Abdominal visceral to subcutaneous fat distribution in dogs: computed tomography accuracy and factors associated with distribution

Abstract

Background

Abdominal fat distribution, particularly visceral fat, is commonly assessed as a marker of obesity-related and metabolic diseases in people. While this relationship may exist, few studies consider the factors related to the relative distribution of visceral and subcutaneous abdominal fat in dogs. This cross-sectional study evaluated associations between measures of abdominal adiposity, visceral and subcutaneous fat distribution (V/SQ), body condition score (BCS), age, sex, neuter status, and breed conformation in 205 dogs presenting to a tertiary veterinary hospital between March 2006 and March 2020. The influence of several disease states on abdominal adiposity and fat distribution was also evaluated. Additionally, the study aimed to assess the criterion validity of average computed tomography (CT) Hounsfield units, and linear and cross-sectional area measures of abdominal adiposity and fat distribution relative to CT-derived volumetric analysis and for intra-observer reliability.

Results

Greater total abdominal adiposity was seen in older dogs, with values highest around 10 years of age and lower after 10 years of age, and was lower in terrier breeds and dogs with neoplasia. Greater V/SQ was observed in older dogs, hounds, and terriers, but decreased with increasing BCS, total abdominal adiposity, and thoracic height-width ratio. Additionally, V/SQ was higher in dogs with hyperadrenocorticism. Body condition score was moderately correlated with total abdominal, visceral, and subcutaneous adiposity. Abdominal fat areas measured at L3 overestimated total abdominal and visceral fat percentages but underestimated subcutaneous fat percentages, with increasing bias at higher fat percentages. Linear fat measurements were moderately correlated with total abdominal adiposity, but only weakly correlated with abdominal fat distribution.

Conclusions

This study supports the association between abdominal adiposity, age, breed category, and potentially certain diseases like neoplasia. Moreover, it highlights the correlation between V/SQ, age, and total adiposity, while emphasising the preferential distribution of fat to the visceral compartment in dogs with hyperadrenocorticism. The study also identified a novel association between V/SQ, specific breed categories, and body conformation (i.e., thoracic height-width ratio). Importantly, CT volumetric measures are more reliable in determining abdominal fat distribution than area and linear measures, supporting the use of CT volumetric measures in the study’s methodology and its implications for future research and clinical practice.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12917-026-05503-x.

Background

Overweight and obesity (OO) are conditions defined as abnormal or excessive fat accumulation [1–4]. Dogs are considered overweight when their body fat exceeds 20–30%, and obese when it surpasses 30% of their ideal body weight [4]. Overweight and obesity are recognised as one of the most prevalent disorders affecting pet dogs, with a 22–60% prevalence reported [5–10]. Various risk factors contributing to or exacerbating OO, include dog-related (e.g., signalment, genetic factors, exercise volume), diet-related (e.g., amount of food intake, frequency of treats, consumption of table scraps), owner-related (e.g., demographic, socioeconomic situation, ability to exercise); disease-related (e.g., hyperadrenocorticism, hypothyroidism), pharmaceutical-related (e.g., glucocorticoids or anticonvulsants administration), and environment-related (e.g., exposure to obesogens) factors [5, 11–13].

Increased adiposity has welfare implications as OO is associated with various disease processes and metabolic derangements [11, 14–16]. A potential contributor to OO pathophysiology and metabolic derangements is the relative distribution of fat to the visceral (central) or subcutaneous (peripheral) regions in dogs. Although there is scant and, at times, contradictory research, increased fat distribution to the visceral compartment surrounding the organs may be more inflammatory and more significantly influence adipokine, cardiovascular, and metabolic dysregulation than general OO [17–20].

In humans, the gold standard methods for measuring visceral fat are CT and MRI, although DXA is commonly used due to its relatively high accuracy, accessibility, and low radiation exposure [21, 22]. Visceral fat area (VFA) established from a single transverse slice through the abdomen at the level of the third lumbar vertebrae is widely used in people [21]. Ultrasound measurements have also been used to estimate human visceral fat to provide greater access and limit radiation exposure, but reproducibility and accuracy are poor [21]. Further, there is a growing consensus focused on the abdominal circumference of people, supported by body mass index (BMI), as a more accurate reflector of central obesity and its relationship with metabolic health dysfunction [23]. Some of these methods have been replicated in dog studies, with CT-determined VFA at the third lumbar vertebra commonly referenced, and CT volume, MRI, and ultrasound methods published [24–29]. However, there is limited validation and standardisation of these techniques.

Current evidence suggests increased visceral-to-subcutaneous fat distribution/ratio (V/SQ) in dogs is associated with increasing age, neuter status, and certain dog breeds, but the association between V/SQ and certain disease processes, such as hyperadrenocorticism (HAC), has not been objectively evaluated in dogs [27, 28, 30–33]. Investigation of these and other associations needs to be established if visceral adiposity’s effect on dogs’ overall health is to be fully elucidated and early intervention initiated.

Thus, this cross-sectional observational study aimed to assess for an association between abdominal visceral and SQ fat distribution, total abdominal adiposity, body condition score (BCS), age, sex, neuter status, and breed conformation in a tertiary veterinary hospital population of dogs. The influence of several disease states on abdominal adiposity and abdominal fat distribution was also evaluated. Finally, mean Hounsfield units, single-slice fat area at the third lumbar vertebra (CTA), and linear measures of fat thickness were assessed for criterion validity against volumetric abdominal fat measurements (CTV), and for intra-observer reliability.

Results

Descriptive statistics

A total of 205 abdominal CTs were analysed from the original 448 CTs retrieved from the hospital records between March 2006 and March 2020. Data were 100% complete for breed, age, sex-neuter status, and body weight, and 80% for BCS.

Supplemental 1 (S1) displays frequency cross-tabulations and univariable comparisons for categorical data. There was a bimodal age distribution, with an overall median age of 8.0 years (IQR: 2.7 to 10.8 years; range: 3 months to 16.3 years), with modes at approximately 1–2 years and 10–11 years. Neutered dogs (mean ± SD = 7.7 ± 4.3 years [95% CI of mean: 7.0, 8.3 years]) were significantly older than entire dogs (4.9 ± 4.2 years [95% CI: 3.4, 6.4 years]). The median body weight was 21.0 kg (IQR: 10.1 to 32.8 kg; range: 1.2 to 67.8 kg), with a median body condition score of 5 out of 9 (IQR: 4 to 6; range: 2 to 9).

There were 131 purebred and 74 mixed-bred dogs, with predominant breeds including Border collies (n = 13), Labrador retrievers (n = 11), Staffordshire terriers (n = 9), golden retrievers (n = 7), German shepherds (n = 7), Jack Russell terriers (n = 6), beagles (n = 5), and pugs (n = 5). No other breed had more than five representatives. When categorised according to ANKC standards, there were 32 working dogs, 24 gundogs, 24 utility dogs, 19 terriers, 11 non-sporting breeds, 11 toy breeds, and 10 hounds. Two American and two Australian Bulldogs were considered mixed breeds due to ANKC recognition limitations, resulting in 74 mixed-breed dogs. Additionally, there were 88 mesocephalic, 25 brachycephalic, and 18 dolichocephalic dogs, with 99 non-chondrodystrophic and 32 chondrodystrophic dogs. The HWR varied across breed groups, as shown in Supplemental Table 2 (S2). Specifically, working dogs and non-sporting ANKC breeds, giant breeds, and dolichocephalic dogs had deeper, narrower cranial abdomens, while chondrodystrophic dogs exhibited shallower, wider cranial abdomens. Hounds exhibited deeper chests in this sample, although the observed difference was not statistically significant.

Abdominal fat distribution

The multivariable associations between each risk factor and fat distribution are displayed in Tables 1 and 2 and Supplemental Files 3 and 4. The CT confounding variables significantly reduced total abdominal, visceral, and SQ fat measurements. Specifically, small abdominal organs and an increasing volume of effusion emerged as the main contributors to CT confounding. Small abdominal organs resulted in an increased V/SQ, while the other CT confounders did not show a statistically significant effect on the V/SQ. Notably, the greatest impact of effusion and small organs was on decreasing SQ fat percentage, with the V/SQ increased when these confounders were not adjusted for total abdominal adiposity.

Table 1.

