Accepted Manuscript

Title: Measuring body composition in dogs using multifrequency bioelectrical

impedance analysis and dual energy x-ray absorptiometry

Author: L.S. Rae, D.M. Vankan, J.S. Rand, E.A. Flickinger, L.C. Ward

PII: S1090-0233(16)30029-6

DOI: http://dx.doi.org/doi: 10.1016/j.tvjl.2016.04.007

Reference: YTVJL 4800

To appear in: The Veterinary Journal

Accepted date: 11-4-2016

Please cite this article as: L.S. Rae, D.M. Vankan, J.S. Rand, E.A. Flickinger, L.C. Ward,

Measuring body composition in dogs using multifrequency bioelectrical impedance analysis and

dual energy x-ray absorptiometry, The Veterinary Journal (2016), http://dx.doi.org/doi:

10.1016/j.tvjl.2016.04.007.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service

to our customers we are providing this early version of the manuscript. The manuscript will

undergo copyediting, typesetting, and review of the resulting proof before it is published in its

final form. Please note that during the production process errors may be discovered which could

affect the content, and all legal disclaimers that apply to the journal pertain.

1

Measuring body composition in dogs using multifrequency bioelectrical impedance analysis 1

and dual energy X-ray absorptiometry 2 3

L.S. Rae a, D.M. Vankan

a, J.S. Rand

a, E.A. Flickinger

b, L.C. Ward

c,* 4

5 a School of Veterinary Science, The University of Queensland, Gatton, Queensland 4343 6

Australia 7 b Procter and Gamble Pet Care, FEI Products Research, Mason, OH 45040 USA 8

c School of Chemistry and Molecular Bioscience, The University of Queensland, St Lucia, 9

Queensland 4072 Australia 10

11 * Corresponding author. Tel.: +61 7 3365 4633. 12

E-mail address: l.ward@uq.edu.au (L.C. Ward). 13

14

Highlights 15

16

Body composition of dogs was measured using multifrequency bioimpedance. 17

Reference body composition was measured by dual-energy X-ray absorptiometry.. 18

When cross-validated bioimpedance predicted mean fat-free mass within 1.5% of measured 19

values. 20

Abstract 21

Thirty-five healthy, neutered, mixed breed dogs were used to determine the ability of 22

multifrequency bioelectrical impedance analysis (MFBIA) to predict accurately fat-free mass 23

(FFM) in dogs, using dual energy X-ray absorptiometry (DXA)-measured FFM as reference. A 24

second aim was to compare MFBIA predictions with morphometric predictions. 25

26

MFBIA-based predictors provided an accurate measure of FFM, within 1.5% when 27

compared to DXA-derived FFM, in normal weight dogs. FFM estimates were most highly 28

correlated with DXA-measured FFM when the prediction equation included resistance quotient, 29

bodyweight, and body condition score. At the population level, the inclusion of impedance as a 30

Page 1 of 22

2

predictor variable did not add substantially to the predictive power achieved with morphometric 31

variables alone; in individual dogs, impedance predictors were more valuable than morphometric 32

predictors. These results indicate that, following further validation, MFBIA could provide a 33

useful tool in clinical practice to objectively measure FFM in canine patients and help improve 34

compliance with prevention and treatment programs for obesity in dogs. 35

36

Keywords: Dog; Bioelectrical impedance; DXA; Body composition; Fat-free mass; Obesity 37

38

Introduction 39

Excess body fat is the most common nutritional disorder of dogs in Western countries, 40

with an estimated prevalence 33 to 44% or higher (German, 2006; Gossellin et al., 2007; Zoran, 41

2010; Laflamme, 2012). Obesity is known to induce insulin resistance, oxidative stress and a 42

chronic, low-grade inflammatory state thought to contribute to the development of osteoarthritis 43

and other diseases (Zoran, 2010; Laflamme, 2012), or osteoarthritis and reduced lifespan (Kealy 44

et al., 2002). A moderately high fat diet has been shown to increase visceral fat two-fold in dogs, 45

with minimal increases in bodyweight (BW; Kim et al., 2003) conducive to the development of 46

insulin resistance; insulin resistance increases with adiposity, even if BW is stable. Prevention of 47

obesity is more effective than its subsequent treatment and is best instituted while animals are 48

just beginning to gain weight (Zoran, 2010; La Flamme, 2012), yet veterinarians often neglect to 49

formally diagnose and discuss an increase in BW (Lund et al., 2005). Once obesity is established 50

it is much more difficult to implement successful weight loss strategies (Gossellin et al 2007; 51

Zoran 2010). For these reasons, some authors have stressed the importance of assessing adiposity 52

Page 2 of 22

3

per se rather than simply BW (LaFlamme, 2012), as the latter does not necessarily reflect body 53

fat content (Stanton et al., 1992). 54

55

Body fat can be accurately measured by various methods (Heymsfield et al., 2005; 56

