Utility of silhouette showcards to assess adiposity in three countries across the epidemiological transition

The Pulvers’ silhouette showcards provide a non-invasive and easy-to-use way of assessing an individual’s body size perception using nine silhouette shapes. However, their utility across different populations has not been examined. This study aimed to assess: 1) the relationship between silhouette perception and measured anthropometrics, i.e., body mass index (BMI), waist circumference (WC), waist-height-ratio (WHtR), and 2) the ability to predict with silhouette showcards anthropometric adiposity measures, i.e., overweight and obesity (BMI ≥ 25 kg/m2), obesity alone (BMI ≥ 30 kg/m2), elevated WC (men ≥ 94 cm; women ≥ 80 cm), and WHtR (> 0.5) across the epidemiological transition. 751 African-origin participants, aged 20–68 years old, from the United States (US), Seychelles, and Ghana, completed anthropometrics and selected silhouettes corresponding to their perceived body size. Silhouette performance to anthropometrics was examined using a least-squares linear regression model. A receiver operator curve (ROC) was used to investigate the showcards ability to predict anthropometric adiposity measures. The relationship between silhouette ranking and BMI were similar between sexes of the same country but differed between countries: 3.65 [95% CI: 3.34–3.97] BMI units/silhouette unit in the US, 3.23 [2.93–3.74] in Seychelles, and 1.99 [1.72–2.26] in Ghana. Different silhouette cutoffs predicted obesity differently in the three countries. For example, a silhouette ≥ five had a sensitivity/specificity of 77.3%/90.6% to predict BMI ≥ 25 kg/m2 in the US, but 77.8%/85.9% in Seychelles and 84.9%/71.4% in Ghana. Ultimately, silhouettes predicted BMI, WC, and WHtR similarly within each country and sex but not across countries. Our data suggest that Pulvers’ silhouette showcards may be a helpful tool to predict anthropometric and adiposity measures in different populations when direct measurement cannot be performed. However, no universal silhouette cutoff can be used for detecting overweight or obesity status, and population-specific differences may stress the need to calibrate silhouette showcards when using them as a survey tool in different countries.


Background
The prevalence of overweight and obesity is increasing in populations spanning the epidemiological transition and may be particularly high in individuals of African-origin. [1][2][3][4] Elevated weight is associated with the development of non-communicable diseases (NCDs), including cardiovascular disease, type 2 diabetes mellitus, hypertension, dyslipidemia, cancers, and sleep apnea. [5][6][7][8] Because of its simplicity and ease of measurement, body mass index (BMI, kg/m 2 ) is widely used to assess a person's adiposity. In addition to BMI, waist circumference (WC) and waist-to-height ratio (WHR) correlate well with fat mass as assessed by accurate methods such as computed tomography (CT). [9][10][11][12] However, BMI does not discriminate well between adipose and lean mass, and waist circumference and waist-to-height ratio have been suggested to predict adiposity better. 9-11 Yet, while they may not predict actual adiposity perfectly at the individual level, these simple adiposity markers may reliably predict mean BMI levels and the prevalence of obesity at the population level. 9,12 Measures of adiposity that do not rely on actual measurements may be useful in some situations, such as in surveys and studies of public health, anthropology, economics, and marketing, particularly when studies must be performed without direct contact with a person (e.g., mail-order or internet-based) or to avoid the burden of asking respondents to remove clothing. Furthermore, self-reported adiposity (e.g., self-reported height and weight) are prone to reporting bias and can also depend on access to home anthropometric tools like scales and varying cultural views on body size. [13][14][15][16][17][18][19] Initially developed by Stunkard and colleagues, sex-speci c silhouette showcards (referred to as "silhouettes" hereafter) can be used to determine one's perception of their body size. This tool relies on presenting to respondents a series of pictures/drawings of distinct body sizes in an increasing sequence, from which respondents select the one they think best re ects their body size. 20 Silhouettes should be ethnically ambiguous enough to be used in different cultures, but still detailed enough to be relatable. A variety of silhouette tools have been developed and validated for different populations. [21][22][23][24] Pulvers and colleagues created culturally relevant body image silhouette showcards for African Americans (Fig. 1). 25 These silhouettes were validated in different populations of African-origin such as Seychelles, the Caribbean, and the USA. [25][26][27][28] While many studies have shown a good association between the silhouettes and adiposity measures, including for the prediction of obesity, most studies have only assessed their validity in a single population at a time. [21][22][23][24][25][26][27][28][29][30][31][32] Also, only a few studies have directly compared the associations of silhouette ranking between different populations with diverse ethnic backgrounds or with different population mean BMI levels. Assessing the validity of silhouettes to predict adiposity in different populations may be challenging as one's assessment of body image relies on an individual's ability to appraise their current body size and correctly classify their weight relative to objective measurements and, also, considering that cross-cultural evaluation should rely on studies that use the same methodology in different countries. 29,33−36 Therefore, our study aims to assess: 1) the relationship between Pulvers silhouette showcard ranking and measured adiposity markers (BMI, WC, and WHR), 2) the performance of silhouette ranking to predict adiposity makers, particularly overweight and obesity based on elevated BMI, and 3) the performance of silhouette ranking to predict BMI, WC, and WHR, in three different African-origin populations representing different stages of social and economic development and different prevalence of obesity. 25 Page 3/11

