Predicting dry matter intake by growing and finishing beef cattle: Evaluation of current methods and equation development 1,2

Transcription

1 Published November 21, 2014 Predicting dry matter intake by growing and finishing beef cattle: Evaluation of current methods and equation development 1,2 U. Y. Anele,* 3 E. M. Domby,* 4 and M. L. Galyean* *Department of Animal and Food Sciences, Texas Tech University, Lubbock ; and Agriculture and Agri-Food Canada, Lethbridge, Alberta T1J 4B1, Canada ABSTRACT: The NRC (1996) equation for predicting DMI by growing finishing beef cattle, which is based on dietary NEm concentration and average BW 0.75, has been reported to over- and underpredict DMI depending on dietary and animal conditions. Our objectives were to 1) develop broadly applicable equations for predicting DMI from BW and dietary NEm concentration and 2) evaluate the predictive value of using NE requirements and dietary NE concentrations to determine the DMI required (DMIR) by feedlot cattle. Two new DMI prediction equations were developed from a literature data set, which represented treatment means from published experiments from 1980 to 2011 that covered a wide range of dietary NEm concentrations. Dry matter intake predicted from the 2 new equations, which were based on NEm concentration and either the ending BW for a feeding period or the DMI per unit of average BW (End BW and DMI/BW, respectively), accounted for 61 and 58% of the variation in observed DMI, respectively, vs. 48% for the 1996 NRC equation. When validated with 4 independent data sets that included 7,751 pen and individual observations of DMI by animals of varying BW and feeding periods of varying length, DMI predicted by the 1996 NRC equation, the End BW and DMI/BW equations, and the DMIR method accounted for 13.1 to 82.9% of the variation in observed DMI, with higher r 2 values for 2 feedlot pen data sets and lower values for pen and individual data sets that included animals on lower-energy, growing diets as well as those in feedlot settings. The DMIR method yielded the greatest r 2 values and least prediction errors across the 4 data sets; however, mean biases (P < 0.01) were evident for all the equations across the data sets, ranging from as high as 1.01 kg for the DMIR method to 1.03 kg for the 1996 NRC equation. Negative linear bias was evident in virtually all cases, suggesting that prediction errors changed as DMI increased. Despite the expanded literature database for equation development, other than a trend for lower standard errors of prediction with the DMI/BW equation, the 2 new equations did not offer major advantages over the 1996 NRC equation when applied to the validation data sets. Because the DMIR approach accounted for the greatest percentage of variation in observed DMI and had the least root mean square error values in all data sets evaluated, this approach should be considered as a means of predicting DMI by growing finishing beef cattle. Key words: beef cattle, dry matter intake required, feed intake, prediction 2014 American Society of Animal Science. All rights reserved. J. Anim. Sci : doi: /jas Supported by the Jesse W. Thornton Chair in Animal Science Endowment and Paul Whitfield Horn Professorship funds at Texas Tech University. 2 The authors are grateful to M. Brown, G. Erickson, T. Klopfenstein, J. MacDonald, B. Nuttelman, P. Beck, T. McAllister, N. DiLorenzo, C. Krehbiel, and A. DiCostanzo for providing the datasets used to evaluate equations. 3 Corresponding author: uchenna.anele@agr.gc.ca 4 Present address: Cargill Anim. Nutr., Amarillo, TX Received December 31, Accepted March 5, INTRODUCTION Optimizing cattle performance is essential for beef producers to remain competitive, and DMI is possibly the single most important factor influencing productivity in growing finishing beef cattle operations. Accurate prediction or measurement of DMI is a fundamental prerequisite in calculating nutrient requirements, balancing diets, and predicting performance with NE equations (NRC, 1996). The NRC (1996) equation for predicting DMI by growing finishing cattle, which is based on dietary NEm concentration and average BW 0.75, has been

