IntroductionTop
Body mass index (BMI) is the most common objective measure used to classify excess adiposity in human beings (Lau et al., 2007). Although BMI has also been applied in horses and ponies (Donaldson et al., 2004; Carter et al., 2009a; Thatcher et al., 2012; Banse & McFarlane, 2014), the main system to assess body condition in horses is based on assigning a subjective body condition score (BCS). This method consists in evaluating the deposition of subcutaneous fat in specific body regions and the subsequent assignment of a score considering established criteria through a palpation and visual assessment (Carter & Dugdale, 2013). Even though it has become a universally accepted method to estimate the degree of fatness (Dugdale et al., 2012), BCS possesses well-known limitations at the individual level as occurs with the BMI, including the inability of both systems to directly distinguish between lean and fat tissue (Frankenfield et al., 2001; Geor & Harris, 2013). Therefore, with the same BCS, substantial variation in adiposity can occur (Dugdale et al., 2011a). Furthermore, in the case of BCS, its inherent subjectivity and, thus, the semi-quantitative nature of this evaluation, lead to the belief that these scoring systems are unreliable and are necessary clinically more applicable and useful subdivisions to differentiate horses with higher scores (Burkholder, 2000; Dugdale et al., 2012).
Considering that the limitations are common not only in horses but also in all animal species where BCS has been utilized, several studies have developed alternative methods to differentiate body condition using objective criteria (Bewley et al., 2008; Azzaro et al., 2011; Halachmi et al., 2013; Das & Paksoy, 2015; Pfeifer et al., 2017). However, some of them require specialised equipment, software and personnel to interpret the results making them less applicable in clinical settings. Moreover, most of them have been implemented only for their use in cattle (Bewley et al., 2008; Azzaro et al., 2011; Halachmi et al., 2013; Das & Paksoy, 2015; Pfeifer et al., 2017) and therefore, the necessity to find a suitable method to objectively evaluate the body condition in horses is still lacking. Against this perspective, taking into account that body weight alone is not a good indicator of relative adiposity (Carter & Dugdale, 2013), a good assessment of body fat reserves, minimizing the influence of body dimensions and intestinal contents, can be achieved by evaluating ultrasonographically the amount of subcutaneous fat.
Therefore, the aims of this study were: 1) to develop an alternative body scoring index based on the ultrasonographic evaluation of localized subcutaneous fat deposits; 2) to assess the agreement between BCS and the new index applied to Andalusian horses; 3) to adjust the BCS cut-off values (if necessary) for overweight and obesity in this breed.
MethodsTop
Animals
From a population of over 1,500 Andalusian horses located in south-eastern Spain (comprising the provinces of Albacete, Alicante and Murcia), 166 (6.7 ± 3.7 years, 78 stallions and 88 females) were utilized in this cross sectional study for the development of a new body fat scoring index (BFSI) built on the application of ultrasonographic measurements. The sample was randomly selected in order to ensure a sufficient number of horses from both genders representing their score characteristics, and hence testing the reliability of this method.
Body and fat measurements
Using the nine point scale described by Henneke et al. (1983), two independent and trained evaluators assigned the scores on each horse and the average value was used as final BCS. Considering previous descriptions (Thatcher et al., 2012) but also the phenotypic characteristics of the Andalusian horses, the following four body categories were established: thin horses (if BCS = 4.5), normal body condition (if BCS 5- 6.5), overweight (if BCS 7-7.5) and obese horses (if BCS = 8).
Afterwards, the amount of fat reserves were measured evaluating the subcutaneous fat thickness (SFT) by real-time ultrasound at seven localized fat deposits which included: three equidistant areas along the neck crest (SFT-N25%, SFT-N50%, SFT-N75%), behind the shoulder (SFT-S), the ribs (SFT-Rb), the rump (SFT-R) and the tailhead (SFT-TH) as previously described (Martin-Gimenez et al., 2016). Ultrasonography was carried out using a portable Honda Electronics HS-1500V (Aichi, Japan) ultrasound device in B-mode with a linear transducer at 7.5 MHz frequency. BCS evaluation and the ultrasonographic measurements were taken for each horse on the same day. Measurements were obtained by freezing the image on the screen and measuring the position of maximal fat thickness. All measurements of SFT were performed in triplicate by the same researcher. To assess the reliability of repeated measurements, the intraclass correlation coefficients were calculated showing a significant repeatability (p<0.001) of 96.8% for SFT-N25%, 95.6% for SFT-N50%, 97.2% for SFT-N75%, 98.6% for SFT-S, 98.6% for SFT-Rb, 99.4% for SFT-R and 99.6% for SFT-TH. Because the agreement between different measurements was good (Fleiss, 1986) mean values of the three measurements were used for statistical analyses.
