1st Page

Journal of Animal Science - Article



This article in JAS

  1. Vol. 89 No. 12, p. 4054-4067
    Received: Sept 07, 2010
    Published: December 4, 2014

    2 Corresponding author(s):


Effects of dietary chromium supplementation on performance, carcass characteristics, and meat quality of growing-finishing swine: A meta-analysis1

  1. J. Sales 2 and
  2. F. Jančík
  1. Department of Nutrition and Feeding of Farm Animals, Institute of Animal Science, Uhříněves, 104 00 Prague 10, Czech Republic


Dietary Cr supplementation has potential to decrease fat and increase lean in carcasses of growing-finishing swine. However, effects of Cr supplementation on performance and economically important carcass and meat quality characteristics varied considerably among studies. Therefore, a meta-analysis was designed to quantitatively describe effects obtained in several independent studies. To accommodate differences in methodology among studies, standardized effect sizes (Hedges’s g) were calculated for results from 31 studies, in which Cr was supplemented as complexes of Cr Met chelate, Cr nanocomposite, Cr nicotinate, Cr propionate, Cr tripicolinate, or Cr yeast in diets for growing-finishing swine. Summary statistics were calculated by frequentist fixed and random effects, and hierarchical Bayesian models. With characteristics related to carcass quality, observed heterogeneity (P < 0.10) could not adequately be explained in a meta-regression by differences in initial BW and amount of Cr supplemented. Random effects and Bayesian models to summarize effect sizes for these characteristics showed similar results. According to random effects models, dietary Cr supplementation decreased (P < 0.05) 10th-rib fat thickness (mean effect size = −0.479; 95% confidence intervals = −0.680 to −0.279; 24 studies; 59 comparisons), whereas percentage carcass lean (mean effect size = 0.614; 95% confidence intervals = 0.366 to 0.863; 22 studies; 52 comparisons) and LM area (mean effect size = 0.571; 95% confidence intervals = 0.364 to 0.778; 29 studies; 72 comparisons) increased. Average daily gain and G:F, which did not present heterogeneity, were improved by Cr supplementation, whereas no effects were detected in characteristics (CIE color, drip loss, cook loss, shear force) related to meat quality. Some publication, or other small-study bias, was evident in results on growth and feed efficiency. However, directions of mean effect sizes were not changed by application of the trim-and-fill method to correct for bias.


Evidence has been presented that dietary Cr supplementation may improve growth rate, carcass characteristics, and reproductive performance of swine under some circumstances. However, responses have been inconsistent among studies (NRC, 1997; Jackson et al., 2009). Chromium has the potential to increase carcass lean without an increase in dietary protein intake. Consequently, it would prevent the potential negative environmental impact of a greater excretion of nitrogen after greater protein intakes (Lindemann, 1999).

Chromium in nature mostly exists in the stable trivalent (Cr+3) form. The primary metabolic role of Cr is to potentiate the action of insulin by facilitating insulin binding to receptors at the cell wall (NRC, 1997). Although earlier research has associated Cr activity with its presence in an organometallic molecule called the glucose tolerance factor, recent research indicated that glucose tolerance factor activity is independent of a unique Cr compound, and it has been proposed that the glucose tolerance factor is simply a decomposition product of chromodulin (Pechova and Pavlata, 2007). The involvement of Cr in the uptake of AA by skeletal muscle (Evans and Bowman, 1992), and in metabolism of nucleic acids (Okada et al., 1989) and lipids (Riales and Albrink, 1981), are well recognized.

Although complexes of Cr chloride, Cr Met chelate, Cr nanocomposite, Cr nicotinate, Cr propionate, and Cr yeast have been included in diets for livestock, organic Cr tripicolinate has received most interest as a supplement in swine diets to increase carcass leanness (Wang et al., 2007). Chromium status of animals, the amounts of bioavailable Cr in the feed, and exposure to certain environmental stresses are some factors that might have an influence on the need of Cr supplementation of practical swine diets (NRC, 1997).

Meta-analysis procedures to combine data from independent studies for a quantitative analysis are particularly useful when there is a conflict in results reported and when small sample sizes limit statistical power of individual studies to detect differences (Carriquiry et al., 2008). The current study was conducted to quantitatively describe the effects of Cr supplementation of growing-finishing swine diets on growth performance, carcass characteristics, and meat quality.


