Pictogram Is a Valid Instrument to Classify At-Risk Adult Population Based on Abdominal Obesity: Results from Pars Cohort Study based results

Background: Abdominal obesity is associated with increased risk of myocardial infarction and death events. Thus, obtaining data on the status of abdominal obesity is important in risk factor assessment and prevention of non-communicable diseases. This study aimed to evaluate the validity of using pictograms to classify abdominal obesity indices (waist circumference [WC], waist-hip ratio [WHR], and waist-height ratio [WHtR]) into normal and at-risk categories and determine the effects of demographic characteristics on this validity. Methods: This cross-sectional study used data from Pars Cohort Study (PCS). Participants chose the most similar pictogram scores to their body size at 15, 30 years, and current age. Optimal normal/at-risk cut-off values for pictograms were calculated using sensitivity/specificity plots. Receiver operating characteristic curves were used to evaluate the validity of pictograms. Validity measures were analyzed across different subgroups of demographic characteristics. Results: A total of 9263 participants (46% males) were included in the study. The estimated area under the curves were 84% for WC, 77% for WHR, and 89% for WHtR in males, and 84% for WC, 73% for WHR, and 90% for WHtR in females. Optimal pictogram cutoffs to classify central obesity for WC, WHR, and WHtR were 4, 4, and 5 in males and 4, 4, and 6 in females, respectively. The majority of demographic characteristics were not associated with the validity of pictograms. Conclusion: Using pictograms to determine normal and at-risk categories of abdominal obesity indices is valid among adult population with a wide range of demographic characteristics. However, the results need to be interpreted with caution in those with a positive history of weight fluctuation.


Introduction
Maldistribution of total body fat is associated with adverse metabolic and cardiac outcomes. 1,2 Visceral body fat increases the risk of insulin resistance and metabolic syndrome. 3 Furthermore, previous studies show that correlations of abdominal obesity indices (waist circumference [WC], waist-hip ratio [WHR], and waistheight ratio [WHtR]) with myocardial infarction and death events are higher than the correlation of body mass index (BMI) with these events. 4,5 Thus, obtaining data on the status of abdominal obesity is important in risk factor assessment and prevention of non-communicable diseases. 6 Evaluation of abdominal obesity indices in field studies can be problematic, especially in low-resource settings, due to several reasons. First, it is expensive and time-consuming to provide the required instruments for accurate measurement on a large scale. 7 Second, measuring these indices requires experienced technicians. Finally, in line with the development of telemedicine, alternative assessment methods should be provided that do not require the physical presence of the participants.
Body shape pictogram (silhouette) can be a potential alternative of using direct measurements to classify abdominal obesity status. Pictograms are sets of body images ranging from very lean to extremely obese for each gender. 8,9 Participants are required to select the picture of the pictogram most similar to their body shape. This instrument has been previously used in several studies for the main purpose of BMI classification. 10,11 A previously designed pictogram by Stunkard et al 11 has also been validated to estimate BMI in a study conducted in northern Iran. 9 To the best of our knowledge, the validity of using pictograms in the classification of abdominal obesity status has not been previously studied in the Iranian population. In this study, we aimed to evaluate the validity of using pictograms to classify WC, WHR, and WHtR categories and also to determine the demographic factors that may alter its classification accuracy in a representative adult population from southern Iran.

Study Design and Population
This cross-sectional study was conducted using the baseline data obtained from Pars Cohort Study (PCS). PCS is an ongoing population-based prospective study that started in the fall of 2012 in southern Iran, on a semiurban multi-ethnic population aged between 40 and 75 years. Details of the study design are published elsewhere. 12

Variable Measurement
PCS data were collected using personal interviews, physical examination, and biological sampling based on the study manuals using standardized and calibrated tools by well-experienced personnel. In this study, we used data on age, gender, marital status, ethnicity, education, physical activity, list of household appliances, assets and entertainments, and anthropometric measures including a single measure for height, weight, waist, and hip circumferences. Anthropometric measurements were done when participants wore light clothing, emptied their pockets completely, and did not wear shoes. These measurements were performed once. Height, WC, and hip circumference were measured to the nearest centimeter. WC was measured at the midpoint between the lower margin of the last palpable rib and the top of the iliac crest. The widest portion of the buttocks was considered for measurement of the hip circumference. 13 Participants' weight was measured to the nearest 100 g. One kilogram of measured weight was deducted, equivalent to the average weight of light clothing.
