Prediction of intake in growing dairy heifers under tropical conditions

A meta-analysis was conducted to develop models of the prediction of dry matter intake (DMI) in growing dairy heifers [postweaning to 390 kg of body weight (BW)] under tropical conditions. The adequacy of the models was assessed in a comparison with the 4 US models currently used to predict DMI [Quigley; National Research Council; and 2 Hoffman models]. The data set was created using 95 treatment means from 28 studies published in journals. The data set (studies) was randomly divided into 2 data subsets for the statistical analyses. The first data subset was used to develop the prediction equations for DMI (17 studies; 58 treatment means), and the second data subset was used to assess the adequacy of the predictive models (11 studies; 37 treatment means). The models were developed using nonlinear and linear mixed analyses. Breed (Bos taurus vs. Bos taurus × Bos indicus), BW (240.2±62.2 kg), and average daily gain (ADG, 0.83±0.28 kg/d) were the independent variables. No significant effects of the breed or the interactions between the breed and metabolic BW (BW(0.75)) or breed and ADG were detected. Thus, nonlinear [DMI=0.1175 × BW(0.75) - 3.4984 × e((-2.4690 × ADG))] and linear models [DMI=8.7147 - 0.2402 × BW(0.75) + 0.0027 × (BW(0.75))(2) + 3.6050 × ADG - 1.4168 × ADG(2)] were proposed for both breeds. The nonlinear model explained 81% of the variation in the DMI, over-predicted the DMI by 0.21 kg/d and predicted the DMI with a higher accuracy and precision than the linear model [root mean square error of prediction (RMSEP)=8.82 vs. 10.71% of the observed DMI, respectively]. The Quigley model explained only 54% of the variation in the DMI and was the fourth most accurate and precise model (RMSEP=11.21% of the observed DMI). The National Research Council model explained 69% of the variation in the DMI but under-predicted the DMI by 0.53 kg/d, with an RMSEP of 12.72% of the observed DMI and presence of systematic constant bias. The Hoffman exponential model I (BW as the input) adequately predicted the DMI with an accuracy that was similar to the proposed nonlinear model. The equation of the Hoffman exponential model I explained 75% of the variation in the DMI and over-predicted the DMI by 0.07 kg/d, which was the second most accurate and precise equation (RMSEP=9.35% of the observed DMI). However, the Hoffman exponential model II (BW and diet NDF as the inputs) did not adequately predict the DMI, because it explained only 54% of the variation in the DMI, under-predicted the DMI by 0.72 kg/d, and had a high RMSEP (17.96% of the observed DMI). The use of nonlinear models increase the accuracy and the precision of the prediction of DMI compared with the linear models. Only the models proposed in the present study, the Hoffman exponential model I (BW as the input), and the Quigley model were adequate for the prediction of the DMI of growing dairy heifers under tropical conditions.