The derivation of chemical prediction equations for monitoring energy declarations.

Abstract

294 C. FISHER* The use of direct bioassays or of in vitro simulation of digestion for feeds have raised questions about mnitoring and verifying the values. Propsah for the introduction of energy declarations on mixed this parpose is briefly discussed, but chemical prediction equations are most likely to be used in Europe. Prediction equations are derived fran experimzntal data on the masured ME values of feeds of varying chemical camposition. The selection of interpretative mode.& is briefly discussed, the min distinction being between pely empirical statistical descriptions of data and mdels which incorporate known biological concepts. EQuations tiich are determined eqirically but are subsequently found to agree with theoretical expectations find widespread suppoti. The derivation of equations in the U.K. and Europe is briefly The resulting range of alternative equations offers a described, choice 'between accuracy of prediction and canplexity, and cost, of the chemical analyses. Selection of an quation should take account both of its predictive properties and the reproducibility of the analyses Several equations are available with residual standard involved. deviations, incorporating both these sources of variation, b&men 0.30 and 0.40 MJ/kg. By canparison the standard error of a bioassay result is 0.15 to 0.20 MJ/kg. A ring-test of one equation is described. Within+boratory repeatability ms gocx3 (s.d. = 0.40% of man) but b&men-laboratory reproducibility was mch poorer (s.d. = 4.48% of meanL Better standardisation of analytical mthc& might improve the latter value. The corrdation between observed and predicted ME values obtained with this quation was high (r = 0.98) but there was a large bias which variedsignificantlybetv mck~elsandyoungchicks. l In mst countries of the world trade in canpomd animal feeds is governed by legal regulations. In particular these define the information on canposition and nutritional value which mst be provided by the seller and also mzans by tiich the values given can be monitored and verified. The regulations differ in detail and the iqact they have on the pattern of trade will vary according to the organisation of the industry e.g. the importance of integration, awqerative trading etc. , However there are also a lot of elemants which arecmmntomst countries, In .the U.K. the nrain points of the regulations are a) tha* ingredients should be f8wholesm suitable for their purpose and free fran associated hazards~~~ b) declarations of oil, crude protein, fibre, ash levels; total vitamin A, E and D contents plus indications of *Agricultural Reseamh Council, P~I&Y Research Centre, Rcslin, MidlothianEi259pS,Scotland. 295 storage life: copper (if >50 mg/kg), magnesium (>O.S%), rrplybdma and selenium (if added) f urea, biuret, urea phosphate and IBDU (expressed as protein equivalent), uric acid (expressed as protein equivalent if colourants, preservatives and greater than 1%); antioxidants, These regulations have evolved over time, the recent medicammts. developments being an interpretation for use in the U.K. of various directives fran the European Econtic Camunity. It is of course an eventual. purpose of the EEX: #at uniform regulations should govern trade throughout the camunity. Although this information is obviously useful it does not define the most inportant nutritional factors which determine the econanic value of a feed - energy and amino acid levels, Furthenmrethereisa view held by sane farmers that the declarations do not advise tha sufficiently about the use of unusual ingredients, especially industrial by-products. It is not the plrposse of this paper to reflect in any way the political arguments in the U.K. about declarations nor to represent the views of the wders or their custtirs. Suffice it to say that a strong appeal to goverxmznt fran the Farmers Unions to legislate for open declarations (i.e. a listing of each formula) has led, not to agreement on this point, but to an undertaking that the amount of nutritionally useful information should be increased. In the first place energy declarations will be introduced and this runs in parallel with a similar decision in the m. The introduction of energy declarations is not Uncontrov~si~ and since no legislation has been announced I till outline briefly how the topic has developed fron a technical point of view. 1 should also stress that the views pressed here are personal ones. Once agreement was reached to introduce energy declarations then important questions zose as to how the values were to be defined and how they could be mnitored and verified. Most of the discussion was about chemical prediction quations since this is the mzthal of control most likely to be used. Alternatives such as rapid bioassays and in vitro digestion methods do exist but at this stage it is clear that legislation in ntrope will be based on equations which relate the AME value of a feed (corrected to zero N-retention, Mn) to readily defineable and rrreasurable chemical canponents, Existing eqmtions (Table 1) seemd insufficient for the purpose of verification for two *main reasons. Firstly they did not consider a very wide range of chemical