Waist circumference, waist‐to‐height ratio and BMI percentiles in children aged 5 to 19 years in India: A population‐based study

Abstract Objective Nationally representative percentiles for waist circumference (WC), waist‐to‐height‐ratio (WHtR), and body mass index (BMI) are not available for children and adolescents in India. Methods Using LMS method, age‐ and gender‐specific reference growth charts were constructed for WC (n = 68,261), WHtR (n = 68,261), and BMI (n = 67,741) from children/adolescents aged 5–19 years who participated in a nationally representative survey. General obesity, indicating overall obesity, was defined as age–sex‐specific BMI z‐scores ≥ 95th percentile. Central obesity was defined in three ways: WC ≥ 90th percentile, WHtR ≥ 0.5, and both WC ≥ 90th percentile and WHtR ≥ 0.5. Findings WC and BMI percentiles for boys and girls are lower than those previously reported from India and several other countries. The BMI percentiles are lower than the WHO 2007 reference population. The prevalence of general obesity using India specific BMI centiles was 2.9% (95% CI: 2.6–3.2). The prevalence of central obesity was 6.1% (95% CI: 5.7–6.6) using WC ≥ 90th percentile, 5.3% (95% CI: 5.0–5.7) using WHtR ≥ 0.5, and 3.6% using both criteria. Three‐fourth of children with general obesity also had central obesity based on WC ≥ 90th. Conclusions Indian children are thinner than Caucasian and other Asian children, and the global WHO reference population. Using India specific reference, the prevalence of central obesity is higher than general obesity with a significant overlap between the two.

children are prone to develop central obesity highlighting the need for early intervention. 16 Over the last decade, several groups have applied the LMS model to create smoothed percentile charts for WC, and in some countries WHtR in children and adolescents from Malaysia, 10 Pakistan, 9 Poland, 11 and Turkey 17 among others. In India, most studies have created LMS percentile curves for WC and WHtR among children and adolescents from urban schools. 18 Similarly, BMI percentiles have been created with children from urban, middleto-upper socio-economic strata. 19,20 India is a diverse country with 70% rural population and varied economic status and ethnicity.
Presently, percentile curves for WC, WHtR, and BMI drawn from a national representative population of Indian children/adolescents are not available. Obtaining representative, normative information for WC, WHtR, and BMI is necessary for reliable identification and prevention of overweight and obesity and associated cardiometabolic risk factors among Indian children.
The LMS method was used to create age-and gender-specific reference growth charts for WC, WHtR, and BMI in Indian children and adolescents aged 5-19 years using data from the nationally representative Comprehensive National Nutrition Survey (CNNS 2016(CNNS -2018. Based on India-specific percentiles, the prevalence of general and central obesity was determined, and socio-demographic differentials examined with the objective of informing national policy and programmes, and to serve as a baseline for future comparisons.

| METHODS
The CNNS was conducted under the aegis of the Ministry of Health and Family Welfare (MoH&FW) in collaboration with UNICEF and the Population Council. The CNNS was designed to provide nationally representative and comprehensive nutritional profiling of preschoolers (0-4 years), school-age children (5-9 years) and adolescents (10-19 years), based on biological sample assessment and multiple anthropometric measures. This paper focuses on schoolage children and adolescents (5-19 years).

| Study design and participants
The survey design and methodology are published elsewhere. 21 Briefly, the CNNS used a multi-stage, stratified, probability proportion to size cluster sampling design to select a nationally representative sample of households and individuals aged 0-19 years across all 29 states of India and the capital Delhi.
Households with individual(s) between 0 and 19 years were randomly selected from rural and urban primary sampling units (PSU); children/adolescent members were classified into three strata (0-4, 5-9, and 10-19 years), and only one child/adolescent was selected from each stratum per household. The sample size was set at 122,100 (40,700 in each age group) from 2035 PSUs to provide national, state-level, and rural-urban estimates. 21 Children/ adolescents who had a chronic illness, physical deformity, mental illness or cognitive disability, or any ongoing current illness (fever, infection) were excluded from the survey. The survey collected socio-demographic data: place of residence, wealth index, religion, caste, mother's education, safe water, and sanitation from questionnaires and anthropometry data. 21 SARNA ET AL. -393

| Study sample
Children/adolescents aged 5-19 years were included in this analysis.
Participants for whom data on height, weight, or WC was missing were excluded from this analysis (Figure 1). At the time of anthropometric measurement, a few participants were detected to have a physical deformity that was not evident at the time of recruitment into the survey (e.g., scoliosis, kyphosis, bow-legs etc.); these participants were excluded from the analysis (n = 215). We created three analytical samples. Our first analytical sample included all the eligible participants. As a sensitivity analysis, we constructed two reference populations: (i) after excluding very thin (<−3SD) and very obese (>+3SD) participants (analytical sample 2) and (ii) after excluding thin (<−2SD) and obese (>+2SD) participants (analytical sample 3) based on age-and sex-adjusted BMI z-scores using the WHO 2007 growth reference chart (Figure 1).