Animal Care and Use Committee approval was not obtained for this study because the data were obtained from an existing database.


An electronic search was used to identify English-language papers and abstracts that have evaluated the effects of dietary Cr supplementation on performance, carcass characteristics, and meat quality of growing-finishing swine at slaughter. Further search and reviews were conducted based on citations in those papers. For inclusion in the database, studies had to include 1) a control treatment without any dietary Cr supplementation, 2) a measurement of variation, and 3) did not restrict any nutrient in the basal diet. Because of the known poor availability of inorganic Cr chloride (Ohh and Lee, 2005), results obtained with inorganic Cr chloride as a supplement were not included in the database.

Based on the aforementioned criteria, 31 studies (Appendix Table A1) were identified and included in the database. Results from abstracts were included if they could be supported by a complete published report or tabulated values. A minimum of 6 studies was arbitrarily set as a requirement for any response variable to be evaluated in the meta-analysis. According to this requirement, performance could be described by ADG (kg), ADFI (kg), and G:F. Dressing percentage (DP), percentage carcass lean (PCL), LM area (cm2), percentage carcass fat (PCF), 10th-rib fat thickness (TRFT, cm), and average backfat thickness (BFT, cm) were response variables related to carcass characteristics. Meat quality could be identified with CIE L* (reflectance), a* (redness), and b* (yellowness), drip loss (DL, %), cook loss (CL, %), and shear force (SF, kg).

In all studies, corn- and soybean-based diets were used. It was indicated in 9 studies that Lys in excess of NRC (1988, 1998) requirements (105 to 125%) was provided. With the exception of studies by Wang and Xu (2004) and Wang et al. (2007, 2009), all swine received Cr supplementation during the growing and finishing phases.

Some studies have included different independent experiments. Others have evaluated dietary Cr supplementation at different amounts or with different sources in the same study. Similarly, breed and dietary Lys have been included as variables in some studies (Appendix Table A1). In these studies, individual treatment means were used for comparison to the control. O’Quinn et al. (1998) have reported separate mean values for barrows and gilts because swine were fed different diets to more accurately meet their Lys requirements. To prevent possible bias caused by an excessive amount of values from a single study, these values were averaged. The variability that was removed from their SD by the dividing factor was recovered according to DeCoster (2004) to get a pooled SD. In studies where feed efficiency was reported as feed:gain, the values were changed to G:F.


With different experimental methodologies among studies and different methods used to determine characteristics such as PCL, PCF, BFT, DL, CL, and SF, a standardized effect size (Hedges’s g) was applied within studies to identify differences caused by dietary Cr supplementation. This unitless effect size (ES) estimates the difference between the means for control and Cr-supplemented groups divided by the pooled SD, and is corrected for bias with small sample sizes (Hedges and Olkin, 1985). Replicates were used for sample size, except when indicated that individual animal data were used to calculate treatment means in studies.

Fixed effects, random effects, and hierarchical Bayesian models were used to calculate mean effect sizes across studies. Under the general fixed effects model, it is assumed that there is 1 true effect that underlies all the studies in the analysis, with all differences in observed effects because of within-study variability (sampling error):where Ti is the effect of the ith study; θ is the true effect; and εi is the error term for the ith study.

By contrast, a random effects model allowed that the true effect could vary from study to study and included between-study variability (true heterogeneity) as well as sampling error (Borenstein et al., 2009):where μi is the random effect of the ith study.

For fixed effects models, summary statistics were calculated by inverse variance methods (Hedges and Olkin, 1985), whereas mean effect sizes and their related precision in random effects models were estimated by the method of moments (DerSimonian and Laird, 1986). Calculations, as described in detail by Sales (2011), were done with Comprehensive Meta-Analysis software (Biostat, Engelwood, NJ).