variables and therefore the potential pay-off between the canplaity of equations and their accuracy could not be fully explored. Secondly, they did not take account of recent develomts in ME systems, in particular the introduction of WE. The Poultry Research Centre were therefore asked to undertake new experiments and the results have been published (Fisher, 1982a). At about the samz time technical discussions were taking place in EWope to establish a basis for legislation in the EzM3. In this wider forum there was mturally a range of opinions and, to try and reconcile these, experimental data fran four (more recently five) laboratories were analysed. The 'best' quation fran this canbined analysis was similar to that suggested by H&W. et al. (1977) and it has nw been adopted as the basis of an m direct= The results of a ring-test to establish its reproducibility and m evaluation of its predictive properties have been published (Fisher, 1983). At present these issues are being debated by governumt and by the interested parties. Theeventualoutca~is notcertainbutthe 296 equation finally adopted within the EM3 will probably be used throughout Europe and sag alternative equations suggested in the work of Fisher (1982a1, although mre efficient and cheaper are likely to be overlooked. There are still scmz technical matters to be resolved. PREDICTIONE!QUA!TIONSFoRMEl!N3OLIsABLE ENERGY The calculation of energy values fran the chemically defined constituents of a feed is well established. It is nearly 100 years since Atwater defined his 'factors' stating that protein, fat and carbohydrate, when digested and absorb&,' yield 4, 9 and 4 kcal./g 'available* energy. Fran this starting point a variety of chemical prediction equations for poultry have been proposed (Table 1) and widely us&i in practical feed formulation. It is interesting that m=>st of these equations, although derived independantly and in different ways, are extremly similar when recalculated on a canparable basis. The coefficients for fats, proteins and carbohydrates are also realistic if it is assumzd that, when digested, these nutrients yield 38.5, 18.5 and '17.2 W/g respectively (H&-tel et &. 1977). 297 The derivation of prediction qmtions has been reviewed elsewhere (Fisher 1982b) but tw points should be stressed. Having assembled data on the ME values of a range of feeds of varying -sition they cm be interpreted in different ways and judicious selection of an interpretative mdel can overcane sag of the inherent limitations of An equation may be judged to be more the experimntal approach. 'robust' for practical use if, in addition to being an effective mpirical descriptor of the data frcxn which it was derived, it is also consistent with the external evidence and expectations about the underlying relationships. Thus, for example, m can consider whether equations should contain constant terms, whether they should contain negative predictors and whether the coefficients agree with theoretical energy values. -* Models which are a sumnation of the energy yielding canponents of a feed are attractive. If they include all such caqqnents a constant term should not be required, and if each canponent can reasonably be represented by a single coefficient for all feeds, then other dietary characteristics and interactions should not be required, Such arguments led Hkrtel (1979) to propose an equation with fat, protein, starch azd sugar as energy~yielding variables and with no constant. Obviously this argurtmt has limitations. If the energy value of a digested nutrient is constant, then the use of single coefficients to describe crude nutrients is quivalent to assuming constantdigestibility for all feeds, This is clearly untrue and a factor such as fibre level my feature in a prediction equation as an empirical index of digestibility rather than an energfsource per se. The second general issue is the implication for prediction equations of Sibbald's ideas about true and apparent ME. Sibbald (1976) and elsewhere has argued that the constant excretion of endogenous energy fran birds in ME experiments leads to the observed AME being reduced as food intake declines. The result is that AME values may be systematically underestimated in same feeds and if these - effects are correlated with any chemical variable e.g. high fibre levels, then this will lead to a spurious relationship between energy values'\ and chemical canposition and the true relationship will be concealed, This problem can be oversaw in several ways e.g. by controlling intake (Fisher 1982a; HSrtel et al. 1977) but it should not be ignored. These have been reported by Fisher (1982a). Wnty eight feeds made fran practical feed ingredients and varying in fat (20-160 g/kg), crude protein (120-250 g/kg) and calculated AME (9-15 M/g) were used. Each was tested both as a Cal and pelleted but as extrusion had little effect the dat.a were dined to give 56 estimates of ME for feeds of known canposition. Each feed was analysed for a range of chem&cqJ variables in at least three laboratories to provide an estimate'& reproducibility of the chemical analyses. Metabolisable energy determinations were made with adult cockerels using-, a mdification of Sibbald's (1976) ?ME assay. Six replicate birds were given 