Evaluation of methods for calibrating in vitro digestibility estimates of ruminant feeds

Abstract

The present study summarizes a calibration study in which 104 samples, representing reliable and diversified data from all over the world selected for reliability, were analyzed for in vitro digestibility in three different laboratories. Special attention was given to the following topics: the variability between laboratories and methods of calibration between laboratories; contribution of standards; and source of error for both in vitro and in vivo data simultaneously. Correlation between the in vitro sets of the different laboratories was high ( r 2 from 0.931 to 0.975; residual standard deviation (RSD) from 0.037 to 0.025). Correlation between laboratories and in vivo data was also high, with similar r 2 and RSD for dry matter digestibility and somewhat lower for organic matter digestibility. Indirect calibration of each laboratory on in vivo DMD data by its regression on one of the other laboratories instead of directly on in vivo data, reduced r 2 by no more than 0.0009–0.0012. This was despite the fact that each laboratory had a different regression equation on the same in vivo data. Correction of in vitro values by means of a run factor based on three reference feeds with digestibility of 0.5, 0.6 and 0.8 did not improve the calibration. Correlating the residuals from the regression analyses of in vivo on in vitro data for pairs of similar feeds in the same run (one serving as a ‘standard’ for the other) resulted in very low and mostly insignificant r values. This indicates that in two similar feeds, the deviations of the in vitro digestibilities from their respective in vivo digestibilities were not consistent. One must therefore conclude that the inclusion of many standards of this nature does not improve the accuracy of the digestibility estimates. A few constant well spaced standards can serve as a check for large deviations that indicate something went wrong in a given run. A different calibration method using a functional relationship model instead of a regression model was also tested. It was found to be appropriate for in vivo/in vitro calibration because it not only takes into account errors of observation in both variables, but also calculates the relationship between in vivo (OMD) and in vitro data from all the laboratories simultaneously.