Body fat percentage (BF%) was also calculated to assess the new BFSI as monitor parameter of body fat stores. This variable was estimated from the equation of Kane et al. (1987) where: BF% = 2.47 + 5.47 * (rump fat in cm). The site to measure rump fat was determined by placing the probe over the rump at approximately 5 cm lateral from the midline at the centre of the pelvic bone (Westervelt et al., 1976).
All the measurements and body condition estimations were collected following informed consent from the owners.
Statistical analysis
Data were expressed as mean ± standard deviation (SD) or percentage, as appropriate. Normality of quantitative variables was checked using the Kolmogorov-Smirnov test.
To test the usefulness of SFT measurement technique, two different approaches were performed: 1) analysis of variance (ANOVA) to determine the relationship among the different scores included in the BCS system and the SFT values and; 2) association between BCS, BF% and SFTs using the Pearson’s and Spearman’s correlation coefficients. Besides, Student’s t test for independent samples was used to evaluate the association among the gender and the SFTs (Daniel, 2000).
Construction of the new body fat scoring index
Mean and SD of each SFT measurement were calculated. Depending on the number of SDs from the average SFT value, a standardized score (equal to the integer part of the standardized residual) was assigned to every SFT on each horse. The difference between an individual SFT value and the mean divided by the SD corresponds to the standardized residual (Daniel, 2000). To simplify these calculations and taking into account the mean and SD, different intervals were established attributing to them the following standardized scores: -2 (- 8, mean – 2*SD], -1 (mean – 2*SD, mean – SD], 0 (mean – SD, mean + SD]), +1 [mean + SD, mean + 2*SD), +2 [mean + 2*SD, mean + 3*SD) and +3 [mean + 3*SD, + 8). The overall objective score of a horse resulted from the sum of all scores obtained at each anatomical area. Differences between genders in overall objective scores were assessed using Student’s t test or Mann-Whitney test depending on normality. Mean and SD of the overall objective scores and their intervals, similarly to what was done with each of the SFT measurements, were estimated defining five possible BFSI categories. To analyse the association between BCS and the BFSI, Pearson’s Chi-square test, Spearman’s correlation and Cohen’s kappa coefficients were calculated (Cohen, 1968; Daniel, 2000; Thrusfield, 2005).
Spearman’s correlations were used to evaluate the association between BF%, BCS and BFSI. To determine changes in BF% across the BCS and BFSI categories, an ANOVA was performed. Duncan post hoc test was used to separate between significant means.
The reliability of the BCS to distinguish between horses that did and did not exhibit an overweight or obesity state defined by the new BFSI, was estimated by calculating the area under the curve (AUC). The coordinates of the receiver operating characteristic curve (ROC) were used to set the cut-off values that maximised the accuracy (proportion of true results) (Greiner et al., 2000). Their confidence intervals (CI) were calculated using the Wilson’s score method (Wilson, 1927).
The analyses were carried out with the statistical software program IBM SPSS for Windows Vers. 19, except for the calculation of weighted Cohen’s kappa coefficient that was used StatsToDo (https://www.statstodo.com/CohenFleissKappa_Pgm.php). Values of p<0.05 were considered significant.
ResultsTop
The global mean of BCS was 6.12 ± 1.05 without significant differences between both genders (p=0.695). Agreement between the two body condition evaluators was moderate (Cohen’s kappa weighted =0.493, CI95%: 0.423, 0.564; p<0.001). Mean values ± SD of the seven SFT measurements are described in Table 1. Regard to the gender, males presented significantly higher values at four SFT measurements (SFT-N25%, SFT-N50%, SFT-S and SFT-Rb) than females (Table 1). Subcutaneous fat thicknesses (with the exception of SFT-N25% and SFT-N50%) were significantly correlated with BCS and BF% (Table 1). Likewise, most of SFT measurements were significantly different depending on the BCSs assigned in the sample (Table 1), increasing their values as the BCS does, despite the low correlations (Fig. 1).
