Inferring pandemic growth rates from sequence data

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

Owing to the way in which BSPs are computed, a slowing or even flattening off at later times will often be present and can easily be misinterpreted as evidence for a slowing of epidemic spread. While this may not be a problem for computing the demographic history of many populations, it will be most significant when applied in real-time to an epidemic when the infected host population is increasing exponentially. If one knows a priori that the number infected is growing exponentially, then coalescent estimation can be improved by a parametric exponential model (as for analysing H1N1 influenza in Fraser et al. [8] ) or even by using a more detailed epidemic model [35] . Another interesting approach that has recently been proposed is to replace Kingman's coalescent by a birth -death model, which perhaps better describes the process of infectious disease transmission [36] . This method would probably have the same advantage in accuracy as the exponential growth model, and may also avoid the problems observed when the infectious population is small (or more precisely when the proportion sampled is not small). This method should be included in future simulations.