In addition, to reflect the uncertainty in the estimates of the and the variability among studies, a hierarchical Bayesian approach was applied (Normand, 1999; WinBUGS, Imperial College and Medical Research Council, London, UK), which implements Markov chain Monte Carlo sampling (Lunn et al., 2000). Noninformative priors for the population mean and variance were assumed. In total, 60,000 iterations were used for parameter estimation, of which the first 1,000 were discarded as burn-in (initial random sequence or chain). Diagnostics (Winbugs) were used to access satisfactory convergence. Significance from 0 (2-tailed) was declared when 95% confidence intervals (CI) for random models and 95% credible intervals for Bayesian models did not include zero.


The presence of true heterogeneity among studies in fixed effects models was identified with Cochran’s Q-tests (Borenstein et al., 2009), and the degree of heterogeneity was quantified with the I 2-index proposed by Higgins and Thompson (2002), as detailed by Sales (2011). Where significant (P < 0.10) heterogeneity was detected, fixed weighted regression analysis with GLM (SAS Inst. Inc., Cary, NC) was done to identify the contribution of explanatory continuous variables to variability. Standard errors and associated probabilities of regression coefficients were corrected as described by Lipsey and Wilson (2000).

Publication Bias

Funnel plots, Begg’s rank correlation (Begg and Mazumdar, 1994), and Egger’s linear regression asymmetry (Egger et al., 1997) 2-tailed tests (Comprehensive Meta-Analysis) were used to evaluate publication bias. Statistical significance was declared at P < 0.10 (2-tailed). The trim-and-fill method (Duval and Tweedie, 2000) was applied to estimate the number of potentially missing studies and recalculate summary statistics with estimated missing values included.



Mean effect sizes calculated according to fixed effects models did not present heterogeneity for ADG, ADFI, and G:F. This was illustrated by nonsignificant Q-values, I 2-values of less than 16% (Table 1), and forest plots in which most studies included 0 in their CI (illustrated for ADG in Figure 1a). The absence of heterogeneity in these response variables was also accentuated by values comparable with those obtained with fixed effects models when and their precision were calculated according to frequentist random and hierarchical Bayesian models (Table 2).

Figure 1.

Forest plots of the effects (Hedges’s g) obtained in fixed effects models for differences in a) ADG and b) percentage carcass lean when Cr-supplemented (B) and nonsupplemented (A) diets were fed to growing-finishing swine. The size of the squares illustrated the weight of each study relative to the mean effect size, which is indicated by the diamond at the bottom. CI = 95% confidence interval. Color version available in the online PDF.


With ES categorized by Cohen (1988) as small, medium, and large at values of 0.2, 0.5, and 0.8, respectively, Cr supplementation had a small positive effect (P < 0.05) on ADG (Table 1). The obtained for G:F was more profound than for ADG and could not be related to ADFI, in which no effect was found.

In a summary of 31 studies by the NRC (1997), ADG in growing-finishing swine was improved (P ≤ 0.10) in 11, and feed efficiency in 8, by dietary Cr supplementation from Cr picolinate at 200 to 500 μg of Cr/kg. Based on feed efficiency obtained in studies, in which diets were either formulated to 100% of NRC (1998) Lys requirements or in excess of requirements, Lindemann (1999) concluded that there is apparently no need for greater dietary protein when Cr is supplemented. This assumption was supported by Ward et al. (1997) and Lien et al. (1998). The involvement of Cr in RNA and DNA synthesis (Ward et al., 1997) or an increase in AA uptake when diets are underformulated or feed intake is restricted relative to protein synthetic capabilities (Ward et al., 1997; Lindemann, 1999) might be responsible for greater responses. Unfortunately, several studies included in the current evaluation lacked information needed to evaluate this assumption. Furthermore, studies included in the current evaluation have been selected based on suggested optimal diet formulation to avoid any possible suboptimal nutrient intakes. This has excluded possible limitations in some nutrients, which would be needed to investigate the interaction between Cr supplementation and dietary protein.

Although both Begg’s and Egger’s tests did not detect any bias with ADG and G:F, 9 and 13 studies, respectively, were needed to regain funnel plot asymmetry according to the trim-and-fill method (Table 3; illustrated for G:F in Figure 2a). In contrast, ADFI showed bias according to the 2-tailed tests; however, only 1 study was needed to adjust values. Results obtained with the trim-and-fill method should not be used for final conclusions. Whereas funnel plot asymmetry could be attributed to factors other than publication bias, adjusted values should primarily be used as a form of sensitivity analysis to assess the potential effect of missing studies (Sutton et al., 2000).

