Predictive Analytics for Disease Outbreak Early Warning in Kogi State

Early warning systems for disease outbreaks improve public health response speed and reduce mortality, and machine learning applied to historical epidemiological data can enhance predictive capacity in resource-limited Nigerian health settings. This study developed a predictive model for disease outbreak early warning in Kogi State, North Central Nigeria, focusing on cholera, meningitis, and Lassa fever. Weekly reported case counts from 2014 to 2022 across 21 LGAs were obtained from the Kogi State Primary Health Care Development Agency. Environmental variables including rainfall, temperature, river flooding extent from satellite data, and population displacement indicators were incorporated. Negative binomial regression for count data and LSTM sequence models were both tested. The LSTM model achieved the best performance for cholera prediction, with F1-score of 0.81 for detecting outbreak-level events defined as exceeding two standard deviations above baseline. Meningitis prediction was more challenging due to lower case counts and irregular epidemic cycles. Flooding events were the strongest predictor of cholera outbreaks with a four-week lead time. Lassa fever predictions benefited most from incorporating rodent habitat index data. The study contributes a pilot early warning framework for a state without a dedicated outbreak prediction system. Recommendations include integration of the model into the Kogi State Integrated Disease Surveillance and Response system, training of state epidemiologists in model output interpretation, and collaboration with Africa CDC for validation of the approach across comparable states.

Keywords: outbreak early warning, predictive analytics, Kogi State, epidemiological data, LSTM