Prediction equations for the energy values of meat and bone meal for poultry using the meta-analysis principle

The demand of animal-origin protein human consumption keeps growing. This encourages research in poultry, in the search of alternatives aiming to reduce production costs. Feeding represents nearly 70% of the total production cost, and it has shown to be an important factor whenever attempts are made to reduce such cost. Poultry nutrition research is being conducted aiming to improve the use of rendered products.

The energy values of feedstuffs used in poultry are essential for a successful formulation of efficient rations, meeting the nutritional requirements of poultry in their various production stages. Two classification methods exist for feed energy determination i.e., direct and indirect methods. The direct or conventional methods require a calorimetric pump and metabolic assays. These methodologies are labor intensive, time communing, and expensive. On the other hand, the indirect methods include prediction equations based on the proximal composition of feeds, and they are routinely obtained in the laboratory.

Prediction equations are considered a rapid, convenient, inexpensive means for the nutritional evaluation of feeds. The indirect method uses prediction equations to evaluate the chemical composition of feeds. Such equations are generated based on simple chemical analyses such as nitrogen, crude energy, ether extract, mineral matter, calcium and phosphorus for animal-origin feedstuffs. Prediction equations can be obtained through meta analysis, which has been defined as the "analysis of the analysis", or "the statistical analysis of a large collection of results from analysis of individual studies, aiming to complement the discoveries" (Glass, 1976). Several meta-analysis studies are being conducted in order to collect data from different conditions, comparing different though related results. The purpose of this effort was to prepare prediction equations using the meta-analysis principle to generate energy values for rendered products commonly used for poultry feeding.

the information used herein refers to the nitrogen balance-corrected apparent metabolizable energy (AMEn) values, and the chemical composition of rendered products commonly used in poultry feed formulation. This data was taken from a broad, meticulous bibliographic review, intended to include the maximum number of studies on this particular topic. For this purpose, congress/symposium proceedings, libraries, and data basis in the CAPES podcast, including the Web of Science, and several others. Literature review included research performed and reported in Brazil in the last 30 years, aiming to obtain the maximum number of data that could influence the analysis. Forty three (43) reports of meat and bone meal experiments were included. After the review, the information was tabulated as per the feedstuff, the metabolic analytical methodology to determine the energy value, the gender and age of the animals involved in the experiments, as well as the chemical composition of each feed. Once tabulated and categorized, the data was analyzed by applying the meta-analysis principles, identifying the best suited equation for AMEn calculation, as a function of the chemical composition. Given that not all publications included all the chemical composition details, the information was analyzed separately as either complete or incomplete data. Complete data was that containing all the information regarding the chemical composition, while incomplete data included those reports not showing Ca or P data. The variation found in the experiments gathered in groups by the meta-analysis, typically corresponds to differences in the methodology employed, animal gender, and in the feed, amongst others. This diversity must be considered at the time of the analysis. The effects considered were: animal gender, animal age and methodology used in the metabolism assay (either total collection or forced feeding + total collection), and the feed used. After identifying such effects, codes were assigned to each particular effect, and then the codes were grouped as to determine the groups to be included in the weighed least square analysis (that will be explained below), in order to consider the above-mentioned effects and consider the meta-analysis procedure. For the effect of the gender, 3 groups were defined: males = 1; females = 2; as hatched (both sexes) = 3. Regarding the age, 4 groups existed: 1 = first and second weeks of life; 2 = third and fourth weeks of life; 3 = fifth and sixth weeks of life;  4 = >6 weeks and undefined ages. As far as the methodology is concerned, only two groups were included: 1 = total excreta collection; 2 = forced feeding + total excreta collection. Regarding the feedstuffs, five groups were included: 1 = meat and bone meal; 2 = offal meal; 3 = fishmeal; 4 = mixed meal; 5 = pork meal. Given that the multiple linear regression statistical model was used, parameter estimates were determined using the weighed least square method (Hoffmann and Vieira, 1977). In order to the weighing factor, the groups predetermined in the study were considered, but the method used for such purpose was the weighed least square method where the inverse variance (1/s²i) is considered for each group. With the purpose of learning about the structure of the relationships among the variables chemical composition and per-feedstuff energy values, the correlations recommended by Pearson, Draper and Smith (1981) were calculated among all pairs possible, using the Proc Corr software by SAS (Statistical Analysis System, 2000).