Putative causal diagram-guided logistic regression analysis outputs for variables associated with CTV total abdominal fat percentage in 205 dogs presenting to a tertiary veterinary hospital between March 2006 and March 2020

Variable		Adjustment set (total effect)	Unadjusted estimates (95% CI)	Adjusted estimates (95% CI)	p-value	Model estimated marginal means for levels (95% CI)
Age (years, quadratic)	Age, centred	CT Confounders, Confirmed Hyperadrenocorticism	1.683 (1.299, 2.067)	1.670 (1.293, 2.046)	< 0.001
	(Age, centred)2	Age, centred, CT Confounders, Confirmed Hyperadrenocorticism	-0.314 (-0.417, -0.211)	-0.222 (-0.312, -0.132)	< 0.001
Sex	Females	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	46.98 (42.74, 51.21)
	Males		-2.112 (-6.054, 1.831)	-1.125 (-4.257, 2.007)	0.482	45.85 (41.52, 50.19)
Neuter Status	Entire	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	43.56 (38.01, 49.11)
	Desexed		7.943 (2.600, 13.286)	3.310 (-1.095, 7.714)	0.142	46.87 (42.86, 50.87)
Sex-Neuter Status	Female entire	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	44.80 (36.04, 53.55)
	Female spayed		8.539 (-1.669, 18.747)	2.351 (-6.073, 10.775)	0.585	47.15 (42.83, 51.47)
	Male entire		-0.274 (-11.567, 11.018)	-1.799 (-10.936, 7.337)	0.700	43.00 (36.89, 49.11)
	Male neutered		7.075 (-3.054, 17.203)	1.651 (-6.661, 9.964)	0.697	46.45 (42.02, 50.88)
ANKC Breed Group	Toys	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	47.77 (39.94, 55.61)
	Terriers		0.634 (-10.024, 11.293)	-6.996 (-15.417, 1.424)	0.105	40.78 (34.32, 47.24)
	Gundogs		4.605 (-5.637, 14.848)	-1.109 (-9.139, 6.921)	0.787	46.67 (40.75, 52.58)
	Hounds		0.598 (-11.693, 12.89)	-1.460 (-11.064, 8.144)	0.766	46.31 (38.32, 54.31)
	Working Dogs		1.703 (-8.130, 11.535)	-0.976 (-8.690, 6.738)	0.804	46.80 (41.30, 52.30)
	Utility		6.037 (-4.206, 16.28)	2.178 (-5.870, 10.226)	0.596	49.95 (43.94, 55.96)
	Non-sporting		-0.533 (-12.528, 11.462)	3.358 (-6.081, 12.796)	0.487	51.13 (43.36, 58.91)
	Mixed breed		3.732 (-5.359, 12.822)	-1.805 (-8.965, 5.354)	0.622	45.97 (41.71, 50.23)
ANKC Terrier	Terriers	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	40.66 (34.26, 47.06)
	Non-terriers		2.517 (-4.212, 9.246)	6.184 (0.838, 11.531)	0.024	46.84 (42.89, 50.80)
ANKC Hound	Hounds	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000		45.98 (37.96, 54.00)
	Non-hounds		2.439 (-6.626, 11.503)	0.500 (-6.672, 7.671)	0.892	46.48 (42.48, 50.48)
				0.5 (-6.672, 7.671)	0.892
Skull Shape	Brachycephalic	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	4.797 (-3.839, 13.432)	3.100 (-3.756, 9.955)	0.377	49.33 (43.39, 55.26)
	Mesocephalic		5.491 (-1.735, 12.718)	0.047 (-5.887, 5.980)	0.988	46.27 (41.67, 50.88)
	Dolichocephalic		0.000	0.000	Reference category	46.23 (39.68, 52.78)
	Mixed breed		5.877 (-1.464, 13.219)	-0.138 (-6.189, 5.913)	0.964	46.09 (41.79, 50.38)
						46.59 (42.01, 51.17)
Chondrodystrophy Status	Chondrodystrophic	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	46.72 (41.20, 52.25)
	Non-chondrodystrophic		1.827 (-3.865, 7.520)	0.161 (-4.371, 4.694)	0.944	46.88 (42.37, 51.39)
	Mixed breed		2.654 (-3.268, 8.576)	-0.602 (-5.331, 4.127)	0.803	46.12 (41.83, 50.41)
HWR		Age‡, CT Confounders, Confirmed Hyperadrenocorticism	-12.041 (-24.563, 0.480)	-9.928 (-20.196, 0.340)	0.060
Size Category (Nominal Categorisation with Cross Breed Excluded)		Age‡, CT Confounders, Confirmed Hyperadrenocorticism	1.931 (0.136, 3.725)	0.973 (-0.3612, 2.307)	0.156
		Age, CT Confounders, Hyperadrenocorticism
Size Category (including Mixed Breed)	Extra Small	Age‡, CT Confounders, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	51.61 (40.95, 62.27)
	Small		-7.373 (-21.312, 6.566)	-7.860 (-18.932, 3.213)	0.166	43.75 (37.26, 50.24)
	Medium-small		4.690 (-9.988, 19.367)	-2.322 (-14.021, 9.377)	0.698	49.29 (41.79, 56.78)
	Medium-large		-2.385 (-15.369, 10.599)	-6.775 (-17.104, 3.553)	0.200	44.83 (39.75, 49.92)
	Large		2.979 (-10.341, 16.298)	-4.367 (-15.002, 6.269)	0.422	47.24 (41.70, 52.78)
	Giant		4.035 (-9.752, 17.822)	0.296 (-10.676, 11.269)	0.958	51.90 (45.56, 58.25)
	Mixed breed		1.150 (-11.591, 13.891)	-5.368 (-15.538, 4.801)	0.302	46.24 (41.99, 50.49)
BCS (nominal)		Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	5.764 (4.596, 6.932)	3.908 (2.816, 5.000)	< 0.001
BCS (categorical)	2	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	0.000	0.000		29.95 (20.87, 39.03)
	3		4.771 (-5.521, 15.064)	3.354 (-5.488, 12.195)	0.458	33.31 (27.06, 39.55)
	4		13.495 (3.818, 23.171)	9.184 (0.771, 17.596)	0.034	39.14 (34.06, 44.22)
	5		17.039 (7.422, 26.656)	11.114 (2.709, 19.519)	0.011	41.07 (35.93, 46.21)
	6		22.67 (12.429, 32.91)	14.65 (5.534, 23.766)	0.002	44.60 (38.60, 50.60)
	7		29.338 (18.597, 40.078)	19.854 (10.32, 29.389)	< 0.001	49.81 (43.23, 56.39)
	8		37.876 (24.412, 51.34)	28.425 (16.493, 40.357)	< 0.001	58.38 (48.89, 67.87)
	9		38.573 (22.851, 54.295)	24.628 (10.676, 38.581)	< 0.001	54.58 (42.24, 66.93)
Abdominal Adiposity	Not applicable
CTV Mean Hounsfield Units		Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	-0.506 (-0.525, -0.488)	-0.494 (-0.518, -0.470)	< 0.001
Disease Chronicity	Acute (< 3 months)	Age‡, Sex-Neuter Status, ANKC Terrier, CT Confounders.	0.000	0.000	Reference category	40.90 (37.41, 44.39)
	Chronic (≥ 3 months)		-6.396 (-10.264, -2.528)	-0.705 (-4.002, 2.593)	0.676	40.20 (36.29, 44.11)
Disease Category	Endocrine	Age‡, Sex-Neuter Status, ANKC Terrier, CT Confounders.	0.000	0.000	Reference category	43.31 (37.84, 48.78)
	Neoplasia		-5.392 (-11.430, 0.645)	-5.498 (-10.664, -0.332)	0.038	37.81 (33.96, 41.66)
	Gastroenteric		-3.973 (-13.337, 5.392)	0.318 (-7.887, 8.523)	0.940	43.63 (36.46, 50.80)
	Trauma/Musculoskeletal		-11.641 (-27.248, 3.966)	-3.370 (-17.034, 10.294)	0.629	39.94 (26.62, 53.26)
	Congenital Portal Vascular Anomaly		-20.914 (-27.814, -14.015)	-0.792 (-8.465, 6.882)	0.840	42.52 (36.36, 48.68)
	Inflammatory		-7.630 (-15.043, -0.217)	0.959 (-6.007, 7.925)	0.788	44.27 (38.70, 49.84)
	Other		-5.202 (-13.037, 2.633)	-2.473 (-9.26, 4.314)	0.476	40.84 (35.04, 46.63)
Confirmed Hyperadrenocorticism	No evidence of hyperadrenocorticism	Age‡, Sex-Neuter Status, ANKC Terrier, CT Confounder.	0.000	0.000	Reference category	40.61 (37.33, 43.89)
	Confirmed Hyperadrenocorticism		9.017 (-0.439, 18.473)	4.137 (-3.523, 11.797)	0.291	44.75 (36.50, 52.99)
Drug Length	No drugs	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	0.000	0.000	Reference category	42.40 (37.10, 47.70)
	Acute (< 3 months)		-3.756 (-7.972, 0.460)	-0.366 (-3.794, 3.062)	0.834	42.03 (36.76, 47.31)
	Chronic (≥ 3 months)		2.994 (-2.833, 8.821)	3.048 (-1.672, 7.768)	0.207	45.45 (39.17, 51.72)
Prednisone /Phenobarbitone	No	Age‡, Sex-Neuter Status, ANKC Terrier, CT Confounder.	0.000	0.000	Reference category	40.32 (37.03, 43.62)
	Yes		4.425 (-3.297, 12.147)	5.231 (-0.867, 11.329)	0.094	45.55 (38.97, 52.13)
CT Confounding Findings	Normal	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	45.40 (39.79, 51.01)
	Abnormal		-6.143 (-10.256, -2.030)	-5.445 (-8.842, -2.049)	0.002	39.95 (35.12, 44.79)
Presence of Effusion	No	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	41.20 (36.25, 46.15)
	Yes		-2.354 (-7.035, 2.326)	-3.656 (-7.422, 0.110)	0.059	37.54 (31.74, 43.35)
Volume of Effusion		Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	-2.225 (-3.923, -0.528)	-2.318 (-3.682, -0.954)	< 0.001
Presence of Mass	No	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	42.41 (37.36, 47.45)
	Yes		2.154 (-2.017, 6.326)	-4.285 (-7.820, -0.749)	0.019	38.12 (32.78, 43.46)
Organomegaly	No	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	42.40 (37.06, 47.73)
	Yes		-5.040 (-9.237, -0.843)	-2.718 (-6.253, 0.816)	0.133	39.68 (34.56, 44.79)
Mass and Organomegaly	No	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000		43.83 (38.45, 49.20)
	Yes		-3.245 (-7.130, 0.640)	-4.255 (-7.480, -1.030)	0.010	39.57 (34.64, 44.51)
Small organ	No	Age‡, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	41.88 (36.92, 46.84)
	Yes		-15.901 (-20.677, -11.124)	-5.679 (-10.585, -0.773)	0.024	36.20 (29.91, 42.49)

^‡ Wherever age was entered into a model it was centred by its mean and entered in quadratic form as age, centred + (age, centred)²

Bold indicates statistical signficance (p < 0.05). CTV – Volume computed tomography; Abd. Fat – total abdominal fat percentage; SQ – Subcutaneous abdominal fat percentage; Visc. – Visceral abdominal fat percentage; V/SQ – Visceral-to-Subcutaneous Fat Ratio; LgV/SQ – log₁₀(V/SQ)

ANKC Australian National Kennel Council, BCS Body condition score

Table 2.

