Non-parametric approaches to the impact of Holstein heifer growth from birth to insemination on their dairy performance at lactation one

Parametric approaches have been used widely to model animal growth and study the impact of growth profile on performance. Individual variation is often not considered in such approaches. However, non-parametric modelling allows this. Such an approach, based on spline functions, was used to study the importance of growth profiles from age 0 to 15 months (i.e. insemination) on milk yield and composition in primiparous cows. A dataset of 447 heifers was used for analysis of growth performance; 296 of them were also used to study impact on lactation. All of them originated from a French experimental herd and were born between 1986 and 2006. Clustering methods were also tested. Comparison of spline methods showed that a cubic spline interpolation method, with no smoothing parameter, was best suited to studying heifer growth. Similarly, partitioning around medoids proved the most accurate clustering method for classifying heifer growth into groups. The results of these analyses agreed with those previously published, supporting the utility of these methods. A final study on the impact of breakdowns in the growth curves was performed. A breakdown was considered only when the derivative of the interpolation function was negative or zero. Of the 447 heifers initially used, 125 (Gr0), 175 (Gr1) and 147 (Gr2) had no, one, or two or more breakpoints during the 0-15 months of age period. Milk yield on a 305 d basis was significantly reduced with an increased number of breakpoints (6548 v. 6828 and 6905 kg for Gr2, Gr1 and Gr0 animals, respectively). Fat content was also higher in Gr2 than in Gr0 groups, but overall, no difference in total fat or protein-corrected milk production was noted. The intersection between groups for growth and groups for breakdowns confirmed that animals with two or more breakdowns belonged more frequently to the group with the lowest growth performance. These results offer the possibility of analysing large databases, originating from an automatic collecting system (e.g. milking robots) or from different herds, breeds, genetics, etc. These approaches could also be used for studies on body score index, girth development, lactation profiles, etc. and in other species, such as dairy goats or beef cattle. They could find use in the development of new models of prediction, e.g. the probability of heat appearance on an animal basis, which could be included among useful management tools.