EUVIMEDEUVIMED entdecken
CrossrefInfokarte

Bayesian Forecasting as the Analytical Core of an AI-Assisted Early Warning System for Dengue Outbreaks during the 2023 Epidemic: Implementation and Emergency Management in a Dominican Pediatric Hospital

Manuel Colomé-Hidalgo, Robert Paulino-Ramírez, Parthena Savvidou, María Belén Martín-Sanz, Demian Arturo Herrera Morban, Mariana Dauhajre Pimentel, Jonathan Patricio Baldera, Alanna Yasmin Segura-Peña, y Marlenis Mercedes Martínez

Oxford Open Digital Health · 2026-11-02

DOI: 10.1093/oodh/oqag027 ↗
AbstractBibliografische Angaben

Abstract

Abstract Dengue remains a major cause of pediatric morbidity in the Dominican Republic, with increasingly intense epidemic cycles. Advances in statistical modeling, probabilistic forecasting, and AI-assisted digital health systems offer new opportunities to enhance early warning capabilities and strengthen hospital preparedness. This study evaluated a Bayesian negative binomial regression forecasting model as the analytical core of an AI-assisted early warning system (EWS) for predicting weekly dengue cases nationwide, using strict out-of-sample validation during the 2023 epidemic and applying its outputs operationally in a high-demand pediatric hospital. National surveillance data (PLISA/PAHO), hospital records from the Dr. Hugo Mendoza Pediatric Hospital (HPHM), and climatic and satellite indicators (NOAA, MODIS) for 2014–2024 were integrated. The model incorporated climatically plausible lags, harmonic seasonality, two vegetation indices (NDVI_lag4 and EVI_lag4), a long-term spline trend, and weakly informative priors, achieving excellent convergence (Rhat = 1.00). Climatic and environmental predictors exhibited small and uncertain effects, whereas the temporal spline accounted for most long-term epidemic variability. In the 2023 national validation, the model captured the broad epidemic trajectory and improved RMSE and MAE compared with the historical seasonal baseline (RMSE = 783.3; MAE = 423.4; MAPE = 80.2%), although proportional accuracy remained limited during abrupt epidemic acceleration. The national model also captured ascending-phase signals that were temporally aligned with the HPHM hospital outbreak, including its sharp rise during epidemiological weeks 36–40. These outputs were used alongside routine surveillance and hospital indicators to support operational situational awareness, capacity planning, patient-flow reorganization, strengthened triage, and emergency preparedness during the epidemic. The model demonstrated temporal signal detection and operational relevance, positioning Bayesian probabilistic forecasting, embedded within an AI-assisted digital health workflow, as a useful analytical component for epidemic monitoring and hospital preparedness, subject to limitations in peak magnitude calibration and proportional accuracy.

Wissen, das zu Ihrer Arbeit passt.

Vom kostenlosen Einstieg über die persönliche Recherche bis zur vollständigen Research Suite.

Tarife ansehen →

Sind alle Verlagsvolltexte enthalten?

Nein. Bibliografische Nachweise, Abstracts und Volltexte sind unterschiedliche Inhalte. Ob ein Volltext verfügbar ist, hängt von der Quelle, Open Access und Ihren vorhandenen Lizenzen ab. Ein EUVIMED-Tarif ersetzt keine zusätzlichen Verlagsrechte.

Tarife ansehen →