Supplemental material for "Evaluation of ruminal amino acid outflow predictions by nutritional models in dairy cattle"

The objective was to evaluate the fit statistics for the predictions of ruminal AA outflows by NRC (2001), NASEM (2021), and NittanyCow (NC) models in dairy cattle. Sixty-seven studies published in English between 1984 and 2020 with 252 treatment means were considered for analysis. Lactational performance (DMI, milk yield, and BW data), dietary nutrient composition (CP, NDF, ADF, ether extract, and ash), and ruminal outflow of MP supply, microbial CP (MicP), RUP, and EAA determined by omasal or duodenal sampling techniques were extracted and used in the analysis. Production and dietary nutrient composition data reported in the studies were also used as inputs to predict ruminal outflows of MP supply, MicP, RUP, and EAA using NRC, NASEM, and NC models. Model fit statistics (i.e., observed – predicted means) and performances of nutritional models were assessed using root mean squared error (RMSE, g/d and % of observed mean), and Lin’s concordance correlation coefficient (CCC). Predictions were considered very accurate, accurate, acceptable, or compromised if mean or linear biases (% of observed mean) were < 5, 5 to 10, 10 to 15, or > 15%, respectively. Overall and based on mean biases, predictions of all response variables were very accurate or accurate across all nutritional models. Models performed similarly when predicting MP supply with RMSE (% of observed mean) ranging from 18 to 19% and CCC ranging from 68 to 72%. Predictions of MicP and adjusted RUP (RUP + endogenous CP) outflows had RMSE ranging from 26 to 36% and CCC ranging from 35 to 58%. Predictions of Lys, Met, and His had RMSE ranging from 23 to 30% and CCC ranging from 42 to 64%. The NASEM was superior to NRC model when predicting Arg, His, Thr, and Val outflows. Compared with observed values, NC seems to predict MicP, Arg, His, Met, Phe, and Thr outflows well. Additionally, NC and NASEM (2021) predicted MP supply, Arg, His, Ile, Leu, Lys, Phe, and Thr similarly. All models presented some limitations if an acceptable threshold for the linear bias is set at < 5% observed mean to avoid Type I error. These biases should be considered by nutritionists when balancing rations for AA.