Association of body-shape phenotypes with imaging measures of body composition in the UK Biobank cohort: relevance to colon cancer risk

Background Body mass index (BMI), waist and hip circumference are strongly correlated and do not reflect body composition. A Body Shape Index (ABSI) and Hip Index (HI) define waist and hip size among individuals with the same weight and height and would thus reflect body density. We examined differences in body composition between body-shape phenotypes defined with ABSI and HI and used this information to propose explanations for associations between body-shape phenotypes and colon cancer risk. Methods We used data from the UK Biobank Resource for 15,520 men, 16,548 women with dual-emission X-ray absorptiometry (DXA) measurements; 3997 men, 4402 women with magnetic resonance imaging (MRI) measurements; 200,289 men, 230,326 women followed-up for colon cancer. We defined body-shape phenotypes as: large-ABSI-small-HI (“apple”), small-ABSI-large-HI (“pear”), small-ABSI-small-HI (“slim”), large-ABSI-large-HI (“wide”). We evaluated differences in body composition in linear models and associations with colon cancer risk in Cox proportional hazards models adjusted for confounders and explored heterogeneity by BMI. Results Among individuals with the same height and weight, visceral adipose tissue (VAT) was lowest for “pear” and highest for “apple”, while abdominal subcutaneous adipose tissue (ASAT) was lowest for “slim” and highest for “wide” phenotype. In the gynoid region, differences between “apple” and “pear” phenotypes were accounted for mainly by fat mass in women but by lean mass in men. In men, lean mass was inversely associated with waist size, while the pattern of gynoid fat resembled ASAT in women. Lean and fat mass were higher for higher BMI, but not hand grip strength. Compared to normal weight “pear”, the risk of colon cancer in men (1029 cases) was higher for “apple” phenotype for normal weight (hazard ratio HR = 1.77; 95% confidence interval: 1.16–2.69) and comparably for overweight and obese, higher for “wide” phenotype for overweight (HR = 1.60; 1.14–2.24) and comparably for obese, but higher for “slim” phenotype only for obese (HR = 1.98; 1.35–2.88). Associations with colon cancer risk in women (889 cases) were weaker. Conclusions ABSI-by-HI body-shape phenotypes provide information for body composition. Colon cancer risk in men appears related to ASAT quantity for “slim” and “wide” but to factors determining VAT accumulation for “apple” phenotype. Supplementary Information The online version contains supplementary material available at 10.1186/s12885-021-08820-6.


S6
Magnetic Resonance Imaging (MRI) MRI VAT was defined as the adipose tissue within the abdominal cavity, excluding adipose tissue outside the abdominal skeletal muscles and adipose tissue and lipids within and posterior of the spine and posterior of the back muscles, as described in reference [14]. MRI  Values for Arms and Legs were calculated as the sum of the right and left variable. Fat-free Mass includes lean and bone mass. Fat Mass and Fat-free Mass add up to the total mass of the region. Note that this differs from DXA measurements, where lean and bone mass are measured separately and Regional Mass comprises the sum of Regional Fat, Lean and Bone Mass.

Hand Grip Strength
We considered hand grip strength in the context of body composition as an indicator of muscle mass functionality. This was calculated as the maximum of Field [46-2.0] (left) and Field [47-2.0] (right) OR, when values in one of the two fields were missing, as the available value.
Measurements were considered missing when values were missing in both Fields.

