Evidence is presented that suggests strongly that hormone growth promotant (HGP) implantation has a negative effect on beef palatability. This is based on a meta-analysis of results reported in refereed papers that have appeared in the meat-science literature. To be included in this analysis, a paper must have reported results for control samples (no HGP) and treatment samples (HGP) for either objective testing (Warner-Bratzler shear-force) or consumer preference (tenderness score). The paper must also have reported estimates and standard errors. Further, we consider only the case of steers, and the M. longissimus dorsi (striploin). While most of these studies yielded non-significant differences, most gave an estimate indicating that the HGP treatment had a negative effect on beef palatability. When these results are combined using a meta-analysis, they provide significant evidence that the use of HGP implants negatively influences palatability.

A fundamental component of both scientific research and policy decision making is the review of existing information on the methodology or intervention of interest. This information is commonly quantitative in nature and arises from a variety of sources. Inherent in this information is some variability both in the information itself and in the perceived reliability of the information supplied from each source. Questions arise therefore as to how this information is to be utilised to best aid decision making or further research on the intervention of interest. This quantitative approach, combining the results from each information source, together with the collection and selection of information has come to be widely known as meta-analysis: ‘the analysis of analyses’ (Glass 1976, p. 4). A meta-analysis can take many forms, corresponding to the field of research in which it is applied. These fields are many, including environmental sciences (Hilborn and Lierman 1998; Gurevitch and Hedges 1999), management (Forza and Dinuzzo 1998; Phillips 1998), education and the behavioural sciences (Dunn et al. 1995; Quinn et al. 1999) and medical research (Halpem et al. 1998; Bent et al. 1999). It can also be applied in meat science. The appeal of meta-analysis is that it presents a scientific, objective means of reviewing and analysing existing quantitative information. It is presented as ‘a rigorous alternative to the casual, narrative discussions of research studies’ (Glass 1976).

The intention of this paper is to assemble a complete summary of the published effects of HGP use on beef palatability as measured by tenderness scores and by Warner-Bratzler shear-force measurements for M. longissimus dorsi in steers.

The studies considered in this meta-analysis are papers that have appeared in refereed journals for which estimates and standard errors are available, including the recent Australian studies (Hunter et al. 2001; Thompson et al. 2008a, 2008b; Watson et al. 2008b; McIntyre et al. unpubl. data) summarised in the Results section. This resulted in 20 studies, with 30 treatment–control comparisons for the tenderness evaluations; and 16 studies, with 22 treatment–control comparisons for the shear-force measurements.

About 70% of these treatment–control comparisons yielded non-significant differences, but more than 90% gave an estimate indicating that the HGP treatment had a negative effect on beef palatability. When these results are combined using a meta-analysis, they provide significant evidence that the use of HGP implants negatively influences palatability. It is found that the estimate of the HGP effect is to increase shear-force by 0.27 kg, with standard error 0.03; and to reduce tenderness by 5.4 points on a 100-point scale, with standard error 0.8.

Publication bias may exist where there is a tendency for papers with significant results to be published. In most papers considered in this meta-analysis, palatability was a subsidiary result to the other reported results, usually the effect of HGP on animal growth. So the significance or otherwise of the palatability result was not important to the paper’s publication. As a consequence publication bias is unlikely to occur in this case. Indeed, in most cases, the HGP effect on palatability was found to be non-significant.

A meta-analysis was conducted of results reported in refereed papers that have appeared in the meat-science literature. Such papers were found using searches on (beef or steer) and (palatability or shear-force or tenderness) and (HGP or hormone/hormonal or implants or shear-force). To be included in this meta-analysis, a paper must have appeared in a refereed English-language journal and it must have reported estimates and standard errors for either objective testing (Warner-Bratzler shear-force) or consumer preference (tenderness score). A paper may contain more than one treatment–control comparison. To be included in the meta-analysis, such a comparison must have a specific control sample (no HGP) and a treatment sample or samples (HGP).

To avoid possible sex effects and muscle differences, attention was restricted to papers reporting on steers, and the M. longissimus dorsi (striploin). This choice ensured a wide range of results, as most trials for which a comparison has been made between control (no implant) and treated (with HGP implant) include steers, striploins, and shear-force or tenderness scores. This resulted in 20 eligible papers for the tenderness evaluations; and 16 for the shear-force measurements.

In order to make the results of the objective tests and the sensory tests comparable, a ‘toughness’ measure is used in each case: in the objective case the shear-force, and in the sensory case a negative perceived tenderness. In each case then the HGP effect is the mean increase in toughness when HGP is used, as compared with a control where HGP is not used.

Sensory scores are reported on varying scales. To ensure that all sensory scores are comparable, they are converted to a scale from 0 to 100. If x denotes the tenderness score on the scale from a to b, then the following definition is used:

In performing the meta-analysis, it is assumed that all HGP treatments have the same palatability effect: i.e. that each is sampled from a random effect distribution with a common mean. This assumption is not contradicted by the published results. Further, in cases where a number of HGP treatments have been compared against a common control, the HGP effect estimates are not independent as each has a common control. In such cases, the HGP treatments are pooled before comparison with the control.

In combining independent estimates, the optimal linear combination of estimates was used, i.e. with weights inversely proportional to the square of the standard error. Given independent estimates, est_i with standard error s.e._i (i = 1, 2,...,k), the combined estimate is given by

and for this estimate the standard error is such that

The number of HGP effect estimates reported here is determined by the number of independent control groups. The relevant estimates, standard errors and confidence intervals were extracted, combining results as necessary. These results were included in the meta-analysis of published results given below.