30 g of each test feed by intubation after a 40 h starvation period. EXcreta were collected for 48 h. EMogenous energy losses were determinedin birds treated similarly but given 30 g 299 glucose Over a series of experimznts endagenous losses of energy, balance, wzre found to be very constant and a corrected to zero single value 32,s M/48 h UAS used in all calculations, Starting with the observed AMEn values for birds receiving 30 g food, values corresponding with an intake of 80 g were calculated via TME as follows and Fisher, 1982). l N(McNab being derived frm a number of published and unpublished experimmts. The rationale of these procedures was to use the rapid and accurate Sibbald mthod and to avoid bias due to variations in food intake, with the benefit of hindsight and of recent developznts in our technique we would now use a higher intake, 50 or 60 g, with a consequent reduction in inherent errors. The data fran this experiment consisted of 56 sets of AMEn(80) values and corresponding results for 14 ,analytical variables; the proxirrate wnehts, neutral and acid detergent fibre, acid detergent lignin, Christian lignin, starch (by enzymic hydrolysis and polari~~try), sugar, fatty acid ratios, gross eneqy and a masure of in vitro 'digestibility. These were analysed by conventional regression mthds which produce an estimate of residual standard deviation (s) by tiich equations can be assessed. However in a related study on ruminant feeds (Wainman et al. 1981) it was pointed out that the reproducibility of the chgnicalxyses used should also be considered when assessing each equation. Thus of two equations with the sam s value, the one with mre reproducible analyses will be preferred for practical use. The cost of the analyses will also be important but this has not been formally incorporated into the assessmant of quations. ways 0 Therefore the 'accuracyt of quations can be looked at in three 1. by s, the conventional residual standard deviation. This masures how well an equation described the observed variation in AME. 2. by s', a standard deviation which includes both the unexplained variation and the analytical variability. The derivation of this will be found in Wainman et al. (1981) or Fisher (1982a). 3. by SW, a standard deviation which includes only the analytical . variability. The selection of an equation should be based on s, and, in particular, on s', Once select&, reproducibility is a function only of s'., and this forms the basis, for example, of tolerance limits. Several thousand prediction equations tre computed during this wxk but only the six shown in Table 2 will be discussed. Equation 6 is the best descriptor of these data that was found; best in the sense that it had the lowest s value and all of the regression coefficients were individually significant t-2). It explained 98.5% of the observed variation in the observed AME values and has an s value of 300 Oe24 MT//kg/ increasing to 0.33 MJ/kg when the variability of the analytical x&hods is considered WL This is a fairly cmplex and The calculated energy value for fat (33.6 N/g), costly equation. starch (17.2 kJ/g) and protein (15.3 kJ/g) are realistic but there is a highly significant negative effect of NDF which is assuI113d to be acting as an index of digestibility. The inclusion of this negative term for a 'fibre' fraction necessitates the positive and significant constant term. The unsaturated to saturated fatty acid ratio has a small, kxt statistically significant, effect on goodness of fit. Bquation ? is the sam as w. 6 except that the expensive fatty acid analysis is anitted. This has orily a minor effect and gives a slight improvemnt in the s' value. Bquation 8 is the ~~JIIF~ as 7 except that CE' is substituted for NDF. A general finding in this work was that NDF t)ckls a mre effective predictor of ME than CF. Equation 9 contains the four major energy sourcest fat, protein, starch and sugar as predictors. The constant term was not significant in this equation confirming the absence of other energy sources. This very straightforward rmdel was proposed by H8rtel (1979) and is also used in the EEC-quation discussed below. It is a very effective predictor of ME values but is neither the best nor the cheapest. -Cons 10 and 11 are both based on the proximate canponents and could therefore be implemnted in the UK without additional analytical Egu, 10 canbines the canponents in a conventional way and costs accounts, in these data as in those of H%tel et aI.. (1977) for about 95% of the total variation. In this case the sic term for fat is The rather complicated significant but the effect of CP is not. re-arrangement of the proximate caqonents in equ. 11 stems fran the hypothesis that the effect of fibre is to reduce the energy value of the other canponents of the frr3. This is supported by the significant interactions between fibre and fat, protein and NFE tit the equation is empirical because there is still a highly significant positive coefficient for fibre and a large negative intercept. However this equation has a smaller s value than equation 9 which requires starch and sugar analyses although it falls down on the theoretical variance of the andytic mthods because of the interaction terms. The advantage of E93F wer