Estimating feed efficiency: Evaluation of mathematical models to predict individual intakes of steers fed in group pens

To evaluate feed efficiency using residual feed intake (RFI), it is necessary to measure and record daily feed intake for each animal. This can be accomplished by housing them in individual pens or by using sophisticated electronic feeders in group pens. All the available options are very expensive and very laborious; therefore, several researchers have developed methods to predict individual DMI of cattle fed in group pens. Three intake models were tested with a data set of 60 Angus x Hereford steers fed a corn-based finishing diet in both group and individual pens. After the first 60 d (period 1) of the study, animals were switched from group to individual pens, and then vice versa for another 60 d (period 2); thus, the entire feeding trial was 120 d long. No difference was observed in DMI between periods for steers fed individually (period 1 = 10.9 kg/d and period 2 = 11.2 kg/d, P = 0.44), but a difference was observed in group pens (period 1 = 12.7 kg/d and period 2 = 10.9 kg/d, P < 0.01). In addition, no difference (P ≥ 0.15) was observed in carcass characteristics, such as HCW, dressing percentage, quality grade, LM area, KPH percentage, yield grade, or backfat between RFI groups (low, medium, and high). Average daily gain and G:F were not different between RFI groups within each period (P ≥ 0.06), but there were period differences (P < 0.001). Models 1 and 2 were based on growth, carcass composition, and nutrient requirements, whereas model 3 was based on the heterogeneity of pen intakes when cattle were rotated through the pens on a daily basis. Models 1 and 2 were forced through the mean observed DMI, so the mean bias was zero, but they were not precise, with a slope bias greater than 50%. Model 3 showed low accuracy (mean bias = 20%), but it was precise, with a slope bias of 21%. Because RFI is the error of the DMI equation, any inaccuracy when estimating intake will lead to a bias in the prediction of RFI. In conclusion, these models could be used to predict mean DMI, but they were not adequate for estimating RFI.