Recent Australian studies

CRC trial: Hunter et al. (2001), Thompson et al. (2008b)

This trial has been reported in Hunter et al. (2001) where a preliminary report of the sensory results from 203 animals was given. Thompson et al. (2008b) presented further results from the same trial for 486 animals, though the results given there did not specifically elaborate on the estimates and standard errors of the HGP effects. These are specified below. In the trial, 509 Brahman and F1 Brahman crossbred steers were used. At weaning, these animals were allocated to three market endpoints (Domestic = 220 kg, Korean = 280 kg, Japanese = 320 kg) by two nutritional finishing strategies (grain-based feedlot, pasture-fed). About half of each group was allocated as control (no implant) and the other half as treated (20 mg estradiol-17 about every 100 days). Fuller details of the experimental procedure can be found in Hunter et al. (2001) or Thompson et al. (2008b). The numbers of animals in the six groups are indicated in Table 1. The completed study contains more than twice the number used in Hunter et al. (2001). However, as pointed out in the previous papers, there were some problems with processing some of the carcasses, resulting in cold-shortening: Fig. 1 indicates that there were a number of extremely tough samples. There were also a number of extreme ultimate pH values. To avoid distortion from outliers, meat for which pHu > 5.7 were deleted from the sample. As a result, meat corresponding to the points to the right of Fig. 1 were discarded from the analysis.

Estimates and standard errors of the effect of HGP implants on tenderness and shear-force are presented in Table 2.

WA-1 trial: Thompson et al. (2008a)

This trial, utilising three cooking methods, covered a wide range of muscles for heifers and steers. The details are covered in full in Thompson et al. (2008a). Here only the details for the M. longissimus for steers are extracted. There were 47 control steers and 44 treated steers with results given in Table 3.

Meta-analysis

The papers used in the meta-analysis are listed in Figs 2 and 3. The results of the meta-analysis are summarised in Table 4. The details of the meta-analysis procedure are summarised in the Appendix.

In neither case does the homogeneity test show any significant difference between the studies. In the case of shear-force, the best estimate of the inter-study variance is zero, so that the random-effects model reduces to the fixed-effects model. In the case of the sensory data, there is some indication of a difference between studies, although this is not significant (P = 0.119). Despite this, the random effects model seems preferable, so as to allow for the possible differences between studies. It is a more general model, and it includes the fixed effects model as a special case. Both fixed-effects and random-effects model results are reported: the random-effects confidence interval is slightly wider, to allow for the possible variation between studies.

The meta-analysis is illustrated in Figs 2 and 3. In these figures, each line represents a treatment–control comparison, from the specified paper. The (reported or calculated) estimate and standard error are given and, on the same line, the estimate and the 95% confidence interval are shown: the dot representing the estimate and the line the confidence interval, with respect to the specified scale. The overall estimates and confidence intervals (for both the fixed effects and random effects models) are given at the bottom of the diagram.

The confidence intervals for the separate experiments are usually quite wide, and mostly include zero, indicating non-significance. But when the results are accumulated the result is a narrow confidence interval which clearly excludes zero, and is therefore highly significant.

There is significant evidence here that HGP affects palatability negatively. Initially it was felt that some time limitation should be put on the studies considered. However, although there is some (non-significant) indication of a variation between studies, the estimates show no trend with year of publication (see Fig. 4). In fact the results for tenderness after 1980 yield a slightly larger estimate of HGP effect. There seemed to be no good reason for excluding any of the published studies.

There was no a priori indication for a publication bias, and most results were non-significant in any case. A funnel plot for each of the sets of estimates is show in Fig. 5. There is no indication of asymmetry, which would suggest a publication bias.

The evidence presented here suggests very strongly that HGP has a negative effect on beef palatabilty. Evidence for the sensory results dates back 50 years and little seems to have changed: the results for 1955–1980 are quite similar to the results for 1981–2008.

The question which must now be asked is what might be done to ameliorate it. Or is it just a price to be paid for efficient beef production?

The author is grateful to two anonymous referees for their thorough reviews and helpful criticisms.

Consider a collection of k studies, each of which measures the effect of a particular intervention. It can be assumed that from each of these studies a point estimate of the effect of intervention and a standard error of that estimate are available. This Appendix considers two standard methods by which these studies may be combined in order to obtain an estimate and confidence interval for the overall effect of the intervention of interest.

For each of the k studies let (i = 1,2,...,k) denote the estimated effect of intervention and θ_i the true effect of intervention. A general model is then specified by

and the e_i are assumed to be independent. The estimated effect size can be any measure of effect provided the assumption of normality is (at least approximately) appropriate.

In practice the σ_i² are unknown and so estimated variances are used. It is widely assumed, however, that the individual studies provide reasonable estimates of the σ_i². This assumption is almost universally made when the studies are analysed individually and should be the case provided the studies are at least moderate in size. In order to emphasize that the within study variances are estimated, the notation is used, where is the standard error of . The general model is therefore rewritten as

For the meta-analyses considered here, the parameter of interest is the overall effect of intervention, denoted μ. In the remainder of this Appendix outlines both the fixed and random effects models and describes how each relates μ to the θ_i. In both cases the estimates are point estimates of μ. Mention is also made of a test of homogeneity, widely used to select either the fixed or random effects model.

The fixed effects model for meta-analysis assumes that the k studies are homogeneous—each having the same true effect of intervention. Therefore θ_i = μ for all i = 1,2,...,k, giving the model

This model only allows for within-study variation.

The overall effect μ is commonly estimated using a weighted average. Ignoring the sampling error in the , it is optimal to use weights proportional to giving

The notation is used to indicate that the weights use estimates , although the variation associated with these estimates is ignored in practice. Under the assumptions of independence and normality,

and an approximate 95% confidence interval for μ is given by