Evaluating Performance of Mathematical Models to Generate Short-term Forecasts of Chlamydia and Gonorrhea in Georgia, 2017-2019

BACKGROUND: Sexually transmitted infections (STIs) represent a widespread and financially burdensome public health issue in the United States. Real-time forecasting of STIs could offer opportunities to respond to STI outbreaks in a timely manner, inform STI screening efforts, and strengthen prevention and allocation efforts. The objective of this study is to evaluate the performance of mathematical models to generate short-term forecasts of chlamydia and gonorrhea in the state of Georgia using retrospective data for the years 2017 to 2019. METHODS: We obtained data on laboratory-confirmed cases of gonorrhea and chlamydia from the Georgia Department of Health. We compared forecast performance between the Auto-Regressive Integrated Moving Average Models (ARIMA), Generalized Logistic Model (GLM), Generalized Additive Models (GAM), and sub-epidemic models in predicting a decline, an increase, and no change in cases for the years 2017-2019. RESULTS: The ARIMA model outperformed all other models with the lowest average mean squared error and highest 95% prediction interval coverage for both gonorrhea and chlamydia. In a scenario analysis, the ARIMA model does well in predicting an increase in cases. CONCLUSION: Disease modeling and forecasting can provide actionable insights for public health planning and intervention. ARIMA was most reliable for stable and increasing trends, while sub-epidemic models better captured declines, highlighting the need to select models based on recent epidemic behavior. GAM and GLM performed competitively in certain contexts but were less robust to sudden changes. Together, these findings support the utility of forecasting methods to improve preparedness and inform STI prevention strategies.