month with high accuracy. Depending upon climatic factors, the previous month's dengue cases had a significant effect on the dengue incidences of the current month. The simple, precise and understandable forecasting model developed could be used to manage limited public health resources effectively for patient management, vector surveillance and intervention programmes in the district.
登革热是斯里兰卡的主要卫生问题之一，给该国造成了巨大的社会和经济负担。一个准确的早期预警系统可以提高预防措施的效率。本研究旨在为斯里兰卡的甘帕哈区开发和验证一个简单准确的预测模型。使用 2012 年 1 月至 2015 年 11 月期间的每月降雨量、降雨天数、温度、湿度、风速和回顾性登革热发病率，开发了三种时间序列回归模型。分析了各种滞后时间，以确定包括多个滞后相互作用的最佳预测期。使用 2015 年 12 月至 2017 年 11 月的流行病学数据对模型进行了验证。基于赤池信息量准则、贝叶斯信息量准则和残差分析对模型进行了比较。
所选模型在训练和验证期间的平均绝对误差分别为 0.07 和 0.22，均方根误差分别为 0.09 和 0.28，预测正确。在训练期间，该地区没有发生登革热疫情，在预测期间发生了 9 次疫情。所提出的模型捕捉到了 5 次疫情，并在 24 个月的测试期内正确拒绝了 14 次。该模型的皮尔斯技能得分（skill score）为 0.49，接收者操作特征（receiver operating characteristic）为 86%和 92%的敏感性。
所开发的基于天气的预测模型能够提前一个月准确地发出登革热爆发和流行的警报。根据气候因素，上月的登革热病例对本月的登革热发病率有显著影响。开发的简单、准确和易于理解的预测模型可以有效地管理有限的公共卫生资源，用于该地区的患者管理、病媒监测和干预计划。
