Reviews on the future of aquaculture production(Reference Naylor, Goldburg and Primavera1) and development of fish diets(Reference Hardy2–Reference Lim, Webster and Lee9) have centred around the replacement of fish meal, on which most fish diets are based, by economically viable and environmentally friendly plant protein alternatives. Defatted soyabean meal has received considerable attention due to a high protein content, reasonably balanced amino acid profile, consistent availability, cost effectiveness and palatability to most fish species(Reference Watanabe4, Reference Gatlin, Barrows and Brown6).
Soyabeans, although evaluated after heat treatment in the whole form in fish diets, are predominantly processed into defatted soyabean meal with or without hulls, but also into soya flour, soya protein concentrate and soya protein isolate. However, costs hamper the use of the latter processed products for effective replacement of fish meal in fish diets(Reference Gatlin, Barrows and Brown6).
As with all potential plant protein sources, the nutrient and antinutritional profiles of soyabean meal are currently not ideal for inclusion in fish diets(Reference Gatlin, Barrows and Brown6, Reference Francis, Makkar and Becker10). Furthermore, as summarised by, among others, Chou et al. (Reference Chou, Her and Su11) and Lim et al. (Reference Lim, Choi and Wang12), there generally appears to be large variability among fish species in the maximum dietary levels of soyabean meal tolerated, indicating different sensitivities to soyabean meal inclusion. Barrows et al. (Reference Barrows, Stone and Hardy13) concluded that the upper dietary inclusion levels of soyabean meal before fish performance or health will be deteriorated is 10–15 % (25 % fish meal replacement) for carnivorous species such as Atlantic salmon (Salmo salar), rainbow trout (Onchorynchus mykiss), sea bass (Dicentrarchus labrax) and yellowtail (Seriola quinqueradiata). However, Heikkinen et al. (Reference Heikkinen, Vielma and Kemiläinen14) stated upper inclusion levels of 20–30 % for carnivorous salmonids. In contrast, omnivorous and carnivorous freshwater fish such as common carp (Cyprinus carpio), tilapia (Oreochromis spp.), blue catfish (Ictalurus furcatus) and channel catfish (I. punctatus) seem to grow well on high percentages (70–100 %) of fish meal replaced by soyabean meal(Reference Chou, Her and Su11, Reference Kikuchi15). Factors causing discrepancy among researchers on the use of soyabean meal as a protein source for fish might be related to quality, processing and inclusion levels of soyabean meal, variation in diet formulation, and differences in fish species, fish size and culture system(Reference El-Sayed3, Reference Elangovan and Shim16–Reference Romarheim, Skrede and Gao17).
Although several reviews(Reference Hardy2–Reference Gatlin, Barrows and Brown6, Reference Lim, Webster and Lee9–Reference Francis, Makkar and Becker10) on feed ingredients for use in fish diets have included the replacement of fish meal by soyabean products, they were concentrated on summative descriptions of results obtained from research studies. These narrative reviews consider all studies with equal weight, without an account for measures of dispersion. Meta-analysis, the review of scientific literature with the emphasis on providing a quantitative synthesis of data, allows the evaluation and integration of results from a group of studies, even those with seemingly contradictory results(Reference Fernandez-Duque18).
The objective of the present study was to analyse, with the use of meta-analytic techniques, available published growth results obtained in fish due to the replacement of dietary fish meal by soyabean products. The outcome would provide a numerical measurement of the extent of growth differences.
Materials and methods
Selection of studies
A comprehensive literature search was conducted on the Internet with the use of several search engines and publishers' websites. Cook et al. (Reference Cook, Guyatt and Ryan19) concluded that unpublished results should not be completely excluded from meta-analysis, but be subjected to the same rigorous methodological evaluation than published peer-reviewed data, and results being presented with and without inclusion of unpublished results. However, inclusion of the latter can be problematic, especially when coming from interested sources. Willingness of investigators related to outcome, with favourable results being provided more easily, and hidden unpublished results even after extensive consultation, could result in an unpresentative sample of unpublished studies. This causes doubt if the inclusion of unpublished studies increases or decreases bias in meta-analyses(Reference Davey Smith and Egger20). Taking the above into consideration, studies have been selected for evaluation in the present study that: (1) had replaced fish meal in diets by soyabean products, (2) presented a measurement of fish growth, (3) appeared in peer-reviewed journals, and (4) been published in English in order to extract all relevant information. Studies fulfilling the above were further subjected to evaluation for inclusion in meta-analyses according to criteria presented in Table 1.
