Forecast of dengue incidence using temperature and rainfall

Area of Science:

• Epidemiology
• Environmental Health
• Public Health Informatics

Background:

• Dengue fever poses a significant public health challenge, necessitating accurate early warning systems.
• Effective epidemic prediction enhances the efficacy of preventive measures against dengue.

Purpose of the Study:

• To develop and validate a forecasting model for predicting dengue cases in Singapore.
• To provide timely early warnings for dengue epidemics.

Main Methods:

• A time series Poisson multivariate regression model was developed using weekly mean temperature and cumulative rainfall data (2000-2010).
• Weather data were modeled using piecewise linear spline functions, considering autoregression, seasonality, trend, and various lag times.
• Model validation involved forecasting dengue cases using weather data alone and assessing performance with Akaike's Information Criterion, RMSE, and ROC curves.

Main Results:

• The optimal dengue forecast period was determined to be 16 weeks.
• The model demonstrated high accuracy, with low errors (0.3 and 0.32 SD) during training and validation.
• Forecast sensitivity for distinguishing outbreaks was high (96-98%), with a low false alarm rate (<3% in 2011).

Conclusions:

• A weather-based dengue forecasting model was successfully developed, providing 16-week advance warnings.
• The model is a simple, precise, and low-cost tool for enhancing decision-making in vector control and resource allocation.
• This approach can significantly improve the management and mitigation of dengue epidemics.