Predicting probability of cure in bovine clinical mastitis.

Knowledge of the probability that a dairy cow will cure of mastitis when treated with antimicrobials would facilitate informed treatment decisions. A generalised linear mixed model was created for predicting the probability of cure of cases of clinical mastitis. Cure was defined as the absence of subclinical and clinical mastitis over the subsequent two or three milk-recordings. The data described 5845 cases of mastitis on 49 U.K. commercial dairy farms between 2003 and 2005, alongside cow- and herd-level variables proposed to be associated with cure. Data were partitioned into two subsets, for training and validating the model respectively. Further model validation was performed against modern data collected between 2020 and 2026. The final model was selected through backwards variable selection, guided by Akaike's information criterion (AIC). Variables significantly associated with cure described the historical occurrence of subclinical or clinical mastitis. Elevated SCC at the two most recent tests was associated with reduced odds of cure, as was the occurrence of three or more cases of clinical mastitis in the previous lactation. The proportion of tests during the previous lactation at which the SCC was greater than 100,000 cells/ml was inversely associated with cure probability. On the validation data, probabilistic predictions were well-calibrated although balanced accuracy of classifications was modest at 0.67. The model provides calibrated predictions of cure probability without the need to identify causal pathogens; its deployment will be of value to farmers wishing to make evidence-based mastitis treatment decisions consistent with the desire to minimise antimicrobial usage.