Establishment of a Bayesian Network Model for Real-Time Zoonotic Disease Risk Prediction in High-Risk Communities of Nigeria

Real-time zoonotic disease risk prediction is a frontier challenge in public health intelligence, and Bayesian network models offer a promising computationally efficient approach integrating multiple dynamic risk variables. This study developed and validated a Bayesian network model for real-time zoonotic disease risk prediction in high-risk communities in Nigeria. Model development used five years of retrospective data from 15 Nigerian states encompassing animal disease event reports, meteorological records, market activity data, wildlife surveillance outputs, and human disease notifications. Bayesian network structure was learned using a combination of expert elicitation workshops with 25 veterinary and public health experts and data-driven algorithms. Model parameters were estimated by expectation-maximization from the retrospective dataset. Prospective validation was conducted over a 12-month period in six pilot communities, with weekly risk score outputs compared against observed disease events using receiver operating characteristic analysis. The model achieved an area under the curve of 0.81 for predicting Lassa fever risk, 0.76 for brucellosis outbreaks, and 0.79 for avian influenza events. False-positive rates were acceptable for routine public health decision-making. Weekly risk alerts triggered appropriate response in 14 of 17 validated outbreak events. The study presents an original computational methodology for zoonotic disease risk prediction adapted to Nigerian data environments and infrastructure constraints. The model is proposed for integration into Nigeria's Electronic Integrated Disease Surveillance and Response platform. Keywords: Bayesian network, zoonotic risk prediction, real-time surveillance, Nigeria, computational epidemiology