Developing early warning systems using predictive models to identify Lassa fever threats in Ondo State, Nigeria

Technical Summary: Developing Early Warning Systems for Lassa Fever in Ondo State, Nigeria

Problem Statement
Lassa fever (LF), a severe hemorrhagic disease caused by the Lassa virus, remains a persistent public health burden in West Africa, particularly in Nigeria. Ondo State is identified as a major hotspot, accounting for over 30% of confirmed national cases. The disease exhibits a distinct seasonal pattern, with outbreaks typically peaking during the dry season (December–March) due to increased contact between humans and the Mastomys natalensis rodent reservoir. Despite existing surveillance efforts, challenges such as underreporting, misdiagnosis due to symptom overlap with malaria, and delayed detection hinder effective outbreak response. The study addresses the critical need for an effective early warning system capable of forecasting outbreak patterns to guide proactive public health decision-making and resource allocation in Ondo State.

Methodology
The study employed a time-series forecasting approach using routinely collected weekly surveillance data of confirmed Lassa fever cases in Ondo State, spanning 382 weeks from January 2018 to April 2025. The data was sourced from the Nigeria Centre for Disease Control and Prevention (NCDC).

1. Data Partitioning: The dataset was split into a training set (70%) for parameter estimation and a testing set (30%) for model evaluation.
2. Model Selection: The researchers utilized the Seasonal Auto-Regressive Integrated Moving Average (SARIMA) model, specifically designed to capture both trend and seasonal behaviors in time-series data. Given the weekly nature of the data and the annual seasonality of LF, a seasonal period ($s$) of 52 weeks was established.
3. Model Identification and Estimation: The auto.arima() function in R was used to automate the selection of candidate models based on the Akaike Information Criterion (AIC). Six candidate models were compared.
4. Evaluation Metrics: Model accuracy was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Squared Error (MSE).
5. Diagnostics: The adequacy of the selected model was verified through residual analysis, including checks for normality and the Ljung-Box test to ensure the absence of significant autocorrelation in the residuals.

Key Results

• Optimal Model: The ARIMA(3,0,2)(1,0,1)[52] model was identified as the best-fitting model, achieving the lowest values across all error metrics (RMSE, MAE, MSE) and the lowest AIC value among the candidates.
• Model Performance: Diagnostic checks confirmed that the residuals were approximately normally distributed and randomly scattered around zero, with no significant autocorrelation (Ljung-Box test $p > 0.05$), indicating the model effectively captured the temporal structure of the data.
• Forecasting Horizon: The model was used to generate a 34-week forecast (from the 19th to the 52nd week of 2025).
  • The forecast predicts an average of approximately 4 Lassa fever cases per week.
  • The model projects a relatively stable transmission trend without catastrophic surges, with most forecasted values falling between 5 and 15 cases per week.
  • Seasonal Peaks: The forecast indicates that transmission peaks are likely to occur between the 48th and 52nd weeks of the year, aligning with the historical dry season pattern.
• Visual Validation: Comparisons between the fitted values and the actual test set data demonstrated that the model successfully reproduced major seasonal spikes and the general epidemic trend of Lassa fever in the region.

Significance and Claims
The paper claims that the developed SARIMA model provides a statistically robust framework for understanding the transmission dynamics of Lassa fever in Ondo State. By capturing both short-term variations and long-term seasonal patterns, the model offers a reliable tool for anticipating probable outbreaks.

The authors emphasize that the findings support the development of proactive early warning systems. Specifically, the study suggests that:

• Lassa fever transmission in Ondo State is likely to persist at moderate levels with predictable seasonal peaks.
• Public health policymakers can utilize these forecasts to strengthen surveillance and optimize resource allocation ahead of peak periods (specifically late in the year).
• The model serves as a foundational step toward operationalizing predictive algorithms for real-time outbreak control, moving beyond retrospective analysis to forward-looking public health planning.

The study concludes that while the model predicts moderate transmission under stable conditions, the recurring seasonal nature of the disease necessitates continuous preparedness and targeted interventions during the identified high-risk windows.