Definition of covariates
Extensions Field-0.X correspond to the initial assessment visit at enrolment and extensions Field-2.X correspond to the imaging visit. Fields for variables evaluated both at enrolment and at the imaging visit are marked as Field-0(2).X.
Date of birth was constructed from Field [34-0.0] "Year of birth" and Field [52-0.0] "Month of birth", using 15 as the day of birth for all participants.
Age at enrolment or at the imaging visit was calculated in years as the difference between Field [53-0(2).0] "Date of attending assessment centre" and the date of birth (as defined above), divided by 365.25 and rounded to an integer. Six five-year categories for age at enrolment were S7 used for stratification of Cox proportional hazards models when examining associations with colon cancer risk (40 to <45 years, 45 to <50 years, 50 to <55, 55 to <60 years, 60 to <65 years and 65 to 70 years). Age at the imaging visit was used as a continuous adjustment variable when examining associations with body composition.
Weight change during last year preceding the initial assessment at enrolment or the imaging visit was self-reported in Field [2306-0(2).0] "Weight change compared with 1 year ago"; Question: "Compared with one year ago, has your weight changed?" with three valid answers: 0 "No -weigh about the same" (category Stable weight), 2 "Yes -gained weight" (category Weight gain), 3 "Yes -lost weight" (category Weight loss). Participants with missing information were assigned to the median sex-specific category (Stable weight).
Alcohol consumption at enrolment or at the imaging visit was based on Field [1558-0(2).0] "Alcohol intake frequency"; Question: "About how often do you drink alcohol?" as follows: Up to 3 times a month -Answer 4: "One to three times a month"; or 5: "Special occasions only"; or 6: "Never"; Up to four times a week -Answer 2: "Three or four times a week"; or 3: "Once or twice a week"; Daily or almost daily -Answer 1: "Daily or almost daily". Participants with missing information were assigned to the median sex-specific category (Up to four times a week).
S8 Physical activity at enrolment or at the imaging visit was defined as follows: Very activewas based on Field [816-0(2).0] "Job involves heavy manual or physical work"; Question: "Does your work involve heavy manual or physical work?"; Answer 3: "Usually" or 4: "Always"; OR Field [904-0(2).0] "Number of days/week of vigorous physical activity 10+ minutes"; Question: "In a typical WEEK, how many days did you do 10 minutes or more of vigorous physical activity? (These are activities that make you sweat or breathe hard such as fast cycling, aerobics, heavy lifting)"; Answer (numerical) 3-7; Moderately active -was based Field [904-0(2).0] Answer 1-2 OR Field [884-0(2).0] "Number of days/week of moderate physical activity 10+ minutes"; Question: "In a typical WEEK, on how many days did you do 10 minutes or more of moderate physical activities like carrying light loads, cycling at normal pace? (Do not include walking)"; Answer (numerical) 3-7; OR Field [864-0(2).0] "Number of days/week walked 10+ minutes"; Question: "In a typical WEEK, on how many days did you walk for at least 10 minutes at a time? (Include walking that you do at work, travelling to and from work, and for sport or leisure)"; Answer (numerical) 7, when participants were not already included in category very active; Less active -was based on Field [904-0(2).0] Answer (numerical) 0 OR Field [884-0(2).0] Answer (numerical) 0-2 OR Field [864-0(2).0] Answer (numerical) 0-6 or -2: "Unable to walk", when participants were not already included in category moderately or very active. Participants with missing information were assigned to the median sex-specific category (Moderately active).
Townsend deprivation index (available only at enrolment) was used as an indicator of socioeconomic status and was based on Field [189-0.0] "Townsend deprivation index at recruitment", which was calculated by UK Biobank as a score corresponding to the output area in which the participant's postcode was located immediately prior to joining UK Biobank, based on the preceding national census output areas. A greater score implies a greater degree of material deprivation. This variable was categorised in sex-specific tertiles. The tertile boundaries were: -3.229 and -0.797 for men and -3.220 and -0.880 for women in the cancer risk dataset; -3.526 and -1.547 for men and -3.451 and -1.410 for women in the DXA dataset; -3.512 and -1.489 for men and -3.451 and -1.470 for women in the DXA VAT subset; -3.494 and -1.506 for men and -3.464 and -1.434 for women in the MRI dataset (see Supplementary Figure S1 for definition of datasets).
Participants with missing information were assigned to the median sex-specific category (Middle tertile).
Family history of cancer at enrolment was based on three variables: Fields [20107-0.0/9] "Illness of father", Question: "Has/did your father ever suffer from? (You can select more than one answer)", Fields [20110-0.0/10] "Illness of mother", Question: "Has/did your mother ever suffer from? (You can select more than one answer)" and Field [20111-0.0/11] "Illness of siblings", Question: "Have any of your brothers or sisters suffered from any of the following diseases? (You can select more than one answer)". Category Yes (bowel) was based on Answer: 4 "Bowel cancer" to any of the three sets of fields; Category Yes (lung, breast, prostate) was based on Answer: 3 S9 "Lung cancer", 5 "Breast cancer" or 13 "Prostate cancer" to any of the three sets of fields and category No included the remaining participants.
Fresh fruit and vegetable intake at enrolment was based on the sum of two fields: Field [1309-0.0] "Fresh fruit intake" (continuous), Question: "About how many pieces of FRESH fruit would you eat per DAY? (Count one apple, one banana, 10 grapes etc as one piece; put '0' if you do not eat any)" and Field [1299-0.0] "Salad / raw vegetable intake" (continuous), Question: "On average how many heaped tablespoons of SALAD or RAW vegetables would you eat per DAY?