Figure 2.

Funnel plots of the effects (Hedges’s g) obtained in fixed effects models for differences in a) G:F and b) CIE L* (reflectance) when Cr-supplemented and nonsupplemented diets were fed to growing-finishing swine against the SE of the estimates. An unequal number of studies on one side of the vertical line might be an indication of publication bias. Empty circles indicate observed values, and filled circles indicate possible missing values calculated according to the trim-and-fill method of Duval and Tweedie (2000). Mean effect sizes of observed values are presented by empty diamonds, whereas filled diamonds present the estimated mean effect sizes when potential missing values were included. Color version available in the online PDF.


Carcass Characteristics

Mean effect sizes for all response variables related to carcass quality showed (P < 0.10) heterogeneity greater than 28% (Table 1). Effect sizes that varied from study to study were also evident from forest plots, as demonstrated for PCL in Figure 1b. This considerable heterogeneity rendered the use of fixed effects models to calculate summary statistics inappropriate.

Analyses to determine the sensitivity of to any individual study by leaving 1 study out, calculating the of the remaining studies, and comparing the results with the based on all the studies (Normand, 1999) did not identify any influential studies. Source of Cr, breed, sex, phase when supplemented, type of housing, period of supplementation, and final BW, which could be used as independent variables to explore heterogeneity, were characterized by unbalanced data because of missing information. This could result in false conclusions of its effect on the total response variable spectrum. Therefore, only initial BW and Cr supplemented could be included as continuous variables in a meta-regression to explore possible sources of heterogeneity. With the exception of PCF, these explanatory variables could explain few of the heterogeneity obtained (Table 4). However, PCF included only 16 comparisons, and results should be treated with caution. The relation between the number of studies and number of explanatory variables to reach reliable conclusions with meta-regression is still unresolved (Thompson and Higgins, 2002).

In the current evaluation, ES for PCF and TRFT were negatively (−0.053 ± 0.010 and −0.014 ± 0.005, respectively) and LM area (0.012 ± 0.005) positively related (P < 0.10) to initial BW. This implied that fat decreased and lean increased in the carcass to a larger extent when Cr supplementation was started at a greater initial BW. This is in contrast to studies by Page et al. (1993) and Lindemann et al. (1995), where the magnitude of the increase in LM area and decrease in fat depth was greater with decreased initial BW of swine. However, effects in those studies were confounded by initial BW at the time of supplementation and the total length of the supplementation period (Lindemann, 1999). These factors could not be differentiated in the current meta-analysis. Boleman et al. (1995) found that dietary Cr supplementation from Cr picolinate improved linear carcass measurements when fed during the finishing period and not when included during the growing period.

Effect sizes decreased (P < 0.10) with increasing dietary Cr supplementation with LM area (−0.00013 ± 0.00006), PCF (−0.005 ± 0.003), and BFT (−0.002 ± 0.001). Although this indicated that fat was decreased more when amount of Cr inclusion was increased, it also showed that the positive effect of Cr on LM area was decreased. Few studies have included quantities other than 200 μg/kg of Cr, which is permitted as the maximum inclusion amount in the United States (Lindemann et al., 2008), in the same study. In the study of Shelton et al. (2003), different Cr quantities did not present any relation to BFT, whereas positive linear (Khajarern et al., 2006) and increased quadratic (Page et al., 1993; Lindemann et al., 1995) trends were observed between Cr and LM area in other studies. The absence of any relationships between Cr supplementation and ES obtained with TRFT and PCL, and a lack of information on the mode of action of Cr on nutrient distribution in the animal, prevented further speculation on the relationships.

As a result of unexplained heterogeneity, random effects models, which included between- and within-study variability, were needed to combine ES. The inclusion of the additional source of uncertainty (between-study variance) in a random effects model will always result in a wider CI, except when this source is zero (Borenstein et al., 2009). This widening effect was especially evident in PCL and PCF, in which a high degree of heterogeneity was detected. With PCF, application of a random effects model to calculate summary statistics resulted in a nonsignificant effect (0 included in the CI; Table 2), compared with a significant negative effect when a fixed effects model (Table 1) was used. Hierarchical Bayesian models presented similar results than frequentist random effects models (Table 2). Although there is a 95% chance that the true value of the parameter will be within the interval with a 95% Bayesian credible interval, a 95% frequentist CI, which is based on the concept of an indefinite number of samples, does not contain the true parameter with 95% probability. When intervals differ between frequentist random effects and hierarchical Bayesian models when uninformative priors are used in the Bayesian models, only the credible intervals provide logical results (McCarthy, 2007).