Table 1 Selection of studies for inclusion in meta-analysis
Whereas some studies included only one level of fish meal replacement, others contained multiple replacements. Furthermore, different products(Reference Wee and Shu21–Reference Hemre, Sanden and Bakke-McKellep28), similar products subjected to different processing treatments(Reference Olivia-Teles, Gouveia and Gomes23, Reference Nengas, Alexis and Davies25, Reference Boonyaratpalin, Suraneiranat and Tunpibal27), supplementation with amino acids(Reference Lim, Choi and Wang12, Reference Brown, Twibell and Jonker26, Reference Shiau, Chuang and Sun29–Reference Tibaldi, Hakim and Uni41), effects at different dietary protein levels(Reference Shiau, Chuang and Sun29, Reference Shiau, Pan and Chen42), and the influence on different fish species(Reference Refstie, Korsoen and Storebakken43) and fish sizes(Reference Brown, Twibell and Jonker26, Reference Khan, Jafri and Chadra37–Reference Choi, Wang and Park38, Reference Gallagher44–Reference Hernández, Martínez and Jover46), were often evaluated in the same study. Due to the apparent effect of all of the above variables, data were not pooled for individual studies, but used in individual comparisons. Although this might caused dependence on one another for some effect sizes, exclusion of non-independent comparisons may bias results more than their inclusion(Reference Hedges and Olkin47, Reference Gurevitch, Morrow and Wallace48). The above resulted in a coding system based on trial identification numbers.
Data analysis
Fish growth in studies selected for inclusion in the meta-analysis has been presented in different units: total weight gain (g), weight gain (%), specific growth rate (%), and daily and thermal growth coefficients. This necessitated the use of a common metric independent of differences in unit measurements. Effect size was measured with Hedges' d (Reference Hedges and Olkin47), based on the difference between the means (
) for treatment (T) and control (C) groups, standardised by dividing by the pooled standard deviation (s p), and corrected for bias (J) for small sample sizes (n):
with
and
The asymptotic se of the effect size was estimated by Hedges(Reference Hedges49):
Precision of d was illustrated with the 95 % CI:
Summary statistics were calculated using a random-effects model(Reference Hedges and Vevea50), which takes into account between-trial variability (true heterogeneity) as well as within-trial variability (sampling error).
A fail-safe number (Nfs)(Reference Orwin51) has been calculated to indicate the number of unpublished comparisons with null effects needed to reduce the observed d to a negligible level:
where n is the number of treatment v. control comparisons, 
is the weighted mean d of comparisons, 
is the desired minimal mean d and 
is the mean d of additional comparisons.
Results and discussion
Soyabean meal
Of ninety-nine studies presenting information on the influence of replacement of dietary fish meal by defatted soyabean meal on fish growth, 47 % were found suitable for inclusion in a meta-analysis (Table 1). Absence of a diet without any soyabean meal, which could serve as a true control group for calculation of an effect size, was the single factor resulting in the highest amount (n 18) of rejected studies.
Comparisons of the replacement of fish meal by soyabean meal at different levels without dietary supplementation of amino acids, extracted from different studies and coded as trials, are presented in Table 2(Reference Chou, Her and Su11, Reference Lim, Choi and Wang12, Reference Kikuchi15–Reference Romarheim, Skrede and Gao17, Reference Wee and Shu21–Reference Boonyaratpalin, Suraneiranat and Tunpibal27, Reference Shiau, Chuang and Sun29, Reference Shiau, Kwok and Hwang30, Reference Davies and Morris32, Reference Keembiyehetty and Gatlin33, Reference Khan, Jafri and Chadra37–Reference Ai and Xie39, Reference Tibaldi, Hakim and Uni41, Reference Refstie, Korsoen and Storebakken43, Reference Hernández, Martínez and Jover46, Reference Robaina, Izquierdo and Moyano52–Reference Akiyama, Unuma and Yamamoto61).