Putative causal diagram-guided logistic regression analysis outputs for variables associated with log of computed tomographic volume abdominal visceral-to-subcutaneous fat distribution (lgV/SQ) in 205 dogs presenting to a tertiary veterinary hospital between March 2006 and March 2020

Variable		Adjustment set (total effect)	Coefficients (β) (exponentiated) (95% CI)	Adjusted estimates (exponentiated) (95% CI)	p-value (adjusted)	Model estimated marginal means for levels (exponentiated) (95% CI)
Age (years) [V/SQ % per year]	Age	CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	1.008 (1.003, 1.013) [0.8% (0.3%, 1.3%)]	1.017 (1.011, 1.022) [1.7% (1.1. 2.2%)]	< 0.001
Sex	Females	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.000	0.000	Reference category	0.69 (0.60, 0.78)
	Males		1.025 (0.978, 1.073)	1.029 (0.987, 1.072)	0.184	0.73 (0.64, 0.83)
				1.025 (0.984, 1.069)	0.236
Neuter Status	Entire	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.000	0.000	Reference category	0.74 (0.63, 0.88)
	Desexed		0.985 (0.923, 1.050)	0.975 (0.920, 1.035)	0.410	0.70 (0.62, 0.79)
Sex-Neuter Status	Female entire	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.000	0.000	Reference category	0.63 (0.48, 0.82)
	Female spayed		1.085 (0.960, 1.226)	1.042 (0.932, 1.165)	0.470	0.69 (0.61, 0.79)
	Male entire		1.146 (1.001, 1.312)	1.105 (0.980, 1.247)	0.106	0.79 (0.66, 0.95)
	Male neutered		1.093 (0.969, 1.234)	1.059 (0.949, 1.182)	0.307	0.72 (0.63, 0.82)
ANKC Breed Group	Toys	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.000	0.000	Reference category	0.66 (0.52, 0.84)
	Terriers		1.153 (1.023, 1.301)	1.103 (0.989, 1.230)	0.080	0.83 (0.68, 1.00)
	Gundogs		1.026 (0.914, 1.152)	1.009 (0.909, 1.120)	0.865	0.67 (0.56, 0.81)
	Hounds		1.121 (0.976, 1.288)	1.129 (0.996, 1.279)	0.059	0.87 (0.69, 1.11)
	Working Dogs		0.980 (0.877, 1.095)	0.985 (0.891, 1.089)	0.763	0.64 (0.54, 0.75)
	Utility		1.019 (0.908, 1.144)	1.049 (0.945, 1.165)	0.366	0.74 (0.62, 0.89)
	Non-sporting		0.915 (0.799, 1.047)	0.945 (0.836, 1.068)	0.366	0.58 (0.46, 0.73)
	Mixed breed		1.055 (0.952, 1.169)	1.035 (0.943, 1.136)	0.468	0.71 (0.63, 0.81)
						0.72 (0.61, 0.85)
ANKC Terrier	Terriers	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.000	0.000	Reference category	0.83 (0.68, 1.00)
	Non-terriers		0.889 (0.822, 0.961)	0.928 (0.865, 0.997)	0.041	0.70 (0.62, 0.79)
ANKC Hound	Hounds	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.			Reference category	0.88 (0.69, 1.12)
	Non-hounds		0.920 (0.827, 1.024)	0.904 (0.823, 0.993)	0.037	0.70 (0.62, 0.79)
Skull Shape	Brachycephalic	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.942 (0.852, 1.043)	0.943 (0.861, 1.033)	0.208	0.65 (0.54, 0.78)
	Mesocephalic		1.016 (0.933, 1.106)	0.980 (0.906, 1.060)	0.614	0.71 (0.61, 0.81)
	Dolichocephalic		0.000	0.000	Reference category	0.74 (0.61, 0.90)
	Mixed breed		1.028 (0.944, 1.121)	0.983 (0.907, 1.065)	0.669	0.71 (0.62, 0.81)
Chondrodystrophy Status	Chondrodystrophic	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.000	0.000	Reference category	0.73 (0.61, 0.86)
	Non-chondrodystrophic		0.965 (0.903, 1.032)	0.979 (0.923, 1.040)	0.496	0.69 (0.60, 0.79)
	Mixed breed		1.002 (0.935, 1.074)	0.992 (0.932, 1.055)	0.792	0.71 (0.62, 0.81)
HWR		Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.837 (0.722, 0.969)	0.863 (0.755, 0.988)	0.034
Size Category (Nominal Categorisation with Cross Breed Excluded)		Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.983 (0.963, 1.003)	0.994 (0.975, 1.013)	0.512
Size Category	Extra Small	Age, CT Confounders, Confirmed Hyperadrenocorticism, Abdominal Adiposity.	0.000	0.000	Reference category	0.73 (0.52, 1.02)
	Small		1.090 (0.923, 1.287)	1.018 (0.877, 1.183)	0.813	0.76 (0.62, 0.93)
	Medium-small		1.008 (0.846, 1.201)	0.990 (0.846, 1.159)	0.902	0.71 (0.56, 0.90)
	Medium-large		0.999 (0.856, 1.167)	0.966 (0.841, 1.111)	0.630	0.67 (0.58, 0.79)
	Large		1.020 (0.870, 1.196)	0.978 (0.847, 1.129)	0.759	0.69 (0.58, 0.82)
	Giant		0.963 (0.817, 1.136)	0.978 (0.844, 1.134)	0.771	0.69 (0.57, 0.85)
	Mixed breed		1.041 (0.894, 1.212)	0.989 (0.862, 1.134)	0.872	0.71 (0.62, 0.81)
BCS (nominal)		Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	0.993 (0.975, 1.011)	0.982 (0.964, 1.000)	0.050
BCS (categorical)	2	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	0.000	0.000	Reference category	0.92 (0.65, 1.32)
	3		1.051 (0.894, 1.234)	1.005 (0.864, 1.169)	0.950	0.93 (0.73, 1.19)
	4		1.026 (0.882, 1.194)	0.963 (0.835, 1.111)	0.607	0.85 (0.69, 1.03)
	5		0.994 (0.856, 1.156)	0.926 (0.803, 1.068)	0.293	0.77 (0.63, 0.95)
	6		1.000 (0.852, 1.174)	0.907 (0.776, 1.059)	0.217	0.73 (0.58, 0.93)
	7		0.998 (0.844, 1.181)	0.911 (0.776, 1.071)	0.261	0.74 (0.57, 0.96)
	8		1.075 (0.871, 1.327)	0.965 (0.790, 1.179)	0.726	0.85 (0.59, 1.23)
	9		0.939 (0.735, 1.201)	0.899 (0.709, 1.140)	0.380	0.72 (0.44, 1.17)
Abdominal Adiposity		Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	0.997 (0.996, 0.999)	0.995 (0.993, 0.996)	< 0.001
CTV Mean Hounsfield Units		Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, CT Confounder	1.001 (1.000, 1.002)	1.002 (1.001, 1.003)	< 0.001
Disease Chronicity	Acute (< 3 months)	Age, Sex-Neuter Status, ANKC Terrier, Abdominal Adiposity, CT Confounder.	0.000	0.000	Reference category	0.63 (0.56, 0.70)
	Chronic (≥ 3 months)		1.008 (0.962, 1.056)	1.020 (0.976, 1.067)	0.374	0.66 (0.58, 0.74)
Disease Category	Endocrine	Age, Sex-Neuter Status, ANKC Terrier, Abdominal Adiposity, CT Confounders.	0.000	0.000	Reference category	0.72 (0.61, 0.86)
	Neoplasia		0.977 (0.906, 1.055)	0.949 (0.884, 1.018)	0.145	0.64 (0.57, 0.72)
	Gastroenteric		0.902 (0.801, 1.015)	0.920 (0.824, 1.027)	0.139	0.60 (0.48, 0.75)
	Trauma/Musculoskeletal		0.978 (0.803, 1.191)	1.009 (0.839, 1.212)	0.928	0.74 (0.49, 1.12)
	Congenital Portal Vascular Anomaly		0.942 (0.864, 1.028)	0.971 (0.880, 1.072)	0.561	0.68 (0.56, 0.81)
	Inflammatory		0.864 (0.787, 0.949)	0.920 (0.838, 1.010)	0.083	0.60 (0.50, 0.71)
	Other		0.884 (0.801, 0.976)	0.886 (0.809, 0.971)	0.010	0.55 (0.46, 0.66)
Confirmed Hyperadrenocorticism	No evidence of hyperadrenocorticism	Age, Sex-Neuter Status, ANKC Terrier, Abdominal Adiposity, CT Confounder.	0.000	0.000	Reference category	0.63 (0.57, 0.70)
	Confirmed Hyperadrenocorticism		1.171 (1.048, 1.308)	1.174 (1.061, 1.299)	0.002	0.92 (0.71, 1.18)
Drug Length	No drugs	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism, Abdominal Adiposity, CT Confounder	0.000	0.000	Reference category	0.77 (0.65, 0.90)
	Acute (< 3 months)		1.009 (0.959, 1.061)	1.004 (0.960, 1.051)	0.848	0.77 (0.66, 0.91)
	Chronic (≥ 3 months)		0.968 (0.903, 1.038)	0.979 (0.919, 1.042)	0.502	0.73 (0.60, 0.88)
Prednisone /Phenobarbitone	No	Age, Sex-Neuter Status, ANKC Terrier, Abdominal Adiposity, CT Confounder.	0.000	0.000	Reference category	0.64 (0.58, 0.71)
	Yes		0.945 (0.863, 1.036)	0.983 (0.905, 1.069)	0.693	0.62 (0.50, 0.76)
CT Confounding Findings	Normal	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism.	0.000	0.000	Reference category	0.74 (0.61, 0.89)
	Abnormal		1.048 (0.998, 1.101)	1.027 (0.979, 1.077)	0.283	0.78 (0.67, 0.92)
				1.027 (0.979, 1.077)	0.283
Presence of Effusion	No	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	0.77 (0.65, 0.90)
	Yes		1.021 (0.966, 1.080)	1.016 (0.963, 1.071)	0.567	0.80 (0.66, 0.96)
				0.997 (0.949, 1.048)	0.913
Volume of Effusion		Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	1.017 (0.997, 1.038)	1.018 (0.999, 1.038)	0.069
				1.018 (0.999, 1.038)	0.069
Presence of Mass	No	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	0.77 (0.65, 0.91)
	Yes		1.036 (0.987, 1.089)	1.003 (0.954, 1.055)	0.893	0.78 (0.65, 0.93)
Organomegaly	No	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	0.74 (0.63, 0.89)
	Yes		1.049 (0.998, 1.102)	1.030 (0.980, 1.081)	0.246	0.80 (0.68, 0.94)
Mass and Organomegaly	No	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	0.74 (0.62, 0.89)
	Yes		1.058 (1.011, 1.107)	1.026 (0.981, 1.074)	0.266	0.79 (0.67, 0.93)
Small organ	No	Age, Sex-Neuter Status, ANKC Terrier, Confirmed Hyperadrenocorticism	0.000	0.000	Reference category	0.75 (0.64, 0.88)
	Yes		1.018 (0.957, 1.083)	1.077 (1.008, 1.150)	0.029	0.89 (0.73, 1.09)

