Archives of Animal Nutrition Vol. 63, No. 6, December 2009, 442–454
Development of equations for predicting metabolisable energy concentrations in compound feeds for pigs Michael Bulanga* and Markus Rodehutscordb a Institut fu¨r Agrar- und Erna¨hrungswissenschaften, Martin-Luther-Universita¨t HalleWittenberg, Halle (Saale); bInstitut fu¨r Tiererna¨hrung, Universita¨t Hohenheim, Stuttgart, Germany
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
(Received 25 February 2009; accepted 20 May 2009) In 2006 in Germany new recommendations for the supply of energy and nutrients to pigs were published, including a modification of the energy evaluation on the basis of metabolisable energy for pigs (MES). It was the objective of this metaanalysis to calculate equations that can be used for predicting MES in compound feeds for pigs on the basis of their nutrient concentrations. Data from digestibility studies with a total of 290 compound feeds were provided by different research institutions. Feeds for both piglets and growing-finishing pigs were contained in the data set. The MES concentration ranged from 10.4–16.6 MJ/kg DM, with the majority of data ranging between 15 and 16 MJ/kg DM. The data were processed with a cross validation method and a multiple regression approach by using MES as the independent variable. A corrected Akaike-Information-Criterion (AIC_ cor) was used for model evaluation. Various models were developed and validated independently. Calculations were made both for piglet and growing-finishing pig feeds, separately and together, for all feeds. It was shown that the MES concentration in compound feeds can be predicted with good accuracy if the concentrations of crude protein, crude fat, crude fibre, and starch are known. Prediction equations were presented. The lack of data for highly fibrous feeds as well as the analysis of different fibre fractions was identified as a challenge for further improvement of the prediction equations. Keywords: pigs; compound feeds; metabolisable energy; Akaike-InformationCriterion; prediction; digestibility
1. Introduction The energy evaluation of feedstuffs for pigs is based on different systems in different countries and comprises specific digestible, metabolisable and net energy systems. Because the measurements of gaseous energy losses and heat expenditure are expensive and cannot be made routinely, many evaluation systems use the digestibility of nutrients to calculate the respective energy value (GfE 2005). The routine evaluation of compound feeds, however, cannot usually be made this way, because the ingredient composition of the compound feed or the digestibility, or both, are unknown. For these feeds, equations were developed that predict the energy value from the analysed chemical composition of the feed. Predicting
*Corresponding author. Email:
[email protected] ISSN 1745-039X print/ISSN 1477-2817 online Ó 2009 Taylor & Francis DOI: 10.1080/17450390903217317 http://www.informaworld.com
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
Archives of Animal Nutrition
443
equations are needed for routine feed evaluation both by the feed industry and the extension services, as well as by the official authorities that are responsible for controlling the declared energy value of a compound feed. Predicting equations are specific to the respective evaluation system that they were developed from. Modifications in the energy evaluation system therefore also require a revision of the prediction equations. In 2006 the Committee for Requirement Standards of the Society of Nutrition Physiology presented a new edition of the recommendations for energy and nutrient supply for pigs (GfE 2006; English version published in 2008). This included an updated equation for the determination of metabolisable energy for pigs (MES) from digestible nutrients. In comparison with the former equation (GfE 1987, 1996), this equation has changed the consideration of different carbohydrate fractions in particular and, therefore, the energy value of feeds. As a consequence, the equation used to predict MES in compound feeds from analysed nutrients also has to be changed. So far, only a few equations for predicting the energy value of a pig diet on the basis of analysed crude nutrients have been published – all of them with a wide span of prediction accuracy (Kirchgessner and Roth 1983; Just et al. 1984; Morgan et al. 1987; Noblet and Perez 1993) and linked to their respective reference energy system. Furthermore, these prediction equations do not differentiate between piglets and growing-finishing pigs, although differences exist in nutrient digestibility, ingredients and diet formulation as well as differences in the rate of passage, for instance, of fibrous feeds, depending on body weight (Le Goff and Noblet 2001; Le Goff et al. 2002; Noblet and van Milgen 2004). Prediction equations calculated for different categories of pigs might therefore be more precise than general equations. The objective of the present study was to develop a simple equation for the accurate prediction of MES in compound feeds for pigs from analysed crude nutrients based on the German energy evaluation system. A meta-analysis of data originating from several institutions was conducted. Part of the objective was to investigate whether separate equations for piglets and growing-finishing pigs could be derived from the dataset and whether these separate equations were more accurate than a general equation for growing-finishing pigs and piglets together. In addition, new prediction equations were to be compared with previously published equations. 2.