Table 2 Highest levels of defatted soyabean meal inclusion and fish meal replacement at which growth obtained did not differ from that with a fish meal control diet (P>0·05), and trials where soyabean meal inclusion could not maintain a similar growth to a fish meal control diet (P<0·05)
C, carnivorous; S, salt water; N, not indicated; WG, weight gain (%); W, warm water ( ≥ 20°C); TWG, total weight gain (g); SGR, specific growth rate (%); CO, cold water ( < 20°C); F, fresh water; O, omnivorous; TGC, thermal growth coefficient.
* Not included in meta-analysis.
A total of 67 % of trials evaluated carnivorous species, with separation according to water type (fresh v. salt) and water temperature (cold v. warm). Only one saltwater omnivorous species (sharpsnout seabream; Diplodus puntazzo)(Reference Hernández, Martínez and Jover46) has been included, and all omnivorous species had been reared in warm ( ≥ 20°C) water. Evaluation periods varied from 33 to 182 d, although 80 % of trial periods were between 8 and 12 weeks. Dietary crude protein levels, converted, if possible, to dry weight when presented on a wet weight basis, varied from 250 to 612 g/kg. Fish meal replaced included brown, Chilean, menhaden, Norwegian, Peruvian and white sources. However, information on the processing status of soyabean meal evaluated was extremely limited. Available data indicated the ranges of crude protein and lipid of fish meal evaluated as 614–750 and 35–152 g/kg, respectively, with 448–544 and 10–141 g/kg, respectively, reported for soyabean meal. In trials 14 and 15 replacement of fish meal by soyabean meal presented higher (P < 0·05) specific growth rate values than the fish meal control diet. This could probably be related to the quality of the fish meal used(Reference Wee and Shu21, Reference Olivia-Teles, Gouveia and Gomes23).
In the calculation of Hedges' d, referred to as effect size hereafter, at individual replacement levels (Fig. 1), comparisons from trials 1, 5, 15, 16 and 34 (Table 2) were excluded due to the absence of a measurement of dispersion of the means. Although effect size can be calculated from P values if the direction of the finding is known, P values in the above five trials were reported as less or more than a number. Such significance levels are often treated as if they were an exact P value (0·05) if P < 0·05, with effect size set to zero if results are reported as non-significant (P>0·05). However, doing this causes poor estimates(Reference DeCoster62), and so was omitted in the present study. An additional trial(Reference Krogdahl, Bakke-McKellep and Baeverfjord63), which evaluated inclusion (76, 117, 153, 194, 270 g/kg) of toasted solvent-extracted soyabean meal as replacement (12, 18, 24, 30, 42 %) for low-temperature dried fish meal in diets (958–962 g/kg crude protein) with Atlantic salmon (fish size: 280 g) over a 60 d period, was included in the meta-analysis. Although this trial did not indicate significance levels among individual replacements, it presented a pooled sem. Estimation of an effect size failed in trial 20 due to sd values of 0·0.
Fig. 1 Effect sizes (Hedges' d, as defined in the Data analysis section) for growth with 95 % CI as influenced by level of fish meal replacement by defatted soyabean meal for carnivorous (○; n 52) and omnivorous (●; n 25) fish species.
Limited values and overlapping of 95 % CI demonstrated no gain in separation of species according to feeding habits (Fig. 1). Furthermore, dietary crude protein levels, which could be categorised accordance to feeding habit (carnivorous, 360–612 g/kg; omnivorous, 225–433 g/kg; Table 2) in the present study, were not linearly related to effect size (Fig. 2), as illustrated by a weighted Pearson correlation coefficient (r) of 0·1334 (95 % CI − 0·0934, 0·3471; P = 0·2474). This eliminated the suggestion(Reference El-Sayed3) that dietary crude protein level, despite some contradictory results, might have an influence on the effect of replacement of fish meal by soyabean meal.
Fig. 2 Effect sizes (Hedges' d, as defined in the Data analysis section) for growth with 95 % CI when replacing fish meal by defatted soyabean meal as influenced by dietary crude protein levels (n 77).
Figure 1 illustrates that the influence of fish meal replacement level prevented the calculation of a cumulative mean effect size across all levels. In addition, effect sizes did not follow a distinct trend with increasing replacement levels. The absence of a strong linear relationship was displayed by a weighted Pearson r of − 0·4271 (95 % CI − 0·5943, − 0·2246; P = 0·0001). This presented an R 2 value of 0·1824, with little of the variation explained by a linear model, and little predictive value.