Bold indicates statistical significance (p < 0.05). CTV – Volume computed tomography; Abd. Fat – total abdominal fat percentage; SQ – Subcutaneous abdominal fat percentage; Visc. – Visceral abdominal fat percentage; V/SQ – Visceral-to-Subcutaneous Fat Ratio; LgV/SQ – log₁₀V/SQ

ANKC Australian National Kennel Council, BCS Body condition score

The following results are presented with adjustments made for age (linear or quadratic, as stated), total abdominal adiposity, sex-neutered status, ANKC terrier breed, cHAC, and CT confounders, as applicable. Further details regarding the relationship between BCS and the total abdominal adiposity and fat distribution are provided below.

After adjusting for confounders, total abdominal adiposity was associated with four variables: age, ANKC terrier breed, BCS, and disease category (see Table 1). Total abdominal adiposity showed a quadratic association with age, with higher values observed in middle-aged dogs, peaking at approximately 10 years, and lower values observed in dogs older than this age (J-shaped distribution) (see Fig. 1). Additionally, abdominal adiposity was lower in terrier breeds, and it varied with disease categories. On average, dogs with neoplasia had the lowest abdominal adiposity.

Fig. 1.

Scatter plots showing pairwise comparison and quadratic relationship between age and computed tomographic volume (CTV) total abdominal fat percentage (A), age and CTV abdominal visceral fat percentage (B), age and CTV abdominal subcutaneous (SQ) fat percentage (C), and age and log of CTV abdominal visceral-to-subcutaneous fat distribution (log V/SQ) (D) in 205 dogs presenting to a tertiary veterinary hospital between March 2006 and March 2020. Solid lines represent model-estimated relationships, and dashed lines represent the corresponding 95% confidence intervals

Abdominal visceral fat percentage was associated with six variables after adjusting for confounders: age, HWR, BCS, total abdominal adiposity, disease category, and cHAC (see S3). The visceral fat percentage increased with increasing total abdominal adiposity. The relationship between abdominal visceral fat percentage and age displayed a limited quadratic pattern, with a gradual plateau around 13 years of age. Based on this analysis and the scatter plot, the visceral fat and age relationship was considered linear for the multivariable analysis (see Fig. 1). Deep-chested dogs, or those with an increasing HWR, exhibited decreasing visceral fat percentages. Dogs with cHAC displayed higher visceral fat percentages than dogs without cHAC, and dogs with inflammatory or “other” diseases had lower visceral fat percentages, on average, than other disease categories.

After adjusting for confounders, SQ fat was associated with six variables: age, HWR, BCS, total abdominal adiposity, disease category, and cHAC (see S4). The SQ fat percentage increased with increasing total adiposity. Subcutaneous (SQ) fat percentage showed a quadratic association with age, with higher values observed in middle-aged dogs, peaking at approximately 8–10 years, and lower values observed in older dogs (J-shaped distribution) (see Fig. 1). Deep-chested dogs, based on increasing HWR, were associated with increasing SQ fat percentages. Dogs with HAC displayed lower SQ fat percentages than dogs without HAC. In contrast, dogs with inflammatory or “other” diseases had the highest average SQ fat percentages.

After adjusting for confounders, the log(V/SQ) showed associations with eight variables: age, ANKC hounds, ANCK terrier, HWR, BCS, total abdominal adiposity, disease category, and cHAC. Increasing BCS and total abdominal adiposity were associated with decreasing V/SQ (see Fig. 2), while higher V/SQ values were observed with increasing age. Deep-chested dogs had lower V/SQ, although this relationship was not evident in ANKC hounds, which had higher V/SQ than non-hounds. Terriers and dogs with cHAC also favoured visceral fat distribution over SQ fat distribution. Dogs with “other” diseases had, on average, the lowest V/SQs of any group, but no clear pattern was identified within this group. The “other” diseases category included respiratory, neurological, urinary, haematological, and non-neoplastic mass conditions.

Fig. 2.

Scatter plots showing pairwise comparison and linear relationship between CT volume (CTV) total abdominal fat percentage and BCS (A), L3 CT area (L3 CTA) total abdominal fat percentage and BCS (B), CTV visceral-to-subcutaneous fat ratio (CTV V/SQ) and BCS (C), and CTV V/SQ and CTV total abdominal fat percentage (D) in 205 dogs presenting to a tertiary veterinary hospital between March 2006 and March 2020. Solid lines represent model-estimated relationships, and dashed lines represent the corresponding 95% confidence intervals

Validity of methods predicting abdominal adiposity and fat distribution

Tables 3 and 4; Figs. 5 and 6 display the validation of L3 CTA, mean Hounsfield units (HU), and ultrasound fat indices to predict CTV fat indices.

Table 3.

Validity of L3 computed tomographic area (CTA) measures of abdominal fat indices relative to their corresponding computed tomographic volume (CTV) measures

	r	rc lower 95% CL	B-A Difference Mean (SD)	B-A Difference 95% LOA	B-A Percentage Difference Mean (SD)	B-A Percentage Difference 95% LOA
L3 CTA total abdominal fat % relative to CTV total abdominal fat %	0.971	0.919	-3.6 (4.3)	-12.0, 4.9	− 6.5% (10.4)	-26.9, 13.8%
L3 CTA visceral fat % relative to CTV visceral abdominal fat %	0.931	0.566	-6.9 (4.6)	-16.0, 2.2	-32.3% (18.8)	-69.1, 4.5%
L3 CTA subcutaneous fat % relative to CTV subcutaneous abdominal fat %	0.971	0.900	3.3 (2.4)	-1.4, 8.1	15.3% (10.4)	-5.1, 35.8%
L3 CTA V/SQ relative to CTV V/SQ	0.893	0.425	-0.4 (0.3)	-1.1, 0.2	-46.5% (20.6)	-86.9, -6.1%

r = Pearson’s correlation

r_c= Lin’s Concordance Correlation Coefficient

B-A Difference mean = mean difference [CTV – CTA values] versus the average of both values

B-A Percentage Difference Mean = mean of the percentage difference [100*(CTV – CTA values)/average)] versus the average of both values

(SD) = standard deviation

95% LOA = 95% Limits of Agreement

95% CL – lower 95% confidence limit

L3 CTA – transverse cross-sectional computed tomographic area at the cranial margin of the third lumbar vertebrae on CT

CTV - computed tomographic volume of the abdomen from the cranial margin of the 10th thoracic to the cranial margin of first sacral vertebrae

V/SQ – visceral-to-subcutaneous fat ratio

Table 4.

Linear regression statistics of mean Hounsfield units and ultrasound measures to estimate computed tomographic volumetric (CV) measures of abdominal fat

	r	Regression Equation	Gradient 95% Confidence Interval	y-intercept 95% Confidence Interval
CTV mean HU relative to CTV total abdominal fat %	-0.966	y = -1.843x + 73.29	-1.911, -1.774	70.18, 76.40
L3 CTA mean HU relative to CTV total abdominal fat %	-0.948	y = -1.954x + 67.01	-2.044, -1.863	62.89, 71.14
TAT relative to CTV total abdominal fat %	0.575	y = 0.639x − 3.666	0.510, 0.767	-9.557, 2.225
VAT relative to CTV visceral abdominal fat %	0.440	y = 0.896x + 2.382	0.637, 1.155	-2.126, 6.890
SAT relative to CTV subcutaneous abdominal fat %	0.444	Y = 0.223x + 1.220	0.159, 0.286	-0.624, 3.064
VAT/SAT relative to CTV V/SQ	0.257	y = 2.473x + 1.428	1.158, 3.787	0.509, 2.346

r = Pearson’s correlation

95% LOA = 95% Limits of Agreement

HU – Hounsfield units

L3 CTA – transverse cross-sectional computed tomographic area at the cranial margin of the third lumbar vertebrae on CT

CTV - computed tomographic volume of the abdomen from the cranial margin of the 10th thoracic to the cranial margin of first sacral vertebrae

V/SQ – visceral-to-subcutaneous fat ratio

VAT – linear measure of the visceral adipose thickness measured on and perpendicular to the linea alba from the linea alba’s parietal surface to the ventral margin of the most caudal aspect of liver that crosses midline

SAT – linear measure of the subcutaneous adipose thickness at the same level of VAT

VAT/SAT – VAT-to-SAT ratio

TAT – linear measure of the total adipose thickness at the same level of VAT (VAT + SAT)

Fig. 5.