Gossellin et al., 2007), although many of these, e.g. computed tomography (Purushothaman et 57

al., 2013), quantitative magnetic resonance imaging and dual- energy X ray absorptiometry 58

(DXA; Zanghi et al., 2013) require specialised equipment and/or general anaesthesia, and are not 59

practical or available for many research and clinical applications. Body condition scoring 60

(BCS), morphometric measurements, and bioelectrical impedance analysis (BIA) offer non-61

invasive, practical methods for estimating body composition. BCS, using a validated 62

methodology, offers a semi-quantitative assessment that correlates well with percent body fat 63

(Mawby et al., 2004; Shoveller et al., 2014), but is a somewhat subjective measure, relying on 64

visual appraisal and palpation that requires some level of training for competency. BIA, by 65

contrast, is an objective technique that measures the electrical resistance of body water (TBW) 66

that relates directly to the fat-free mass (FFM) of the body (Stone et al., 2009; German et al., 67

2010). 68

69

This study aimed to evaluate the ability of multifrequency bioelectrical impedance 70

analysis (MFBIA), BCS scoring and morphometric measures to predict FFM in dogs compared 71

with FFM measured by DXA. 72

73

Materials and methods 74

Dogs 75

Page 3 of 22

4

Thirty-five neutered (18 male, 17 female) mixed breed research dogs were used. All 76

dogs were clinically healthy based on history, physical examination findings, recent 77

haematological and biochemical blood analyses and were determined to be free from internal 78

parasites by faecal analysis after deworming. The dogs’ ages were unknown but, based on 79

physical characteristics and dentition, all were estimated to be > 7 months of age. Bodyweights 80

were 12.1 - 43.0 kg and BCS was 3 - 4.25 (average of two assessors) on a scale of 1 - 5 (1, 81

emaciated; 5, obese; Table 1). 82

83

The study was performed according to The University of Queensland Animal Ethics 84

Committee’s Policies and Guidelines and study protocols were approved by The University of 85

Queensland Animal Ethics Committee SVS/307/04, approved 1st July 2004 and annually until 86

16th

July 2007. 87

88

Bodyweight and body condition score 89

Bodyweight was measured (to the nearest 10 g using a veterinary scale (SK-Vet-150; 90

Accuweigh) and BCS was determined by two experienced research technicians following 91

standardised assessment protocols (McGreevy et al., 2005). BCS was graded into 0.5 increments 92

as proposed by Baldwin et al. (2010). BCS data were averaged for the two assessors; individual 93

ratings did not vary more than 0.5 unit. BCS has been validated against DXA for assessment of 94

body composition in dogs (LaFlamme, 1997). 95

96

Morphometric measurement – length 97

Page 4 of 22

5

In order to generate MFBIA-based prediction equations for body composition, a 98

measurement of the current path is required. Since the precise path is unknown, a surrogate 99

measurement is used, e.g. height in humans (Foster and Lukaski, 1996), or simple linear 100

measurement between the sense electrodes in animals (Ward and Battersby, 2009). After dogs 101

were sedated and placed in left lateral recumbency, body length to the nearest mm was measured 102

from the middle of the right eye to the anus using a flexible tape measure. 103

104

Page 5 of 22

6

105

Dual energy X-ray absorptiometry and multifrequency bioelectrical impedance analysis 106

Dogs were fasted for at least 20 h prior to DXA followed by MFBIA measurements, 107

performed on the same day. Dogs were sedated with SC methadone (0.3 mg/kg; Methone 108

Injection, 10 mg/mL, Ceva Animal Health) and acepromazine (0.03 mg/kg; ACP2, 2 mg/mL, 109

Delvet) 30 min prior to anaesthesia with IV alfaxalone (1-2 mg/kg; Alfaxan CD-RTU, 10mg/mL, 110

Jurox). 111

112

Dogs were scanned (Hologic QDR-4500A) and scans were analysed using 113

manufacturer’s software (Hologic). Dogs were positioned in a standardized fashion, aided by 114

gridlines on the scanner bed, in dorsal recumbency with the head extended, forelegs bent and 115

taped away from the body and the hind legs extended. A single scan (2-3 min duration) was 116

performed by an experienced DXA technician. Tissue quantification was achieved by measuring 117

the differential attenuation by lean, fat and bone mineral of two X-ray beams of different energy 118

levels to provide measurements of whole body lean mass, fat mass (FM) and bone mineral mass 119

(Heymsfield et al., 2005). FFM was calculated as the sum of lean and bone mineral content 120

(BMC). FFM determined by DXA was comparable with that determined from measurement of 121

TBW by tracer dilution (Heymsfield et al., 2005). 122

123

Whole body impedance was measured using a tetrapolar multifrequency bioimpedance 124

spectrometer (SFB7, ImpVet, ImpediMed), which measured resistance (R) and reactance (Xc) at 125