Study Populations and Ethics Approval
This study is a subset analysis from the METS-Microbiome study (R01DK111848) initiated in 2017, for which the protocol has been published. 37 The METS-Microbiome study continues yearly measurements of participants initially recruited for the Modeling the Epidemiological Transition Study (METS; R01-DK080763) in ve African-origin populations spanning the epidemiologic transition varying by the United Nations Human Development Index (HDI) 2010. 38,39 The current data were collected between 2018-2019 from participants in metropolitan Chicago, IL, USA (HDI: 0.92), the mixed urban/rural Seychelles islands The study sample consisted of men and women aged 20-68 years old who were of African-origin except for Seychelles, where both Black-African participants and participants of mixed racial ancestry were included. Approximately 66% of the whole sample identi ed as female.

Survey and Body Size Silhouette Showcards
The survey component of the METS-Microbiome study consisted of a face-to-face interview performed by centrally trained personnel, capturing participants' sociodemographic status, health-related behaviors, and medical history. Participants were also presented with sex and ethnicity-speci c silhouette showcards created by Pulvers (Fig. 1). 25,27 This nine-image tool displayed sex-speci c body sizes in increasing order ranging from very thin to severely obese. To measure participants perceived body size, participants were asked, "In the drawing, which gure best re ects how you think you look with regards to your body shape?".
Participants' responses were recorded on a scale from 1 (representing the thinnest silhouette) to 9 (representing the most obese silhouette).

Anthropometric and Adiposity Measurements
Participants completed a health examination, including measured height (m), weight (kg), and waist circumference (cm). Across all sites, standardized equipment and protocols were used, as previously described. 37 Body mass index (BMI, weight/height 2 ) was calculated and classi ed as underweight (BMI < 18.5 kg/m 2 ), normal weight (BMI 18.5-24.9 kg/m 2 ), overweight (BMI 25.0-29.9 kg/m 2 ) or obese (BMI ≥ 30 kg/m 2 ). 40 A dichotomous waist circumference (cm) variable was used to classify the presence of central obesity as de ned by the International Diabetes Federation (≥ 94 cm in men, ≥ 80 cm in women) for European or African-origin individuals. 11 WHR (waist in cm/ height in cm) was calculated and dichotomized using a widely used cut-off point for normal (WHR ≤ 0.5) or increased central obesity (WHR > 0.5). 41

