Better targeting treatments against Bovine Respiratory Disease by combining dynamic generalized linear models and mechanistic modelling

Bovine Respiratory Disease (BRD) is a major health challenge for young bulls. To minimize economic losses, collective treatments have been widely adopted. Nevertheless, performing collective treatments involve a trade-off between BRD cumulative incidence and severity, and antimicrobial usage (AMU). Therefore, we propose a proof of concept of a decision support tool aimed at helping farmers and veterinarians make informed decisions about the appropriate timing for performing collective treatment for BRD. The proposed framework integrates a mechanistic stochastic simulation engine for modelling the spread of a BRD pathogen ( Mannheimia haemolytica ), and a hierarchical multivariate binomial dynamic generalized linear model (DGLM) which provides early warnings based on infection risk estimates. Using synthetic data, we studied 48 scenarios, involving two batch sizes (small and large), four farm risk levels of developing BRD (low, medium, balanced, and high), two batch allocation systems (sorted by risk level or randomly allocated), and three treatment interventions (individual, conventional collective, and DGLM-based collective). In most scenarios, collective treatments triggered by the DGLM were associated with a reduction of BRD cumulative incidence and disease severity. DGLM-based collective treatments typically exhibited either lower or equivalent AMU compared to conventional ones, being particularly advantageous in medium, balanced, and high-risk scenarios. Additionally, DGLM estimates of infection risk aligned well with empirical risks during the first time steps of the simulation. These findings highlight the potential of the proposed decision support tool in providing valuable guidance for improving animal welfare and AMU. Further validation through real-world data collected from on-farm situations is necessary.