256 frequencies from 3 - 1000 kHz at a constant drive current of 200 µA. Ag-AgCl gel EKG-126

style (24 x 22 mm) skin electrodes (ImpediMed) were used. Hair at the electrode site was clipped 127

Page 6 of 22

7

closely to the skin and cleaned with an alcohol wipe. Based on preliminary reproducibility and 128

reliability studies, the following electrode locations were used: voltage sense electrodes were 129

placed cranially at the right stifle and right elbow with current drive electrodes 10 cm distal, 130

similar to the protocol used in other studies (Scheltinga et al., 1991). Measurement time was <1 s 131

and data (10 consecutive readings) were downloaded to a computer for analysis. 132

133

Multifrequency bioelectrical impedance analysis theory and data analysis 134

MFBIA data were uploaded to a computer and analysed (Bioimp software, version 135

4.15.0.0, ImpediMed). The software fitted the recorded resistance and reactance data to a semi-136

circular plot of resistance vs. reactance, after the Cole model of biological impedance (Thomas et 137

al., 1998) that represents the body as a resistor representing the extracellular water (ECW), in 138

parallel with a resistor representing intracellular water and a capacitor representing the cell 139

membranes. According to this circuit model, the resistance measured at infinite frequency, or 140

other high frequency, (Kyle et al., 2004; McGree et al., 2007), is that of the overall conductive 141

volume, i.e. TBW, while the resistance at zero frequency is that of the ECW (Cornish et al., 142

1993). The impedance at the frequency of maximal reactance, the characteristic frequency or fc, 143

had special significance, since at fc current flow is dependent only on the resistances of the water 144

compartments and not on membrane capacitance. Hence the impedance (Zc) at fc should also be 145

an appropriate frequency from which to predict TBW (Cornish et al., 1996). 146

147

TBW volume was related to impedance or resistance and length according to the 148

following equation: 149

Page 7 of 22

8

where TBW is the volume of TBW, is the specific resistivity of ohm.cm , L is 150

conductive length (cm) and Z or R is impedance or resistance (ohm), respectively (Thomas et al., 151

1998). FFM was readily obtained from TBW by dividing by the hydration constant of FFM, 152

assumed to be 0.732 (Schoeller, 1996); consequently FFM could be substituted for TBW in the 153

above equation. 154

155

Statistical analysis 156

Descriptive data are presented as mean ± standard deviation (SD) with group differences 157

assessed using Students t test following normality testing D’Agostino-Pearson test; MedCalc, 158

version 12.7.0, MedCalc Software). Body composition prediction equations were produced using 159

multiple linear regression techniques (Zar, 1999), using a backward stepwise method (MedCalc, 160

version 12.7.0, MedCalc Software); FFM by DXA was the dependent variable. Independent 161

variables examined were sex, BW, BCS, length (L), and the impedance indices, R50 index 162

(L2/R50); R500 index (L

2/R500); R∞ index (L

2/R∞); Zc index (L

2/Zc). The coefficient of 163

determination adjusted for multiple independent variables (r2 adjusted) and the root mean square 164

error (RMSE) were determined with alpha level of significance set at 0.05. 165

166

A ‘split-group’ cross-validation procedure was used in which prediction equations were 167

generated in a randomly selected by sex ‘prediction’ group (12 males and 11 females). These 168

equations were then used to predict FFM in the remaining one-third of the population (six 169

females and six males), the ‘validation’ group. The validation and prediction groups were not 170

significantly different (P > 0.05) in any characteristics. Predicted FFM was compared to that 171

measured by DXA using the concordance correlation coefficient, rc (Lin, 1989), Pearson 172

Page 8 of 22

9

correlation coefficient, rp (Zar, 1999) and agreement between the two methods was assessed 173

using limits of agreement (LOA) analysis (Bland and Altman, 1986). 174

175

Results 176

Dogs enrolled in this study (Table 1) ranged from small to large breed crosses of varying 177

lengths (61 - 98 cm) and weights (12.1 - 43 kg). BCSs, however, were more uniform, with 89% 178

of dogs in ideal condition (BCS 3), 11% classified as overweight (BCS 4) and none classified as 179

obese or underweight (Table 1). Male animals were significantly heavier and had significantly 180

greater lean (P < 0.05), BMC (P < 0.001) and FFM (P < 0.05) than female animals. FM was 181

30% greater in males than females, although this difference was not significant (P = 0.124), 182

reflecting the difference in BW. When expressed as %BW, %FM was 17.7 ± 4.4% and 17.3 ± 183

4.3% in males and females, respectively. 184

185

Fat-free mass (FFM) prediction equations. 186

Seven equations to predict FFM were generated by the regression analyses (Tables 2 and 187

3). Preliminary analyses showed that sex was not a significant predictor in any analysis; 188

therefore, it was removed from all equations. The resistance indices, L2/Zc and L