Transverse CT images illustrating various abdominal adiposity patterns and visceral-to-subcutaneous fat distribution/ratio (V/SQ) in a tertiary veterinary hospital population of dogs. The images are arranged from left to right, indicating increasing total abdominal fat adiposity, and from bottom to top, representing increasing V/SQ. The four patients depicted are as follows: (A) 1.5-year-old female-spayed mixed breed with a left divisional intrahepatic portocaval shunt, having 14% total abdominal fat and a V/SQ of 1.35; (B) 11-year-old male-neutered mixed breed undergoing staging for soft tissue sarcoma of the left thoracic limb, exhibiting 52% total abdominal fat and a V/SQ of 1.22; (C) 1.4-year-old female-spayed Boxer diagnosed with intestinal adenocarcinoma, displaying 28% total abdominal fat and a V/SQ of 0.379; (D) 7-year-old male-neutered Golden Retriever undergoing staging for insulinoma, demonstrating 57% total abdominal fat and a V/SQ of 0.287

Fig. 6.

Transverse and sagittal CT reconstruction of a dog’s abdomen showing the location of the abdominal height and width (hashed lined), the visceral adipose thickness (VAT), and subcutaneous adipose thickness (SAT) (solid lines) measurements. The cranial abdominal height and width were measured at the widest internal diameter of the cranial abdomen constrained within the costal arch and used to determine the cranial abdominal height-width ratio (HWR). The VAT and SAT were performed perpendicular to the parietal surface of the linea alba, aligning with the most caudal margin of the liver at the midline

In univariable analysis, BCS had poor to moderate correlation with CTV total abdominal fat percentage (r = 0.604), CTV visceral fat percentage (r = 0.509), CTV SQ fat percentage (r = 0.529), L3 CTA total abdominal fat percentage (r = 0.605), L3 CTA visceral fat percentage (r = 0.514), and L3 CTA SQ fat percentage (r = 0.476). There was marked variability of CTV total abdominal adiposity for every BCS unit, with, at best, 36% of the variation in adiposity being explained by BCS (r² = 0.365, [95%CI:0.231,0.529]) and for every unit increase in BCS, the CTV total abdominal adiposity increased by 5.76% (95% CI: 4.60, 6.93) (see Fig. 2) and L3 CTA abdominal fat changed by 6.62% (95% CI:5.29, 7.96). Additionally, BCS had a poor, negative correlation with V/SQ (r=-0.066) when not correcting for signalment or hyperadrenocorticism. There was a moderate negative correlation between CTV and L3 mean HU with BCS (r=-0.638 and r=-0.626, respectively). See Tables 1 and 2 and Supplemental Files 3 and 4 for further analyses.

Overall, L3 CTA abdominal fat measures showed consistent biases relative to the CTV measures, generally overestimating total abdominal fat and visceral fat percentages but underestimating SQ fat percentages. The variations worsened with increasing fat percentages (drift), most significantly observed in visceral fat percentages. This drift resulted in a more significant disparity in the L3 CTA V/SQ estimation in overweight/obese dogs (see Table 3; Fig. 3).

Fig. 3.

Paired plots comparing CT cross-sectional area–derived abdominal fat measurements obtained at the cranial margin of L3 with equivalent CT-derived volumetric (CTV) measurements between the cranial margins of T10 and S1 in 205 dogs presenting to a tertiary veterinary hospital between March 2006 and March 2020. For each comparison, the left panel shows Lin’s concordance correlation plot, and the right panel shows the corresponding Bland–Altman plot. A - Total abdominal fat (bias = -6.5%, 95% LOA: -26.9%; 13.8%); B - visceral abdominal fat (bias = -32.3%, 95% LOA: -69.1%; 4.5%); C - subcutaneous abdominal fat (bias = 15.3%, 95% LOA: -5.1%; 35.8%); and D - visceral-to-subcutaneous fat distribution (V/SQ) (bias = -46.5%, 95% LOA: -86.9%; -6.1%). In the Lin’s concordance plots, the solid line represents the fitted line of agreement with the dashed lines indicating the 95% confidence interval, and the dotted line represents the line of identity (1:1 agreement). In the Bland–Altman plots, the solid line represents the mean percentage difference (bias), and the dashed lines represent the 95% limits of agreement (LOA)

The CTV and the L3 CTA mean HU strongly and negatively correlated with the overall abdominal fat percentage (r=-0.966 and r=-0.948, respectively). Ultimately, the CTV mean HU could explain more of the variance than the L3 CTA mean HU (see Table 4; Fig. 4).

Fig. 4.

Scatter plots showing correlations between alternative measures of abdominal adiposity and CT-derived volume (CTV) reference measures in 205 dogs presenting to a tertiary veterinary hospital between March 2006 and March 2020. A - Mean Hounsfield units (HU) recorded within the entire abdominal volume (CTV Mean HU) compared to the CTV total abdominal fat percentage. B - Mean Hounsfield units (HU) recorded within the entire transverse abdominal area measured at the cranial margin of L3 (L3 Mean HU) compared to the CTV total abdominal fat percentage. C – Ultrasound total adipose thickness (TAT) compared to CTV total abdominal fat percentage. D – Ultrasound visceral adipose thickness (VAT) compared to CTV visceral fat percentage. E - Ultrasound subcutaneous adipose thickness (SAT) compared to CTV subcutaneous fat percentage. F - Ultrasound VAT/SAT compared to CTV visceral-to-subcutaneous fat distribution (CTV V/SQ). The abdominal volume was defined as tissue volume between the cranial margins of T10 and S1, with tissue thresholded at − 250 to 2000 HU and fat at − 250 to − 25 HU. VAT and SAT were measured perpendicular to the linea alba at the most caudal aspect of the liver at the midline. Solid lines represent model-estimated relationships, and dashed lines represent the corresponding 95% confidence intervals

The linea alba fat measurements showed poor to moderate correlation with the corresponding CTV fat indices (see Table 4; Fig. 4), while TAT provided a moderate correlation with total abdominal adiposity. The poor association of VAT and SAT measurements to CTV visceral and SQ fat measures resulted in a weak correlation between VAT/SAT and V/SQ (r = 0.257). Total adipose thickness (TAT) correlated poorly with BCS (p = 0.42).

Reliability of abdominal fat segmentation

Fat segmentation using the − 250/-25 HU thresholding extracted fat with a mean of -99.40 HU (SD ± 11.10 HU, [95% CI: -100.90, -97.88HU) and median of -102.00 HU (range: -118.10 to − 67.40 HU), with a slightly positive skew (skewness = 0.79 ± 0.17).

There was minimal variation when the CTV measurements were repeated, with a near-perfect intra-observer agreement in the segmentation of total abdominal volume (− 250/-25HU: r_c lower 95% CL = 0.998, B-A 95% LOA [− 3.5, 3.9%]); total abdominal fat (− 250/-25HU: r_c lower 95% CL = 0.999, B-A 95% LOA [− 4.0, 4.3%]); and visceral abdominal fat (− 250/-25HU: r_c lower 95% CL = 0.996, B-A 95% LOA [− 7.2, 2.5%]) when repeated after more than six months of performing the original measurements. The visceral abdominal fat was slightly overestimated on the repeat measurement (bias ~ 2.3%) and resulted in a near-perfect intra-observer agreement, but a slightly overestimated repeat V/SQ measurement (V/SQ: r_c lower 95% CL = 0.990, B-A 95% LOA [− 9.8, 1.5%]).

The intra-abdominal CTA measurements varied more than CTV when repeated more than six months after the original measurements. There was near perfect intra-observer agreement in the segmentation of L3 abdominal volume (− 250/-25HU: r_c lower 95% CL = 0.999, B-A 95% LOA [− 1.7, 2.0%]) and L3 abdominal fat (− 250/-25HU: r_c lower 95% CL = 0.998, B-A 95% LOA [− 5.8, 7.3%]). However, despite a near-perfect intra-observer agreement for L3 visceral fat, there was greater variability (− 250/-25HU: r_c lower 95% CL = 0.992, B-A 95% LOA [− 15.8, 13.4%]). This resulted in relatively variable L3 V/SQ calculations (− 250/-25HU: r_c lower 95% CL = 0.940, B-A 95% LOA [− 47.0, 32.2%]).

Discussion

To the authors’ knowledge, this is the first study to evaluate the relationship between signalment and disease on CTV abdominal fat distribution in dogs. This study, using abdominal adiposity as a proxy for total adiposity, is consistent with previous associations found between adiposity, age, breed, and certain diseases [34]. Additionally, it confirms the correlation between V/SQ and age, as established in earlier research [3, 8, 13, 27, 34]. Notably, the findings also introduce a novel relationship between V/SQ, abdominal adiposity (contrary to existing literature), and body conformation (HWR) [27]. Our results further support the supposition that dogs with HAC preferentially distribute fat to the visceral space [13, 35]. These findings emphasise the importance of shifting our focus from general adiposity to a more nuanced examination of the relationship and causal direction of regional fat distribution on health, metabolic dysfunction, and disease pathogenesis [18].

