Nowcasting by Bayesian Smoothing: A flexible, generalizable model for real-time epidemic tracking

Sarah F. McGough; Michael A. Johansson; Marc Lipsitch; Nicolas A. Menzies

Here, we introduce Nowcasting by Bayesian Smoothing (NobBS), a simple and flexible generalized Bayesian model for nowcasting infectious diseases in different settings. We demonstrate the robustness of this approach in two very different disease surveillance contexts and identify the conditions that favor its application, especially when the reporting delay distribution is time-varying. Specifically, NobBS allows for both uncertainty in the delay distribution and the time evolution of the epidemic curve, producing smooth, time-correlated estimates of cases. We demonstrate that NobBS performs well for weekly nowcasts of (1) dengue cases in Puerto Rico and (2) influenza-like illness (ILI) cases in the United States, requiring no disease-specific parameterization despite the two pathogens being very different (vector-borne vs. directly transmitted) and exhibiting substantially different reporting delays. Lastly, we test NobBS against a previous Bayesian nowcast method (9) and find that NobBS outperforms this benchmark for both diseases and multiple time periods. In particular, we show that while point estimates of the models are similar when time-to-report distributions are relatively fixed over time, NobBS improves the estimation of uncertainty and accommodates temporal variation in delay probabilities. We present an R package, "NobBS," as a tool to complement both routine public health surveillance as well as forecasting efforts.