Profiles of dairy cows with different productive lifespan emerge from multiple traits assessed at first lactation: the case of a grassland-based dairy system

Abstract

This study focuses on extensive grassland-based dairy systems, where cows may be exposed to various hazards (climatic, health…) and limiting environments (feeding conditions) over their life. In such livestock systems, the ability of cows to remain in the herd (i.e. lifespan) reveals their ability to cope with such perturbations and to continue meeting the farmer's expectations. Productive lifespan was thus considered as a proxy of robustness in dairy cows. We hypothesized that productive lifespan of dairy cows can be predicted by a set of life function traits recorded during first lactation. A dataset composed of variables describing the entire productive life of 185 (Montbéliarde and Prim'Holstein) dairy cows managed in an experimental farm was used to test our hypothesis. This dataset contained information on 32 variables related to productive and reproductive performances, health status, average daily dry matter intake, body weight and body condition score describing the first lactation of all multiparous dairy cows, together with their productive lifespan (difference between age at culling and age at first calving). Clustering and multi-trait profile analyses of dairy cows revealed the existence of trade-offs between the productive lifespan of dairy cows and milk productivity, fertility and body weight at first lactation. For instance, cows with good functional performance (in terms of health and reproduction traits) but moderate milk production during first lactation had longer productive lifespan, whereas cows with moderate milk production but poor reproductive performance had a shorter one. We applied linear discriminant (LD) modelling technique to predict to which productive lifespan (short, average, long) profile an individual cow would belong, based on life function and production traits measured during the first lactation. K-fold cross-validation (k=4) of the LD model yielded an accuracy of 89% and precision ranged between 89 and 97%. Prediction quality of the cross-validated LD model suggests that our approach of classifying individual dairy cows is relevant and advantageous compared to classifications done per average cluster or profile. Studying traits of cows at first lactation enables early identification of cows that have a greater probability of living longer and thus have a better robustness.