Statistical Analyses
Participant characteristics were summarized using means and 95% con dence intervals. Proportions were calculated and presented as a percent (%) and 95% con dence intervals for categorical variables. Spearman's rank correlation coe cients were used to describe the associations between the self-reported perceived silhouette ranking and BMI, WC, and WHR.
Mean BMI and 95% con dence interval for each silhouette rank was determined by sex and by country. To assess whether the slopes of the relation between silhouette ranking and adiposity markers differed between countries and sex, we estimated the linear regression coe cients (i.e., the change of the three adiposity markers corresponding to 1 silhouette ranking change) by sex and country with accompanying 95% con dence intervals.
The self-reported silhouette showcards were assessed for accuracy in predicting widely used dichotomized adiposity markers, e.g., overweight and obesity (BMI ≥ 25 kg/m 2 ) or obesity alone (BMI ≥ 30 kg/m 2 ), elevated waist circumference (cm, (≥ 94 cm in men, ≥ 80 cm in women) and elevated waist-to-height ratio (WHR > 0.5) using sex and country-speci c receiver-operator curve (ROC) analysis. 26 We used the area under the curve (AUC, i.e., the c-statistic) and sensitivity and speci city associated with different cut-offs of the silhouettes to predict these dichotomous adiposity categories.
All statistical analyses were performed using STATA SE 12 (StataCorp, College Station, TX, USA). Table 1 shows the main characteristics of the 751 participants from the three countries. Mean age differed slightly across countries and was highest in men in the USA (47.1 years) and lowest in women in Ghana (41.4 years).  Table 2 shows the Spearman's correlation coe cients of the relationship between the perceived self-reported silhouette rankings with BMI, WC, and WHR, by country and sex. These coe cients ranged between 0.71 and 0.80 in men and women in all countries, except in men in Ghana (0.55-0-58), (p < 0.001 for all coe cients). Relationship between silhouette ranking and measured BMI Table 3 shows a graded increase in mean BMI according to silhouette ranking by sex and country. The table also depicts the least-squares linear regression coe cients by sex and country between participants' measured BMI and the self-reported silhouettes. Regression coe cients (i.e., slopes of the regression lines) were higher in women compared to men in all three countries. Regression coe cients were signi cantly lower in Ghana than in the other two countries for both men and women. In the USA and Seychelles, an increase in 1 silhouette unit was associated with an increase of 3.05-3.75 BMI units (kg/m 2 ) but only 1.15-2.06 BMI units in Ghana. Nearly identical trends were observed for WC and WHR (Supplementary tables 1 and 2). A robust regression analysis, which lessens the in uence of outliers on the regression coe cient estimates, was also performed, and estimates were almost identical as those in the least-squares linear regression. Self-reported silhouette as a discriminator of overweight and/or obesity   Performance between silhouette ranking to BMI, waist circumference, and waist-to-height ratio in detecting adiposity Table 5 shows the sex and country-speci c AUCs (i.e., c-statistic) of silhouette ranking to predict overweight or obesity status (BMI ≥ 25 kg/m 2 ) or obesity alone (BMI ≥ 30 kg/m 2 ). AUCs ranged between 0.79 and 0.92 in men and between 0.87 and 0.97 in women, with little differences by sex or country. Similar AUC values were found for silhouette ranking to predict elevated WC and WHR.