However, according to their distribution (Fig. 1), effect sizes tended to be grouped into three replacement level categories: 4–40 %, with several mean effect sizes higher than 0 and most 95 % CI overlapping with zero; 42–83 %, with all mean values less than 0 and limited overlapping of 95 % CI with zero; and 100 % with values, although limited (n 5), including extremes. Trials presenting effect sizes that deviated to a large extent from zero in the 4–40 % replacement category included: 37 % fish meal replacement evaluated with hybrid striped bass in trial 7 ( − 4·8717; 95 % CI − 7·6320, − 2·1115), and 40 % replacement with rainbow trout in trial 36 ( − 8·8314; 95 % CI − 14·0782, − 3·5847).
As mentioned above, factors related to ingredients, diet, fish species and rearing might have an influence on the outcome of dietary fish meal replacement by soyabean meal. With information on these sources of variability seldom reported, and all sources of variation most often unidentified, the logic of the analysis in the present study was that effect sizes have been sampled from a distribution of effect sizes with a true effect that could vary from study to study. Therefore a random-effects model was the appropriate model to compute the mean of the effect sizes(Reference Borenstein, Hedges and Rothstein64). Mean effect sizes for different fish meal replacement categories are presented in Table 3. To be compatible with further comparisons, categories were classified as 4–40, 41–95 and 100 %. This strategy should not be confounded with subgroup analysis, which can be described as an analogue of the ANOVA(Reference Lipsey and Wilson65), and is used to identify heterogeneity among studies when fitting a fixed-effects model.
Table 3 Mean effect sizes* for fish meal replacement by defatted soyabean meal at different levels
* Hedges' d (as defined in the Data analysis section).
Interpretation of effect sizes is controversial, but the most accepted opinion is that of Cohen(Reference Cohen66), who proposed values of 0·2, 0·5 and 0·8 to be considered as indicative of small, medium and large standardised effect sizes, respectively, in social sciences. However, biological importance is more objective than practical or clinical importance in which subjective judgements are needed, and biologists should evaluate effect sizes according to their hypotheses(Reference Nakagawa and Cuthill67). In the present study an effect size was considered as statistically significant from no effect (0) at the the 5 % level (two-tailed) if the the 95 % CI did not overlap with zero(Reference Steidl, Thomas, Scheiner and Gurevitch68). According to the above, growth obtained with diets in which 4–40 % of fish meal (inclusion levels of 150 to 756 g/kg) was replaced by soyabean meal (inclusion levels of 71 to 366 g/kg) did not differ from growth when feeding a fish meal control diet (Table 3). However, with an upper 95 % CI of − 1·1625, fish meal replacement at 41–95 % caused a cumulative effect size substantially different from zero. The effect size calculated for 100 % fish meal replacement should be treated with caution, as it becomes impossible to estimate the between-trials variance with any precision when sample sizes become limited(Reference Borenstein, Hedges and Rothstein64).
Due to most studies evaluating the effect of soyabean meal inclusion at several fish meal replacement levels, the occurrence of the tendency to only publish positive results causing publication bias, the so-called ‘file drawer problem’(Reference Rosenthal69), is unlikely to have had any importance in the present study. However, a Nfs was calculated to estimate the robustness of each cumulative effect size, with 
chosen as − 0·2000 and 
as 0(Reference Rosenberg70). With a mean effect size of − 0·1142, calculation of the number of unpublished comparisons with null effects to reduce the observed effect size to − 0·2000 was irrelevant for replacement of fish meal at 4–40 %. However, with replacement of 41–95 % fish meal, 183 additional studies with an effect size of 0 would reduce the mean effect size to − 0·2000. With Nfs considered as strong if greater than 5n+10, with n the original number of studies(Reference Rosenthal71), the above value illustrates the stability of the latter calculated mean effect size. Although seventy-four null effects would be needed to reduce the effect size to − 0·2000 at 100 % fish meal replacement, care should be practised with the interpretation of this calculated mean effect size, as described above.