2 Predicting dry matter intake by beef cattle 2661 found inadequate under certain conditions (Patterson et al., 2000; Block et al., 2001; McMeniman et al., 2009). For example, Patterson et al. (2000) evaluated the NRC equation for accuracy of DMI prediction using data from 7 beef cattle growing studies in which 54 diets were fed. They reported that the equation overpredicted DMI with low-quality diets and underpredicted DMI with high-quality diets, with the overall conclusion that the equation did not accurately predict intake by cattle on low-quality roughage diets. Previous results from our laboratory (McMeniman et al., 2009) indicated significant mean and linear biases across the range of observed DMI when NRC (1996) equations were evaluated with a commercial feedlot database consisting of 3,363 pen means collected from 3 feedlots over a 4-yr period. For the NRC model to be used more effectively to determine nutrient requirements and predict performance by beef cattle, new DMI prediction equations should be developed to address these limitations. One alternative to predicting DMI from inputs such as BW and dietary NE concentration is to calculate the DMI required (DMIR) to achieve a specified level of performance. This method involves using the NRC (1996) equations to determine the NEm and NEg required for a given BW and ADG and then calculating the DMIR by dividing the NE requirements by their respective dietary NE concentrations. This method has been used to identify feed efficiency differences in group-fed cattle (Tedeschi et al., 2006). Our objective was to 1) develop more robust equations for predicting DMI from BW and dietary NEm concentration and 2) evaluate the utility of the DMIR approach for predicting DMI by growing finishing cattle. MATERIALS AND METHODS Data were obtained from commercial feedlot databases and from the results of unpublished experiments conducted by university scientists at various locations around the United States and Canada. Thus, no live animals were used by the authors in conducting this project. Databases. Data sets were from 5 different sources: 1) Literature = a data set of 531 treatment means complied from studies with growing finishing beef cattle reported in the Journal of Animal Science from 1980 to 2011, 2) Alpharma = a commercial feedlot database described by Galyean et al. (2011), 3) McMeniman = a commercial feedlot database described by McMeniman et al. (2009), 4) Unpublished pen data = a database of pen means obtained from unpublished experiments conducted by 7 different university and government research units in the United States and Canada, and 5) Unpublished individual data = a database of individual animal observations obtained from unpublished experiments conducted by 3 different university and government research units in the United States and Canada. Descriptive statistics for the 5 databases used in the experiment are shown in Table 1. The minimum prerequisite for a study to be included in a data set was that diet composition, duration of the feeding period, sex, initial and final BW, ADG, and DMI were reported. For all data sets used, the diet composition on a DM basis was used to calculate dietary NEm and NEg concentration (as well as RDP) based on the concentrations reported for each ingredient in the NRC (1996) feed composition database. This approach was used to minimize the effect of feed composition data as a source of variation among studies. Only the Literature data set was used for development of new equations, whereas the other data sets were used to validate the developed equations. The Literature data set covered a wide range of dietary NEm concentrations and included treatment mean data used in development of the NRC (1996) NEm-based equation (1980 to approximately 1992; M. L. Galyean, personal communication) as well as additional treatment mean data from experiments conducted between approximately 1993 to Sauvant et al. (2008) highlighted the importance of considering additional sources of systematic variation in meta-analysis studies. To minimize these sources, we excluded from the Literature data set receiving study data in which cattle experienced a high rate of morbidity or data from studies in which the treatments involved any type of long-term stress; thus, the data set was limited to healthy animals. We also excluded studies in which the animals were limit fed. Efforts were made to include as many results as possible from experiments involving low-quality diets and thereby to address one of the limitations of the NRC (1996) model noted by Patterson et al. (2000). In addition to developed equations, the NRC (1996) equation for predicting DMI by growing finishing beef cattle and the DMIR method were tested in the 4 validation data sets. For the NRC (1996) equation, consistent with McMeniman et al. (2009), an arbitrary calf vs. yearling adjustment was made by assuming animals with an initial BW of 320 kg were calves, whereas those >320 kg were considered yearlings. The DMIR was calculated using the average shrunk BW of the pen of animals (or individual animal when applicable) and the shrunk ADG for the feeding period to determine the NEm and NEg (Mcal) required based on NRC (1996) equations. Because data were not typically available in the published studies to do so, no adjustments were made for effects of environment or previous plane of nutrition. Likewise, no adjustment was made to the dietary NEm concentration for the use of ionophores. Dietary NEm and NEg concentrations calculated for each study from feed composition data were divided into the dietary