Age

Despite the heterogeneity of the data set, abdominal adiposity displayed a quadratic distribution pattern (J-shape) with age, consistent with the trends observed in total adiposity in dogs [5, 36, 37]. Notably, abdominal fat increases until around 10 years of age and subsequently decreases [5]. Moreover, our study affirms that ageing is also linked to the redistribution pattern of adiposity [27]. Specifically, there is a preferential rise in visceral fat compared to the quadratic distribution of SQ fat, resulting in an overall increase in the V/SQ with age [38]. This preferential distribution of fat to the visceral space is considered metabolically detrimental. In people, age is one of the leading contributors to visceral redistribution and the development of adipose tissue dysfunction [39, 40]. Human age-related adipose tissue dysfunction is a complex interaction between adipose cell differentiation and senescence, immune cell infiltration, and the release of proinflammatory cytokines [39]. This intricate interplay ultimately leads to the progressive impairment of the SQ adipose tissues to store lipids (“triglyceride sink”), redistributing fat to visceral and ectopic regions and chronic low-grade inflammation. While there is some debate within the veterinary literature, insulin resistance has been associated with increasing visceral mass, suggesting a similar mechanism may occur in dogs [18, 41–43]. Consequently, the pro-inflammatory cytokines are associated with insulin resistance, as well as endothelial, metabolic, and inflammatory dysregulation, and ultimately contribute to the pathophysiology of several diseases [18]. This increasing V/SQ, with decreasing total and SQ fat, may be associated with geriatric-related morbidities and reduced lifespan in dogs, but further evaluation is required [44].

Sex and neuter status

Although no statistically significant association was identified, neutered dogs tended to have higher total abdominal adiposity, consistent with established associations between desexing and overweight and obesity [45, 46]. In contrast, sex and neuter status were not significantly associated with visceral or subcutaneous fat distribution, in agreement with previous work [27]. Entire male dogs showed a tendency toward greater visceral fat distribution compared with entire females, although this did not reach statistical significance.

These trends may reflect the influence of sex hormones on fat distribution, as described in people, where testosterone is associated with greater visceral fat deposition and oestrogen with preferential subcutaneous fat storage [47, 48]. In this study, male-neutered and female-spayed dogs exhibited intermediate visceral-to-subcutaneous fat distribution between intact males and females, consistent with prior canine studies [5, 30, 38, 46, 48]. However, the limited number of intact dogs and the absence of data on age at neutering restrict robust inference, and further investigation of hormonal influences on fat distribution in dogs is warranted.

Breed and breed conformation

Specific breeds are predisposed to being OO, and it has been recently alluded that certain toy breeds may favour visceral fat distribution compared to SQ fat [5, 30]. The influence of specific breeds and breed conformations on abdominal adiposity and fat distribution was identified in terriers, and with the abdominal height-width ratio (HWR) conformation. Terriers had lower abdominal adiposity than other breeds, as previously demonstrated, and favoured a distribution to the visceral over the SQ compartment [8, 49]. Hounds appeared to favour distribution to the visceral over the SQ compartment, despite having relatively high HWR. These findings may reflect breed-specific differences, such as an increase in lean muscle mass in terriers that may limit the risk of obesity seen with desexing and may suggest a breed-protective trait from obesity [49, 50]. In our study, HWR reflected breed conformation well, and its use in CT is the first time this has been described [9]. Dogs with deep, narrow conformations (increasing HWR) typically favoured reducing total adiposity with SQ over visceral fat distribution, although, as stated, this was not reflected in the Hounds group. However, the dolichocephalic dogs and a few Great Danes probably exaggerated this effect and skewed the data. These variations warrant further investigation, as they may reflect or affect breed-specific body conformation and standards, and may contribute to the predisposition for obesity-related disorders observed in certain breeds.

Visceral-to-subcutaneous fat relative to adiposity

The relationship between V/SQ and total abdominal adiposity differed from our previous findings, likely due to a more significant power in this study [22]. This relationship has been noted previously and is consistent with the anticipated physiological orthotopic fat distribution pattern, with a preference for the SQ space. It is crucial to emphasise that the observed relationship or effect is relatively weak, a factor that is likely diminished with advancing age [25, 31, 38, 39]. The low contribution of BCS and total abdominal adiposity to the variation in V/SQ demonstrates that focusing on measures of total adiposity is not a sensitive or specific means of evaluating the effect of fat distribution on health (Figs. 2 and 5). This is particularly relevant to BCS, which has a stronger positive relationship with subcutaneous fat mass than visceral fat mass [51].

Similarly, in people, there is now a growing consensus to focus on measures of central obesity, specifically visceral obesity, such as waist circumference, more so than metrics of total body obesity, such as BMI [23]. These metrics of visceral adiposity have been shown to provide both independent and additive information for predicting morbidity and risk of death in people and are a means of stratifying metabolically unhealthy from metabolically healthy obesity. Additionally, substantial evidence emphasises the importance of considering lean muscle mass in conjunction with adiposity [31, 50]. Sarcopenic obesity appears to have a more profound impact than obesity accompanied by adequate muscle mass. This underscores the necessity for research methods that effectively assess fat distribution and lean muscle mass, shedding light on the influence of compartmental fat distribution on overall health.

Fat distribution in OO dogs may contribute to the variations in their health metrics seen in the literature [41, 52]. Some studies have found that not all OO dogs show clinical metabolic issues, but small subgroups exhibit obesity-related metabolic dysfunction (ORMD) [52–54]. Although not evaluated, this heterogeneous metabolic outcome related to OO may be attributable to regional fat distribution and relative muscle mass. Subcutaneous fat deposition may be “protective” and, combined with greater muscle mass, as seen in younger OO dogs, may limit metabolic derangement and the manifestation of ORMD [31, 55]. Morphological methods of assessing obesity, like BCS, do not assess fat redistribution patterns well and thus limit their applicability in exploring these metabolic subsets in dogs [18, 31].

Disease and fat distribution

The effect of the hypothalamic-pituitary-adrenal (HPA) axis, particularly HAC, on both generalised and visceral obesity has been described [3, 32, 33]. However, to the authors’ knowledge, this is the first objective evaluation to demonstrate the preferential distribution of visceral fat in dogs with HAC. Interestingly, HAC was not related to total abdominal adiposity in this study. The predisposition for general obesity in dogs with HAC is likely due to polyphagia and increased caloric intake, coupled with reduced energy expenditure, but is not a strongly recognised component of the disease, which is supported in our findings [32]. However, an increase in visceral fat is commonly noted in dogs with HAC, although compounded by the concurrent hepatomegaly and weakening of the abdominal muscles. This is supported by the documented increase in serum leptin concentrations in dogs with HAC, with leptin preferentially released by visceral fat [33]. This has potential diagnostic application from a diagnostic imaging perspective, particularly AI-generated radiomics, as it may provide differential diagnosis stratification based on physiological manifestations, such as differentiating functional adrenocortical adenocarcinoma from phaeochromocytomas. This may be further augmented by using other measures of body composition, such as lean tissue mass and distribution.

Dogs with neoplasia tended to have lower total adiposity than dogs with other disease categories. However, the effect of diseases on fat distribution was uncertain in our study, apart from HAC. The category of “other” disease was a random assortment of diseases, and no clear pattern emerged. Generally, endocrinopathies tended to favour higher visceral fat and V/SQ than inflammatory and neoplastic diseases, likely skewed by the HAC cases. Further, this may be compounded by organ steatosis, such as hepatic lipidosis, which may increase the visceral adiposity on CT by reducing the Hounsfield units of the liver [56]. This was seen in one case of a dog that presented with diabetic ketoacidosis and hepatic lipidosis, which reduced the liver attenuation to -45HU. Thus, with more specific clinical questions, more specific segmentation, and the support of deep neural networking, V/SQ and body composition analysis may be used to provide differential diagnosis stratification [57]. Ultimately, body composition and tissue distribution may determine health-associated outcomes in dogs, like in people [39].

Congenital portal vascular anomalies formed a large cohort of young dogs and contributed to the bimodal age distribution. This was due to the high intravascular interventional radiology caseload at the hospital. Typically, these dogs had the lowest total, visceral, and SQ fat percentages, which resulted in marked variations in V/SQ. However, these findings were not present after correcting for age and the presence of ascites, as seen in cases of portal venous hypertension. Further studies evaluating young dogs without portosystemic shunts should be performed to ensure this was not a significant contributor to the age-related variation in V/SQ seen in this study.

Finally, the administration of phenobarbitone and prednisolone did not statistically significantly alter total abdominal adiposity, but there was a trend toward increasing adiposity. This is expected due to the side effect of polyphagia secondary to these medications and should be considered in future investigations [3].

Reliability

A significant consideration in using CT segmentation is the threshold values set for segmenting specific body tissues and the presence of confounding diseases that may alter the attenuation of the regions of interest. Regarding the CT threshold values, the authors used the previously published ranges of -250/-25 HU for fat [34]. This has shown a slightly higher correlation and agreement with DXA measure of fat than Ishoika’s -135/-105 HU threshold, and is similar to that used in cat and human thresholds [24]. The distribution of fat HU in this paper was, on average, higher than that proposed by Ishioka. This might suggest that more accurate measures of fat thresholding may be achieved with a narrower but slightly higher threshold value (97% CI for fat attenuation was − 101/-98 HU) than that used in this study and by Ishioka, but our measurements were not standardised against a CT quality assurance phantom.

