Forecasting dengue transmission in Thailand by time and space

Thesis

Dengue causes substantial morbidity and mortality in Thailand annually. Forecasting when and where cases will increase can help public health officials plan vector control and health sector responses. A systematic review of dengue forecasting literature revealed that most studies used fixed effects models that ignore temporal and spatial correlation, with evaluation focused primarily on point predictions rather than probabilistic forecasts. 

Using Thai surveillance data (2013-2022), this thesis characterized dengue dynamics and developed forecasting models. Spatial and temporal patterns showed strong heterogeneity. Wavelet analysis revealed annual seasonality and multi-year interannual cycles, likely driven by different mechanisms. Serotype data indicated serotype 1 dominance, but major gaps in serotype and genomic sampling highlighted the need for more systematic surveillance. 

Bayesian spatio-temporal models were developed to forecast dengue cases using an expanding window framework. Climate variables (temperature, Standardized Precipitation Evapotranspiration Index, wind speed), environmental variables (Enhanced Vegetation Index), at no lag up to 6-week lags, and genetic diversity and mobility data were tested individually to identify optimal lags, then incorporated into autoregressive and seasonal baseline models. Models were evaluated at 124-week horizons using the Continuous Ranked Probability Score.

Forecasting skill decreased with longer horizons. The autoregressive model with seasonality performed best overall. Covariates improved only the autoregressive baseline, suggesting the seasonal component already captured covariateassociated information. All models performed poorly during outbreak periods compared to endemic periods. 

This work demonstrates that seasonal models using surveillance data alone can inform planning during endemic periods. However, improving outbreak forecasting remains a critical challenge requiring further methodological development to better support public health decision-making during high-transmission periods.

Authors: Jafari, Y

• DOI: 10.5287/ora-qrnpnn2gg
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: dengue; forecasting
• Subjects: Epidemiology