Abstract
Predicting pathogen distributions is vital for risk assessment and disease control in wild and domestic animal populations. Co-infections are of particular clinical and epidemiological significance, as they can often be associated with increased disease severity and pathogen shedding, increasing both the burden of disease and the risk of spillover (Peel et al., 2019). Moreover, pathogens can also interact and affect the likelihood of subsequent infections (Fountain‐Jones et al., 2019). Yet, when predicting pathogen distributions, research overwhelmingly focuses on single infections caused by individual pathogens. This univariate focus means that important biotic effects are often excluded in distribution models, affecting both the reliability of inferences and the accuracy of predictions. Multivariate models help overcome this hurdle by accounting for biotic associations between pathogens. To explore avenues for improving distribution modelling in veterinary epidemiology, our primary objective was to understand the wider use of multivariate models in co-infection research. From these findings, we make recommendations on how Joint Species Distribution Models could be used by veterinary epidemiologists.
To understand the wider use of multivariate models in co-infection research, we conducted a systematic review of the literature using a suite of terms relating to both multi-response models and co-infections. This yielded 746 unique peer-reviewed research articles. Using seven pre-defined exclusion criteria, we identified 75 peer-reviewed primary studies that jointly measured infection patterns with two or more endo-pathogens of humans or animals in natural environments. Only 27% studied co-infections in animal hosts. Using a generalised linear model to explore how model choice may vary with study goals and purpose (i.e. inference or prediction), we found a strong association between model choices and study goals. Studies that sought to predict the spatial distributions of co-infections typically employed multinomial models, ignoring pathogen associations. Clustering analysis based on study features and a citation analysis to investigate rates of knowledge exchange both identified distinct clusters for multinomial versus multivariate model users. We conclude that biotic associations are often ignored when predicting co-infections and that there exists a lack of knowledge exchange among co-infection researchers with different research agendas. Moreover, we show the untapped potential for using Joint Species Distribution Models in veterinary epidemiology and how one can go about selecting the most appropriate model for complex data for improving predictions of co-infection distributions (Powell-Romero et al., 2023), as well as highlight the importance of interdisciplinary collaboration.
References
Fountain‐Jones, N. M., Packer, C., Jacquot, M., Blanchet, F. G., Terio, K., Craft, M. E., & Ezenwa, V. (2019). Endemic infection can shape exposure to novel pathogens: Pathogen co‐occurrence networks in the Serengeti lions. Ecology Letters, 22(6), 904-913. doi:10.1111/ele.13250
Peel, A. J., Wells, K., Giles, J., Boyd, V., Burroughs, A., Edson, D., . . . Clark, N. (2019). Synchronous shedding of multiple bat paramyxoviruses coincides with peak periods of Hendra virus spillover. Emerging Microbes & Infections, 8(1), 1314-1323. doi:10.1080/22221751.2019.1661217
Powell-Romero, F., Fountain-Jones, N. M., Norberg, A., & Clark, N. J. (2023). Improving the predictability and interpretability of co-occurrence modelling through feature-based joint species distribution ensembles. Methods in ecology and evolution, 14(1), 146-161. doi:https://doi.org/10.1111/2041-210X.13915