3 2662 Anele et al. NEm and NEg requirements, respectively, with the results summed to yield the DMIR. Statistical Analyses. The mixed model procedure (St-Pierre, 2001) was used for equation development because the Literature dataset was compiled from multiple published studies that potentially varied in experimental design, experimental locations, year, and season as well as various other experimental conditions. Therefore, the data source (e.g., a study in a published article) was included as a random effect in the model, which not only accounted for the fixed effects of the dependent variables but also the random variation associated with studies. For development of new equations, an a priori decision was made to select dependent variables that were consistent with those used in the NRC (1996) equation (e.g., initial, final, and average shrunk BW and dietary NEm concentration). Nonetheless, other potential classification and independent variables (e.g., sex, duration of the feeding period, and dietary RDP concentration) were considered during model development. All statistical analyses were performed with Mixed and Reg procedures of SAS (SAS Inst. Inc., Cary, NC). For validation of developed equations as well as the NRC (1996) and DMIR approaches, observed DMI was regressed on predicted DMI for each equation. The coefficient of determination (r 2 ) was obtained as a measure of the strength of the relationship between observed and predicted DMI. Root mean square error (RMSE) was used to evaluate model precision. Shah and Murphy (2006) noted that mean bias is useful to test of the robustness of models, whereas linear bias can be used to test inadequacy in model structure. Both mean and linear biases were calculated by regression of residuals (observed predicted DMI) on mean-centered predicted DMI (St-Pierre, 2003). St-Pierre (2003) reported that by centering predicted DMI to the mean value, the intercept of the linear model is estimated at the mean value of the independent variable rather than at a value of 0. The intercept term at the mean value is a measure of the mean prediction bias, with a t test on the estimate of the intercept used to determine the statistical significance of this bias. The slope of this mean-centered regression is an estimate of the linear prediction bias, which reflects whether the prediction error is consistent across the range of data, again using a t test to evaluate significance. RESULTS AND DISCUSSION Equation Development. As noted previously, our desire was to develop new equations that used similar input variables to the NRC (1996) equations (e.g., some measure of BW and dietary NEm concentration). Because the NRC (1996) equation actually predicts NEm intake (Mcal/BW 0.75 ), and DMI is derived by dividing NEm Table 1. Descriptive statistics for the literature and validation data sets Data set 1 Mean SD Range Literature No. of observations 531 Steers, % Heifers, % Steers and heifers mixed, % 3.01 Avg. initial BW, kg to 480 Avg. final BW, kg to 663 DMI, kg/d to Days on feed to 226 Dietary NEm, Mcal/kg of DM to 2.31 Alpharma No. of observations 781 Steers, % Heifers, % Avg. initial BW, kg to 473 Avg. final BW, kg to 689 DMI, kg/d to 11.9 Days on feed to 289 Dietary NEm, Mcal/kg of DM 2.20 McMeniman No. of observations 3,323 Steers, % Heifers, % Avg. initial BW, kg to 451 Avg. final BW, kg to 663 DMI, kg/d to Days on feed to 315 Dietary NEm, Mcal/kg of DM to 2.25 Unpublished pen data No. of observations 2,286 Steers, % Heifers, % Avg. initial BW, kg to 531 Avg. final BW, kg to 751 DMI, kg/d to Days on feed to 273 Dietary NEm, Mcal/kg of DM to 2.39 Unpublished individual data No. of observations 1,361 Steers, % Heifers, % 2.87 Bulls, % Avg. initial BW, kg to 576 Avg. final BW, kg to 724 DMI, kg/d to Days on feed to 176 Dietary NEm, Mcal/kg of DM to Data sets were as follows: Literature = treatment means from studies with growing finishing beef cattle reported in the Journal of Animal Science from 1980 to 2011; McMeniman = a commercial feedlot database described by McMeniman et al. (2009); Alpharma = a commercial feedlot database described by Galyean et al. (2011); Unpublished pen data = a database of pen means obtained from unpublished experiments conducted by 7 different university and government research units in the United States and Canada; Unpublished individual data = a database of individual animal observations obtained from unpublished experiments conducted by 3 different university and government research units in the United States and Canada.