Data on the individual perception of body size and appearance were previously collected using a self-reported questionnaire. The body silhouettes used in this study were previously designed by Stunkard et al. 11 The participants were asked to choose their body image perception at the ages of 15 and 30 years, and at the time of the interview. The pictograms showed increasing body size in ascending order and included 7 and 9 pictures for male and female individuals, respectively ( Figure 1).

Data Cleaning and Preparation
The data were assessed for internal and external consistency and also rechecked with the hard copy if needed. Cleaned data were prepared for the analyses.
Socioeconomic status (SES) was determined by assets analysis using multiple correspondence analysis of data on the properties. Several items were used to determine the SES of the participants, including phone at home, cell phone, washing machine, dishwasher, microwave device, camcorder, car, household residential area, number of rooms per household members, the main cooling devices and the main cooking device. Then, the participants were categorized into four ordered SES groups. 14 Physical activity (PA) was assessed based on the Metabolic Equivalent of Task (MET) score. We implemented International Physical Activity Questionnaire (IPAQ) and used the standard method for its analysis. Then, MET scores were divided into three thirties, named low, medium and high PA. 15 The long-term weight cycling variable was defined based on the selected pictograms at three cut-points of 15 years, 30 years, and current age. Individuals who selected the 30-year pictogram as their lowest or highest score (demonstrating a fluctuation in the pattern of pictograms over time) were labeled as cyclers. The remaining participants were classified as non-cyclers.
BMI was divided into underweight/normal (< 25), overweight (≥ 25 and < 30), and obese (≥ 30) similar to the cut-off used by Keshtkar et al. 9 Age was categorized into three separate groups of < 50 years, ≥ 50 and < 60 years, and ≥ 60 years. Education and ethnicity variables were classified into 3 groups each (illiterate, ≤ 12 years of education, and university degree), and (Persian, Turk, and others), respectively. WC was categorized into two groups of at-risk and normal based on a limit of ≥ 90 cm for both genders. 16 WHR was classified into two groups of at-risk and normal with a cut-off value of ≥ 0.95 for males and ≥ 0.90 for females. 17 Finally, WHtR ≥ 0.51 for both genders was classified as at-risk based on a previous study in the Iranian population. 18,19 Statistical Analysis Descriptive statistics including frequencies, median and interquartile range (IQR) were estimated. Genderspecific directly age-standardized proportions of at-risk WC, WHR and, WHtR along with their 95% confidence intervals (CIs) were estimated using world standard population 2000-2025. The linearity assumption for the association between successive levels of the pictogram's ordinal scores and different central obesity indices was evaluated for each gender. Pearson's correlation between the pictogram score and different measures of abdominal obesity was assessed for each gender.
Previously defined cut-off values for WC, WHR, and WHtR in the Iranian population were used as gold standard measures for classifying people into normal and at-risk categories of central obesity. Sensitivity and specificity values and their 95% CI were calculated for each of the pictogram pictures to discriminate participants with at-risk values of WC, WHR, and WHtR from normal values. The best cut-off value was determined based on sensitivity/specificity plots and optimized Youden Index which is defined as sensitivity + specificity -1. 20 The classification accuracy of using pictogram to divide people into normal and at-risk categories based on each central obesity measure was assessed using the area under the receiver operating characteristic curve (AUC) [%, 95% CI] obtained from logistic regression models. To assess the effect of background variables of the target population on the validity of pictograms, stratified statistical analyses were done using the test for the equality of AUCs based on an algorithm suggested by DeLong et al. 21 Stata software version 14.1 (College Station, TX: Stata Corp LLC) was used and a P value < 0.05 was considered as statistically significant.