Material and methods
2.1. Data collection Research institutions in Germany were invited to contribute their data from digestibility trials using compound feeds for pigs. The trials had been conducted in accordance with the guidelines for pig studies on the energy value of feeds (GfE 2005), with the exception that no restriction in body weight (BW) had been implemented. Each dataset should contain information about the ingredient composition of the compound feed, its dry matter (DM) concentration and the concentrations of crude nutrients (organic matter [OM], crude protein [CP], ether extract [EE], crude fibre [CF], N-free extractives [NfE]), starch (St) and sugar (Su). Dietary concentrations of neutral detergent fibre (NDF) and acid detergent fibre (ADF) were also requested. However, the number of datasets that contained NDF and ADF was so low that these fibre fractions could not be considered in the further procedure. Other requested data were the breed, sex, and number of pigs used in the digestibility trial,
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
444
M. Bulang and M. Rodehutscord
the BW of pigs and dry matter intake (DMI), the length of adaptation and faeces collection periods, and the measured digestibilities of the nutrients. Only the mean values for digestibility determined in each trial were considered. Energy lost with urine was not the subject of this study and urine was not collected in most of the trials. A total of 290 datasets from digestibility studies with compound feeds was provided from the following institutions: the Institut fu¨r Tiererna¨hrung und Stoffwechselphysiologie in Kiel, Institut fu¨r Tiererna¨hrung in Braunschweig, Institut fu¨r Tiererna¨hrung und Futterwirtschaft in Grub, Thu¨ringer Landesanstalt fu¨r Landwirtschaft in Jena, Institut fu¨r Agrar- und Erna¨hrungswissenschaften in Halle, and Institut fu¨r Nutztierwissenschaften und Technologie in Rostock. Data from Beste (1988), Kleine Klausing (1990) and Gru¨newald (1992) were also considered. The chemical analyses of feeds and faeces were done according to the official methods VDLUFA (2007). The concentration of MES in all compound feeds was not requested but was calculated for all feeds in the same way from digestible nutrients, using Equation 3 of GfE (2006). The MES concentration was further used as the dependent variable for multiple regression analysis in calculating the prediction equations. 2.2. Data description Pigs used in the digestibility trials were mainly commercial crossbreeds of Deutsche Landrasse, Deutsches Edelschwein and Pietrain. In some trials, maternal crossbreeds of Deutsches Edelschwein, Duroc, Large White, Pietrain or Hampshire were used. The BW of pigs ranged between 8 and 117 kg, with most pigs within the range of 30– 110 kg BW. Nearly all digestibility trials had been conducted with male pigs. Seven trials were done with female pigs and in some other cases the sex was not specified. The number of pigs used in each trial ranged between 2 and 13, and in most cases it was 4 (n ¼ 142). The length of adaptation and collection periods ranged between 2 and 10 days (mean: 7 days) and 4 and 8 days (mean: 7 days), respectively. 2.3.
Diet composition and nutrient content and digestibility
A large proportion of the compound feeds consisted of wheat grain (160), barley grain (54) and a mixture of both (41) as the main ingredients (Table 1). The ‘main ingredient’ was the feedstuff with the highest inclusion level or a combination of two feedstuffs with the same relative proportion or a maximum difference of 1% between them. The composition of seven diets was unknown, because they were complete compound feeds purchased for the digestibility study. The high proportion of diets mainly based on wheat grain or wheat/barley grain was found among the feeds for both growingfinishing pigs and piglets. A total of 78% of the 92 diets for piglets were wheat grain based, 13% a mixture of wheat and barley grain and nearly 5% barley grain based. The variation in analysed composition, digestibility and MES for piglets and growing-finishing pig diets is summarised in Table 2. All criteria showed great variation. However, distribution was not equal, as indicated for some nutrients in Figure 1, and the majority of diets for piglets and growing-finishing pigs contained between 15 and 16 MJ MES per kg DM. The number of diets for piglets and growing finishing pigs in MES classes 10, 11, 12, 13, 14, 15, 16 and 17 MJ per kg DM were 0, 2, 0, 0, 0, 70, 19, 1 as well as 1, 5, 4, 3, 10, 143, 31 and 1, respectively.
445
Archives of Animal Nutrition Table 1.