2/R50 were also 189

not significant, but because L2/R50 was approaching significance (P = 0.057) an equation was 190

generated using this index to provide a point of comparison with previously published studies. 191

An equation was also generated to include only the most significant BIA variable, L2/R∞, and the 192

most significant morphometric measure, BW. Therefore, equations were generated using three 193

BIA resistance indices: L2/R50, L

2/R500 and L

2/R∞. As BW was the most highly significant of all 194

Page 9 of 22

10

variables, another equation was generated using only BW and BW combined with the other 195

morphometric variables. 196

197

Testing the predictor in the validation group of dogs 198

Predicted FFM values were calculated from the equations for every dog in the validation 199

group and compared with the FFM measured by DXA. Concordance and Pearson’s correlation 200

coefficients, and bias and LOA are presented in Table 4. 201

202

For every equation, values for rc and rp were very similar and > 0.98 for all equations, 203

indicating that the predicted data were highly correlated and were close to the line of identity 204

with the measured data. For all equations, biases were small (0.6 - 1.3%) and positive, indicating 205

the equations slightly underestimated FFM compared to DXA-measured FFM. Equations that 206

included a resistance index as a predictor variable exhibited the smallest LOA, varying from ± 207

6.9% for equation 3 to ± 7.5% for equation 1 (Figs. 1 and 2). In contrast, morphometric-based 208

predictors generally exhibited smaller biases than those that included an impedance predictor 209

variable but had larger LOA, ranging from ± 8.2% (equation 5) to ± 11.6% (equation 7). 210

Although differences between equations in bias, LOA (Table 4), r2 and RMSE (Table 2) were 211

generally small, equation 3 was deemed to have the best predictive performance for impedance-212

based predictors, while equation 5 was considered to be the best performing morphometric-based 213

predictor. 214

215

Thus, the final predictive equation (Equation 8) based on these variables, BW, BCS and 216

L2/R∞, using the data for all dogs, was determined: 217

Page 10 of 22

11

where the measurement units were as follows: FFM (g), BW (g), L (cm), and ∞ (ohm). This 218

equation was used to predict FFM in all dogs and %FM by difference with BW (Fig. 3a). 219

Correlation with DXA-determined %FM was high (rp = 0.789; rc = 0.730; Fig. 3b), although 220

LOA were wide with a significant slope, indicating that the BIA-based prediction overestimated 221

body fat percentage in animals with low %FM, but underestimated above approximately 24% 222

FM (Fig. 3b). 223

224

Discussion 225

This is the first study to use and validate MFBIA analysis to estimate body composition 226

in relatively lean dogs. In this cohort of mixed-breed dogs, MFBIA-based predictors provided an 227

accurate (within 1.5%) measurement of FFM when compared to DXA-derived FFM. FFM 228

estimates correlated most highly with DXA-FFM when the prediction equation included 229

resistance quotient, BW, and BCS. Notably, however, the inclusion of impedance as a predictor 230

variable did not add substantially to predictive power compared with prediction based upon 231

simple morphometric measurements, as judged by the magnitude of the correlation coefficient 232

and bias, although LOA were generally smaller, indicating greater predictive accuracy at the 233

individual subject rather than population level. 234

235

Comparisons of our observations with existing published data (Stone et al., 2009; 236

German et al., 2010; Jeusette et al., 2010) were difficult, as different impedance devices were 237

used, and either the prediction equations were not published or cross-validations were not 238

performed. In addition, data were presented as derived FM rather than measured FFM. SFBIA-239

Page 11 of 22

12

predicted FM correlated well (r = 0.819) with BCS in 46 dogs (Stone et al., 2009), although 240

agreement was not determined, and comparison against a reference technique such as DXA was 241

not performed. German et al. (2010) compared FM by DXA with that predicted by SFBIA in 24 242

dogs, but found a poor correlation (r2

= 0.44) and poor LOA (-16% to 21%). Both of these 243

studies presented data for body FM only, not FFM, and in neither case was the predictive 244

algorithm or its validation provided. Neither study compared impedance based predictors against 245

simple morphometric measurements. The more recent study of Jeusette et al. (2010), using an 246

SFBIA device, found good correlation of FFM and DXA-FFM (r = 0.84 - 0.87) in a population 247

of mixed breed dogs (n = 19); addition of morphometric measurements (height, BW, length and 248

pelvic circumference) to the regression improved correlation (r = 0.92), although LOA remained 249

poor (-6 - 5 kg; 25 - 21%). MFBIA prediction of FFM in our study provided better predictive 250

performance than previously published single frequency predictions. 251

252

FM was predicted less accurately than FFM. Correlation of FM by DXA and impedance 253

was high (r = 0.79), but LOA were wide; an observation in accord with the observations of 254

