A district-level ensemble model to enhance dengue prediction and control for the mekong delta region of Vietnam

The Mekong Delta Region (MDR) of Vietnam faces increasing vulnerability to severe dengue outbreaks due to urbanization, globalization, and climate change, necessitating effective early warning systems for outbreak mitigation. This study developed a probabilistic forecasting model to predict dengue incidence and outbreaks with 1–3-month lead times, incorporating meteorological, sociodemographic, preventive, and epidemiological data. A total of 72 models were evaluated, with top performers from spatiotemporal models, supervised PCA, and semi-mechanistic hhh4 frameworks combined into an ensemble. Using data from 2004-2011 for development, 2012–2016 for cross-validation, and 2017–2022 for evaluation, the ensemble model integrated five individual models to forecast dengue incidence up to three months ahead. Performance was assessed using Brier Score, Continuous Ranked Probability Score (CRPS), bias, and diffuseness, and we evaluated performance by horizon, geography, and seasonality. Using the 95th percentile of the historical distribution as the epidemic threshold, the ensemble model achieved 69% accuracy at a 3-month horizon during evaluation, surpassing the reference model’s 58%, though it struggled in years with atypical seasonality, such as 2019 and 2022, possibly due to COVID-19 disruptions. By providing critical lead time, the model enables health systems to allocate resources, plan interventions, and engage communities in dengue prevention and control.Author summary: Dengue fever poses a significant health threat in Vietnam’s Mekong Delta, where outbreaks are worsening due to urbanization and climate change. This study introduces a forecasting model that predicts dengue cases and outbreaks at the district level up to three months in advance. Unlike previous approaches focused on provincial trends, our model combines multiple advanced methods, including spatiotemporal analysis and machine learning, to deliver localized, actionable forecasts. By integrating weather, population, and past outbreak data, it enables health officials to anticipate outbreaks and take targeted preventive actions, such as mosquito control and public education campaigns.