CE' could 'also be shwn in this type of quation. l % . Three main conclusions were drawn fran these aperimznts. Firstly that, within the range of ?mmal' feed ingredients us& chemical prediction quations could effectively predict the ME values of canpound poultry feeds. The residual standard deviation of the %est' equation was 0.24 MT/kg whilst a man determined value basal on six replicates had a standard error of 0.15 Mf/kg. Prediction was therefore nearly as Qood as direct masuremznt. The final conclusion was that selection of an equation for practical use tJould have to reflect thebalance of accuraw-andmstandthat such abalanceshould take account of analytical variability. Initially five sets of results totalling 177 observations were analysed. However one of these, the only one based on young birds, showed much higher variability than the remaining four and was therefore anitted, The selected equation was based on 141 obsenmtions 301 on adult birds and frcm the following sources: 1) Statens Husdyrbrugsfors#g, Copenhagen. Dr LE. Petersen (29 feeds, cockerels) 2) Institute for Poultry Research, Be&berg=, The Netherlands. Dr C.A. Kan (18 feeds, cockerels) 3) H&e1 et al. (19771, University*of Hohenheim (39 feeds, hens) 4) Fisher (1982a), discussed above. [Note; whilst this paper is in preparation further data have becane available fram am, Nmzilly, France, Dr B. Leclercq and inclusion of these in the final. analysis will probably lead to slightly different equations frm those shown here]. The test feeds used in the different laboratories varied widely, both in canposition and in the range of variables covered. The analytical mthods also differed slightly but, of necessity, this had to be ignored, Analytical results were available for FAT, CP, CF, ASH, SIC and SJG with derived values for NFE and what H&e1 et al. (1977) call residual NFE (RNF = WE - SIC - SJGL Prelimi~analyses revealed little evidence of significaxxt differences in regression slopes between laboratories and therefore 'parallel-line' regression mdels were fitted. There we significant differences in intercept values for different laboratories which ware &in&I into a single average figure. In this dined analysis the %est' equation contained, like H&&l% (1979) equation and equation 9 (Table 21, FAT, CP, STC and SIG as predictors. The intercept values ranged frost -0.22 to +0.72 W/kg, the average 0.077 MJ/kg being cmbined into the coeff icier&s to yield the equation. It is this equation that has been adopted provisionally within the . EEL Unlike the PRC data these pooled results yielded no equations which are superior either as predictors or on the basis of cost. Data mre not available for EilDF tit there was no benefit in adding CE' to the equation above. A canbined starch and sugar figure was about as effective as the separate analyses which might reduce analytical costs but no canbinations of the proximate caqonents were found which had any prmise. It is not clear Qhy these results differed in this respect fran those found at PE3C. RING+STEV?WA!I'IONOFEEEQUA!I!ION ' As already pointed once equation has its naccuracym or reproducibility is a reflection only of the analytical. mehods gnployed. To determine this for the proposed EEC equation four feeds were made up and circulated to 21 laboratories throughout Europe for analysis for FAT, CP, SIC and SWG. AME values were also . determined on the same feeds using both tube-fed%ckerels and young chicks. The results of this exercise shuwed that repeatability of the chemical deteminations, and therefore of the predicted AME values, w excellent within a laboratory tiilst the reproducibility bettJleen laboratories, was &much poorer. The repeatability and reproducibility standard deviations (Steiner 1975) illustrate this quite clearly (Table 3). TABLE 3 Repeatability and reproducibility of chemical. analyses When expressed as a percentage of the mean values the repeatability s.d.'s for duplicate determinations range fran 0.49% for r&sture to The reproducibility s.d/s, again for duplicate 4.6% for sugar. determinations in two randdy selected laboratories, range fran 3.8% for crude protein to 23.2% for sugar. The 95% confidence intervals for pairs of duplicate determinations made in tm laboratories on identical feeds are 1.6% moisture, 2.0% fat, 2.3% protein, 6.5% starch and 2.2% sugar (all on a d.m. basis). These laboratories were all asked to use the sala[13 EEC analytical procedures but even so, there is reason to argue that better standardisation of methods would probably reduce the estimates of reproducibility. In a mxe limited study in which each a@ysis was done in three out. of 8 laboratories, all within the U.K., the reproducibility limits were 0.85% fat, 1.3% protein, 4.3% starch and 0.7% sugar (Fisher 1982aL Nevertheless over the whole of Europe it is clear that interlaboratory differences in analytical results are going to be an important smrce of variability in predicted ME values. When 'these ring-test data were used to calculate AME values with the proposed EEC equation the repeatability s.d. was 0.052 MJ/kg d.m. or 0,408 of the mean. The reproducibility s-d. was 0.582 W/kg d.m. or 4.48% of the mzan. Thus if a Sanple of feed were to be analysed in duplicate at two randomly selected