Childhood obesity is associated with increased risk of obesity in adulthood and with increased cardiovascular risk factors (CVRF) in children(Reference Freedman, Dietz and Srinivasan1,Reference Paradis, Lambert and O’Loughlin2) and adults(Reference Umer, Kelley and Cottrell3). The prevalence of obesity in children has increased rapidly, particularly in low- and middle-income countries(Reference Lobstein, Jackson-Leach and Moodie4). In Ho Chi Minh City (HCMC), the largest city in Vietnam, a lower to middle-income country, the prevalence of overweight and obesity in schoolchildren (6–18 years old) increased from 12 % in 2002, to 22 % in 2009 and to 41 % in 2014(Reference Tran, Vu and Do5,Reference Hạnh, Diệp and Hoa6) . CVD are a leading cause of death for adults in Vietnam(7). Thus, early detection of childhood obesity and concomitant CVRF is needed to inform public health strategies and prevent increases in the burden of CVD in Vietnam and similar countries in the South-East Asian region.
Anthropometric indicators such as BMI, waist circumference (WC) and waist-to-height ratio (WHtR) can be used as an accurate and cost-effective method for discriminating variations in overweight, obesity and CVD risk in children and adolescents(Reference Alves Junior, Mocellin and Goncalves8). WC is commonly used as an abdominal obesity indicator, which has shown additional benefits to BMI for prediction of cardiometabolic risk factors in children(Reference Samouda, de Beaufort and Stranges9). WHtR is a simple indicator that only requires a tape measure for assessment and is less influenced by age, sex and ethnicity than BMI. It has previously been found to be useful but not better than WC and BMI for identification of CVRF in an international sample of children(Reference Lo, Wong and Khalechelvam10).
Binary cut-offs rather than continuous variables have been promoted most commonly for CVD risk prediction(Reference Zimmet, Alberti and Kaufman11,12) . However, the most optimal cut-offs to be used for this purpose, in particular in children and adolescents of different ethnicities, are not well established(Reference Ford, Patel and Narayan13). A cut-off of 0·5 for WHtR is generally recommended for identification of children at increased risk of CVD(Reference Ochoa Sangrador and Ochoa-Brezmes14), whereas WHtR cut-offs of 0·44 (boys) and 0·43 (girls) are recommended for Korean children(Reference de Onis, Onyango and Borghi15). It is unclear which cut-off is suitable for Vietnamese children(Reference Lo, Wong and Khalechelvam10). In addition, a BMIZ of 2·0 sd is the cut-off used globally to identify children with obesity. However, this cut-off is not necessarily optimal to identify body fatness or health outcomes in specific ethnicities. Asian children tend to have higher TAG levels, insulin resistance and CVD incidence at a lower BMI cut-off compared with their non-Asian counterparts(Reference Hastuti, Kagawa and Byrne16). Furthermore, consideration of anthropometric indicators as continuous variables rather than dichotomised criteria may have additional benefits. The performance of BMI, WC and WHtR in the identification of two specific CVD risk factors, elevated blood pressure (BP) and dyslipidaemia, is therefore evaluated in a large survey of children and adolescents in Vietnam. Optimal cut-offs are identified and compared with commonly used cut-offs, and the usefulness of discrete compared with continuous anthropometric indicators for this purpose is considered.
Methodology
Study population
Data from the Survey of Nutritional Status Among School-aged Children conducted by the HCMC Nutrition Center funded by the Department of Health, HCMC, Vietnam, was used in the present study(Reference Hạnh, Diệp and Hoa6). The primary objectives of this survey were to estimate the prevalence of stunting, thinness and overweight, to estimate the prevalence of hypertension, to estimate the mean height of children in each age, to estimate the prevalence of lipid disorder and the prevalence of Fe deficiency in school-aged children in primary, secondary and high school in HCMC, Vietnam. The sample size for data collection was selected from the largest sample size requirement for each objective. The largest sample size of 10 900 students was from the estimation of mean height for each age group from 6 to 18 years in school-aged children in HCMC. This estimation was calculated from the standard deviation of height for age from the nutritional survey in school-aged children in HCMC in 2009 with 95 % CI, a marginal error of 0·5 (cm) and a design effect of 2·0. The sample size of 950 children for sub-group of children with blood test was calculated from the estimation of the prevalence of Fe deficiency in each school level from the prevalence of Fe deficiency in each school level from a previous survey, with 95 % CI and the marginal error of 5 % and a design effect of 2·0.
