A Bayesian approach to analyze energy balance data from lactating dairy cows1

• Published in: Journal of dairy science Vol. 94; no. 5; pp. 2520 - 2531
• Main Authors: Strathe, A.B., Dijkstra, J., France, J., Lopez, S., Yan, T., Kebreab, E.
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.05.2011
• Subjects: energy metabolism; modeling; modeling; energy metabolism
• ISSN: 0022-0302; 1525-3198
• DOI: 10.3168/jds.2010-3836

The objective of the present investigation was to develop a Bayesian framework for updating and integrating covariate information into key parameters of metabolizable energy (ME) systems for dairy cows. The study addressed specifically the effects of genetic improvements and feed quality on key parameters in current ME systems. These are net and metabolizable energy for maintenance (NEM and MEM, respectively), efficiency of utilization of ME for milk production (kL) and growth (kG), and efficiency of utilization of body stores for milk production (kT). Data were collated from 38 studies, yielding 701 individual cow observations on milk energy, ME intake, and tissue gain and loss. A function based on a linear relationship between milk energy and ME intake and correcting for tissue energy loss or gain served as the basis of a full Bayesian hierarchical model. The within-study variability was modeled by a Student t-distribution and the between-study variability in the structural parameters was modeled by a multivariate normal distribution. A meaningful relationship between genetic improvements in milk production and the key parameters could not be established. The parameter kL was linearly related to feed metabolizability, and the slope predicted a 0.010 (−0.0004; 0.0210) change per 0.1-unit change in metabolizability. The effect of metabolizability on kL was smaller than assumed in present feed evaluation systems and its significance was dependent on collection of studies included in the analysis. Three sets of population estimates (with 95% credible interval in parentheses) were generated, reflecting different degrees of prior belief: (1) Noninformative priors yielded 0.28 (0.23; 0.33) MJ/(kg0.75d), 0.55 (0.51; 0.58), 0.86 (0.81; 0.93) and 0.66 (0.58; 0.75), for NEM, kL, kG, and kT, respectively; (2) Introducing an informative prior that was derived from a fasting metabolism study served to combine the most recent information on energy metabolism in modern dairy cows. The new estimates of NEM, kL, kG and kT were 0.34 (0.28; 0.39) MJ/(kg0.75d), 0.58 (0.54; 0.62), 0.89 (0.85; 0.95), and 0.69 (0.60; 0.79), respectively; (3) finally, all informative priors were used that were established from literature, yielding estimates for NEM, kL, kG, and kT of 0.29 (0.11; 0.46) MJ/(kg0.75d), 0.60 (0.54; 0.70), 0.70 (0.50; 0.88), and 0.80 (0.67; 0.97), respectively. Bayesian methods are especially applicable in meta-analytical studies as information can enter at various stages in the hierarchical model.