Jeusette et al. (2010), where correlations and LOA for FM were weaker than those for FFM. This 255

was not entirely surprising, since impedance is a function of body water content and hence FFM, 256

not FM per se. BIA provides an indirect estimate of FM, as BW-FFM, with consequent 257

propagation of errors (Kyle et al., 2004). In addition, DXA measurement of FM incorporates 258

larger imprecision errors (3 - 4%) than for FFM. 259

260

Our study has both strengths and limitations. Although the population of dogs was large 261

by comparison with previous studies (German et al., 2010; Jeusette et al., 2010), it did not 262

Page 12 of 22

13

include any underweight or obese animals or very small or very large dogs, or dogs with unusual 263

shape, such as Dachshunds. In our study, all dogs were young adults; however, obesity is more 264

common in middle aged pet dogs. Future studies are required in dogs with a range of ages, 265

especially middle-aged dogs (German, 2006), since that is when accurate determination of body 266

composition is most important for making recommendations to clients regarding management of 267

canine obesity. Sex was not a significant predictor variable in this cohort. This was not 268

surprising, since all dogs were neutered and representative of the general population of dogs in 269

Australia, where 78% are neutered (Heady, 2006). The effect of sex on prediction of body 270

composition should be determined in a study of intact animals for comparison. Despite the larger 271

number of dogs used in the present study compared to previous studies (Jeusette et al. 2010), our 272

data provided limited scope for cross-validation. Ideally, the prediction equations generated here 273

for FFM and the derived estimates of FM should be validated in a large, independent, mixed 274

population of dogs. Future studies should also investigate whether measurements taken while 275

unsedated dogs are standing (German et al., 2010) are feasible, as this would make the method 276

truly non-invasive and increase its utility in routine clinical practice. 277

278

The present study confirmed good predictive performance of simple morphometric 279

measurements for body composition, possibly because most of the dogs were lean. These data 280

question the value of impedance for body composition assessment. Our results indicate that when 281

monitoring individual dogs, predictors that included impedance in addition to morphometric 282

measurements were most accurate. Bodyweight is a readily determined objective measurement. 283

Measurement of height is more difficult in a conscious animal, but is nonetheless an objective 284

measurement, while BCS involves subjective assessment. In the present study, BCS was the 285

Page 13 of 22

14

recorded as the mean score of two experienced assessors. Although individual veterinarians are 286

likely to be consistent in their assessment of BCS, the measurement is subjective and variation 287

between veterinarians can occur. In our study, predictive performance was only slightly 288

worsened when BCS was omitted as a predictor variable. Bodyweight alone provided good 289

prediction of FFM in our population of lean dogs, but this does not necessarily mean that it is 290

acceptable for clinical use, particularly in obese animals. A dog classified as overweight on the 291

basis of BW according to a breed standard may not necessarily have excess body fat. Rather, it 292

could have increased lean body mass as, for example, in athletic and working dogs (Crane, 293

1991). 294

295

Conclusions 296

The impedance technique has the potential to provide a tool that veterinarians could use 297

to routinely assess the body composition of dogs in veterinary practice. It is safe, potentially non-298

invasive, portable, and low-cost compared to techniques such as DXA. It can provide a fast and 299

objective quantitative measure of FM and FFM, which is not possible with subjective 300

assessments such as BCS. Despite this, simple morphometric measurements performed equally 301

well in this population of lean dogs. Whether impedance measurements can improve body 302

composition assessment in the target population of overweight and obese animals requires 303

further study. 304

305

Conflict of interest statement 306

L.C. Ward consults for ImpediMed. E.A. Flickinger is an employee of P and G Pet Care, 307

which partially funded the research. ImpediMed did not have any involvement in the execution 308

Page 14 of 22

15

of this study or in the preparation of this manuscript. None of the other authors has any other 309

financial or personal relationships that could inappropriately influence or bias the content of the 310

paper. 311

312

Acknowledgements 313

The authors wish to thank Deanne Waine, Jenny Hall, Libby Jolly, Melita Watkins and 314

veterinary students of The University of Queensland for their assistance with data collection and 315

to Nicole Richards and Linda Oliver for their dedication to the care of the dogs. Our appreciation 316

also goes to Princess Alexander Hospital for their technical assistance with the DXA scans. 317

This work was supported by grant from P and G Pet Care, USA. 318

319

References 320

Baldwin, K., Bartges, J., Buffington, T., Freeman, L.M., Grabow, M., Legred, J., Ostwald Jr. D., 321

2010 AAHA Nutritional assessment guidelines for dogs and cats. Journal of the American 322

Animal Hospital Association 46, 285-296. 323

324

Bland, M.J., Altman, D.G., 1986. Statistical methods for assessing agreement between two 325

methods of clinical measurement. The Lancet 327, 307-310. 326

327

Cornish, B.H., Thomas, B.J., Ward, L.C., 1993. Improved prediction of extracellular and total 328

body water using impedance loci generated by multiple frequency bioelectrical impedance 329

analysis. Physics in Medicine and Biology 38, 337-346. 330

331

Cornish, B.H., Ward, L.C., Thomas, B.J., Jebb, S.A., Elia, M.. 1996. Evaluation of multiple 332

frequency bioelectrical impedance and Cole-Cole analysis for the assessment of body water 333

volumes in healthy humans. European Journal of Clinical Nutrition 50, 159-164. 334

