Short communication: Prediction of intake in dairy cows under tropical conditions

A meta-analysis was conducted to develop a model for predicting dry matter intake (DMI) in dairy cows under the tropical conditions of Brazil and to assess its adequacy compared with 5 currently available DMI prediction models: Agricultural and Food Research Council (AFRC); National Research Council (NRC); Cornell Net Carbohydrate and Protein System (CNCPS; version 6); and 2 other Brazilian models. The data set was created using 457 observations (n=1,655 cows) from 100 studies, and it was randomly divided into 2 subsets for statistical analysis. The first subset was used to develop a DMI prediction equation (60 studies; 309 treatment means) and the second subset was used to assess the adequacy of DMI predictive models (40 studies; 148 treatment means). The DMI prediction model proposed in the current study was developed using a nonlinear mixed model analysis after reparameterizing the NRC equation but including study as a random effect in the model. Body weight (mean = 540 ± 57.6 kg), 4% fat-corrected milk (mean = 21.3 ± 7.7 kg/d), and days in milk (mean = 110 ± 62 d) were used as independent variables in the model. The adequacy of the DMI prediction models was evaluated based on coefficient of determination, mean square prediction error (MSPE), root MSPE (RMSPE), and concordance correlation coefficient (CCC). The observed DMI obtained from the data set used to evaluate the prediction models averaged 17.6 ± 3.2 kg/d. The following model was proposed: DMI (kg/d) = [0.4762 (± 0.0358) × 4% fat-corrected milk + 0.07219 (± 0.00605) × body weight(0.75)] × (1 - e(-0.03202 (± 0.00615) × [days in milk + 24.9576 (± 5.909)])). This model explained 93.0% of the variation in DMI, predicting it with the lowest mean bias (0.11 kg/d) and RMSPE (4.9% of the observed DMI) and the highest precision [correlation coefficient estimate (ρ) = 0.97] and accuracy [bias correction factor (Cb)=0.99]. The NRC model prediction equation explained 92.0% of the variation in DMI and had the second lowest mean bias (0.42 kg/d) and RMSPE (5.8% of the observed DMI), as well as the second highest precision (ρ = 0.94) and accuracy (Cb = 0.98). The CNCPS and AFRC DMI prediction models explained 93.0 and 85.0% of the variation in DMI but underpredicted DMI by 1.8 and 1.4 kg/d, respectively. These 2 models (CNCPS and AFRC) resulted, respectively, in RMSPE of 11.3 and 10.7% of the observed DMI, with moderate to high precision (ρ = 0.81 and 0.82) and accuracy (Cb = 0.84 and 0.89). The remaining 2 models resulted in the poorest results, underpredicting DMI by 2.3 and 1.9 kg/d, with RMSPE of 22.8 and 14.9% of the observed DMI and moderate to low precision (ρ = 0.49 and 0.76) and accuracy (Cb = 0.81 and 0.86). The new model derived from the current meta-analytical approach provided the best accuracy and precision for predicting DMI in lactating dairy cows under Brazilian conditions.