Using WC, WHR, and WHtR as gold standard measures to define central obesity resulted in pictogram score cutoff values of 4, 4, and 5 in males and 4, 4, and 6 in females, respectively (Table 3 and Table S4). In males, the sensitivity and specificity values based on the optimal cut-off values were 74% and 81% for WC, 68% and 74% for WHR, and 80% and 84% for WHtR, respectively. The corresponding values in females were 83% and 72% for WC, 71% and 64% for WHR, and 77% and 86% for WHtR, respectively.
Most of the individual characteristics did not have a statistically significant association with the accuracy of pictogram scores to classify central obesity. For both males and females, SES, PA, age group, ethnicity, marital status, and education had no effects on the classification accuracy of pictograms to determine central obesity. However, weight cycling had a significant association with the accuracy of pictogram scores classifying central obesity defined by WC and WHR (P value < 0.001), but not with WHtR (P value < 0.001 for both; Table 4).
Sensitivity analysis was done based on the cut-off of abdominal obesity indices defined by the WHO and is presented in Supplementary file 1.

Discussion
The present study confirms the validity of using Stunkard's set of pictograms to determine the population at risk of developing adverse health outcomes due to central obesity in the Iranian adult population. In both males and females, the highest AUC for predicting adverse health outcomes with pictograms was achieved when defining central obesity based on WHtR (0.89, 0.90), WC (0.84, 0.84), and WHR (0.77, 0.73), respectively. We also demonstrated the optimal cut-off for differentiating at-risk persons for each central obesity indices. Also, we have shown that the validity of pictograms is not affected by SES, PA, age group, ethnicity, marital status, and education; however, weight cycling influenced this validity when defining central obesity with WC and WHR in both genders.
Most of the previous studies on the assessment of the accuracy of pictograms have focused on BMI. Pictograms were validated to classify obesity based on BMI with AUCs of 0.93 for females and 0.88 for males in the Caucasian population. 22 Another study in the Japanese population validated the use of body shape silhouettes to discriminate between obese and thin individuals based on BMI with AUCs of > 0.80 in both genders. 23 Lo et al 24 found Stunkard's pictograms to be superior to self-reported WC and waist to height ratio in the assessment of weight status (AUCs of > 0.8 in both genders) in Chinese adolescents. In the young healthy Spanish population, pictograms were validated to determine weight status defined by BMI with AUCs of 0.9 and 0.8 for males and females, respectively. 25 Additionally, body silhouettes are shown to be reflective of past obesity based on measured and self-reported BMI in the European population. 10 Body show cards were also validated to determine crude (based on BMI) and central (based on WC, WHtR) obesity in the African population with high accuracy (AUCs of > 0.9 in both genders). 26 To date, only one Iranian study is conducted on the validity of pictograms to classify obesity. This study shows that Stunkard's pictograms are valid to determine crude obesity (defined by BMI) in the Iranian population. 9 Our study validates the accuracy of using the same set of pictograms to determine central obesity in the Iranian population with AUCs of 77% to 89% in males and 73% to 90% in females for different indices of central obesity.
There is great variance in the age-standardized prevalence of obesity in our population using different definitions of central obesity (WC, WHR, or WHtR). Each index evaluates central obesity from a different perspective, and the optimal cut-off values for the classification of at-risk groups of the population using these indices are not necessarily obtained based on the same outcomes. Thus, an individual may be classified as at-risk according to a certain index while categorized as normal using another one at the same time. Despite being generally valid to classify central obesity, pictograms have higher accuracy to classify at-risk groups when central obesity is defined based on WHtR and WC rather than WHR. This finding results from higher pictogram score correlations with WHtR and WC than its correlation with WHR. This is the first study to assess the accuracy of pictograms to determine central obesity defined by WHR. A previous study in the African population revealed high accuracy (AUCs > 0.90) for the classification of central obesity defined by WHtR and WC using a different set of body show cards. 26 These findings propose the currently developed body silhouettes as valid tools to evaluate central obesity defined by WHtR and WC. However, the results may not be readily translated to other measures of central adiposity like WHR. An underlying reason may be attributed to the inherent nature of the currently available show cards with more focus on horizontal and vertical dimensions of pictograms to demonstrate obesity rather than highlighting the ratio of waist to hip. Besides, most people may perceive obesity to be related to the degree of abdominal circumference or its ratio to height rather than other existing definitions of obesity like WHR. Previous studies report that sociodemographic and psychological factors are associated with self-perception of weight status. Dorsey et al 27 found that weight misperception is highly prevalent among the US population  and is associated with ethnicity and educational status. In another study, married women were shown to be more likely to perceive themselves as overweight while such an association was not found in men. 28 Moreover, evidence suggests that the level of education affects the selfperception of obesity in the Korean adult population. 