Table 1. Global mean ± SD values of subcutaneous fat thickness (SFT) and their relationship with gender, body condition score (BCS) and body fat percentage (BF%).
|
|
Due to significant differences in SFT values were observed according to the gender, firstly different sets of intervals had to be created stratified by sex to define the standardized scores corresponding to each fat deposit (Table 2). Secondly, the overall objective score of a horse was calculated adding up the seven standardized scores obtained. Global mean of the new BFSI was 0.265 ± 2.697 being similar between males and females (p=0.910). Based on the absence of differences between genders after standardizing the scores, it was possible to define the final scores common to both genders. In this instance, five body categories (emaciated, thin, normal, overweight and obese) were proposed so that an animal with an overall standardized score equal to or greater than 6 was considered as obese, while an animal with a score between 3 and 5 (both inclusive) was classified as overweight (Table 3).
Table 2. Standardized scores and corresponding intervals for each subcutaneous fat thickness (SFT) measurement according to the gender
Table 3. Body condition classification using the new body fat scoring index (BFSI)
Table 3. Body condition classification using the new body fat scoring index (BFSI)
The application of the BCS system showed that the majority of the samples were distributed among the interval scores of 5 and 6.5 (63.9%) and 7-7.5 (22.9%). In addition, 8.4% and 4.8% were classified as thin horses (BCS = 4.5) and obese (BCS = 8) respectively. Concerning the BFSI, only one horse was classified as emaciated (0.6%) (because of its low representativeness, it was excluded for further calculation of agreement). Most of the horses (74.7%) had a normal body condition and 11.4% were overweight. Fifteen horses (9.0%) were considered as thin horses while seven (4.2%) were categorized as obese (Table 4 & Fig. 2).
Table 4. Association among the body condition score (BCS) and the new body fat scoring index (BFSI) categories.
|
|
The correlation between BCS and the BFSI was significant and moderate attending to the Spearman’s correlation coefficient value (rho=0.428; p<0.001). In addition, the association among the BCS and the BFSI categories was highly significant (p<0.001), however the concordance between both body scoring methods was very low, evidenced by a weighted Cohen’s kappa coefficient of 0.262 ± 0.071 (p=0.004) (Table 4). In this manner, it could be observed as remarkable results, that half of horses with BCS = 8 were classified as having a normal body condition with the BFSI, while the remaining 50% were equally distributed among overweight and obese. Similarly, in the case of the horses with BCS = 7-7.5, more than half (65.8%) were categorized as normal with the BFSI, 23.7% as overweight and 10.5% as obese (Table 4).
The global BF% was 10.36 ± 2.94 without significant differences between both genders (p=0.126). Otherwise, BF% was significantly correlated with BCS (rho=0.491; p<0.001) and BFSI (rho=0.503; p<0.001). Depending on the BCS categorization, significant differences (p<0.001) were found among thin (7.74 ± 2.89) vs normal horses (9.82 ± 2.18) and, these two categories vs overweight (12.10 ± 2.73) and obese (13.84 ± 5.42). However, no differences were observed between overweight and obese horses. On the contrary when BFSI was applied, obese horses (16.03 ± 4.45) presented higher BF% values (p<0.001) respect to overweight horses (12.50 ± 2.78), as well as these two categories respect to normal (9.93 ± 2.36) and thin horses (8.84 ± 2.71). In this case, BF% was similar between thin and normal horses.
Diagnostic accuracy of BCS to distinguish overweight or obese horses from all other horses was assessed by evaluation of areas under the ROC curves. For the first case, BCS had an AUC =0.759 ± 0.055 (p<0.001). In the second case, BCS presented an AUC = 0.878 ± 0.050 (p=0.001) (Fig. 3). The ROC curve analysis was also employed to determine the BCS cut-off values for detecting overweight (BFSI = [3, 5]) and obese (BFSI =6) horses, and these values were 7.5 with an accuracy of 85.54% (CI95%: 79.39%, 90.09%) and 8.5 with an accuracy of 96.99% (CI95%: 93.14%, 98.71%), respectively.
|
|