laboratories it is expected, with 95% probability, that the two man predicted AME values would not The average differ by nwlre than 1.65 MJ/kg or 12.7% of the m. difference for many such canparisons is 0.56 MJ/kg. It is on the basis of results such as these thattolerancelimits for ME declarationshave tobefixed. The canparison of the predicted and determined AME values is sumnarised in Fig. 1. When interpreted by a parallel line statistical rcrodel, &ich was not significantly different fran the two separate tie&, there was a very high correlation between the predictions and ohs-tions (r = 0.98), but a considerable bias. The values for cockerels are underestimated by 0.57 MS/kg and those for chicks werestimateci by 2.24 MI/kg. Thus the equation is predicting relative AME values effectively but corrections must be made to yield accurate absolute values. Since the equation was derived only with data for adult birds the relative magnitude of the bias at the two ages is as expected* DISCUSSION Taken as a whole these various results show that the prediction of ME values from\ four or five chemical variables is reasonably accurate. At best, in the PEE mrk we obtained a standard deviation, including the variability of the analytical procedures of about 0.33 MJ/kg and for several. equations the s.d. was below 0.4 W/kg or about 3% of the IEall. TUerance limits on declared values depend only on the reproducibility of the chemical analyses. In the PRC experinrents it appeared that the appropriate s.d. might be as low as 0.20-0,25 MJ/kg but such an encouraging result was not 6btained in the ring-test over the whole of Europe. Standardisation of analytical procedures is obviously a critically iqortant question. a my c+ison these various esttites'of s-d. my'be Mapared d6h ;the'stand&d:error,of a II& d&mined AME value, using Six replic&&s, 'Of 0615. wfi<g. In -&utine'&&k-rather higher values would probably II& found, deperidirig oh the.techriique used. It is also interesting toinote 'that whC'the ME .&Lues *of the test.diets usd in the`= &periments VIES calculated fran four sets of table values the residual s.d.'s. ... Thus providing the problems of rang&i from 022 to 0,59 MJ/kg. analytical variability can be reduced the chtical prediction equations are virtually as good as direct bioassays and samwhat better than caqlete knwledge of the formulations. Dissatisfactions with prediction equations stem mainly fran the inherent wzakmss of the whole approach rather than fran the present state of develomt of the equations. Within the constraints of a legal declaration system it does not seem likely that new equations will be found by further experimentation. It would be foolish to rule out further develomts in analytical chemistry but again, within the cost constraints of a routiqe declaration scheme, it seens unlikely that mre general and robust predictors of ME values will be found. The one possible exception is Near IR spectrophotmtry but even in this mse sag early premising work with ruminant feeds could not be reproduced with our poulw feeds (A. Hall, private camunicationL An obvious limitation of the approach is that the results cannot be extrapolated beyond the type and range of feeds tested. It has been ass-umd that the feeds used in the experimts are a satisfactory sa@e of the poplation about which predictions are to be made; assuming of course that such a po@ation is hamgenous with respect to the relationships under investigation. A wide range of practical feed ingredients were used in the trials but it is inevitable that sane feeds will not be well described by any one, reasonably siqle, equation. Variations in anti-nutritive factors or the use of feeds containing very high levels of single ingredients are the most likely sources of systematic error and it is only by the use of a direct bioassay that such pssibilities can be cmpletely avoided. Recent developments in the techniques for rapid bioassays prabably make them quite competitive on a cost basis with even relatively siqle arrays of chemical analyses but there would be enonmus organisational costs and problems in using such a bioassay as the basis of a declaration scheme, An alternative arrangement would be to use the bioassay as a final check and arbiter in cases of dispute. The most obvious theoretical deficiency of prediction quations is that they imply that the energy yield of crude nutrients is constant i.e. that digestibility is constant. Thus in the ELEC equation the coefficient for sugar, 11.1 M/g, is considered to be too low by scribe cammmtators, This figure implies a digestibility of the energy fran sugar of 0.71 if digested sugar yields 15.6 l&J/g (H&t& et al. 1977). - Certainly feeds containing sucrose will be underestince this sugar has a digestibility of 0.99 (H&t& et al.. 19771, whilst feeds with milk sugars, digestibility = ca. 0.6, win overestimated. Such problems can be resolved if they are anticipated, what seems to be impossible is to guard against them in general. The energy value of fat is influenced by several kmwn factors and may differ quite widely frm an average assus& figure. In the PRC work this was reflected in the significant effect of fatty acid ratios but this .is an expensive parameter for routine measurmmt. For feeds