Inclusion level of the major ingredient in the compound feeds (n ¼ 290). Inclusion level [%]
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
Main ingredient Faba beans Peas Barley grain Oats grain Maize grain Rapeseed cake/barley grain Rye grain Dried sugar beet pulp/wheat bran Wheat grain Wheat grain/faba bean Wheat grain/barley grain Wheat bran Unknown*
Number of diets
Mean
SD#
Minimum
Maximum
1 1 54 2 6 1 7 2 160 1 41 7 7
53 59 50 66 70 50 43 24 50 48 35 32 –
– – 14.3 – 0 – 14.3 0 11.0 – 6.4 10.8 –
– – 27.5 49 70 – 30 24 30 – 25 26 –
– – 97 82 70 – 72 24 99 – 49 56 –
Notes: #SD, Standard deviation; *No information about the ingredients was available.
2.4. Calculations and statistical methods The dataset was separated into data from piglets (530 kg BW) and from growingfinishing pigs (30 kg BW). Data within these subsets were assigned to two separate data pools for calculation and validation of the regression equations according to a cross-validation method described by Shtatland et al. (2004) and by using the ‘‘rununi’’ function of SAS. Prediction equations were calculated alternatively for the total BW range and separately for piglets and growing-finishing pigs. The comparison was made to answer the question of whether a general equation can be used to predict MES with a sufficiently high level of accuracy or whether only separate equations are adequate. All equations developed were validated separately. Potential regression variables in the models were expressed in g per kg DM. In some cases, the nutrient concentration tended to have a non-linear relationship to MES, especially starch. Therefore, as the first step, a so-called ‘‘quasi-linear’’ multiple regression approach according to the description of Rasch et al. (1996) was tested (equations labelled ‘A’ in each BW category). In this quasi-linear approach, all potential variables were squared and used in the equation development. All equations were calculated without intercept, because the variation in MES should be explained only by the independent variables and a residual error. The independent variables crude ash (CA), CP, EE, CF, St, Su and organic residue (OR), as well as their squares, were used for model building. Organic residue was calculated as the difference between OM and the sum of CP, EE, CF and St, as well as Su in some equations. Correlation (Proc Corr) and multiple regression analysis (Proc Reg), as well as comparison of means (digestibilities of EE by pigs and piglets by Student’s t-test), were conducted with the software package SAS (Version 9.1, 2002–2003, SAS Institute Inc., Cary, NC, USA). For model building, in addition to root mean square error (RMSE), the Akaike-Information-Criterion (AIC_cor) was calculated according to Hurvich and Tsai (1989). The model with the smallest AIC_cor was considered the best. In case of different sample sizes in multiple regression analysis
446
M. Bulang and M. Rodehutscord
Table 2.
Chemical composition and digestibility of the feeds. Range
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
Mean
SD#
CV* [%]
Piglets (530 kg BW, n ¼ 92) Chemical composition [g/kg DM] Crude ash 58.0 7.0 12 Organic matter 942.0 7.0 1 Crude protein 207.9 16.8 8 Ether extract 38.4 9.0 23 Crude fibre 42.0 13.8 33 N-free extractives 653.7 25.7 4 Starch 428.4 54.5 13 Sugar 47.7 19.3 41 Digestibility [%] Organic matter 88.3 3.0 3 Crude protein 85.2 3.7 4 Ether extract 79.2 9.2 12 Crude fibre 44.6 7.7 17 N-free extractives 92.5 2.2 2 MES* [MJ/kg DM] 15.2 0.7 5 Growing-finishing pigs (30 kg BW, n ¼ 198) Chemical composition [g/kg DM] Crude ash 54.5 18.7 34 Organic matter 945.5 18.7 2 Crude protein 193.6 26.0 13 Ether extract 33.9 14.0 41 Crude fibre 45.2 18.8 42 N-free extractives 673.6 42.6 6 Starch 470.5 86.2 18 Sugar 44.8 18.0 40 Digestibility [%] Organic matter 86.8 4.2 5 Crude protein 84.5 4.9 6 Ether extract 71.6 15.2 21 Crude fibre# 37.3 10.4 28 91.3 3.2 3 N-free extractives# MES [MJ/kg DM] 14.9 1.0 6
Minimum
Maximum
725%
þ25%
41.0 919.0 140.5 18.0 31.0 572.0 95.4 24.0
81.0 959.0 240.8 72.0 128.0 735.5 514.6 162.0
44.0 935.9 164.4 22.7 34.1 627.7 382.6 34.2
63.9 956.1 221.9 49.1 65.0 708.5 502.9 70.4
70.3 62.7 31.9 29.3 79.2 11.0
92.1 89.8 93.0 62.6 96.0 16.5
80.8 79.7 55.7 31.1 87.0 14.3
90.1 88.6 85.7 52.2 93.7 15.7
36.0 819.0 119.5 9.0 19.0 523.0 109.0 22.0
181.0 964.0 245.0 107.0 131.0 827.0 668.0 147.8
44.4 925.3 163.0 21.6 33.2 617.3 351.8 32.2
74.7 955.6 225.3 53.2 68.9 715.6 544.4 65.6
66.1 62.0 75.7 8.2 74.5 10.4
94.0 92.3 92.8 74.6 95.0 16.6
81.8 78.6 54.8 26.3 88.1 13.9
90.1 88.3 85.7 51.7 93.5 15.6
Notes: #SD, Standard deviation; *CV, coefficient of variation.