4 Predicting dry matter intake by beef cattle 2663 Table 2. The NRC (1996) NEm-based equation and 2 newly developed equations for predicting intake by growing finishing cattle derived from a literature database 1 Equation Dependent variable Equation components 1 r 2 RMSE NRC (1996) Calves DMI, kg/d [BW 0.75 ( NEm NEm )]/NEm Yearlings DMI, kg/d [BW 0.75 ( NEm NEm )]/NEm End BW DMI, kg/d End BW NEm NEm DMI/BW DMI, % of BW NEm NEm Data for the dependent variables were adjusted for the random effects of the literature source (i.e., study), and the coefficient of determination (r 2 ) and root mean square error (RMSE) were determined from the data after this adjustment. NEm = dietary NEm concentration (Mcal/kg of DM). For the NRC (1996) equation, BW (kg) = average shrunk BW (initial + final shrunk BW divided by 2). For the newly developed equations, End BW = shrunk BW (kg) at the end of the feeding period (not necessarily the final BW associated with slaughter at a quality grade endpoint), and DMI/BW = DMI divided by the average shrunk BW (kg; initial + final shrunk BW divided by 2) for the feeding period. intake by dietary NEm concentration and multiplying by average BW 0.75, an additional desire was to predict DMI more directly using inputs of dietary NEm concentration and BW. As noted previously, however, other independent or classification variables were considered in the process of equation development. Specifically, sex (steer, heifer, and mixed sexes), duration of the feeding period, and dietary RDP concentration were evaluated as potential variables to include in models. Of these variables, sex and RDP concentration proved to be nonsignificant (P > 0.10), whereas duration of the feeding period reported in the studies was significant (P < 0.05). Despite its significance, duration was not included in the final models because its effect on model R 2 and RMSE values was negligible, and duration is obviously confounded with the random effect of a study and not biologically meaningful in the context of predicting DMI to establish nutrient requirements. As expected from previous work with the NRC (1996) equation dietary concentration of NEm, NEm 2, and various measures BW were significant variables (P < 0.01) for predicting DMI. Of the measures of BW evaluated, the greatest R 2 and least RMSE values were obtained with the shrunk BW measured at the end of the feeding period (not necessarily the target BW at a particular body composition endpoint) as opposed to the average BW 0.75 used in the NRC (1996) equations. The new End BW equation developed from this analysis (Table 2) accounted for 88.4% of the variation in DMI in the Literature data set, with a RMSE of 0.44 kg. For this equation, the potential interaction of End BW with dietary NEm concentration was tested and found to be nonsignificant (P > 0.22). In an effort to use an approach somewhat more consistent with the NRC (1996) equation, which uses average BW, we also evaluated DMI per unit of average shrunk BW in an equation that included dietary concentration of NEm and NEm 2. As shown in Table 2, this DMI/BW equation accounted for a somewhat lesser percentage of the variation of observed DMI (61.9%) than the End BW equation, but we believed its simplicity warranted further evaluation. In addition, the greater R 2 with the End BW equation might reflect an association between the ending BW for a feeding period and the length of the feeding period, which, as stated previously, was considered as a possible model component. Equation Validation. The true test of the predictive value of any empirical regression equation is how well the equation works with an independent set of data. In the present study, the DMIR method, along with the NRC (1996) equation and the 2 new equations, were applied to data in the 4 independent data sets to predict DMI. As noted previously, observed DMI was regressed on predicted DMI for each equation or method and within each data set. Results of the validation analyses are shown in Table 3. Using the Alpharma and McMeniman data sets, DMI predicted by the NRC (1996) equation, the End BW and DMI/BW equations, and the DMIR method accounted for 66.6, 55.9, 73.4, and 79.3% and 66.6, 53.3, 74.5, and 82.9%, respectively, of the variation in observed DMI for the 2 data sets. The DMIR approach had the least RMSE value (0.389 and 0.386, respectively) in the Alpharma and McMeniman data sets and was the only equation with positive mean bias, indicating that it underpredicted DMI. The DMIR method had a negative linear bias (P < 0.001), suggesting that the magnitude of underprediction decreased with increasing DMI. The End BW, DMI/BW, and NRC (1996) equations