335

Crane, S.W. 1991. Occurrence and management of obesity in companion animals. Journal of 336

Small Animal Practice 32, 275–282. 337

338

Foster, K.R., Lukaski, H.C., 1996. Whole-body impedance—what does it measure? American 339

Journal of Clinical Nutrition 64, 388S-396S. 340

341

Page 15 of 22

16

German, A.J., The growing problem of obesity in dogs and cats. 2006. Journal of Nutrition 136, 342

1940S–1946S. 343

344

German, A.J., Holden, S.L., Morris, J., Biourge, V.B., 2010. Comparison of a bioimpedance 345

monitor with dual-energy x-ray absorptiometry for non-invasive estimation of percentage 346

body fat in dogs. American Journal of Veterinary Research 71, 393-398. 347

348

Gossellin, J., Wren, J.A., Sunderland, S.L. 2007. Canine obesity: an overview. Journal of 349

Veterinary Pharmacology and Therapeutics 30, 1–10. 350

351

Headey, B., 2006. National people and pets survey. Socially Responsible Pet Ownership In 352

Australia: A decade of progress. Melbourne Institute of Applied Economic and Social 353

Research, Melbourne, Vic Australia, 24 pp. 354

355

Heymsfield, S., Lohman, T, Wang, Z., Going, S., 2005. Human Body Composition. 2nd Edn, 356

Human Kinetics Press, Champaign, IL, USA, 366 pp. 357

358

Jeusette, I., Greco, D., Aquino, F. Detilleux, J., Peterson, M., Romano, V., Torre, C., 2010. 359

Effect of breed on body composition and comparison between various methods to estimate 360

body composition in dogs. Research in Veterinary Science 88, 227-232. 361

362

Kealy, R.D., Stowe, H.D., Lawler, D.F., Ballam, J.M.,Mantz, S.L., Biery, D.N., Greely, E.H., 363

Lust, G., Segre, M., Smith, G., 2002. Effect of diet restriction on life span and age-related 364

changes in dogs. Journal of the American Veterinary Medical Association 220, 1315-1320. 365

366

Kyle, U.G., Bosaeus, I., De Lorenzo, A.D., Deurenberg, P., Elia, M., Gómez. J,M,, Heitmann, 367

B.L., Kent-Smith, L., Melchior, J.C., Pirlich, M., Scharfetter, H., Schols, A.M., Pichard, C., 368

2004. Composition of the ESPEN Working Group. Clinical Nutrition 23, 1226-1243. 369

370

Kim, S.P., Ellmerer, M., Van Citters, G.W., Bergman, R.N., 2003. Primacy of hepatic insulin 371

resistance in the development of the metabolic syndrome induced by an isocaloric moderate-372

fat diet in the dog. Diabetes 52, 2453-2460. 373

374

Laflamme, D.P., 1997. Development and validation of a body condition score system for dogs. 375

Canine Practice 22, 10–5. 376

377

Laflamme, D.P., 2012. Obesity in dogs and cats: What is wrong with being fat? Journal of 378

Animal Science 90, 1653-1662. 379

380

Lin, L.I.K., 1989. A Concordance Correlation Coefficient to Evaluate Reproducibility. 381

Biometrics 45, 255-268. 382

383

Lund, E.M., Armstrong, P.J., Kirk, C.A., Klausner, J.S., 2005. Prevalence and risk factors for 384

obesity in adult cats from private US veterinary practices. International Journal of Applied 385

Research in Veterinary Medicine 3, 88–96. 386

387

Page 16 of 22

17

Mawby, D., artges, J. ., d’Avignon, A., aflamme, D.P., Moyers, T.D.,Cotrell, T., 2004. 388

Comparison of various methods for estimating body fat in dogs. Journal of the American 389

Animal Hospital Association 40, 109-114. 390

391

McGree, J.M., Duffull, S.B., Eccleston, J.A., Ward, L.C., 2007. Optimal designs for studying 392

bioimpedance. Physiological Measurement 12:1465-1483. 393

394

McGreevy, P.D., Thompson, P.C., Pride, C., Fawcett, A., Grassi, T. Jones, B., 2005. Prevalence 395

of obesity in dogs examined by Australian veterinary practices and the risk factors involved. 396