(Include lettuce, tomato in sandwiches; put '0' if you do not eat any)". Answers: -10 "Less than one" were re-coded to 0.5. Answers: -1 "Do not know" and -3 "Prefer not to answer" were consider missing. The total was dichotomised as Less than five portions a day or Five or more portions a day and were used as an indication of a healthy lifestyle. Participants with missing information were assigned to the median sex-specific category (Less than five portions a day).
Processed meat intake at enrolment was based on Field [1349-0.0] "Processed meat intake", Question: "How often do you eat processed meats (such as bacon, ham, sausages, meat pies, kebabs, burgers, chicken nuggets)". Category Less than twice a week included answers: 0 "Never", 1 "Less than once a week" and 2 "Once a week". Category Twice or more a week included answers: 3 "2-4 times a week", 4 "5-6 times a week" 5 "Once or more daily". Answers: -1 "Do not know" and -3 "Prefer not to answer" were consider missing. Participants with missing information were assigned to the median sex-specific category (Less than twice a week).
Red meat intake at enrolment was based on the sum of three fields: Field [1369-0.0] "Beef intake", Question: "How often do you eat beef? (Do not count processed meats)", Field [1379-0.0] "Lamb/mutton intake", Question: "How often do you eat lamb/mutton? (Do not count processed meats)" and Field [1389-0.0] "Pork intake", Question: "How often do you eat pork? (Do not count processed meats such as bacon or ham)". The categorical answers were converted to a continuous scale as follows: Answer 0 "Never" remained 0; Answer 1 "Less than once a week" was coded as 0.5; Answer 2 "Once a week" was coded as 1; Answer 3 "2-4 times a week" was coded as 3; Answer 4 "5-6 times a week" was coded as 5.5; Answer 5 "Once or more daily" was coded as 7. Answers: -1 "Do not know" and -3 "Prefer not to answer" were consider missing. Categories Less than twice a week and Twice or more a week were derived with respect to the total of the three variables. Participants with missing information were assigned to the median sex-specific category (Less than twice a week for women and Twice or more a week for men).
Fibre intake at enrolment was based on the sum of fibre allocated to the consumption of fruit, vegetables, bread and cereal, using estimated quantities of fibre per food type and quantity according to reference [17] as follows: Field [1309-0.0] "Fresh fruit intake" (continuous) Question: "About how many pieces of FRESH fruit would you eat per DAY? (Count one apple, one banana, 10 grapes etc as one piece; put '0' if you do not eat any)" allocating 2g of fibre per piece; Field [1319-0.0] "Dried fruit intake" (continuous) Question: "About how many pieces of DRIED fruit would you eat per DAY? (Count one prune, one dried apricot, 10 raisins as one piece; put '0' if you do not S10 eat any)" allocating 0.5g of fibre per piece; Field [1299-0.0] "Salad / raw vegetable intake" (categorical) Question: "What type of bread do you mainly eat?" allocating 0.68g per slice of "White" bread (code 1), 1.26g per slice of "Brown" bread (code 2), 1.80g per slice of "Wholegrain or wholemeal" bread (code 3), 1.25g per slice of "Other type of bread" (code 4) or missing bread type; Field [1458-0.0] "Cereal intake" (continuous) Question: "How many bowls of cereal do you eat a WEEK?" (divided by 7 to obtain the cereal bowls per day) and Field [1468-0.0] "Cereal type" (categorical) Question: "What type of cereal do you mainly eat?" allocating 7.16g per bowl of "Bran cereal (e.g. All Bran, Branflakes)" (code 1), 2.92g per bowl of "Biscuit cereal (e.g. Weetabix)" (code 2), 1.92g per bowl of "Oat cereal (e.g. Ready Brek, porridge)" (code 3), 4.18g per bowl of "Muesli" (code 4), 0.54g per bowl of "Other (e.g. Cornflakes, Frosties)" (code 5), or 3.34g per bowl if missing cereal type. For quantitative variables, answer -10 "Less than one" was re-coded to 0.5 and answers -1 "Do not know" and -3 "Prefer not to answer" were considered missing. This variable was categorised in sex-specific tertiles. The tertile boundaries were: 11.92 g and 17.13 g for men and 12.52 g and 17.11 g for women. Information for fibre intake was missing when information for all variables used to calculate the quantity was missing. Participants with missing information were assigned to the median category (Middle tertile).
Nonsteroidal Anti-inflammatory Drugs (Aspirin & Ibuprofen) use at enrolment was based on Fields [6154-0.1/5] "Medication for pain relief, constipation, heartburn"; Question: "Do you regularly take any of the following? (You can select more than one answer)"; Answer 1: "Aspirin" OR Answer 2: "Ibuprofen" (for category Yes). The guidance included the message: "Some over the counter medicines are known by other names. Please enter the corresponding name if you take any of the following REGULARLY (that is, most days of the week for the last 4 weeks): Aspirin (Alka Rapid Crystals, Alka-Seltzer XS, Anadin Extra, Anadin, Original, Askit powders, Aspro Clear, Codis 500, Disprin, Disprin Extra) OR Ibuprofen (Anadin Ultra, Anadin Ibuprofen, Cuprofen Plus, Nurofen, Solpaflex, Ibuleve)". Category No was defined as any of the following Answers: 3 "Paracetamol", 4 "Ranitidine", 5 "Omeprazole", 6 "Laxatives" OR -7 "None of the above". Values were considered missing if there was no answer to this question or Answer -1 "Do not know" or Answer -3 "Prefer not to answer" were given. Missing values were replaced with the sex-specific median (No).
(You can select more than one answer)". Women providing Answer 4 "Hormone replacement therapy" were considered Current user. Women with missing information for Field [2814-0(2).0] were assigned to the median category (Never user). Pre-menopausal (Pre-MP) -were classified women who had not been defined as post-menopausal above and had reported pre-menopausal status at the imaging visit with Answer 0: "No" to Field [2724-2.0] OR had age at the imaging visit < 55 years when menopausal status at the imaging visit was unknown, i.e. not defined as post-or pre-menopausal according to the above criteria.