tests, this criterion showed stable and small prediction errors compared to other criteria for model building (Spilke and Mielenz 2006). The content of ME corrected for bacterially fermentable substances (MEBFS) was calculated according to previous recommendations (GfE 1987, 1996). 3.
Results and discussion
3.1. Equation development As the initial step, the correlation between individual nutrients and MES was calculated. For the piglet feeds, the highest significant correlation to MES (p 0.05) was found for CF, followed by St, Su, EE and CP, with correlation coefficients (rpearson) of 70.90, 0.73, 70.64, 0.57 and 0.26, respectively (Table 3). The ranking of correlation coefficients was similar for the growing-finishing pig feeds: CF (70.89),
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
Archives of Animal Nutrition
Figure 1.
447
Distribution of nutrient concentrations in the diets of the main dataset (n ¼ 290).
Table 3. Correlation coefficients (rpearson) and respective p-values between crude nutrients and calculated MES for piglets and growing-finishing pigs (n ¼ 290). OM
CP
EE
Piglets (530 kg BW, n ¼ 92) 0.297 0.259 0.572 MES 0.004 0.013 50.001 Growing-finishing pigs (30 kg BW, n ¼ 198) MES 0.546 0.315 0.157 50.001 50.001 0.027
CF
NfE
Starch
Sugar
70.898 50.001
0.193 0.066
0.734 50.001
70.643 50.001
70.889 50.001
0.387 50.001
0.673 50.001
70.584 50.001
St (0.67), Su (70.58), CP (0.31) and EE (0.16). Thus, in both groups of feeds, the CF fraction negatively affected the MES concentration. The effect of the EE fraction was much greater in the piglet feeds than in the growing-finishing pig feeds. Equations were calculated both on the basis of all feeds (labelled as 1), and separately for the piglet feeds (labelled as 2) and growing-finishing pig feeds (labelled as 3) (Table 4). Letters A to D label the different levels of information that were considered. Within each category of feeds, the A equations showed the lowest RMSE. The lowest RMSE was reached with equation A2 (* 0.18 MJ MES/ kg DM) for piglet feeds, followed by equation A3 for growing-finishing pig feeds
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
448 Estimated regression equations to predict MES (MJ/kg DM) using different combinations of nutrients as independent variables (g/kg DM). All (8–117 kg BW) Equation
A1
#
B1
C1
Piglets (530 kg BW) x
D1
$
A2
B2
C2
Growing-finishing pigs (30 kg BW) D2
A3
B3
C3
D3
CP 0.022860 0.020930 0.021503 0.021041 0.013220 0.019395 0.019329 0.019586 0.021095 0.020984 0.021984 0.021197 EE 0.033930 0.032846 0.032497 0.033050 ne 0.043018 0.043655 0.042760 0.034562 0.031895 0.030506 0.031849 CF ne* 70.023898 70.021071 70.025123 ne 70.014847 70.014559 70.019627 ne 70.023710 70.020134 70.026291 St 0.016554 0.016458 0.016309 0.030816 0.016182 0.016131 0.019988 0.016556 0.016413 Su ne 0.021068 0.014045 0.015015 ne 0.021359 OR{ 0.014893 0.013962 0.014701 0.016164 0.010868 0.014131 0.014313 0.015791 0.014003 0.013495 0.013961 0.016224 CA 70.018451 ne ne ne 70.028456 ne CP6CP ne ne ne EE6EE ne 0.000458 ne CF6CF 70.000210 70.000071 70.000238 St6St ne 70.000021 70.000004 Su6Su 0.000194 ne 0.000293 OR6OR ne ne ne CA6CA 0.000094 ne 0.000140 RMSE 0.2243 0.2426 0.2535 0.2613 0.1763 0.2250 0.2222 0.2233 0.1995 0.2394 0.2464 0.2687 RMSE [%] 1.5 1.6 1.7 1.7 1.2 1.5 1.5 1.5 1.3 1.6 1.7 1.8 { AIC_cor 7254 7235 7225 7218 794 775 777 779 7187 7158 7154 7140 r-square 0.9997 0.9997 0.9997 0.9997 0.9999 0.9998 0.9998 0.9998 0.9998 0.9997 0.9997 0.9997 n 133 133 133 133 43 43 43 43 90 90 90 90
Notes: #Quadratic variables were not considered in the B equations; xQuadratic variables, Su and CA were not considered in the C equations; $Estimation only based on CP, EE, CF and OR (D equations); *ne, the variable was offered but a parameter was not estimated; {OR ¼ OM 7 CP 7 EE 7 CF 7 St 7 Su; if Su or St or both were not considered in the model as an independent variable then they were included in OR; {Akaike-Information-Criterion (AIC_cor) was calculated according to Hurvich and Tsai (1989).