| Anthropometric measurements
The anthropometric parameters included were height, weight, and waist circumference. Trained female health workers collected all anthropometric data. Height was measured in centimeters on a SECA height board (to the nearest 0.1 cm); the mean of two readings was recorded. Weight was measured in kilograms (up to 0.01 kg) using a SECA portable digital weighing scale; only one reading was taken/ recorded. Waist circumference was measured in centimeters (to the nearest 0.1 cm) at the midpoint between the lowest rib and the iliac crest in the mid-axillary line at the end of normal expiration using a non-elastic measuring tape. 22 The mean of two readings was recorded. Rigorous quality monitoring was maintained including weekly calibration of the height board and daily calibration of the weighing scale and repeat measurements by quality monitors (CNNS report). 21 For height measurement, the inter-and intra-technical error of measurement (TEM) scores were within the global cutoffs of 0.95 and 0.69 cm. 23 There are no global TEM cutoffs for WC.

| Statistical methods
The LMS method was used to compute age-and sex-specific percentiles for WC, WHtR, and BMI. WHtR was calculated as waist (centimeters)/height (centimeters), and BMI was calculated as weight (kgs)/height 2 (meters). Each measurement was summarized by three smooth curves plotted against age representing the median (M), coefficient of variation (S) and skewness (L) of its distribution. The Box-Cox-Cole-Green (BCCG) distribution with penalized spline smoothing was used to construct smoothed age-sex specific percentile curves of WC, WHtR, and BMI for the three analytical samples. 24 LMS values and percentiles were calculated using the general additive models for location, scale, and shape(GAMLSS) 4.3-1 library under R3. 1.2. 25 Goodness of fit of the models was accessed by the Bayesian information criterion and by Q-Q plots. 26  Turkey, 17 Malaysia, 10 and Pakistan 9 ), and a study by Khadlikar and colleagues 18 for an Indian population that used the same measurement methods. Similarly, BMI 50th and 90th percentile curves were compared with India (Indian Academy of Pediatrics), 19 Malaysia, 28 Poland, 29 Turkey 30 and the WHO global reference population. 5 For sensitivity analysis, data distribution of the reference populations in analytical samples 1: all eligible participants, sample 2: excluding >+3SD and <−3SD participants and sample 3: excluding > +2SD and <−2SD were compared. Age-and sex-specific 50th and 90th percentiles for WC and 50th and 95th percentiles for BMI were developed with 95 percent confidence intervals (95% CI) using raw values of WC and BMI. 95% CIs were calculated using simple standard error and bootstrap method with 1000 repetitions. The 95% CI in analytical samples 1, 2, and 3 were similar for 90th percentile for WC and 95th percentile for BMI indicating no significant differences at values used to define obesity ( Figure S3).
WHO defines obesity as age-and gender-specific BMI z-scores =>2SD (95th percentile) from percentiles; we used this cutoff based on newly constructed percentile reference charts to define general obesity. 5 The National Health and Nutrition Examination Survey (NHANES) has proposed an age-and sex-specific cutoff of ≥90th percentile of WC for identifying central obesity; this cutoff has been used by several studies. 1,10,11,31 WHtR has also been used to define central obesity with a fixed cutoff of ≥0.5. 32 Recently, studies have combined WC and WHtR to define central obesity, as age-and sex specific WC percentile ≥ 90th and WHtR ≥ 0.5. 9 Central obesity is reported based on all three indicators: WC ≥ 90th percentile, WHtR > 0.5, and both WC ≥ 90th percentile and WHtR > 0.5.  and 0.1% (307/67,741) was very obese (BMI z-score≥3SD); more boys were thin compared to girls (7.7% vs. 4.1%) (data not shown).
Detailed age-and sex-specific descriptive statistics are provided in Table S1.

| Percentiles
WC and WHtR percentiles are presented in Figure 2. Corresponding percentile values and LMS parameters are presented in Table 2. WC increased with age in both boys and girls; there were marked sex differences in the shape of centile curves. Girls had lower WC values than boys at any age and percentile, and these differences increased F I G U R E 1 Flow chart for the analytical sample which included all participants with height, waist circumference, and BMI measurements SARNA ET AL.  Table 1 and corresponding centile curves presented in Figure 2. BMI increased with age in both boys and girls; however, there were marked sex differences in the shape of the curves. Girls had lower BMI values than boys at younger ages (5-10 years), and higher BMI values thereafter. Among girls, there was a sharper increase in BMI between 11 and 14 years, then gradual plateauing from age 15 years. Boys exhibited a steady increase with a marginally higher increase between 11 and 15 years. There were no significant differences in centile curves and corresponding BMI values with the "reference" population sample 2 (Table S1, Figure S1). For sample 3, there were no differences at higher percentiles, but significant differences were observed at percentiles below the median (Table S3, Figure 2).
The percentiles for WC from CNNS were lower than those from the US, Poland, Turkey, Malaysia, and Pakistan ( Figure 3; Table S2).
The WC centiles were also lower than those previously reported by from India by Khadlikar (2014). BMI centiles from CNNS were lower than those from Poland, Turkey, Malaysia, Pakistan (available for 5-12 years) and the WHO reference population (Figure 3, Table S3).
Additionally, BMI percentiles were also lower than reported previously by the Indian Academy of Pediatrics (IAP, 2015).

| Prevalence of obesity
The prevalence of BMI based general obesity using the WHO reference was 1.1% (95% CI: 1.0-1.3), marginally higher among males ( Table 3). The prevalence of general obesity was significantly higher when based on the Indian population specific centiles (2.9%; 95% CI: 2.6-3.2); there was no difference between males and females. Central obesity based on WC ≥ 90th percentile was 6.1% (Male: 6.0%; 95% CI: 5.5-6.6 vs. Female: 6·.2%; 95% CI: 5.7-6.6; p = 0.095). 76% of the children identified under the general obesity (using India specific BMI centiles) were also identified as having central obesity ( Table 2). Central obesity based on WHtR ≥ 0.5 was 5.3% (95% CI: 5.0-5.7); the prevalence was higher among females (Female: 5.7%; 95% CI: 5. 95% CI: 2.9-3.6; p = 0·.11). 65% of children identified as having general obesity (India specific centiles) were also identified as centrally obese by this criterion. Table 4 shows the mean and standard deviation of WC, WHtR, and BMI z-scores across socio-demographic characteristics of the study population. For all three measures-WC, WHtR, and BMI-the mean z-scores were significantly higher (p < 0.001) in urban subjects, in those who were economically better off; those from higher castes; those with educated mothers and those with access to safe sanitation.

| DISCUSSION
This paper presents age-and sex-specific WC, WHtR, and BMI percentile curves drawn from a nationally representative population of children and adolescents (5-19 years) in India with strong emphasis on quality control and monitoring. In conformity with earlier Indian and global studies, girls had lower BMI values than boys   international studies, the CNNS percentiles centiles were substantially lower than those from the US (WC), 27 Malaysia (WC and BMI), 10,28 Poland (WC and BMI), 11,29 Turkey (WC and BMI), 17,30 Pakistan (WC and BMI available only for 5-12 years), 9,38     Consensus Statement on WC recommends the use of WC in addition to BMI to assess obesity. 8 It has been observed that both BMI and WC,or WHtR perform similarly when predicting a cluster of cardio-metabolic risk factors, with greater effect seen among obese children. 39 Further analyses are needed to assess the relative utility of BMI, WC, and WHtR cut-offs used in this survey for predicting associated cardio-metabolic risk factors.
The study found lower age-and sex-specific z-score values for BMI, WC, and WHtR in rural areas and in poorer households, suggesting that at a national level obesity exists largely in well-off urban pockets. Other indicators pointing towards an association between wealth and overweight/obesity were higher z-scores for BMI, WC, and WHtR in households with better sanitation, higher educational attainment of mothers, and higher caste. This is further supported by evidence from the comparison of centile values from economically better off study populations and the CNNS data, reported above. Changes in lifestyle with urbanization including reduced physical activity, increased sedentary living, and unhealthy diets may be probable underlying causes. Mushtaq and colleagues reported similar findings from Pakistan, a country with an ethnically similar population to India. 9 In conclusion, the conventional metrics recommended for identifying children with general or central obesity are consistently lower with the nationally representative CNNS reference than several international references including that from the WHO. This suggests that Indian children and adolescents are relatively thinner which could be due to a combination of genetic, environmental, and inter-generational factors. It is therefore possible that prediction of cardio-metabolic risk factors associated with central or general obesity would be lower if international cut-offs were employed. Further analyses are required to determine cutoffs associated with biomarker-based cardio-metabolic risk factors in this population. Finally, the nationally representative reference will prove invaluable for documenting and comparing the details, especially the distribution of secular trends in this population.

ACKNOWLEDGMENTS
We would like to thank all the study participants for providing their valuable time. We are grateful to all the reviewers for providing valuable feedback which helped improve the article. The study was funded by Aditya and Megha Mittal, Mittal Foundation, UK.