Veterinary Record 156, 695-702. 397

398

Purushothaman, D., Vanselow, B.A., Wu, S., Butler, S., Brown, W.Y., 2013. Whole body 399

computed tomography with advanced imaging techniques: A research tool for measuring 400

body composition in dogs. Journal of Veterinary Medicine 2013, 1-6. 401

402

Scheltinga, M.R., Helton, W.S., Rounds, J., Jacobs, D.O., Wilmore, D.W., 1991. Impedance 403

electrodes positioned on proximal portions of limbs quantify fluid compartments in dogs. 404

Journal of Applied Physiology 70, 2039-2044. 405

406

Schoeller, D.A., 1996. Hydrometry. In Human body composition, Roche ,A.F., Heymsfield, 407

S.B., Lohman, T.G. ( Eds.) 1st Edn., Human Kinetics Press, Champaign, IL, USA, pp. 25-408

44. 409

410

Shoveller, A.K., DiGennaro, C.L., Spangler, D., 2014. Trained vs untrained evaluator assessment 411

of body condition score as a predictor of percent body fat in adult cats. Journal of feline 412

medicine and surgery. 16, 957-965. 413

414

Stanton, C.A., Hamar, D.W., Johnson, D.E., Fettman, M.J., 1992. Bioelectrical impedance and 415

zoometry for body composition analysis in domestic cats. American Journal of Veterinary 416

Research 53, 251-257. 417

418

Stone, R., Berghoff, N., Steiner, J.M., Zoran, D. 2009. Use of bioelectric impedance device in 419

obese and lean healthy dogs to estimate body fat percentage. Veterinary Therapeutics 10, 420

59-70. 421

422

Thomas, B.J., Ward, L.C., Cornish B.H., 1998. Bioimpedance spectrometry in the determination 423

of body water compartments: accuracy and clinical significance. Applied Radiation and 424

Isotopes 49, 447-55. 425

426

Ward, L.C., Battersby, K.J. 2009 Assessment of body composition of rats by bioimpedance 427

spectroscopy: Validation against dual-energy x-ray absorptiometry. Scandanavian Journal 428

Laboratory Animal Science 36, 253-261. 429

430

Zanghi, B.M., Cupp, C.J., Pan, Y., tissot-Favre, D.G., Milgram, N.W., Nagy, T.R., Dobson, H., 431

2013. Non-invasive measurements of body composition and body water via quantitative 432

Page 17 of 22

18

magnetic resonance, deuterium water, and dual-energy x-ray absorptiometry in awake and 433

sedated dogs. American Journal of Veterinary Research 75, 733-742. 434

435

Zar, J.H., 1999. Biostatistical analysis. 4th Edn. Prentice Hall Inc., Upper Saddle River, NJ USA, 436

663 pp. 437

438

Zoran, D.L., 2010. Obesity in dogs and cats: a metabolic and endocrine disorder. Veterinary 439

Clinics of North America Small Animal Practice 40, 221-239. 440

441

442

Page 18 of 22

19

Figure legends 443

444

Fig. 1. Correlation comparison plots in the validation group of dogs for the equation with the best 445

resistance index, L2/R∞ ([0.57 x BW] + [-1239.593 x BCS] + [75.993 x R infinity index] + 446

4041.392; L, length; R, resistance; equation 3, Table 2) and the morphometric equation ([0.7041 447

x BW] +2045.5622; equation 5, Table 2). Dual energy X-ray absorptiometry -determined FFM is 448

on the x-axis and the predicted FFM on the y-axis. Panel: (a) Equation 3 fat-free mass; (b) 449

Equation 5 fat-free mass. Solid circles, females; solid triangles, males; , 95% confidence 450

interval; , line of identity; ___, line of best fit. 451

452

Fig. 2. Bland and Altman plots in the validation group of dogs for the equation with the best 453

resistance index, L2/Rinfinity (L, length; R, resistance; equation 3) and the morphometric equation 454

(equation 5). (a) Equation 3 fat-free mass; (b) Equation 5 fat-free mass. Solid circles, females; 455

solid triangles, males; , limits of agreement (1.96 × standard deviation); , zero mean 456

difference; , regression line; ___, mean of data. 457

458

Fig. 3. Comparison of measured body fat percentage and predicted body fat percentage for all 459

dogs using the final prediction equation (Equation 8). (a) Box plot. Solid circles, females; solid 460

triangles, males; box, 25th to 75th

percentile and median; I, minium to maximum values (b) 461

Correlation plot. Solid circles, females; solid triangles, males; , 95% confidence interval; , 462

line of identity; ___, line of best fit. (c) Limits of agreement plot. Solid circles, females; solid 463

triangles, males; , limits of agreement (1.96 × standard deviation); , zero mean difference; 464

