Reconstructing prevalence dynamics of wildlife pathogens from pooled and individual samples

Pathogen transmission studies require sample collection over extended periods, which can be challenging and costly, especially in the case of wildlife. A useful strategy can be to collect pooled samples, but this presents challenges when the goal is to estimate prevalence. This is because pooling can introduce a dilution effect where pathogen concentration is lowered by the inclusion of negative or lower-concentration samples, while at the same time a pooled sample can test positive even when some of the contributing samples are negative. If these biases are taken into account, the concentration of a pooled sample can be leveraged to infer the most likely proportion of positive individuals, and thus improve overall prevalence reconstruction, but few methods exist that account for the sample mixing process.

We present a Bayesian multilevel model that estimates prevalence dynamics over time using pooled and individual samples in a wildlife setting. The model explicitly accounts for the complete mixing process that determines pooled sample concentration, thus enabling accurate prevalence estimation even from pooled samples only. As it is challenging to link individual-level metrics such as age, sex, or immune markers to infection status when using pooled samples, the model also allows the incorporation of individual-level samples. Crucially, when individual samples can test false negative, a potentially strong bias is introduced that results in incorrect estimates of regression coefficients. The model, however, can account for this by leveraging the combination of pooled and individual samples. Last, the model en- ables estimation of extrinsic environmental effects on prevalence dynamics.

Using a simulated dataset inspired by virus transmission in flying foxes, we show that the model is able to accurately estimate prevalence dynamics, false negative rate, and covariate effects. We test model performance for a range of realistic sampling scenarios and find that while it is generally robust, there are a number of factors that should be considered in order to maximize performance.

The model presents an important advance in the use of pooled samples for estimating prevalence dynamics in a wildlife setting, can be used with any biomarker of infection (Ct values, antibody levels, other infection biomarkers) and can be applied to a wide range of host-pathogen systems.