Table 1. Correlation coefficients calculated amongst all variables chemical composition and AMEn in meat and bone meals

All the variables showed significant correlations with AMEn. Both CP (P<0.05) and EE (P<0.05) were positively correlated with AMEn. MM, Ca and P showed a negative correlation with AMEn. EE was the only variable showing a positive correlation with all independent variables as well as with AMEn. This emphasizes the importance of EE in the determination of AMEn by prediction equations. MM was the variable with a negative and significant correlation (P<0.05) with AMEn. Therefore, if for any reasons the MM content is high, the energy content in meat and bone meal would be reduced in the same fashion and, also, the opposite is true. Considering the information catalogued, the prediction equation obtained that showed the highest determination coefficient (R², 0.44) for meat and bone meal AMEn was AMEn = 1,839.20 + 105.60 EE -176.24 P. Likewise, with a R² of 44% and a statistically-significant negative correlation (P<0.05) when the pool of variables is considered in the adjusted model, the variable considered as the most important one in the model was P (partial R² = 0.26), followed by EE (partial R² = 0.18) which in turn is positively and significantly correlated (P<0.05). The predicted equations for meat and bone meal can be seen in Table 2.

Table 2. Prediction equations obtained to estimate the AMEn values in meat and bone meal, in function of the chemical composition of meat and bone meals (vales expressed on a dry matter basis)

The adjustment of the equation with a determination coefficient of 0.44 could have been influenced by the number of data. Nevertheless, it can be seen that the equation with the largest number of variables was the one with the highest determination coefficient. The AMEn value found by Rostagno et al. (2005) for meat and bone meal (45%) was 2,445 Kcal/Kg, being that the value estimated by the equation 1,839.20 = 105.60 EE - 176.24 P was 2,364 Kcal/Kg. The value estimated by the equation showed a difference of only 3.32% as compared with that reported by Rostagno et al. (2005). Therefore, the equation showed to be efficient in predicting the energy value of meat and bone meal. considering that the equation AMEn = 1,889.06 + 82.41 EE - 27.87 MM that showed a determination coefficient of 0.37, and that was generated without the independent variables Ca or P, its determination coefficient was decreased, meaning a minor adjustment to estimate the AMEn of meat and bone meal. This fact can be proven by comparing the value reported by Rostagno et al. (2005) i.e., 2,445 kcal/kg with the value estimated by the equation i.e., 2,117 kcal/kg. The difference between these values was 13.41%. Despite of this considerable difference between the estimated vs. the tabulated values, the equation that showed the correlation coefficient of 0.37 can be a viable tool when considering the complexity and cost of the analyses of Ca and P.

For meat and bone meal, the equation AMEn = 1,839.20 + 105.60 EE - 176.24 P with a R² of 44%, was best suited to estimate the AMEn. The prediction equations obtained estimated values very similar to those reported by Rostagno et al. (2005), showing that these equations are a convenient tool for the determination of meat and bone meal AMEn.

Draper NR & Smith H. 1981. Applied regression analysis. 2. ed. New York: John Wiley, 709p.

Glass GV. 1976. Primary, secundary, and meta-analysis of research. Educational Researcher 6(1):3-8.

Hoffmann R & Vieira S. 1977. Análise de regressão: uma introdução à econometria. São Paulo: Hucitec.

Rostagno HS et al. 2005. Tabelas brasileiras para aves e suínos: composição de alimentos e exigências nutricionais. Viçosa, MG: UFV. 186 p.

Statistical Analisys System. 2000. SAS/STAT: user's guide, version 7.0. Cary. 325p.