containing 1 to 2% of vegetable fat fran ingredients and up to 5% of a reasonable quality feed fat variations in fat canposition may not have very large effects. To deal with the question in any general way again threatens the simplicity of the scheme. A general proposition is that the ME of fat declines with level of inclusion and in the PRC wmk the quadratic effect of fat level ms quite often significant, especidlly 305 in the data for pelleted feeds. This effect of pelletting is attributed entirely to chance, In the four sets of data analysed to derive the EX equations non-linear effects of fat level could not be demonstrated. Variations in ME values between different classes of poultry Most of the developt work has been done with adult fowls and the application of the equations to young birds or to other species may or may not be justified. The direct evidence on this topic is extrmly confused and will not be revieti At present there is no, ckar-cut quantitative answer to the here. probla and further experimmtal work is required. The question as to how such mrk should be done to give unequivocal answers is also ccmplex. ' remain as an unresolved problem. In the very limited test of the EEC equation the interpretation shown in Fig. 1 implies that the relative values of the feeds were the WE for chicks and cockerels, Whilst this is a statistically justif ied conclusion it will be seen that, in detail, the low energy feed tended to be underestirrrated for the cockerels and considerably averestimated for the chicks. The situation may therefore not be so siqle, The magnitude of the bias bet- the two tms of bird used in this experimntwasunexpectedlylarge andthismay reflect, ti m extent, the very different experimental techniques that tJ13re used. Particular care is required to &tain valid canparisons of energy availability and the measuremnt of nutrient digestibility rather than of ME might be preferred, The develo-t and standardisation of accurate analytical methods is an dbvious requirement for an energy declaration scheme based on chmical prediction equations. Althotigh it has been armed that equations should be assessed on the basis of their chemical reproducibility there is of course a danger that their relative merits will be changed @T future develomnts in analytical technique. This topic cannot be discussed in detail but a few points should be noted. First, e found no benefit in the ring-test frun adjusting the results to a dry matter basis, although we assum that this would always be done. The standard EE procedures for crude fat analysis specify ether extract for mst feeds and acid ether extract for a range of materials for which ether extract is inmnplete, including cmpound feeds with added fat. In the ring-test three laboratories analysed the feeds by both methods and found an average, 0.82, 0.91 and 0.69% mre fat following acid hydrolysis. For the prediction equation it is assumedthatthe acid ether extract is always used although this is not theoretically appropriate in ail cases. Starch is detemined by % polarimkry after solubilisation in dilute EC1 and correction for sugars extracted in 40% ethanol. In the PRC experimznts this gave, on average, 2.1%' mre starch than a m~thoa using amyloglumsidase/glucuse oxidase, but the tm could notbedistinguished~ their effectivess in predicting ,ME values. . In sue recent data frm France a wider diff&ence, 6.14% starch, was found over 48 feeds (B. LRcleroq, private cmmnication1, Sugars are extracted in ;40% aLcoho1 and deter&n& after inversion, as saccharose using the L;uff-Whorl m&h&. Amngst possible alternatives to chmical prediction quations the use of direct bioassays has already been mentioned, The other main alternative is to replace chemical analyses with an in vitro stilation 306 of digestion and to masure the disappearance of energy in the systm FUruya et al. (1979) described a m-stage assay using US&. pepsin/El and an extract of porcine intestinal fluid. Sakamto et al., (1980) showed that this assay accurately predicted both dry mtt~d We have tested a similar crude protein digestibility in the hen. system but using a camxcial pancreatic en- preparation in the The solubilisation of energy was assessed to give a second stage, nneasure of in vitro DE UVDE). For 28 feeds used in the ME experiments 'tie man IVDE was 14.69 MJ/kg which canpared well with the observ& The correlation was 0.87 with a residual value of 14.20 MJ/kg. standard deviation in the observed values after regression on IVDE of This canpared with an r.s.d. for the better chemical 1.00 w/kg. equations of 0.30 MJ/kg or less, Canbination of the in vitro results with chemical analyses did not produce any improvmt over the chemical analyses alone. At this stage of developwt therefore the in vitro method, although reasonably effective, does not look like z Whether it can be improvd and alternative to chemical prediction. whether it would better detect feeds which were poorly described by a prediction equation will have to await further develo-t. It is concluded that if energy declaration schemes are introduced for cammcial. and political reasons then reasonable solutions can be found to the technical problems raised, by the use of chemical Hcwever canpletdy robust equations are not pre