S12
Menopausal status-HRT at enrolment or at the imaging visit -as the number of premenopausal women HRT users was limited, a combined variable was used in all analyses, separating only post-menopausal women by HRT use. Women were assigned to four categories: Pre-MP, Post-MP -Never HRT user, Post-MP -Former HRT user and Post-MP -Current HRT user.
Use of oral contraceptives at enrolment was determined for women by Field [2784-0.0] "Ever taken oral contraceptive pill"; Question: "Have you ever taken the contraceptive pill? (include the 'mini-pill')"; Answer 0: "No" (for Never user) or Answer 1: "Yes" (for Ever user). Question: "How old were you when you had your child?" (UK Biobank note: "Current Field was collected from women who indicated they had given birth to only one child, as defined by their answers to Field 2734"). Women with missing information for Field [2734-0.0] were assigned to the median category (<30 years).

S13
Supplementary ABSI -a body shape index (cut-off: ≥79.7998 exact median); Apple -large-ABSI-small-HI; BMIbody mass index; HI -hip index (cut-off: ≥49.11381 exact median for HI calculated with coefficients from UK Biobank (used for men in the current paper); ≥60.26223 exact median for HI calculated with coefficients from NHANES [Ref. 5] (the NHANES coefficients were used only for women in the current paper)); n (%) -number of participants (percent from the total per row); NHANES -National Health and Nutrition Examination Survey; Pear -small-ABSI-large-HI; SDstandard deviation; Slim -small-ABSI-small-HI; Wide -large-ABSI-large-HI; ** -BMI for correctly classified and for misclassified men was compared with t-test (per row), p<10 -16 for all comparisons.