overpredicted DMI, but there was no evidence of linear bias (P = 0.113) with the End BW equation. When the 4 equations were evaluated using the unpublished pen and individual data sets, the coefficient of determination (r 2 ) ranged from 0.36 to 0.52 and 0.13 to 0.25, for the pen and individual data sets, respectively. Mean and linear biases were evident (P < 0.01) in all the equations. In addition, all equations underpredicted DMI, but the magnitude of underprediction changed with increasing DMI (linear bias; P < 0.001). Similar to the evaluation using the Alpharma and McMeniman datasets, the DMIR method had the least RMSE values, indicating greater precision of the predictions. Although the DMIR method underpredicted DMI the least (mean bias of 0.14

5 2664 Anele et al. Table 3. Regression equations from 4 different data sets and 4 different methods of predicting intake by growing finishing beef cattle 1 Data set and methods Regression of observed on predicted DMI r 2 RMSE Mean bias P-value Linear bias P-value Alpharma DMI required DMI = Pred DMI < <0.001 NRC (1996) DMI = Pred DMI < <0.001 End BW DMI = Pred DMI < DMI/BW DMI = Pred DMI McMeniman DMI required DMI = Pred DMI < < NRC (1996) DMI = Pred DMI < <0.001 End BW DMI = Pred DMI < <0.001 DMI/BW DMI = Pred DMI < Unpublished pen data DMI required DMI = Pred DMI < <0.001 NRC (1996) DMI = Pred DMI < <0.001 End BW DMI = Pred DMI < <0.001 DMI/BW DMI = Pred DMI < <0.001 Unpublished individual data DMI required DMI = Pred DMI < <0.001 NRC (1996) DMI = Pred DMI < <0.001 End BW DMI = Pred DMI < <0.001 DMI/BW DMI = Pred DMI < < Data sets and equations are described in Table 1. RMSE = root mean square error; Pred DMI = predicted DMI. vs. 0.20, 0.55, and 0.22 kg for the NRC, End BW, and DMI/BW equations, respectively) in the unpublished individual data set, it had the highest mean bias in the pen data set (1.01 vs. 0.42, 0.42, and 0.69 kg for the NRC, End BW, and DMI/BW equations, respectively). Guiroy et al. (2001) noted that underprediction bias of up to 2% in the total DM delivered to feedlot cattle can be expected because of feed that was lost and not consumed by cattle (bunk cleaning, wind losses, etc.). Underprediction bias in our case with the DMIR approach was approximately 1.4% for the individual data set and nearly 10% of the mean DMI for the pen data set. Interactions between numerous dietary, animal, environmental, and management factors are likely to be responsible for the significant biases observed in evaluation of these equations. Control of DMI by ruminants is complex and multifactorial (Forbes, 2003), and using only BW and NEm concentration of the diet as independent variables does not account for these additional sources of variation affecting DMI. Similar to our results for the DMIR approach, Williams et al. (2006) reported that DMIR predicted by Cornell Value Discovery System (CVDS) model (Tedeschi et al., 2004) underpredicted DMI with an average bias of 3.4% using data from 502 individually fed steers produced in a heterosis experiment. This method has been used to identify feed efficiency differences in group-fed cattle. The DMIR for maintenance was adjusted for cold stress in their model; however, in the present analyses, as noted previously, insufficient information was available to make potential adjustments to maintenance requirements suggested by NRC (1996; Fig. 1). The CVDS model accounted for 44% of the variation in the observed DMI in the Williams et al. (2006) study compared with a range of 24.9 to 82.9% for the DMIR method across the 4 data sets in the present study. Contrary to our results, Williams et al. (2006) reported that DMIR predicted using the Decision Evaluator for the Cattle Industry (DECI) model (Williams and Jenkins, 2003a,b) overpredicted DMI with an average bias of 0.4% and accounted for 53% of the variation in the observed DMI. Both the CVDS and DECI models can be used to predict animal performance when DMI and nutrient supply are known, and these models are capable of working in reverse manner to predict the DMI and nutrient supply required for an animal to achieve a known level of performance. We observed lower coefficients of determination and higher RMSE with the unpublished pen and individual data sets than with the Alpharma and McMeniman data sets. The dietary NEm concentration varied little in these latter 2 data sets, and the feeding periods were longer and less variable than in the unpublished pen and individual data sets. In addition, the distribution of sex was more variable in the unpublished data sets, with the individual data set including a high proportion of bulls. Longer feeding periods probably stabilized the DMI data as well as potentially yielding a better estimate of average BW compared with some of the shorter feeding periods in the unpublished pen and individual data sets. This result might also suggest that BW is a more important factor than NEm in predicting DMI, which is consistent

