Using Longitudinal Data of Somatic Cell Counts and Milk Yield to Identify Mastitis-Resilient Cows

Bovine mastitis remains a significant concern for the dairy industry, affecting animal welfare, farm profitability, and milk quality. Despite genetic and non-genetic control efforts, its multi-pathogenic nature makes eradication challenging, highlighting the importance of mastitis resilient cows (animals that are minimally affected or recover quickly after a health challenge). Resilience is commonly quantified using resilience indicators (RIs) based on performance deviations from a hypothetical unperturbed trajectory (i.e., the expected performance in the absence of challenges). To date, RIs for dairy cows have been constructed exclusively from time-series performance data (e.g., milk yield). However, to breed for resilience to mastitis, it would be more useful to develop mastitis-specific indicators than general production-based ones. This study explores the use of daily records of somatic cell counts (SCC) and milk yield to identify and characterise potential perturbations associated with mastitis. Specifically, we evaluated alternative statistical approaches for estimating individuals’ unperturbed trajectories and assessed their consistency in detecting periods of systematic deviation. Data from the Swedish University of Agricultural Sciences cattle infrastructure, Gigacow, were used. The dataset includes daily records of milk yield and online SCC from 1,052 dairy cows across up to 7 lactations, from 6 Swedish dairy farms using automatic milking systems with online cell counters. Three methods were applied separately to the milk yield and SCC data: (i) weighted spline, where data points likely associated with perturbations were given lower weights (e.g., milk yield drops or SCC spikes); (ii) quantile regression (targeting the 0.7 quantile for milk yield and 0.3 quantile for SCC); and (iii) iterative polynomial regression, where large negative (milk yield) and large positive residuals (SCC) were progressively removed and the curve refitted to minimise the influence of perturbed observations. Deviations from the estimated trajectories were computed, and potential perturbations characterised. Preliminary findings suggest that quantile regression and weighted spline produce similar trajectories and identify comparable perturbations. In contrast, the iterative polynomial method generates a different trajectory and detects fewer perturbations, reflecting a conservative approach for identifying perturbations. These results can help determine the most appropriate method for detecting mastitisrelated perturbations and, ultimately, identify the most mastitis-resilient animals.