Multivariate spatial modelling for predicting missing HIV prevalence rates among key populations.

The article focuses on developing a multivariate conditional auto-regressive (CAR) model to improve the prediction of HIV prevalence rates among key populations—specifically injection drug users (IDUs), female sex workers (FSWs), clients of female sex workers (Clients), and men who have sex with men (MSM)—where surveillance data are often sparse. Using HIV surveillance data from Ukraine (2004–2015), the model incorporates both spatial dependence across locations and cross-population correlations to borrow strength from related populations and neighboring areas, resulting in more accurate prevalence estimates than models considering populations independently. The study also investigates how different surveillance data contribute to prediction accuracy and offers practical guidelines for optimizing future data collection under resource constraints. Findings highlight significant cross-population correlations, especially between Clients and IDUs, and emphasize the importance of strategic data collection to enhance HIV epidemic monitoring among marginalized groups.