Epidemiological dynamics of enterovirus D68 in the US: implications for acute flaccid myelitis

Figures

• 

  Figure 1:. Spatiotemporal dynamics of PER-predicted EV-D68 cases across 24 states (2014–2019). (A) Weekly proportion of PER-predicted EV-D68 cases. A clear biennial pattern can be observed in the US time series, but local patterns vary. (B) Map of center of gravity of PER-predicted EV-D68 cases. (C) Latitudinal and (D) longitudinal variation in the center of gravity across 14 regions. Red circles represent 5 regions in panel A. Black circles represent the remaining 9 regions. Areas of the circles (and a triangle) are proportional to the number of RV/EV samples. Solid lines represent the estimated regression lines and the associated 95% CIs from the bivariate linear regression; since the bivariate regression depend on two continuous variables, the estimated univariate relationships are plotted by fixing the remaining covariate at its average values.

• 

  Figure 2:. Comparisons of PER-predicted EV-D68 cases and reported AFM cases. (A) Monthly time series of proportion of PER-predicted EV-D68 cases (black bars) and confirmed AFM cases (red lines) in the US. AFM reports are not available prior to August 2014. (B) Pairwise cross-correlation plot between the two monthly time series at the national level using all available EV-D68 and AFM data. (C) Comparison between the monthly proportion of PER-predicted EV-D68 cases at a one month lag and the monthly number of AFM cases at the national level. The solid line is estimated from a linear regression. (D) Map of correlations between annual proportion of PER-predicted EV-D68 cases and the number of AFM cases between 2014–2019 showing a positive association between the two. (E) Comparison between the annual proportion of PER-predicted EV-D68 cases and the annual number of AFM cases across 14 regions (represented by different shapes for each region). Black line represents the estimated fixed effects from a linear mixed model with random slopes and intercepts. Red lines represents the estimated line for each region from a linear mixed model. All correlations are measured after stabilizing time series…

• 

  Figure 3:. Model fit to data for New York, Ohio, Missouri. (A) A comparison of the PER-predicted (red lines) and model-predicted (grey lines) weekly proportion of positive cases. Predictions were made by simulating forward from 10 particle filtered trajectories. (B) A comparison of the PER-predicted and model-predicted weekly proportion of positive cases, including the outbreak in 2018 into our parameter estimation procedure.

• 

  Figure 4:. Long- and short-term predictions of EV-D68 dynamics. A. Bifurcation diagram of the deterministic SIR model using parameter estimates for New York but by varying R0 and the relative amplitude of the seasonal forcing. Colors indicate the mean periodicity of the cycles. Circles represent the original parameter estimates for each state. The triangle and diamond represent the new parameter estimates that include the 2018 outbreak for New York and Ohio, respectively. B–E. Predictions of EV-D68 outbreak in 2020 using stochastic simulations using parameter estimates for New York (B) and assuming 5% (C), 10% (D), and 20% (E) decrease in transmission rates from the 12th week of 2020 until the end of 2020. For each R0 value, we let S0=1.04/R0 and run 100 stochastic simulations until December 2020. We calculate the normalized incidence by dividing the medians of the model-predicted weekly incidence by the median of the model-predicted peak incidence in 2014.