The CT confounding variables, ascites, abdominal masses, organomegaly, and small organs, significantly influenced the results. This was addressed and corrected within the multivariable analysis of this paper. However, these confounding variables should be considered when using CT to research body composition. Interestingly, small organs were associated with greatly reduced total, visceral, and SQ fat percentages with increased V/SQ and likely reflected the cohort of young dogs with congenital portal vascular anomalies [58]. Effusions, abdominal masses, and organomegaly was associated with a decrease in total abdominal adiposity. Surprisingly, these conditions had no impact or even increased the V/SQ. This was attributed to reductions in both visceral and subcutaneous fat percentages, with a more pronounced decrease in subcutaneous fat. Although the precise cause of this phenomenon remains unclear, this relationship suggests that the underlying disease has a more significant influence on fat distribution than the soft tissue alterations that obscure fat on CT scans. This implies that CT scans might offer a reliable method for evaluating body composition and tissue distribution in the presence of significant abdominal pathology, although further research is necessary to understand these findings fully.

Body condition scoring was moderately correlated with total abdominal adiposity and, as previously demonstrated, showed a stronger relationship with SQ fat distribution than visceral adiposity [38, 51]. However, the weak negative correlation between total abdominal adiposity and V/SQ was weakly replicated in dogs with increasing BCS. Like human body mass index (BMI), BCS may be a crude marker of dog adiposity [23]. This study found that BCS only moderately correlated with volume and CTA L3 total abdominal fat percentages, differing from a study of 38 beagles, which found BCS had a strong correlation with the total adipose area at L3 (r = 0.809) [25]. This difference may be due to the 20% of cases missing BCS data in our study, the heterogeneity of our study population, differing methodologies, or breed-specific differences in fat deposition at L3. Further, in our study, the BCS was performed by a non-standardised set of clinicians, and the considerable variation in adiposity for each BCS unit likely highlights the high BCS inter-observer variation, as seen in other studies [4, 59]. However, our study further validated Laflamme’s original estimates that an increase in 5% adiposity occurs with every unit of BCS [4].

Most abdominal fat distribution papers use a single slice area of the abdomen at L3 or L5 [18, 24, 30, 31, 38]. It has also been shown that visceral fat deposition occurs mainly in the cranial abdomen (L3), while subcutaneous deposition is more caudally distributed (L5) [30, 51]. We showed that CTA L3 fat distribution showed more significant variation than CTV measures and that CTA generally overestimated total abdominal and visceral fat while underestimating SQ fat percentages compared to CTV measures. This bias worsened with increasing abdominal adiposity, resulting in marked variation in V/SQ measures. Adolphe’s method also identified similar measurements and reduced variation, which used averaged visceral and subcutaneous fat areas over multiple slices from the thirteenth thoracic vertebra to the seventh lumbar vertebra [18, 31]. This is expected, as the area measured highly depends on what is within the acquired abdominal CT slice. Conformational variation, relatively mobile organs (e.g., spleen), ectopic organs, and local pathology (e.g., hepatomegaly, ascites, masses) are some factors that may have a more significant influence on the CTA L3 measure compared to CTV measures of adiposity and fat distribution in dogs. Further, the gastric volume and the degree of urinary bladder distension may displace and reduce the visceral fat volume. For this reason, the authors recommend using volume metrics to assess abdominal adiposity and fat distribution for greater accuracy. However, in doing so, understanding the influence on specific fat distributions, such as peri-gluteal subcutaneous fat deposition, is reduced.

Additionally, a potential negative of using CTV methodology as a clinical tool is the time it takes to segment regions of interest and the limited access to semi-automated and automated software assistance. Thus, L3 CTA may still prove a clinically valuable tool for assessing adiposity and fat distribution. Further, the average Hounsfield units were assessed as a marker of overall abdominal adiposity. A radiologist can perform this measure in seconds and provide a reasonable estimate of overall body adiposity (r=-0.967). Further research may show abdominal attenuation values as a simple clinical or radiomic measure of abdominal obesity.

As there is limited access to CT in many veterinary clinics, the authors were also interested in further validating their recently introduced ultrasound method of assessing fat distribution [27]. This would provide a more accessible tool for assessing dog fat distribution, as no sedation or anaesthesia is required. Unfortunately, although a moderate correlation between TAT and total abdominal fat percentage was identified, only a poor correlation was observed between VAT and SAT relative to CTV visceral and SQ fat, and ultimately, a weak correlation between VAT/SAT and V/SQ. The flattening of the ventral abdomen due to sternal recumbency and variable liver size may have influenced the measurements, but this was likely negligible. Although the falciform ultrasound method may help assess total adiposity, the current methodology is not useful for determining V/SQ.

Although the relationship between total body adiposity and total abdominal adiposity was supported by the moderate correlation between total abdominal adiposity and BCS, further validation of abdominal CT and, ideally, whole-body CT metrics of body composition is required. Additionally, this study did not consider the specific regional distribution of fat, such as thoracolumbar, lumbar, inguinal, or peri-gluteal SQ fat distribution, or cranial and caudal visceral fat distribution. Further, it did not evaluate organ fat distribution, such as intracellular (steatosis, e.g., hepatic lipidosis), intra-organ (e.g., intramuscular), or between organs (e.g., intermuscular) fat distribution, nor did it evaluate the relationship between fat and lean muscle mass distribution. All these factors may be related to specific health-associated metrics that must be evaluated. For example, fat distributed around pre-menopausal women’s glutei and legs is associated with positive health metrics compared to central obesity in men [39]. Further, lean tissue mass favours positive health outcomes compared to sarcopenic obesity [50]. Therefore, CT offers a greater means of quantifying body composition metrics and should be evaluated against biomarkers of health, associated morbidities, quality of life, and risk of death.

Limitations

The retrospective and cross-sectional design limit the ability to infer causal relationships between fat distribution and disease. The use of a tertiary referral population may also introduce selection bias and limit generalisability to the broader canine population. Missing BCS data may have reduced the strength of associations involving body condition. Ultimately, prospective evaluation of specific factors affecting fat distribution should be performed, and the current findings provide a framework for future investigations. Ideally, greater numbers of different breeds and morphologies would have been evaluated, but this was limited by case availability and the time for data retrieval and tissue segmentation. Some methodological improvements that could be entertained in future research include the potential accuracy and speed of computer-aided segmentation and the reduced variation introduced by standardised protocols, such as the length of fasting before the study, size of urinary bladder at the time of the study, and standardised positioning, all of which were optimised for the presenting complaint in this study. Every animal usually undergoes a minimum 12-hour fast before CT, although this was not defined, and there were various degrees of gastric distension within the study population. The influence of these physiological variations is unknown, but is likely to affect volume-based metrics less than single-area measurements. Sternal recumbency was used in all these cases due to its standard use in clinical practice, as it limits the effect of respiratory movement on organs; however, its effect on body composition assessment is unknown [60]. Another methodological restraint for this study was that the same scales used to weigh each dog were not tared daily. Finally, including dogs less than a year old in our research sample may have introduced the confounding effect of brown adipose tissue. However, as the distribution, extent, and regression of brown fat in dogs are poorly characterised, young dogs were not excluded from our analysis [61]. As brown adipose tissue may appear like fat or soft tissue on CT, further evaluation of brown adipose tissue may require F18-FDG PET-CT, and possibly MRI, to differentiate it from white adipose tissue [62, 63].

Conclusions

The main finding of this study supports the association between abdominal adiposity, age, breed category, and certain diseases. Moreover, it highlights the correlation between V/SQ fat distribution, age, and total adiposity, while emphasising the preferential distribution of fat to the visceral compartment in dogs with HAC. The study also identified a novel association between V/SQ fat distribution, specific breed categories, and body conformation (HWR). Additionally, CT volumetric measures appear more reliable in determining abdominal fat distribution than area and linear measures.

Methods

Case selection

Medical case records of dogs that underwent abdominal computed tomography (CT) at the U-Vet Werribee Animal Hospital, University of Melbourne, were retrieved from March 2006 to March 2020. Data collated included weight, breed, age, sex, neuter status, 9-point scale BCS, and reason for the CT.

Abdominal fat analysis

Non-contrast volume acquisition of the abdomen with dogs in sternal recumbency was performed using a 16-slice CT scanner (Somatom Emotion 16, Siemens, Erlangen, Germany). Proprietary software (Somaris 5 Syngo CT 2014 A, Siemens AG, Muenchen, Germany) was used for semi-automated body composition and distribution volume quantification.

Based on previously described methods, the total abdominal, visceral, and SQ volumes of interest (VOI) were established between the cranial margin of the 10th thoracic vertebrae and the cranial margin of the first sacral vertebrae, and using a 3 mm slice thickness for reconstruction [27, 34]. The computed tomographic volume (CTV) of all abdominal tissues within this region was calculated using attenuation thresholds of − 250 to 2000 HU, while abdominal fat volume was segmented using a threshold range of − 250 to − 25 HU [34]. Total, visceral, and subcutaneous abdominal adiposity was expressed as abdominal fat percentage, calculated as the proportion of fat volume within the defined VOIs relative to total abdominal tissue volume. Subcutaneous fat volume included subcutaneous, intermuscular, and intramuscular fat external to the peritoneal cavity. Visceral and subcutaneous fat volumes were additionally used to calculate the visceral-to-subcutaneous fat volume ratio/distribution (V/SQ). The CT cross-sectional areas (CTA) of total abdominal, visceral, and SQ fat were calculated at the cranial margins of L3 using the same borders and as previously described [26, 40]. The mean Hounsfield units (HU) were recorded for the volume and area measurements.

