Predicting feed intake and feed efficiency in lactating dairy cows using digesta marker techniques

Direct measurement of individual animal dry matter intake (DMI) remains a fundamental challenge to assessing dairy feed efficiency (FE). Digesta marker, is currently the most used indirect technique for estimating DMI in production animals. In this meta-analysis we evaluated the performance of marker-based estimates against direct or observed measurements and developed equations for the prediction of FE (g energy-corrected milk (ECM)/kg DMI). Data were taken from 29 change-over studies consisting of 416 cow-within period observations. Most studies used more than one digesta marker. So, for each observed measurement of DMI, faecal dry matter output (FDMO) and apparent total tract dry matter digestibility (DMD), there was one or more corresponding marker estimate. There were 924, 409 and 846 observations for estimated FDMO (eFDMO), estimated apparent total tract DMD (eDMD) and estimated DMI (eDMI), respectively. The experimental diets were based mainly on grass silage, with soya bean or rapeseed meal as protein supplements and cereal grains or by-products as energy supplements. Across all diets, average forage to concentrate ratio on a dry matter (DM) basis was 59 : 41. Variance component and repeatability estimates of observed and marker estimations were determined using random factors in mixed procedures of SAS. Between-cow CV in observed FDMO, DMD and DMI was, 10.3, 1.69 and 8.04, respectively. Overall, the repeatability estimates of observed variables were greater than their corresponding marker-based estimates of repeatability. Regression of observed measurements on marker-based estimates gave good relationships (R²=0.87, 0.68, 0.74 and 0.74, relative prediction error =10.9%, 6.5%, 15.4% and 18.7%for FDMO, DMD, DMI and FE predictions, respectively). Despite this, the mean and slope biases were statistically significant (P<0.001) for all regressions. More than half of the errors in all regressions were due to mean and slope biases (52.4% 87.4%, 82.9% and 85.8% for FDMO, DMD, DMI and FE, respectively), whereas the contributions of random errors were small. Based on residual variance, the best model for predicting FE developed from the dataset was FE (g ECM/kg DMI)=1179(±54.1) +38.2(±2.05)×ECM(kg/day)−0.64(±0.051)×BW (kg)−75.6(±4.39)×eFDMO (kg/day). Although eDMD was positively related to FE, it only showed a tendency to reduce the residual variance. Despite inaccuracy in marker procedures, eFDMO from external markers provided a reliable determination for FE measurement. However, DMD estimated by internal markers did not improve prediction of FE, probably reflecting small variability.