Supplementation with amino acids
With defatted soyabean meal limiting in total sulfur amino acids when used in animal feeds, diets with high dietary inclusion levels of soyabean meal are often supplemented with methionine and other amino acids(Reference Gatlin, Barrows and Brown6). Trials that have evaluated this concept are summarised in Table 4(Reference Lim, Choi and Wang12, Reference Brown, Twibell and Jonker26, Reference Shiau, Chuang and Sun29, Reference Shiau, Kwok and Hwang30, Reference Davies and Morris32, Reference Keembiyehetty and Gatlin33, Reference Khan, Jafri and Chadra37–Reference Ai and Xie39, Reference Tibaldi, Hakim and Uni41, Reference Shiau, Pan and Chen42, Reference Gallagher44, Reference McGoogan and Gatlin45, Reference Refstie, Førde-Skjærvik and Rosenlund72–Reference Storebakken, Kvien and Shearer78).
Table 4 Highest levels of defatted soyabean meal inclusion and fish meal replacement, with dietary supplementation of methionine, at which growth obtained did not differ from that with a fish meal control diet (P >0·05), and trials where soyabean meal inclusion together with supplemented methionine could not maintain a similar growth to a fish meal control diet (P<0·05)
C, carnivorous; S, salt water; CO, cold water ( < 20°C); SGR, specific growth rate (%); F, fresh water; TWG, total weight gain (g); O, omnivorous; W, warm water ( ≥ 20°C); WG, weight gain (%).
* Fish meal control diet supplemented with 1·2 % dl-methionine.
† Fish meal control diet supplemented with 0·1 % dl-methionine.
‡ Plus l-lysine.
§ Fish meal control diet supplemented with 0·2 % l-methionine.
∥ Not included in meta-analysis.
¶ Plus arginine, histidine, l-lysine, threonine, tryptophan.
In general, methionine supplementation, varying from 0·12 to 2·70 %, has been applied at higher fish meal replacement levels (Table 4) than used when only soyabean meal was included. Five trials (trials 45, 48, 49, 54, 64) evaluated replacements with methionine supplementation together with lower replacement levels without supplementation (Table 2). Few trials included supplementation of lysine (trials 48, 49, 65) and other essential amino acids (trial 65). Methionine was included either in the dl (trials 40, 41, 42, 44, 46, 47, 56, 61, 62, 63) or l form (trials 45, 52, 53, 54, 55, 60, 64, 65), with some studies not reporting the form of methionine.
With the calculation of combined effect sizes, additional studies that did not present statistically significant differences, but supplied data suitable for meta-analysis, were included. Refstie et al. (Reference Refstie, Helland and Storebakken79) replaced 68 % of fish meal by soyabean meal (inclusion level: 600 g/kg) in diets (447–467 g/kg crude protein) supplemented with 0·5 % dl-methionine to evaluate growth of 33·5 g rainbow trout over 56 d. Refstie et al. (Reference Refstie, Storebakken and Roem80) included 339 g/kg hulled toasted soyabean meal together with 2·30 % dl-methionine to replace 39 % fish meal in diets (388–433 g/kg crude protein) for Atlantic salmon (fish size: 107 g) over a 55 d period The above two studies presented growth parameters over different subperiods of the trial. However, variability associated with the dividing factor removed from standard deviations could be regained to get a pooled standard deviation over the entire period(Reference DeCoster62). Venou et al. (Reference Venou, Alexis and Fountoulaki81) presented data on the replacement (20, 30, 45 %) of fish meal by hulled soyabean meal (before and after extrusion) at inclusion levels ranging from 231–485 g/kg, with dl-methionine supplementation at 0·20–0·30 %. These diets (470 g/kg crude protein) were evaluated with 9 and 50 g gilthead seabream over periods of 60 and 66 d, respectively.
Amino acid supplementation did not substantially change effect sizes in the 0–40 and 100 % fish meal replacement categories, compared with those obtained without supplementation. However, it caused a decrease in the mean effect size and 95 % CI in the 41–95 % group (Table 3). Trial 42, with an evaluation of growth in Atlantic salmon over a 300 d period, presented effect sizes of 1·9352 (95 % CI − 0·4396, 4·3100) and − 6·3832 (95 % CI − 11·2212, − 1·5452) at 18 and 38 % fish meal replacement levels, respectively.
Effect sizes obtained with and without amino acid supplementation in the same study are illustrated in Fig. 3. In trials 49, 56 and 57, with 30, 52 and 75 % fish meal replaced, respectively, methionine supplementation caused overlapping of 95 % CI with zero, compared with no overlapping without supplementation. Although 95 % CI still intersected with zero, methionine supplementation decreased the positive effect size found without supplementation in trial 56 at 39 % fish meal replacement, and to a lesser extent in trial 60 at 33 % replacement.
