Comparing two software programs for fitting nonlinear, one- and two-compartment age-dependent digestion models: a Monte Carlo analysis

Authors: Moffet, Corey; Gunter, Stacey

Submitted to: Livestock Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/21/2020
Publication Date: 6/25/2020
Citation: Moffet, C., Gunter, S.A. 2020. Comparing two software programs for fitting nonlinear, one- and two-compartment age-dependent digestion models: a Monte Carlo analysis. Livestock Science. https://doi.org/10.1016/j.livsci.2020.104153.
DOI: https://doi.org/10.1016/j.livsci.2020.104153

Interpretive Summary: Grazinglands are the largest land use system in the world and the revenue received from these lands is vital to the economic health of rural communities. Understanding the nutrition of the animals that use these lands is paramount to making these systems profitable and providing a livelihood to producers and their families. One facet of managing the nutrition of grazing ruminants is an understanding of the digestion process to identify limiting factors and design management alternatives that meet and augment these production limitations. One tool used by nutritionist to study the nutrition of animals is nutritional models that describe and quantify the nutrients that animals consume and digest. At the USDA, Agricultural Research Service in Woodward, Oklahoma we examined a couple of these models and two statistical software programs used to fit data to these models. This research showed that there are differences in the repeatability of values from these programs, but over a wide range of diet types the software programs generally produced similar values. Further, we showed the more complex model examined was more challenging to fit to the data, but in the end it produced more precise values.

Technical Abstract: Using compartmental modeling of digesta kinetics is a valuable tool to ruminant nutritionist to access and quantify the site and extent of digestion within the gastrointestinal tract. To increase the value of one- (G2) and two-compartment (G2G1), age-dependent models, we characterized the repeatability and agreement of 2 programs for fitting these G2 and G2G1 models. Hence, we constructed replicated datasets of fecal marker concentrations for 81 synthetic animals and diet combinations by generating synthetic measured marker concentration values at 15 nominal times since dosing (0, 9, 12, 15, 18, 24, 32, 40, 48, 60, 72, 96, 108, 108, and 120 hours) by adding random error at these sample times to the hypothetical true fecal marker concentration for an animal and diet combination. Two sources of error were involved; sampling time (uniformly distributed between -3 to 1 hours) and marker concentration measurement (normally distributed with mean 0 and SD = 5 + 0.08CT, SD is a function the true concentration CT). The resulting fecal marker concentration datasets were fit to G2 and G2G1 models with 2 programs, one written for R and the other for SAS software. The resulting coefficients, K0, ' or '1, K2, and t, were used to calculate digesta kinetics parameters: particle passage rate, gastrointestinal DM fill, fecal DM output, gastrointestinal mean retention time, and rumen retention time. We evaluated repeatability of each method and agreement between methods. When fitting the 8100 datasets to the G2 model, all converged for both software. But, when fitting the same datasets to the G2G1 model, a small number did not converge. This is a problem when experiment units available in a study are few. The R software produced more repeatable parameter estimates than SAS software, but the ratios of repeatability to mean values was generally less than 10%. Bias and SD of differences between software packages were small, however G2G1 models produced smaller bias and SD of differences than G2 models. Bias and SD for digestion parameters between models and software packages were also small, however the G2G1 model again had smaller biases and SD of differences than the G2 model. Repeatability with R software was better than the SAS software, but the mean differences for parameters were small; the G2G1 model produced more repeatable results than the G2 model, but differences in parameter estimates between software were small for the G2G1 model.