< Back to Article
Improved estimation of time-varying reproduction numbers at low case incidence and between epidemic waves
Fig 4
Epidemics with multiple waves.
We compare reproduction number estimates (
or 
) and one-step-ahead incidence predictions (
or 
) from APEestim with optimal window k*, EpiEstim with window k and EpiFilter with state noise η. We simulate 200 epidemics with multiple waves of infection using the standard renewal model (Eq (1)) for three scenarios, representative examples of which are given in A-C. The true Rs and Is are in black. All mean estimates or predictions are in red and blue with 95% equal tailed credible intervals. APEestim and EpiEstim use a Gam(1, 2) prior distribution and EpiFilter a grid with m = 2000, Rmin = 0.01 and Rmax = 10. In D we provide statistics of the MSE of these estimates (relative to Rs) and the PMSE of these predictions (relative to Is) for all 200 runs. We find EpiFilter is best able to negotiate troughs between epidemic peaks and hence infer resurging infectious dynamics, achieving significantly smaller MSE (2–10 fold reductions) and comparable PMSE to APEestim (which is optimised for prediction).
doi: https://doi.org/10.1371/journal.pcbi.1009347.g004