Short-Term Forecasting of COVID-19 Positivity Rates in Ontario

Background & Objectives: Accurate forecasting of respiratory virus activity, particularly COVID-19, is vital for effective public health preparedness; yet many existing models fail to capture age-specific dynamics and temporal autocorrelation, limiting their predictive utility. This study aimed to improve short-term forecasting of COVID-19 percent positivity by integrating age-stratified modeling with time-series methods, addressing gaps in adaptability and predictive accuracy. Methods: We developed age-specific Generalized Additive Models (GAMs) with lagged predictors and time-series components—including Random Walk (RW), Autoregressive (AR), and Moving Average (MA) terms—to capture non-linear trends in percent positivity while modeling both long- and short-term temporal dependencies. Models were stratified by age group and assessed under various outcome formulations: negative binomial models on case counts with offsets, logistic regression on aggregated counts, and beta regression on percent positivity. All models were fitted using R-INLA (v24.12.11). Rolling monthly time-series cross-validation was used for validation; Root Mean Squared Error (RMSE) guided model selection, while Mean Absolute Percentage Error (MAPE) assessed prediction accuracy. Results: Age-stratified modeling significantly improved forecasts. For adults, autocorrelation terms enhanced short-term fluctuation capture, with informative priors reducing overfitting. In pediatric populations, seasonal effects related to school cycles were especially influential. Bayesian models outperformed standard methods during periods of trend shifts, particularly in winter months. Conclusions: Integrating age-specific stratification, temporal components, and Bayesian inference improved COVID-19 forecasting accuracy. These approaches offer actionable insights for tailored public health interventions. Future work will extend this framework to influenza and RSV using multivariate time series models to understand interactions between respiratory viruses.