S14
Supplementary # -a sex-specific model regressing: the index was derived as: Note that waist circumference, hip circumference and waist-to-hip ratio scaled for weight and height correspond to A Body Shape Index (ABSI), hip index (HI) and waist-to-hip index (WHI).
## -a sex-specific model regressing: the index was derived as: BIA measurements were provided in kg, DXA measurements were provided in g and were divided by 1000 to convert to kg; MRI measurements were provided in L. Models for anthropometric, BIA and DXA measurements and hand grip strength were based on the DXA dataset (except for VAT and ASAT, which were based on the DXA VAT subset) and for MRI on the MRI dataset (see Supplementary Figure S1 for the definition of datasets).
Weight was obtained in kg from Field [21002-0.0] for the cancer risk dataset and Field [21002-2.0] for the imaging datasets and was used in kg. Height was obtained in cm from Fields [50-0(2).0] "Standing height" and was used in cm for HI and WHI but was divided by 100 and used in m for body composition indices and ABSI. Waist circumference was obtained in cm from Fields [48-0(2).0] and was used in cm for WHI and indices scaled only for height but was multiplied by 10 and used in mm for ABSI. Hip circumference was obtained in cm from Fields [49-0(2).0] and was used in cm.
Note that the regression coefficients ,  and  provided in the Table include the corresponding sign, as derived from the log-linear models, and they were multiplied by -1 in the Index formula.
This change of sign is required because Weight and Height are part of the denominator in the Index equation, while the Index formula is set up only with multiplication. The multiplication by -1 is incorporated in the formulae for ABSI, HI and WHI in the main document.

S20
Supplementary n (%) -number of participants (percentage from the total per column or, for cohort size and colon cancer cases, percentage from row overall); NW -normal weight BMI (86.8% of NW men and 80.5% of NW women were classified as "slim" body-shape phenotype); OB -obese BMI (91.6% of OB men and 96.1% of OB women were classified as "wide" body-shape phenotype); OWoverweight BMI.

S21
Waist and hip circumference were dichotomised at the sex-specific medians (≥96 cm for men and ≥83 cm for women for large waist circumference; ≥103 cm for men and ≥102 cm for women for large hip circumference). Continuous variables are summarised with mean (standard deviation).
The four body-shape phenotypes per sex were compared with one-way ANOVA for continuous variables and chi-squared test for categorical variables. All comparisons were significant at p<0.001.
SD difference (95% CI) -derived from linear regression models with adjustment for age, selfreported weight change within the year preceding the visit, smoking status, alcohol consumption, physical activity, Townsend deprivation index, region (except for VAT, ASAT and MRI) and, in women, menopausal status and use of hormonal replacement therapy (see definition of covariates in Supplementary Methods). Body-composition measurements were converted to allometric indices with scaling for height and weight (see scaling coefficients in Supplementary Table S1) and then to sex-specific z-scores (value minus mean, divided by the standard deviation). Note that DXA lean mass does not include bone mass, which is included in BIA fat-free mass. See Supplementary Methods for the calculation of DXA lean and fat mass.
p<5*10 -16 for all body-composition indices and hand grip strength -p-value obtained from a likelihood ratio test comparing models with and without an ABSI-by-HI variable, i.e. evaluating the significance of the association between each body composition index and body shape overall, accounting for all covariates.