The visceral adipose thickness (VAT), subcutaneous adipose thickness (SAT), and total adipose thickness (TAT) measurements were taken perpendicular to the parietal surface of the linea alba, aligning with the most caudal margin of the liver at the midline, an adaptation from recent ultrasound methodology [27]. The VAT was measured from the linea alba’s parietal surface to the ventral surface of the liver. The SAT was measured from the skin surface to the parietal surface of the linea alba. The TAT was calculated as the combined measurement of both VAT and SAT (see Fig. 6).

Abdominal conformation

Standardised measurements from the transverse CT images assessed the abdominal conformation, modified from a previously established method [64]. The cranial abdominal height and width were measured at the widest internal diameter of the cranial abdomen constrained within the costal arch. The height was measured from the ventral surface of the thoracic vertebrae perpendicular to the parietal surface of the linea alba, and the width between the inner surface of the left and right ribs on the same slice of the CT. The height and width were used to determine the height-to-width ratio (HWR). This body conformation was relatively restrained by skeletal conformation and minimally affected by intra-abdominal anatomy and SQ fat deposition. The floating 12th and 13th ribs were not included in the measurements.

Data stratification and preparation

Data cleaning and preparation for analysis were completed by the primary author and statistician (R.T. and S.M.F.). The definitions of categories are provided in Supplemental 5 (S5).

Dog breeds were categorised based on the Australian National Kennel Council (ANKC) standards (http://ankc.org.au/Breed/Index/1) as: group 1 (toys); group 2 (terriers); group 3 (gundogs); group 4 (hounds); group 5 (working dogs); group 6 (utility); and group 7 (non-sporting) breeds [5]. Mixed breeds were categorised separately (group 8). The dogs were further classified on skull shape as dolichocephalic, mesocephalic, or brachycephalic; and as either chondrodystrophic or non-chondrodystrophic breeds, and as previously described [65–68]. Mixed breeds were excluded from these categories. Dogs were also classified into size categories as extra small (< 6.5 kg), small (6.5 to <9 kg), medium-small (9 to < 15 kg, medium-large (15 to < 30 kg), large (30 to < 40 kg), and giant (> 40 kg) as previously described [15].

The reason for the clinician requesting the CT and final diagnosis was recorded, and the disease was categorised as acute (< 3 months) or chronic (≥ 3 months) [69, 70]. The primary diagnoses were categorised as endocrine (including neoplasia), neoplastic (non-endocrine), primary gastroenteric (not neoplasia), trauma/musculoskeletal, congenital portal vasculature anomaly, inflammatory, or other, based on the primary diagnosis. The specific endocrinopathy was reported, as was the type and location of the neoplasia. Dogs were defined as having confirmed hyperadrenocorticism (cHAC) if confirmed by low-dose dexamethasone stimulation or ACTH stimulation test and supported by appropriate imaging findings. Presumptive hyperadrenocorticism (pHAC) was recorded if the dog had chronic treatment by glucocorticoids (≥ 3 months) at the time of the CT, or an adrenal nodule/mass was observed on the abdominal CT, and no confirmatory dynamic endocrine testing results were available. Sensitivity analyses assessed the influence of considering HAC status as confirmed, presumptive, or a combination of the two (see supplementary S6). Presumptive HAC diluted the effect of cHAC, and cHAC alone was used in the multivariable analyses. The type and length of medications before the CT were recorded; the length of medication use was categorised as chronic if used for over three months.

The first CT was used for dogs that underwent multiple CT examinations, with duplicates excluded from the analysis. No further exclusions were made based on CT confounders, but the data were stratified for analysis as defined below.

The presence of potentially confounding pathology on CT was noted (CT confounders), particularly the presence of abdominal and pleural effusion, abdominal masses, organomegaly (including markedly distended urinary bladders or gastrointestinal tract), and small organs. The volume of effusion was semi-quantified as trace (fat stranding to a thin sliver of fluid), mild (≤ 2 pockets, ≤ 2 cm depth), moderate (> 2 pockets), and marked (diffuse fluid accumulation) [71]. A mass was defined as any focal lesion ≥ 2 cm in diameter, and the mass location and diameter were reported [72]. Organomegaly or small organs were recorded if they were reported in the original radiology report. The presence of effusion, abdominal masses, organomegaly, and small organs was considered a potential confounding variable in abdominal fat measurements. The CT findings were dichotomised into CTs with no confounders and CTs with confounders (effusion, abdominal masses, organomegaly, or small organs).

The literature was reviewed to explore characteristics associated with fat distribution. A putative causal network was constructed as a directed acyclic graph (DAG) and used to generate appropriate adjustment sets of variables for each exposure and outcome of interest (see Fig. 7; S5: Explanatory variables; and definitions S7: Putative causal web model/Directed acyclic graph (DAG) and justification) [73]. The association between each risk factor was assessed using univariable analysis (see S1: Descriptive statistics and univariable analysis of categorical variables in a cross-sectional study of CTV abdominal fat distribution in a tertiary veterinary hospital population of dogs). For reporting, exploratory variables were collapsed into seven risk factors for the putative diagram: age, ANKC Terrier Breed, sex-neuter status, total abdominal adiposity, cHAC, and CT confounding pathology on CT. ANKC Hound group was considered too small for analysis.

Fig. 7.

Putative causal web model/Directed acyclic graph (DAG). Directed acyclic graph showing putative causal paths linking explanatory variables to the measured CTV abdominal visceral-to-subcutaneous fat ratio in an Australian tertiary veterinary hospital population of dogs. A detailed description of each link in the above plot is provided in supplementary materials S2

Power calculation

A power analysis was performed using the data from a previous study [27]. A sample size calculation undertaken in G*Power (version 3.1.9.4) estimated 106 dogs were required, using an f² effect size of 0.095, alpha = 0.05, power = 0.80, and testing for statistically significant associations for two predictor variables of 8 total degrees of freedom [74, 75]. Thus, a minimum sample size of 120 was sought to allow for expected controlling for subgroup analysis.

Statistical methods

Relationships between variables were visualised on scatter plots, and the assumption of normality and group variance was evaluated. V/SQ had a lognormal distribution, which was logarithmically transformed to base 10 for analysis, log(V/SQ), with results back-transformed for interpretation. Descriptive data were summarised as mean and standard deviation for parametric variables, median and interquartile range for non-parametric variables, and frequencies with modes where appropriate for categorical or binomial variables.

Independent t-tests were used to compare continuous variables between groups. As indicated, an appropriate multivariable regression model was constructed to assess associations between explanatory variables and each outcome variable (total abdominal fat percentage, visceral fat percentage, SQ fat percentage, and the log visceral/subcutaneous fat ratio, i.e., log(V/SQ). The directed acyclic graph (DAG, see supplementary materials S7) informed which variables to adjust for based on the exposure and outcome of interest. Model outputs included regression coefficients and their 95% confidence intervals (95% CIs) for both univariable models and models based on the DAG-informed adjustment sets, along with model-estimated marginal means for each level of categorical explanatory variables. When assessing the linearity of associations between age and each of the outcome variables, Lowess (Locally Weighted Scatterplot Smoothed) curves were fit to plots of each outcome variable and age, and the coefficients and Akaike information criterion (AIC) values were compared [76]. As this indicated that the relationship was nonlinear and that the most appropriate fit could be achieved by incorporating age into regression models as a quadratic term centred on its mean. However, for visceral fat and log(V/SQ), the relationship did not seriously depart from linear.

Criterion validity of CT Hounsfield units, L3 cross-sectional area, and ultrasound-derived abdominal fat measures relative to CT-derived volumetric indices (reference method), as well as intra-observer reliability based on repeated segmentation measurements on the same data set after a minimum 6-month interval using Lin’s Concordance Correlation Coefficient (r_c) (Lin’s Concordance Correlation Coefficient; SPSS Syntax; Garcia-Granero, M.; updated 04/2009, https://gjyp.nl/marta/Lin.sps) and Bland-Altman (B-A) limits of agreement (LOA) [77]. The r_c lower 95% confidence limit (CL) was reported as the strength of agreement and described as perfect (r_c=1.00), near perfect (> 0.99), substantial (0.95–0.99), moderate (0.90–0.95), and poor agreement (< 0.90) [78]. Statistical analysis was performed using GraphPad Prism (GraphPad Prism for Mac OS X, version 9.3.1, GraphPad Software, La Jolla, CA, USA, www.graphpad.com), Jamovi (The jamovi project (2022). Jamovi (Version 2.3.21.0) [Computer Software]. Retrieved from https://www.jamovi.org) and the R software package [79].

Directed acyclic graph showing putative causal paths linking explanatory variables to the measured CTV abdominal visceral-to-subcutaneous fat ratio in an Australian tertiary veterinary hospital population of dogs. A detailed description of each link in the above plot is provided in supplementary materials S2.

Abbreviations

BCS

Body condition score

CL

Confidence limit

CT

Computed tomography

DXA

Dual-energy x-ray absorptiometry

HU

Hounsfield units

ROI

Region of interest

SQ

Abdominal subcutaneous fat volume (as measured by CT)

VOI

Volume of interest

V/SQ

Visceral-to-subcutaneous fat distribution/ratio

Authors’ contributions

RT was a major contributor to experimental design, data acquisition, statistical analysis, data interpretation, critical analysis and writing of the manuscript. SF was a major contributor to the experimental design, statistical analysis, data interpretation and critical analysis. CM and FD were major contributors to the manuscript’s data interpretation and critical analysis. All authors have read and approved the final manuscript.

Data availability

The datasets used and analysed for the study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.