, regression line; ___, mean of data. 465

Page 19 of 22

20

Table 1. Mean ± standard deviation data for enrolled dogs. 466

Parameter Neutered males Neutered females All

n 18 17 35

Scale weight (kg) 24.2 ± 7.5 19.1 ± 4.4b 21.6 ± 6.8

Length (cm) 80.4 ± 8.8 74.1 ± 7.5 77.4 ± 8.6

Body composition

BCS 3.2 ± 0.37 3.2 ± 0.38 3.2 ± 0.4

Lean mass (kg) 18.5 ± 5.0a 14.8 ± 3.4

b 16.8 ± 4.6

Fat mass (kg) 4.5 ± 2.3

3.4 ± 1.5 3.9 ± 2.0

Bone mineral content (g) 702.9 ± 199.9a 525.8 ± 129.1

c 616.8 ± 189.4

Fat-free mass (kg) 19.2 ± 5.2a

15.4 ± 3.5b 17.3 ± 4.8

DXA weight (kg) 23.5 ± 7.2 18.8 ± 4.8b 21.3 ± 6.5

DXA weight (% scale

weight)

98.4 ± 0.9 98.4 ± 1.2 98.4 ± 1.1

Whole body impedance

R50 (ohm) 136.3 ± 13.6 138.3 ± 10.6 137.3 ± 12.1

R500 (ohm) 101.7 ± 10.8 102.3 ± 8.0 102.0 ± 9.4

∞ ohm 94.9 ± 10.3 95.7 ± 7.5 95.3 ± 8.9

Zc (ohm) 140.2 ± 15.2 140.4 ± 12.2 140.3 ± 13.6

BCS, body condition score; DXA, dual energy X-ray absorptiometry; R, resistance 467 a Neutered males, P <0.05 468

b Neutered females, P <0.05 469

c Neutered females, P<0.001 470

471 472

Page 20 of 22

21

Table 2. Equations and corresponding adjusted coefficient of determination (r2) and root mean 473

square error (RMSE) for predicting fat-free mass using bioelectrical impedance and 474

morphometric variables in the prediction group of 23 dogs. 475

Equation

number

Equations for fat-free mass prediction (kg) r2

(adjusted)

RMSE

(g)

1 (0.624 x BW) + (-1.579 x BCS) + (0.079 x R50 index) + 5.319 0.973a 0.724

2 (0.578 x BW) + (-1.306 x BCS) + (0.078 x R500 index) + 4.306 0.977 a 0.659

3 (0.570 x BW) + (-1. 39 x CS + 0.076 x ∞ index + 4.041 0.979 a 0.634

4 (0.46 x BW) + 0.098 x ∞ index + 0.905 0.975 a 0.692

5 (0.7041 x BW) +2.046 0.955 a 0.952

6 (0.793 x BW) + (-2.090 x BCS) + 6.895 0.969 a 0.778

7 (0.649 x BW) + (-1.786 x BCS) + (0.114 x Length) + 0.188 0.975 a 0.691

BW, bodyweight; R, resistance; BCS, body condition score 476 a P<0.001; 477

478

Table 3. Significance (P value; variance inflation factor indicated in parentheses) of each 479

variable included in the prediction equations presented in Table 2. 480

481

Equation

Variables 1 2 3 4 5 6 7

Body condition

score

0.021

(2.9)

0.038

(2.4)

0.040

(2.4)

- - 0.003

(1.9)

0.005

(2.1)

Bodyweight

<0.0001

(12.2)

<0.0001

(11.7)

<0.0001

(10.9)

<0.0001

(6.2)

<0.0001

(1.9)

<0.0001

(7.8)

Resistance index

(l2/R)

0.057

(9.7)

0.008

(8.2)

0.003

(7.6)

0.0002

(6.3)

- - -

Length - - - - - - 0.021

(5.8)

482

Page 21 of 22

22

Table 4. Concordance (rc and Pearson’s rp) correlation coefficients, bias, limits of agreement, 483

and significance for the regression slope for all prediction equations (Table 2) for fat-free mass 484

for the validation group of dogs (n = 12). 485

Equation rc rp a Bias

(g)

Limits of agreement

(g)

P for slope

Impedance-based predictors

1 0.993

0.993 95.0

(0.6%)

-1225.1 to 1415.0

(6.9% to 8.2%)

0.975

2 0.993 0.994 153.7

(0.8%)

-1121.7 to 1429

(6.2% to 7.7%)

0.405

3 0.992 0.994 202.7

(1.0%)

-1068.0 to 1473.3

(5.7% to 7.8%)

0.342

4 0.989 0.993 291.4

(1.3%)

-1175.3 to 1758.0

(5.6% to 8.2%)

0.087

Morphometric only (bodyweight, length body condition score)-based predictors

5 0.991 0.992 46.8

(0.8%)

-1300.9 to 1570.9

(-7.4% to 9.1%)

0.830

6 0.992 0.992 39.6

(0.5%)

-1422.0 to 1501.2

(-7.9% to 8.9%)

0.335

7 0.985 0.986 113.7

(1.2%)

-1806.5 to 2033.8

(10.4% to 12.8%)

0.460

a Comparison, dual energy X-ray absorptiometry-predicted 486

487

Page 22 of 22