S25
Supplementary     VAT -visceral adipose tissue; WC -waist circumference; WHR -waist-to-hip ratio.  Figure S1 for the definition of datasets). Measurements were converted to allometric indices with scaling only for height and then to sex-specific z-scores (value minus mean, divided by the standard deviation) (see scaling regression coefficients  formulas, means and standard deviations in Supplementary Table S1).

S32
Supplementary Figure Figure S1 for definition of datasets). Measurements were converted to allometric indices with scaling for weight and height and then to sex-specific z-scores (value minus mean, divided by the standard deviation) (see scaling regression coefficients and , correspondingly, formulas, means and standard deviations in Supplementary Table S1) .
Supplementary Figure S4 Body composition profiles of body-shape phenotypes (DXA VAT subset) ABSI -a body shape index (cut-offs: ≥80 in men; ≥73 in women); Apple -large-ABSI-small-HI; ASAT -abdominal subcutaneous adipose tissue; BIA -bioelectrical impedance analysis measurements; CI -confidence interval; DXA -dual-emission X-ray absorptiometry measurements; HI -hip index (cut-offs: ≥49 in men; ≥64 in women); MRI -magnetic resonance imaging measurements; Pear -small-ABSI-large-HI; SD -standard deviation; Slim -small-ABSIsmall-HI; VAT -visceral adipose tissue; Wide -large-ABSI-large-HI; SD difference (95% CI)derived from linear regression models with adjustment for age, self-reported weight change within the year preceding the visit, smoking status, alcohol consumption, physical activity, Townsend deprivation index, region (except for a single region for VAT, ASAT and MRI) and, in women, menopausal status and use of hormonal replacement therapy (see definition of covariates in S34 Supplementary Methods). Body-composition measurements were converted to allometric indices with scaling for height and weight (see scaling coefficients in Supplementary Table S1) and then to sex-specific z-scores (value minus mean, divided by the standard deviation). Note that DXA lean mass does not include bone mass, which is included in BIA fat-free mass. This is a sensitivity analysis for the restricted DXA VAT dataset, which contains measurements for regional DXA lean and fat mass provided by UK Biobank (see Supplementary Figure S1 for the definition of datasets).
Supplementary Figure S5 Body composition profiles of body-shape phenotypes (minimally adjusted models) ABSI -a body shape index (cut-offs: ≥80 in men; ≥73 in women); Apple -large-ABSI-small-HI; ASAT -abdominal subcutaneous adipose tissue; BIA -bioelectrical impedance analysis measurements; CI -confidence interval; DXA -dual-emission X-ray absorptiometry measurements; HI -hip index (cut-offs: ≥49 in men; ≥64 in women); MRI -magnetic resonance imaging measurements; Pear -small-ABSI-large-HI; SD -standard deviation; Slim -small-ABSIsmall-HI; VAT -visceral adipose tissue; Wide -large-ABSI-large-HI; SD difference (95% CI)derived from linear regression models with adjustment only for age and region (except for a single region for VAT, ASAT and MRI). Body-composition measurements were converted to allometric indices with scaling for weight and height (see scaling regression coefficients in Supplementary   Table S1) and then to sex-specific z-scores (value minus mean, divided by the standard deviation).
HR (95% CI) -derived from cox proportional hazards models: (A) stratified by age and region and adjusted for BMI (for overall) and height; (B) additionally adjusted for self-reported weight change within the year preceding the visit, smoking status, alcohol consumption, physical activity, Townsend deprivation index, diet (consumption of fresh fruit and vegetables, red meat, processed meat, fibre), use of non-steroidal anti-inflammatory drugs, family history of cancer and, in women, menopausal status, use of oral contraceptives and hormonal replacement therapy and age at last live birth (see definition of covariates in Supplementary Methods).