Modelling multi-strain pathogen dynamics: Inferring between-strain interactions and assessing vaccine-induced replacement

Pathogenic bacteria and viruses often consist of multiple genetic variants, also called strains. Despite their differences, strains of pathogens may sometimes affect one another during transmission dynamics and disease development. Due to the interactions between strains, changes in the occurrence of one strain may in turn lead to changes in the occurrence of other strains. Therefore, when targeting a subset of strains of a pathogen with vaccination, the occurrence of the non-targeted strains may also be affected. In particular, if there is competition between the targeted and non-targeted strains, vaccination may lead to an increase in the occurrence of non-targeted strains, which is also called strain replacement.

In this thesis, we studied and further developed methodologies for inferring between-strain interactions and assessing strain replacement. We modified existing estimation methods for inferring between-strain interactions from longitudinal data in the multi-state modelling framework to enable analyses of data with a high number of co-occurring strains. By applying the novel estimation method to a pneumococcal dataset, we were able to derive new insights into how pneumococcal serotypes interact during carriage.

In addition, we investigated the validity of existing methods for estimating between-strain interactions from patterns of co-occurrence in cross-sectional and longitudinal data. While direct interactions in acquisition and clearance can be readily inferred from either cross-sectional or longitudinal data, competition through natural cross-immunity is more evasive, as the ensuing positive association in occurrence of different strains is indistinguishable from synergistic interaction. Furthermore, we showed how common risk factors that influence the occurrence of all strains may confound estimates of between-strain interactions and investigated the usability of various statistical approaches to correct for such confounding.

After studying how between-strain interactions can be inferred from epidemiologic data, we developed a framework to predict strain replacement using the inferred information on between-strain interactions. The proposed framework allows for an integration of simultaneous and possibly heterogeneous interactions in acquisition and clearance between multiple strains into a practical predictor. The framework can be used to predict strain replacement in individual non-vaccine strains as well as for groups of non-vaccine strains. By means of a simulation study, we examined the performance of the proposed predictors under different assumptions of how strain interact.

Finally, we investigated the validity of oft-used measures to evaluate strain replacement from post-vaccination data. We found that, in the presence of vaccine-induced cross-protection, the prevalence of the non-vaccine strains could decrease immediately after the introduction of vaccination, before rebounding into strain replacement. In addition, vaccine-induced cross-protection renders vaccine effectiveness insensitive for detecting strain replacement. In light of these findings, genotype replacement in human papillomavirus should not be ruled out, as its appearance could be consistent with the observed post-vaccination data.

Through this thesis, we are better able to utilize and interpret available data for the purpose of inferring between-strain interactions and assessing the occurrence of strain replacement. In turn, this may improve our ability to anticipate and mitigate the risk of strain replacement and further reduce the disease burden of multi-strain pathogens in human populations.