M. Bulang and M. Rodehutscord
Table 4.
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
Archives of Animal Nutrition
449
(* 0.20 MJ MES/kg DM). This indicates that the accuracy of MES prediction is slightly higher when done with separate equations instead of using one general equation. The interpretation of the estimated coefficients with regard to the theoretical contribution to the energy supply of single nutrients is difficult, because the variation in MES is allocated to nutrients and their squares. Equation A1 contained known dimensions of coefficients for CP, EE, St and OR (similar to published data of Kirchgessner and Roth 1981), but curvilinear scales of CF, Su and CA were also estimated. Noblet and van Milgen (2004) presented similar values for estimated coefficients, except for the organic residue. They calculated contributions to ME from CP, EE, St, Su and OR of 19.7, 32.2, 18.2, 15.9 and 0.5 kJ per g DM, respectively. The consideration of quadratic terms has the disadvantage that the estimated parameters cannot be physiologically interpreted. Also, the equations become complex, which may make their application in practice difficult. In the next step, therefore, equations without any quadratic term were calculated (B equations in Table 4). In comparison with A equations, the RMSE was higher by between 0.02 and 0.05 MJ/kg DM. When B equations are compared amongst each other, the prediction of one common equation (B1) resulted in an only slightly higher error of prediction than the separate prediction for piglet and growing-finishing pig feeds. Routine analyses of feeds do usually not comprise all the nutrients that were considered in this evaluation. Therefore, the next step tested how the accuracy of prediction is affected when the number of considered nutrients is restricted. C equations did not consider Su and CA, and D equations neither considered Su and CA nor St. This reduction in the nutrients considered did not negatively affect the accuracy of prediction of the piglet feeds, but increased the RMSE by up to 0.03 MJ/kg DM in comparison to B equations for the growing-finishing pig feeds (D3) and by 0.02 MJ/kg DM for all feeds (D1). However, all RMSE values found in this study can be regarded as low (Noblet and Perez 1993), and the equations thus sufficiently accurate for predicting MES. For piglet feeds, the estimated coefficients for the EE fraction were around 0.04 and higher than those estimated for the growing-finishing pig feeds (around 0.03). This difference may have been caused by the higher EE level and the higher EE digestibility in the piglet feeds. The difference in mean digestibility of EE between piglets and growing-finishing pigs was 7.6% (p 5 0.0001, Student’s t-test). 3.2.
Validation step
The results of the validation of the prediction equations is exemplarily shown for the equations calculated using all data in Figure 2. With equations A1 and B1, a higher deviation from the line of equality for low energy feeds was found than with equations C1 and D1. This higher deviation for low energy feeds was also found for nearly all other equations. The outcome of the validation is summarised for all predicted equations in Table 5. Results of the validation showed that amongst the equations calculated, for all feeds equation C1 had the lowest mean difference between calculated and predicted MES (0.002 MJ/kg DM) (Table 5). Other parameters for the goodness of validation
450
M. Bulang and M. Rodehutscord
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
Figure 2. Comparison of predicted and calculated MES by the four equations calculated with data from all feeds. The dotted line is the line of equality.
Table 5.
Differences between the calculated and estimated MES in the validation step.
