Forecasting infectious and parasitic disease emergency department attendances using high-dimensional time series data

The increasing global burden of infectious and parasitic diseases (IPDs) presents a significant challenge to public health systems worldwide, particularly to emergency departments (EDs). Accurate forecasting of ED attendances for IPDs is crucial for effective resource planning and public health interventions. We present a novel method for forecasting nationwide IPDs ED attendances incorporating a high-dimensional set of disease and environmental covariates. National ED attendances with IPDs as a cause in Singapore from 2009 to 2018 was taken as the study setting. We proposed simple and trimmed pooling of forecasts using 6 models which can optimize point forecast accuracy over a 1 to 8-week forecast horizon. We found that forecast combinations based on trimmed means provided superior forecast accuracy over sub-models which were used to construct the forecast combinations. Forecast combinations consistently achieved a Mean Absolute Percentage Error below 10.5% in the forecast horizon of 1–8 weeks ahead over the study period. We further demonstrated the robustness of forecast combinations in maintaining good forecast performance even when some sub-model constituents performed significantly worse compared to alternatives. Moreover, epidemiological analysis that employed post-selection inference revealed significant correlations between decreases in past temperature and increases in total precipitation, PM2.5 surface concentration with increased ED attendances for IPDs. The proposed methods can provide forward guidance for potential outbreaks and refine short to long-term resource management for ED admissions. The source code and supplementary material are available at https://github.com/gpeihong/IPDs-ED-Analysis.