Both the random effects and Bayesian models presented the same tendencies: 1) dietary Cr supplementation had no effect on PCF and BFT; 2) it increased DP, PCL, and LM area; and 3) TRFT was decreased. A decrease in TRFT was not manifested in BFT. Mooney and Cromwell (1995) found that a decrease in accretion rate of dissected fat tissue was not associated with a decrease in backfat measurements. However, studies that have addressed the possibility of a redistribution of fat by Cr supplementation in swine, rather than an actual reduction in total fat, could not be sourced.

Increased insulin activity by Cr picolinate supplementation, which stimulates AA transport and protein synthesis in muscle cells, was suggested by Lien et al. (2001) as a possible reason for an increase in LM area. However, Wang et al. (2008) hypothesized that increased serum GH mean concentrations and peak values found in their study could be responsible for the effects of Cr picolinate on protein deposition, as indicated by an increased LM area. Kornegay et al. (1997) suggested that an increase in PCL in the absence of an increase in ADFI under restricted feeding could result from an improvement in dietary nitrogen utilization. However, when energy and nitrogen intake are adequate and lean deposition is maximized, improvements in digestibility could result in the additional energy stored as fat.

Improvements in insulin internalization and insulin sensitivity by Cr picolinate supplementation would be expected to increase fat deposition. However, because this was not supported by research, it was speculated that Cr must be tissue-specific, elicit its effects on carcass composition through various unknown mediators, or both (Boleman et al., 1995; Mooney and Cromwell, 1995). Increased lipolytic activity (Choi et al., 1998; Woodworth et al., 2007) and inhibited lipogenesis (Xi et al., 2001) have been suggested as further possible mechanisms for the action of Cr in swine.

Evaluation of publication bias, or other small-study effects, in the presence of heterogeneity is inappropriate and may lead to false-positive claims (Ioannidis and Trikalinos, 2007). Therefore, no testing for bias was performed with characteristics related to carcass quality.

Meat Quality

Significant heterogeneity was found with calculated for CIE b* and DL (Table 1). With 17 and 13 comparisons for CIE b* and DL, respectively, meta-regression with initial BW and quantity of Cr supplemented as explanatory continuous variables was considered too unstable for realistic conclusions. The use of random effects and hierarchical Bayesian models to summarize (Table 2) for these 2 response variables, however, did not change the findings from fixed effects models that dietary Cr supplementation had no effect. With results obtained through fixed effects models (Table 1), CIE L* and a*, CL and SF were not affected (P > 0.05) by dietary Cr inclusion. Essentially, no evidence of any possible publication, or other small-study bias, was detected in response variables that did not present heterogeneity (Table 3). This is illustrated with ES obtained with CIE L* in Figure 2b.

Research on the effects of dietary Cr supplementation on pork quality is more limited than research on growth and carcass composition, but it is equally variable (Matthews et al., 2005). Matthews et al. (2003) suggested that improved pork quality in swine fed Cr might be attributed to either decreased stress or a change in glycolytic potential. However, results combined from several studies in the current evaluation did not reveal any differences because of Cr supplementation in the response variables investigated.


This quantitative description of results obtained with several studies presented conclusive evidence that Cr supplementation of diets for growing-finishing swine decreases fat and increases lean deposition in the carcass. Supplementation improved G:F, but had no effect on characteristics related to meat quality. Baseline values are presented for the magnitude of effect sizes, which could be used in further studies to evaluate the beneficial effects obtained with Cr inclusion in swine diets. However, to investigate the sources of variability that caused effect sizes to vary from study to study, consistency in variables and detailed reporting of variables will be needed in future studies. A secondary result from this study is that frequentist random effects and hierarchical Bayesian models with uninformative priors presented similar results when used to combine effect sizes.