All (n ¼ 158)
Piglets# (n ¼ 49)
Growing-finishing pigs# (n ¼ 109)
Piglets
Growing-finishing pigs
Calculated MES minus estimated MES [MJ]
95% Confidence limit
Equation
Mean
SD
Minimum
Maximum
Lower
Upper
A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3 A1 B1 C1 D1 A1 B1 C1 D1
70.013 70.020 0.002 0.021 0.009 70.024 70.024 70.012 70.083 70.030 0.001 0.028 0.065 0.056 0.069 0.054 70.048 70.054 70.028 0.006
0.325 0.274 0.260 0.271 0.247 0.219 0.219 0.227 0.373 0.289 0.275 0.290 0.354 0.237 0.229 0.239 0.306 0.283 0.269 0.284
71.759 70.997 70.879 70.826 70.542 70.664 70.656 70.595 72.226 71.023 70.904 70.847 71.759 70.628 70.611 70.572 71.117 70.997 70.879 70.826
0.965 0.731 0.681 0.811 0.708 0.447 0.461 0.566 0.761 0.779 0.821 0.837 0.695 0.626 0.652 0.654 0.965 0.731 0.681 0.811
70.655 70.560 70.512 70.514 70.487 70.464 70.464 70.468 70.822 70.603 70.544 70.547 70.647 70.421 70.390 70.427 70.656 70.614 70.561 70.556
0.629 0.521 0.516 0.556 0.505 0.416 0.416 0.444 0.656 0.544 0.547 0.604 0.777 0.533 0.529 0.536 0.559 0.507 0.505 0.568
Note: #Validation based on the respective data subset.
also were the best for C1. From the differences between equations C1 and D1 it can be concluded that the starch concentration in the feed is a necessary co-predictor of MES when no differentiation between feeds for piglets and growing-finishing pigs is made. If the concentration of starch is not available, the prediction of MES is possible nevertheless by using equation D1, but a lower accuracy of prediction needs to be taken into account. Smaller confidence limits also indicate an increasing
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
Archives of Animal Nutrition
451
accuracy of prediction from equation A1 to C1. Based on these results, it is suggested that equation C1 should be used for prediction of MES in compound feeds if no differentiation is made between feeds for piglets and growing-finishing pigs. When equation 2 (piglet feeds) was separately validated the mean deviation between calculated and predicted values ranged from 70.024 to 0.009 MJ/kg DM across equations A2 to D2. The most precise equations are B2 and C2, suggesting that the consideration of sugar and crude ash is not necessary for predicting MES in these feeds. Consequently, equation C2 is suggested for use if separate predictions for piglet and growing-finishing pig feeds are intended. Validation of equation 3 (growing-finishing pig feeds) also showed that equation C3 was the most precise equation. The standard deviation, minimum and maximum, as well as the confidence limits for the deviation between calculated and predicted MES were the smallest for equation C3. Thus, equation C3 is suggested if separate predictions for piglet and growing-finishing pig feeds are intended. A final validation step tested how the predicted data deviate from the calculated data when the general prediction equations (equation 1) are applied to the validation subsets for piglet and growing-finishing pig feeds. In comparison with the separate prediction equations, the general equations had a higher mean difference and standard deviation in the piglet diets subset, but comparable or somewhat better values were found for the growing-finishing pig feeds subset. These differences might have been caused by the unequal distribution of piglet and pig feeds in the dataset. 3.3.
Comparison with other prediction equations
Only a few equations for ME prediction for pig feeds have been published. They were applied to the present validation dataset (all data). The different equations underpredicted or overpredicted MES from the present dataset (Table 6). This may be explained by the fact that their reference ME values were not identical with the MES defined and calculated by GfE (2006). In the evaluations made by Kirchgessner and Roth (1981, 1983) the range of ME was comparable to the present dataset, but the mean content was 13.9 MJ MEBFS per kg DM compared to approximately 15 MJ MES per kg DM in the present dataset. Differences also existed in the crude nutrient contents of the feeds. Compared to the dataset that was used to predict MEBFS by Kirchgessner and Roth (1981, 1983), the range of OM and EE content in the present dataset was higher, the range of CP and CF content smaller, and that of St was comparable. Equations J and K from Kirchgessner and Roth (1983) showed small under predictions, equation J (70.138 MJ ME per kg DM) more than K (70.016 MJ per kg DM). Equation K showed similar results as equation L from Just et al. (1984). Equation L showed the best validation parameters among the equations considered, except the mean deviation, which was slightly higher (þ0.056 MJ) compared to equation K (70.016 MJ). In the dataset of Just et al. (1984), a higher number of diets (n ¼ 321) with a wider range of crude nutrients compared to the present dataset was used. For CP, EE, CF and NfE, the means and standard deviations (in parentheses) were 206 (48), 49 (32), 65 (33), and 615 g (75 g) per kg DM, respectively. Further external equations with intercept were validated and a tendency for under prediction of MES was found.