6 Predicting dry matter intake by beef cattle 2665 Figure 1. Comparison of the 2 new equations with the NRC (1996) equation for predicting DMI using data from the Literature data set from which the new equations were derived. (A) End BW equation; (B) DMI/BW equation. with the importance of initial BW in predicting DMI by feedlot cattle (Galyean et al., 2011). Archer et al. (1997) reported 70 d as the optimal minimum duration for measurement of growth rate, feed intake, feed conversion, and residual feed intake using postweaning feed intake and BW data from 760 Angus, Hereford, Polled Hereford, and Shorthorn heifer and Angus bull progeny from 78 sires. Therefore, the longer feeding periods and associated greater stability in DMI and BW measurements, and possibly differences in the distribution of sexes, likely resulted in the increased ability to account for variation in DMI in the Alpharma and McMeniman data sets compared with the 2 unpublished data sets. Because dietary NEm concentration is an important component of all the approaches we tested to predict DMI, inaccuracies in estimating NEm concentration likely contributed to prediction errors. As noted previously, in an effort to decrease variation associated with use of multiple databases, we used tabular values from NRC (1996) to determine the NEm concentration of diets in the Literature dataset as well as the 4 evaluation datasets. Perhaps the dietary NEm concentrations were more accurately estimated for the Alpharma and McMeniman datasets than for the 2 unpublished datasets. For future efforts such as ours that use literature data for development of equations, it is important for authors who publish data from feeding studies to fully describe the diets used and to provide relevant digestibility data and energy concentration values when possible. To assess whether the greater range in NEm concentration noted with the unpublished pen and individual datasets (Table 1) affected the predictive ability of the equations we tested, prediction errors were plotted against dietary NEm concentration for the 4 validation datasets (data not shown). These plots suggested no major differences in prediction errors for diets of lower vs. higher NEm concentration. It should be noted that although the range in dietary NEm concentrations was greater with the 2 unpublished datasets, the majority of observations with lower NEm concentrations were in the individual dataset (67.8% less than a NEm concentration of 1.8 Mcal/kg vs. 2% less than 1.8 Mcal/kg in the pen dataset). Therefore, a greater range in energy concentration does not likely explain the lower predictive value with the unpublished datasets. In addition to the possible effects of dietary NEm concentration on our predictions, we did not attempt to remove study or treatment effects from the 4 evaluation datasets. Although our approach might have increased the prediction errors, the variation associated with studies and treatments in these datasets reflects situations to which these equations would be applied in practice. Therefore, adjusting such variation out of the analyses would likely result in an artificial improvement in the predictive ability of the equations, which might not be realized in practice. The observed vs. predicted DMI in the 4 validation data sets for 2 of the approaches we evaluated are shown in Fig. 2. The DMIR approach was chosen for this graphical presentation because it consistently had the greatest r 2 values in the validation data sets as well as the least RMSE values. The new DMI/BW equation was chosen because it generally provided superior fits to data than the End BW equation and also yielded equal or smaller RMSE values to the NRC (1996) equation across the 4 validation data sets. The DMI/BW equation accounted for 73 and 75% of the variation in the observed DMI vs. 79 and 83% by the DMIR method using the Alpharma and McMeniman data sets, respectively. Overall, the proportion of variation in DMI accounted for between observed and predicted DMI in the DMI/ BW equation was similar to the validation results presented by Galyean et al. (2011) using the Alpharma and McMeniman data sets with an equation based on initial shrunk BW. When the 2 equations were evaluated using the unpublished pen and individual data sets, the DMIR approach had a greater coefficient of determination than the DMI/BW equation (0.52 and 0.25 vs and 0.16).