29 In addition, another study including 5440 US adult population shows that race, SES, and level of education are associated with self-perception of being overweight. 30 On the other hand, investigators demonstrated that weight misperception is associated with psychological distress and anxiety in the Australian and Chinese populations. 31,32 Thus, there is the possibility that misperception of weight status affects the accuracy of self-reported body silhouettes to determine central obesity. Our results indicate that the utility of this instrument is not greatly influenced by the sociodemographic properties of individuals. SES, physical activity, age group, ethnicity, marital status, and level of education did not affect the validity of body silhouettes to classify central obesity. Although previous studies found an association between these factors and self-perception of weight, almost all of them have used self-reported height and weight or asked whether the participants have the feeling of being overweight. [27][28][29][30]33,34 Selecting the closest silhouette to body shape appears to be more realistic and less prone to subjective errors observed with previous study designs.  Most of our population belonged to Persian and Turk ethnicity, two major ethnic groups living in Iran. Genderspecific validity of body show cards to determine central obesity was almost the same between different ethnicities. In cases of large differences in performance, there might be a necessity to design new sets of body show cards that more closely resemble the morphologies observed in specific ethnicities. However, the results of our study are in favor of the generalizability of Stunkard's pictograms to detect central obesity among different Iranian ethnicities. Alterations in body appearance may occur as a result of aging. 35 Accordingly, we tested the potential influence of age group on the accuracy of pictograms classifying central obesity. Our results showed an almost similar classification accuracy among different age groups that support using the same set of pictograms across different age groups of the adult population. However, our study population was older than 40 years; the age group in which people's (especially females') perception of their weight status is probably less affected by social values and individual expectations. 36 This may result in less bias and a more realistic interpretation of anthropometric status than that observed in a younger age group. Although weight cycling is used to point to previous fluctuations in weight status, [37][38][39][40][41] it is a term with relative meaning depending on the context in which it has been described. Previous studies on weight cycling have mostly focused on its psychosocial and metabolic consequences. [38][39][40]42,43 Overall, there has been controversy on the association of weight cycling with the risk of adverse metabolic outcomes 39,43 ; however, an association has been shown between a history of weight cycling and central fat accumulation in the body. 38,44 Furthermore, previous evidence suggests the association between the history of weight cycling and the accuracy of self-reported weight status. 41 Our findings indicate that the history of weight cycling negatively impacts the performance of body pictograms to accurately predict central obesity status. Hence, caution should be made on the interpretation of pictogram ratings in those who report previous fluctuations in their weight. It can be assumed that those with weight cycling were probably not satisfied with their actual weight. The discrepancy between one's actual and ideal body weight as well as the effects of weight cycling on central adiposity can both account for the decreased accuracy of pictograms in cyclers.
To date, this is one of the largest studies using the pictogram ratings for the evaluation of obesity in the Iranian population. The population-based nature of our study with more than 9000 participants and adherence to precise measurement protocols make our results generalizable to other nations in developing countries. However, this study has certain limitations. First, our study population was older than 40 years and future studies are needed to assess the validity of body show cards in children and young adults. Second, this study was conducted in a semi-urban area; thus, the sociodemographic characteristics of our population may not be fully representative of those living in high social classes of urban areas. Moreover, participants self-determined the closest body show card to their actual body shape. There has been a higher correlation between pictogram scores selected by expert anthropometry observers than those selected by one expert and one less skilled observer. 45 However, our results have better generalizability to the situations where individuals have to self-determine their scores with implications in large studies and telemedicine.
In conclusion, the pictogram is a valid tool to classify central obesity in the adult population and the ratings can be better interpreted with respect to the sociodemographic context of the target population. This sets a framework to access anthropometric information on a large scale that would have been otherwise difficult to obtain due to the problems associated with human resources, device expenses, and lack of direct access to people.