– – – – 17.4 6.9 5.4
0.0226 0.0223 0.0203 0.0197 0.00586 16.4 0.0143
0.0319 0.0341 0.0252 0.0194 0.0172 160 0.0194
Ether extract 70.0129 70.0109 70.0178 70.0125 0.0255 – 70.0245
Crude fibre Sugar
OR#
0.0166 0.0184 0.0097 0.017 0.0168 0.0074 – – – – – – – – – 11.7 8.2 – – – –
Starch
Crude ash
– – – – 0.0162 – 0.0158 – – 70.0515 – 717.0 0.0106 –
NfE 0.37 0.28 –{ –{ 0.34 0.41 –{
RMSE 2.6 2.1 –{ –{ 2.4 –{ –{
RMSE [%]
70.138 70.016 0.041 0.386 70.342 70.354 0.071
Mean
0.332 0.437 0.301 0.362 1.324 0.523 0.297
SD
70.795 70.881 70.555 70.332 72.965 71.390 70.517
0.519 0.850 0.638 1.104 2.282 0.681 0.660
Confidence limits (p ¼ 0.975) low to up
Validation (n ¼ 109)
Notes: *Equations J and K: Kirchgessner and Roth (1983); equations L, M and P: Just et al. (1984), table XV; equation N: Noblet and Perez (1993); equation O: Morgan et al. (1987); #OR, organic residue; OR ¼ organic matter 7 crude protein 7 ether extract 7 starch 7 sugar 7 crude fibre; {not given; {ME was calculated according Hoffmann and Schiemann (1980) as follows: ME [MJ/kg DM] ¼ 0.0210 dCP þ 0.0374 dEE þ 0.0144 dCF þ 0.0171 dNfE (where d, digestible amount in g per kg DM; CP, crude protein, EE, ether extract; CF, crude fibre; NfE, nitrogen free extractives); kMEBFS was calculated according to Hoffmann and Schiemann (1980) and was corrected by bacterial fermentable substance (BFS) with an additional term in the equation from the above footnote: 7 0.0068 (BFS 7 100) if the content of BFS is 4100 g per kg DM, and MEBFS was also corrected by sugar if the content of sugar was higher than 80 g/kg DM; xME was calculated by digestible energy (DE) minus the energy in urine (Just 1982; Morgan et al. 1987); aME was calculated as described by Noblet et al. (1985): ME ¼ energy of food intake minus the sum of energy of faeces, urine and methane; $ coefficients were expressed in kg/kg DM.
J Kk Lx Mx Na Ox$ Px
{
Crude protein
Regression coefficients of the variables
Application of published equations for ME prediction to the present validation subset of data.
Equation* Intercept
Table 6.
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
452 M. Bulang and M. Rodehutscord
Archives of Animal Nutrition
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
4.
453
Conclusions
We conclude that it is possible to predict MES as defined by GfE (2006) of compound feeds for pigs on the basis of nutrient concentrations. Different options exist for this prediction. If it is intended to use only one equation for all compound feeds, equation C1 should be chosen. Based on the inclusion of both starch and crude fibre, it showed a good accuracy of prediction for the entire range of feeds. However, the application of the equation should be restricted to feeds with nutrient concentrations falling within the range covered by this study. If separate predictions for piglet and growing-finishing pig feeds are intended, equation C2 is recommended for piglet feeds and C3 for growing-finishing pig feeds. Among the previously published equations, none yielded a better prediction than the new equations. Data in the present evaluation were not equally distributed within the range of MES concentration. For the low energy feeds in particular, too few digestibility data were available. It is suggested to improve the data basis especially for highly fibrous compound feeds. It is also suggested that future digestibility studies are extended to NDF and ADF in order to investigate whether an improvement in the accuracy of MES prediction can be achieved by using these fibre fractions. Acknowledgements The authors are grateful for the generous data contributions made by the institutions mentioned above. Also, the authors gratefully acknowledge the impact given by the members of the Committee for Requirement Standards of the Society of Nutrition Physiology.
