Integrating diverse data sources to predict disease risk in dairy cattle

Livestock farming is currently undergoing a digital revolution and becoming increasingly data-driven. Yet, such data often reside in disconnected silos making it impossible to leverage their full potential to improve animal well-being. Here, we introduce a precision medicine approach, bringing together information streams from a variety of life domains of dairy cattle to predict eight common and economically important diseases. Dairy cows are part of a highly industrialised environment. The animals and their surroundings are closely monitored and environmental, behavioural and physiological observations are readily accessible yet seldomly integrated. We use random forest classifiers trained on data from 5,828 animals in 166 herds in Austria to predict occurrences of lameness, acute and chronic mastitis, anoestrus, ovarian cysts, metritis, ketosis (hyperketonemia) and periparturient hypocalcemia (milk fever). To assess the importance of specific cattle life domains and individual features for these predictions, we use multivariate logistic regression and feature permutation approaches. We show that disease in dairy cattle is a product of the complex interplay between a multitude of life domains such as housing, nutrition or climate, and identify a range of features that were previously not associated with increased disease risk. For example, we can predict anoestrus with high sensitivity and specificity (F1=0.72) and find that housing, feed and husbandry variables such as barn design and time on pasture are most predictive of this disease. We also find previously unknown associations of features with disease risk, for example humid conditions, which significantly decrease the odds for ketosis. Our findings pave the way towards data-driven point-of-care interventions and demonstrate the added value of integrating all available data in the dairy industry to improve animal well-being and reduce disease risk.