Discussion
This study continues on the foundation established by Pulver and colleagues in creating the silhouette showcards and subsequent validation in populations of African-origin. [25][26][27] Our data suggest that the Pulvers' silhouette showcards may be a useful tool for predicting objective body size such as BMI, WC, and WHR, in different populations of mainly African-origin. However, the relationship between silhouettes and adiposity markers differed according to the country.
Overall, our data suggest that silhouettes may be a useful tool to predict actual adiposity measures, conditional to adequate calibration for a speci c population.
BMI and other adiposity measures correlated strongly with silhouette ranking in all populations. However, the magnitude of the linear regression coe cients between silhouette ranking and actual adiposity markers differed between the three countries in this study. For example, an increase of 1 silhouette unit was associated with an increase of 3-4 BMI units (kg/m 2 ) in the USA and Seychelles but only 1-2 BMI units in Ghana. This difference suggests varying perceptions of one's body shape according to mean population BMI. One may speculate that in the USA and Seychelles, where mean population BMI is high, individuals with adiposity are more inclined to view a large body shape as normal compared to populations (e.g., Ghana) where mean population BMI is lower.
Again, this altered view suggests that silhouette showcards need to be speci c (i.e., calibrated) to different populations when used for predicting individuals' actual adiposity. From a prevention perspective, the differences in perceptions of one's body size across populations may suggest larger tolerance for larger body shapes in populations with high adiposity levels. Overall, this underlies that silhouettes can have a role for assessing adiposity in populations when direct measurements cannot be made (i.e., for surveillance purposes, as evaluated in this study), but also for assessing perceptions and attitudes of people for weight control programs.
The relationship between silhouettes and adiposity markers can differ according to sex in the same population. Using different silhouette showcards,  21,22,26 It is therefore likely that the same linear regression models can be used in men and women for calibration of the association between silhouettes and BMI (or other adiposity markers) within the same population, as long as mean BMI in the population is similar in both sexes. Inversely, as our data in Ghana suggest, different predictive models may need to be developed in men and women when mean BMI markedly differs between men and women in the same population. Differences in the slopes of the associations between silhouettes and BMI (and other adiposity markers) may also partly depend on different sex-speci c perceptions of body shape, and this question necessitates further studies.
The country and sex-speci c associations between silhouettes and adiposity markers were quite similar when using BMI, WC, and WHR. This relationship is not unexpected as BMI, WC, and WHR quite strongly and similarly inter-correlate with each other, e.g., correlation coe cients of 0.77 to 0.96 in our study, which is consistent with correlations found in other studies. 42 However, the fact that these associations between silhouettes and BMI, WC, and WHR (and the associations between these adiposity markers and objectively measured fat mass) are still not extremely strong, implies that silhouettes would not be a reliable tool to predict adiposity at the individual level (sensitivity and speci city are not optimal), but they can be useful when assessing adiposity levels (e.g., the prevalence of obesity, mean BMI) at a population level, conditional on appropriate calibration in a speci c population. More generally, our data suggest that a subjective two-dimensional pictorial body size assessment (silhouette drawings) can be a useful tool for predicting a volumetric dimension (adiposity), at least at the population level.
This study's main strength was the use of the identical methodology in the three countries, allowing us to make direct comparisons between populations of the same racial origin and that the three populations differed largely according to mean adiposity levels and socioeconomic development stages. However, the study also has limitations. First, although the study was designed to include participants of African-origin in all sites, in order to control for ethnic differences, persons from mixed origin were also included in varying but small proportions, particularly in Seychelles. Second, the study included middle-aged adults, and the ndings may not necessarily extend to older or younger individuals. Third, Pulver's silhouette tool presents body size silhouettes from thinnest to heaviest, which could lead to reporting bias. Future studies should examine if presenting the silhouettes in random order would gather different results. Fourth, survey administrators presented silhouettes to the participants; further studies should assess if results would differ if participants had assessed their silhouettes in the absence of assisting personal. Finally, our analysis, according to sex, was limited because of the limited sample size.

Conclusions
This study supports the utility of Pulvers' silhouette showcards as a useful tool to predict adiposity in populations in settings where body size cannot be measured directly, conditional to adequate adjustment (i.e., calibration) of the associations between silhouette ranking and actual adiposity markers. Although this was not the aim of this study, our results also emphasize potential bene ts of using silhouettes to assess individuals' perceptions and attitudes in the context of weight control programs at clinical or public health levels. 25 Declarations Ethics approval and consent to participate:   Proportion with normal weight, overweight, and obese within each silhouette category in the USA, Seychelles, and Ghana. Notes: N weight: normal weight (BMI 18.5-24.9 kg/m2); Overweight (BMI 25.0-29.9 kg/m2); Obese (BMI ≥ 30 kg/m2); Sey: Seychelles.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementalTables.docx