References Beste R. 1988. Untersuchungen zur Bewertung des Einsatzes von Sojaextraktionsschrot, Ackerbohnen, Weizen und Roggen sowie von synthetischen Aminosa¨uren in der Schweinemast [Dissertation agrar]. Universita¨t Bonn. GfE. 1987. Energie- und Na¨hrstoffbedarf landwirtschaftlicher Nutztiere. Nr. 4. Schweine. Frankfurt am Main: DLG-Verlag. GfE. 1996. Anwendung der Scha¨tzformel fu¨r die Kontrolle des Energiegehaltes von Mischfutter fu¨r Schweine. Proc Soc Nutr Physiol. 5:157–158. GfE. 2005. Determination of digestibility as the basis for energy evaluation of feedstuffs for pigs. Proc Soc Nutr Physiol. 14:207–213. GfE. 2006. Empfehlungen zur Energie- und Na¨hrstoffversorgung von Schweinen. Frankfurt am Main: DLG-Verlag. GfE. 2008. Recommendations for the supply of energy and nutrients to pigs. Frankfurt am Main: DLG-Verlag. Gru¨newald KH. 1992. Untersuchungen an Mastschweinen zur Verminderung der NAusscheidungen durch Einsatz freier Aminosa¨uren [Dissertation agrar]. Universita¨t Bonn. Hoffmann L, Schiemann R. 1980. Von der Kalorie zum Joule: Neue Gro¨ßenbeziehungen bei Messungen des Energieumsatzes und bei der Berechnung von Kennzahlen der energetischen Futterbewertung. Arch Anim Nutr. 30:733–742. Hurvich CM, Tsai CL. 1989. Regression and time series model selection in small samples. Biometr. 76:297–307. Just A. 1982. The net energy value of balanced diets for growing pigs. Livest Prod Sci. 8:541–555. Just A, Jorgensen H, Fernandez JA. 1984. Prediction of metabolizable energy for pigs on the basis of crude nutrients in the feeds. Livest Prod Sci. 11:105–128. Kirchgessner M, Roth FX. 1981. Zur Scha¨tzung energetischer Futterwerte von Mischfuttermitteln fu¨r Schweine. Zeitschr Tierphysiol Tiererna¨hr Futtermittelk. 45:100–108. Kirchgessner M, Roth FX. 1983. Scha¨tzgleichungen zur Ermittlung des energetischen Futterwertes von Mischfuttermitteln fu¨r Schweine. Zeitschr Tierphysiol Tiererna¨hr Futtermittelk. 50:270–275.
Downloaded By: [Universitaetsbibliothek] At: 20:59 28 October 2009
454
M. Bulang and M. Rodehutscord
Kleine Klausing H. 1990. Untersuchungen zur Bewertung des Einsatzes von Hafer, Ackerbohnen und Erbsen in der Schweinemast [Dissertation agrar]. Universita¨t Bonn. Le Goff G, Noblet J. 2001. Comparative total tract digestibility of dietary energy and nutrients in growing pigs and adult sows. J Anim Sci. 79:2418–2427. Le Goff G, Van Milgen J, Noblet J. 2002. Influence of dietary fibre on digestive utilization and rate of passage in growing pigs, finishing pigs and adult sows. Anim Sci. 74:503–515. Morgan CA, Whittemore CT, Phillips P, Crooks P. 1987. The prediction of the energy value of compounded pig foods from chemical analysis. Anim Feed Sci Technol. 17:81–107. Noblet J, Le Dividich J, Bikawa T. 1985. Interaction between energy level in the diet and environmental temperature on the utilization of energy in growing pigs. J Anim Sci. 61:452–459. Noblet J, Perez JM. 1993. Prediction of digestibility of nutrients and energy values of pig diets from chemical analysis. J Anim Sci. 71:3389–3398. Noblet J, van Milgen J. 2004. Energy value of pig feeds: Effect of pig body weight and energy evaluation system. J Anim Sci. 82(E):229–238. Rasch D, Herrendo¨rfer G, Bock J, Victor N, Guiard V. 1996. Verfahrensbibliothek Versuchsplanung und Auswertung. Mu¨nchen-Wien: R. Oldenbourg Verlag. Shtatland ES, Kleinman K, Cain EM. 2004. A new strategy of model building in proc logistic with automatic variable selection, validation, shrinkage and model averaging. SUGI ’29 Proceedings, Paper 191–29, Cary, NC: SAS Institute, Inc. Spilke J, Mielenz N. 2006. Vergleich von Kriterien und Verfahren zur Modellwahl bei der multiplen linearen Regression. Paper presented at: 10. Konferenz der SAS-Anwender in Forschung und Entwicklung; 2006 February 23–24; Hamburg. VDLUFA [Verband Deutscher Landwirtschaftlicher Untersuchungs- und Forschungsanstalten] (editor) 2007. Handbuch der landwirtschaftlichen Versuchs- und Untersuchungsmethodik, Band 3, Die chemische Untersuchung von Futtermitteln. 3rd ed. 1976 with supplements 1–7 (1983, 1988, 1993, 1997, 2004, 2006, 2007), VDLUFA-Verlag, Speyer.