7 2666 Anele et al. Figure 2. Observed vs. predicted DMI in 4 validation data sets (Alpharma, McMeniman, Unpublished Pen Data, and Unpublished Individual Data). Predicted DMI was determined using the new equation (DMI/BW equation) or the DMI required to achieve the observed performance (DMI required). The solid line in each figure represents perfect agreement between observed and predicted values, and the dashed line represents the regression of observed on predicted values. (A), (B), (C), and (D): results for the DMI/BW equation; (E), (F), (G), and (H): results for the DMI required method. The DMI predicted over a range of NEm concentrations for the End BW, DMI/BW, and NRC (1996) equation is shown in Fig. 3. For this illustration, a 450-kg average shrunk BW (310 and 590 kg initial and final shrunk BW, respectively) fed diets ranging in NEm concentration from 1 to 2.40 Mcal/kg was used. Given the average BW, the yearling intercept term of was used for the NRC (1996) calculation. The End BW and DMI/

8 Predicting dry matter intake by beef cattle 2667 Figure 3. Comparison of DMI predictions for a medium-frame (450-kg average BW) steer using the 2 new literature-derived equations and NRC (1996) equation. BW curves across the range in NEm are of similar shape; however, the End BW equation yields greater predicted DMI values across the range. In contrast, the NRC (1996) equation predicted the least DMI at lower dietary NEm concentrations, intermediate DMI at mid-range NEm concentrations, and higher DMI than the other 2 equations at the upper end of the range in NEm concentrations, reflecting the observations reported by Patterson et al. (2000) in their evaluation of the NRC (1996) equation. Despite differences among the equations, it is clear that when applied to independent data sets for validation, the new equations provided only modest improvements at best and in some cases offered no real advantage over the NRC (1996) equation. This lack of improvement in predictive ability is disappointing, but it illustrates the difficulty in developing accurate predictions of DMI by growing finishing beef cattle, no doubt reflecting the complex factors controlling DMI and the inability to adequately account for these factors using regression methods and a few independent variables. Given that the DMIR approach was as useful as the 3 regression equations, accounting for the greatest percentage of variation in observed DMI with the least RMSE values in all data sets evaluated, efforts to further investigate and refine this method of predicting DMI by growing finishing cattle should be considered. LITERATURE CITED Archer, J. A., P. F. Arthur, R. M. Herd, P. F. Parnell, and W. S. Pitchford Optimum postweaning test for measurement of growth rate, feed intake, and feed efficiency in British breed cattle. J. Anim. Sci. 75: Block, H. C., J. J. McKinnon, A. F. Mustafa, and D. A. Christensen Evaluation of the 1996 NRC beef model under western Canadian environmental conditions. J. Anim. Sci. 79: Forbes, J. M The multifactorial nature of food intake control. J. Anim. Sci. 81(E. Suppl. 2):E139 E144. Galyean, M. L., N. DiLorenzo, J. P. McMeniman, and P. J. Defoor Alpharma beef cattle nutrition symposium: Predictability of feedlot cattle growth performance. J. Anim. Sci. 89: Guiroy, P. J., D. G. Fox, L. O. Tedeschi, M. J. Baker, and M. D. Cravey Predicting individual feed requirements of cattle fed in groups. J. Anim. Sci. 79: McMeniman, J. P., P. J. Defoor, and M. L. Galyean Evaluation of the National Research Council (1996) dry matter intake prediction equations and relationships between intake and performance by feedlot cattle. J. Anim. Sci. 87: NRC Nutrient requirements of beef cattle. 7th ed. Natl. Acad. Press, Washington, DC. Patterson, T., T. J. Klopfenstein, T. Milton, and D. R. Brink Evaluation of the 1996 beef cattle NRC model predictions of intake and gain for calves fed low or medium energy density diets. Nebraska Beef Cattle Report, Paper 385. University of Nebraska, Lincoln, Nebraska. Sauvant, D., P. Schmidely, J. J. Daudin, and N. R. St-Pierre Meta-analysis of experimental data in animal nutrition. Animal 2: Shah, M. A., and M. R. Murphy Development and evaluation of models to predict the feed intake of dairy cows in early lactation. J. Dairy Sci. 89: St-Pierre, N. R Invited review: Integrating quantitative findings from multiple studies using mixed model methodology. J. Dairy Sci. 84: St-Pierre, N. R Reassessment of biases in predicted nitrogen flows to the duodenum by NRC J. Dairy Sci. 86: Tedeschi, L. O., D. G. Fox, M. J. Baker, and D. P. Kirschten Identifying differences in feed efficiency among group-fed cattle. J. Anim. Sci. 84: Tedeschi, L. O., D. G. Fox, and P. J. Guiroy A decision support system to improve individual cattle management. 1. A mechanistic dynamic model for animal growth. Agric. Syst. 79: Williams, C. B., G. L. Bennett, T. G. Jenkins, L. V. Cundiff, and C. L. Ferrell Using simulation models to predict feed intake: Phenotypic and genetic relationships between observed and predicted values in cattle. J. Anim. Sci. 84: Williams, C. B., and T. G. Jenkins. 2003a. A dynamic model of metabolizable energy utilization in growing and mature cattle. I. Metabolizable energy utilization for maintenance and support metabolism. J. Anim. Sci. 81: Williams, C. B., and T. G. Jenkins. 2003b. A dynamic model of metabolizable energy utilization in growing and mature cattle. II. Metabolizable energy utilization for gain. J. Anim. Sci. 81: