Inferring pandemic growth rates from sequence data

de Silva, Eric; Ferguson, Neil M.; Fraser, Christophe

Now that we have some understanding of why the BSPs produced from sampling simulated data show a slowdown in effective population size for exponentially growing epidemics, we return to the case of the real H1N1 epidemic. The MCC phylogenetic tree and LTT plot can be examined to see whether the slowdown in the effective population size seen in figure 1 is likely to be an artefact of inference method (as seen in our analysis of simulated data), or instead reflects a real slowing of the epidemic. The dates on the LTT and BSP shown in figure 6 are aligned and the pink-shaded region represents the time over which the BSP estimates of N e stops increasing to the point where there are no new lineages in the LTT. There are many coalescent events over this period in the MCC tree in agreement with the cumulative LTT, which increases over this period. This, therefore, validates the changes seen in the BSP in figure 1 , and suggests that there may well have been For non-parametric estimates, the region over which growth rate is estimated from the BSP is restricted to the region for which near exponential growth is seen. CIs not shown for clarity (table 3) . (b) Same methodology as previous but this time randomly sampling proportionally from generations that have more than 100 sequences. a slowing in epidemic spread in late April and early May 2009, during the period when public concern about the new virus was most pronounced. Of course, other biases in the sampling that could confound the analysis, as described earlier, cannot be ruled out.