Speaker
Description
Most pathogen phylodynamics analyses are done in the context of unclear sampling denominators, low sampling density, and limited participant meta-data. In contrast, the Rakai Health Sciences Program has embedded deep-sequencing of all HIV-viremic individuals into population-based surveillance, enabling linkage of genomic data with a large range of sociodemographic and behavioural covariates.
Our inferential targets are fractions of transmission attributable to specific subpopulations, and relative transmission rates (%sources/%infected in subpopulation). Alternative to birth-death or structured coalescent approaches, we constructed a panel of phylogenetically likely transmission pairs. We interpreted these pairs as realisations of a multi-type point process on a compact age-age domain, with the type encoding an unknown latent state: either truly unlinked, linked from i to j, or linked from j to i. We used Bayesian post-stratification to allocate age-age specific transmission probabilities by many covariates (age, gender, lifetime partnership history, primary occupation, community setting and sexual behaviour), providing a scalable approach for high-dimensional inference.
From 4,260 HIV-positive, successfully deep-sequenced participants, we compiled a list of 625 potential transmission pairs. Of these, the multi-type process model estimated 495 actual transmission events (posterior median). Transmission rates varied substantially within age groups by partnership status (never married, married, separated), typically ≽2-fold. Across inland, fishing, and trading communities key transmission flows (≽10%) originated from married men, both to partners within and outside households. Key transmission flows also included never married women aged 15-29 in trading communities, and married and separated women aged 15-29 in fishing communities. Partnership-specific underreporting of sexual behaviour data posed challenges in quantifying flows by self-reported sexual behaviour. By occupation, flows tended to mirror underlying occupational population structure.
Fitting well-developed statistical models to transmission pair data sets enables easy investigation of population-level transmission flows and rates across many individual-level covariates, and, unlike other approaches, remains scalable to data sets comprising >100k genomes.
| Expedited Notification | No thanks, I do not require Expedited Notification |
|---|