Fig. 3 Effect sizes (Hedges' d, as defined in the Data analysis section) for growth with 95 % CI when replacing fish meal by defatted soyabean meal for amino acid supplementation (□; n 17) compared with non-supplementation (▒; n 12) evaluated in the same study.
Different supplementation levels of dl-methionine at a constant fish meal replacement level presented similar results with Southern catfish (trial 56). However, supplementation with multiple amino acids resulted in a significantly higher growth than supplementation of only methionine and lysine in rainbow trout (trial 65). In trial 57, Keembiyehetty & Gatlin(Reference Keembiyehetty and Gatlin33) evaluated different forms (l-, dl-, acetyl-, dl-hydroxyl analogues) of methionine at the same fish meal replacement level with sunshine bass, but did not find any significant growth differences among l-, dl- and acetylmethionine, and a fish meal control diet.
The 95 % CI ( − 2·7177, − 0·3164) of the cumulative mean effect size ( − 1·5171) with non-supplementated diets (n 12) of trials indicated in Fig. 3 did not include zero. However, when amino acids were supplemented, the 95 % CI moved to − 1·2308 to 0·0015, with a mean effect size of − 0·6146. It should be stressed that supplemented crystalline amino acids are suggested to be prone to faster uptake and catabolism(Reference Cowey and Walton82), and to leaching in aquatic environments(Reference Zarate and Lovell83), compared with those in intact protein.
Soya protein concentrate
Trials that replaced dietary fish meal by soya protein concentrate, produced through aqueous ethanol or methanol extraction of defatted soya flakes, with a typical crude protein content of 650–700 g/kg(Reference Lusas and Riaz84), are presented in Table 5(Reference Rumsey, Siwicki and Anderson24, Reference Nengas, Alexis and Davies25, Reference Kaushik, Cravedi and Lalles31, Reference Médale, Boujard and Vallée34–Reference Day and Plascencia González36, Reference Deng, Mai and Ai40, Reference Kissil, Lupatsch and Higgs85–Reference Stickney, Hardy and Koch89). However, crude protein content of the product used in trial 74(Reference Stuart and Hung86) was indicated as approximately 900 g/kg. Soya protein concentrate has been evaluated with only seven fish species, of which two (Atlantic halibut, white sturgeon) were omnivorous, and twenty-nine from thirty-six comparisons used it to replace ≥ 50 % of fish meal.
Table 5 Highest levels of soya protein concentrate (SPC) inclusion and fish meal replacement, with and without dietary supplementation of methionine, at which growth obtained did not differ from that with a fish meal control diet (P>0·05), and trials where soya protein concentrate could not maintain a similar growth than a fish meal control diet (P<0·05)
C, carnivorous; F, fresh water; CO, cold water ( < 20°C); DGC, daily growth coefficient; N, not indicated; SGR, specific growth rate (%); S, salt water; W, warm water ( ≥ 20°C); TWG, total weight gain (g); O, omnivorous; WG, weight gain (%).
* Not included in meta-analysis.
† Plus l-lysine and l-threonine.
‡ Plus mixture of l-leucine, l-lysine, dl-methionine, l-threonine, l-valine as crystalline amino acids.
§ Plus mixture of l-leucine, l-lysine, dl-methionine, l-threonine, l-valine as cellulose acetate-encapsulated amino acids.
∥ Supplemented with 0·22 % dl-methionine.
¶ Supplemented with 0·42 % dl-methionine.
Replacement of fish meal by soya protein concentrate caused a significant growth decrease in most trials. However, its value as a fish meal substitute was substantially increased when supplemented with amino acids (Table 5). Deng et al. (Reference Deng, Mai and Ai40) evaluated a mixture of amino acids, included as either a crystalline amino acid mixture (trial 80) or encapsulated by cellulose acetate phthalate (trial 81), at a similar fish meal replacement level, but did not find any significant differences in the growth of Japanese flounder between treatments. Methionine supplementation at 100 % fish meal replacement decreased (P < 0·05) growth compared with a fish meal control diet in rainbow trout (trials 82, 83, 84). However, at lower replacement levels in the latter studies soya protein concentrate without amino acid supplementation resulted in similar growth between diets (trials 67, 68).
The absence of a measurement of variance eliminated trials 66, 67, 77, 78 and 82 (Table 5) from the calculation of effect sizes (Fig. 4). An additional trial(Reference Storebakken, Shearer and Roem90), which indicated statistical significance for differences in specific growth rate of Atlantic salmon (fish size: 106–111 g) over different phases of a 84 d period, was included in the calculation of effect sizes. In the latter study soya protein concentrate (inclusion level: 480 g/kg) replaced 75 % of low temperature dried fish meal in diets with crude protein levels of 430–457 g/kg.
Fig. 4 Effect sizes (Hedges' d, as defined in the Data analysis section) for growth with 95 % CI as influenced by level of fish meal replacement by soya protein concentrate without (○; n 16) and with (●; n 10) amino acid supplementation.
Limited values and non-significant weighted Pearson r's between effect sizes and replacement levels found without (r − 0·1055; 95 % CI − 0·5714, 0·4118; P = 0·6973; n 16) and with (r − 0·4541; 95 % CI − 0·8428, 0·2459; P = 0·1874; n 10) amino acid supplementation eliminated any further evaluation of relationships. Cumulative mean effect sizes did not differ substantially between trials without ( − 2·7306; 95 % CI − 3·7991, − 1·6620) and with ( − 2·4373; 95 % CI − 3·9004, − 0·9742) amino acid supplementation, and 95 % CI did not overlap with zero in either.
The evaluation of effect sizes obtained with fish meal replacement by other soyabean products, for example, full-fat soyabeans and soya flour, was prevented by a lack of studies presenting appropriate values, as illustrated in Table 1.
Conclusions
The present study quantified the magnitude and precision of the effect caused by the replacement of dietary fish meal by soyabean products on fish growth. The absence of standardisation in units for measurement of growth in fish resulted in the application of Glassian meta-analysis, based on standardised effect sizes calculated between a control (fish meal) and treatment (fish meal replacement) diet. An important contribution from the study could be ascribed to the identification of deficiencies in reporting of results. Failure to report a measurement of variation, as found with numerous studies evaluated for inclusion, rendered results unsuitable for meta-analysis. Standardising in experimental protocol regarding, among others, replacement levels, evaluation period, measurement units and reporting of variance, are of utmost importance for evaluation of trends with information supplied by different studies.
Data used in the current study presented evidence that the effect of the replacement of fish meal by defatted soyabean meal did not display a definite trend with replacement level. However, replacement of up to 40 % fish meal caused similar growth to that obtained with diets based solely on fish meal as a protein source, irrespective of dietary protein content, in a wide range of fish species. Amino acid supplementation of diets, mostly as crystalline methionine, aided in decreasing the negative effect caused by the replacement of fish meal at levels higher than 40 %. Despite the fact that the above has been indicated by narrative reviews, it was based on summative results obtained with null hypothesis significance testing in individual studies. With limited replicates, as often is encountered in fish nutrition studies, the latter testing technique has low statistical power to detect differences, and gives no indication of the size of differences. The present study is the first to put numerical values to the above differences, and to indicate the direction of effects as obtained across studies.
Evaluation of the influence of fish species, and the influence of stratification of fish species according to feeding habit, water type and water temperature on growth differences due to the replacement of fish meal by soyabean products, are hampered by a lack of suitable values for analysis. A similar lack of values prevented searching of trends at replacement levels higher than 40 %. Further research in order to provide results suitable for meta-analysis is urgently needed.
Baseline values are presented in the currrent study for the magnitude of effect sizes due to replacement of fish meal with soya products, which could be utilised not only in further meta-analyses, but also for comparative purposes in research on individual fish species. Furthermore, the present study illustrates the use of Glassian-based meta-analytic techniques to quantify responses in studies on fish nutrition.
Acknowledgements
The present study was financially supported, including the salary of J. S., by research plan no. MSM 6007665809 of the University of South Bohemia Ceske Budejovice, Research Institute of Fish Culture and Hydrobiology (Vodnany, Czech Republic). It received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
J. S. was the only contributor to all (research, analysis, writing, etc) of the paper.
There are no conflicts of interest involved in this paper.