Between October 2014 and January 2015, data were collected from school-aged children in HCMC using a two-stage cluster sampling design, with stratification by school level, location and sex. Sample size for each school level was calculated based on nutritional status distributions from a previous survey. All schools in HCMC were categorised by school level (primary, secondary and high school) and location (urban and rural). Probability-proportion-to-size sampling was used to select schools from these school categories, and for selection of students within the schools(Reference Hạnh, Diệp and Hoa6). Approximately, 20 % of students in HCMC live in rural districts within the city boundaries(17). A total of thirty schools were selected: eleven primary schools including nine urban and two rural; ten secondary schools including eight urban and two rural and nine high schools including eight urban and one rural. Students were selected from each school’s student list stratified by sex and selected from each grade using systematic random sampling. In total, 500 students were selected from each primary school, 300 from each secondary school and 250 from each high school. For a blood lipid test, fifty students were selected at random per school in seven primary schools, and in six secondary schools, and ninety students per school in five high schools. All children and their parents provided informed written consent prior to data collection. Children with disorders affecting their ability to be accurately weighed and measured such as severe scoliosis, and urgent medical conditions such as high fever or diarrhoea, were excluded from the study before data collection. Ethics was approved by the Scientific Committee at Ho Chi Minh City Nutrition Centre (approval number: 404/QD-TTDD). The secondary analyses received an ethics exemption from the Queensland University of Technology Office of Research Ethics and Integrity (approval number: 1800000880).
Anthropometric measurements
Height, weight and WC were measured by trained health officers using standardised WHO guidelines(18). Children wore light clothes and no shoes during measurement. Weight was measured to the nearest 0·1 kg using electronic scales (HD313; TANITA). Height was measured using a wooden stadiometer and WC using non-elastic tape-measures against the skin at the midpoint between the lower costal border and the top of the iliac crest at the end of expiration, to the nearest 0·1 cm. The circumference at the umbilicus was used if the anatomical landmarks could not be identified. BMI was calculated as weight (kg) divided by height squared (m2). WHtR was calculated as WC (cm) divided by height (cm).
Cardiovascular risk factors
Obesity, elevated BP and dyslipidaemia are CVRF in children(Reference Friedemann, Heneghan and Mahtani19,Reference Rodrigues, Abreu and Resende20) . In the present study, anthropometric indicators were used to identify obesity; thus, we did not include obesity as one of CVRF in the analysis. Overweight was defined as age–sex-specific BMI z-score (BMIZ) > 1·0 sd, and obesity was defined as age–sex-specific BMIZ > 2·0 sd(Reference de Onis, Onyango and Borghi15). All subjects had BP measured in the morning, seated in a quiet room after 10 min rest, on the right arm using electronic sphygmomanometer (HEM-7121; OMRON) with the appropriate cuff-size according to the arm circumference(Reference Arafat and Mattoo21). BP was measured for a second time after an additional 10 min rest. If the first BP reading was abnormal (≥120/80 mmHg), the second reading was used. Elevated BP was defined as systolic BP and/or diastolic BP ≥90th percentile for sex, height and age (6–12 years) and ≥120/80 mmHg (≥13 years)(Reference Flynn, Kaelber and Baker-Smith22). Blood tests for the concentration of total cholesterol, HDL, LDL and TAG were performed in a sub-group of 1009 children after 8 h of fasting. Dyslipidaemia was identified as having one of following: high cholesterol (total cholesterol ≥ 5·18 mmol/l); hypertriacylglycerolaemia (TAG ≥ 1·13 mmol/l (6–9 year) or ≥1·47 mmol/l (10–18 years); low HDL (HDL < 0·91 mmol/l) or high LDL (LDL ≥ 3·37 mmol/l)(Reference Jolliffe and Janssen23). The summary measure ‘combined cardiovascular risk factors’ (CVRF) was created to indicate children who had at least three out of five cardiovascular disorders including elevated BP, high total cholesterol, hypertriacylglycerolaemia, low HDL and high LDL.
Statistical analysis
Analyses were performed using the ‘svy’ prefix in survey analysis in Stata package version 15.1 (StataCorp) to account for the two-stage cluster sampling design. The ‘weight’ value in survey analysis was used separately for the total sample of 10 949 subjects and the sub-sample of 1009 subjects due to the different method of sampling for the sub-sample. These values were calculated from the total number of children in each school, total number of children in each stratum, number of school in each strata and sample size for each school. General characteristics were presented as mean or prevalence and standard errors. The differences in general characteristics between boys and girls were tested using two-sided independent t test in survey analysis. P <0·04 was used to indicate statistical significance. To avoid the influence of age and sex on BMI, BMIZ were used, derived from the mean and standard deviations of the WHO reference population, using WHO Anthro-plus(Reference de Onis, Onyango and Borghi15). WC z-scores (WCZ) were derived from the mean and standard deviations of the study population.
Receiver-operating characteristic analysis was used to evaluate the capacity of anthropometric indicators to identify elevated BP, dyslipidaemia and the combined CVRF. Predicted probabilities were derived, and receiver-operating characteristic analysis was carried out in Stata using the ‘roctab’ command, treating the anthropometric variables as continuous variables. AUC with 95 % CI was reported with higher values of AUC indicating a higher capacity of prediction, AUC = 1·0 indicates perfect prediction(Reference Swets24) and AUC > 0·5 indicates a capacity to discriminate a health outcome. This capacity could be classified as follows: 0·5 < AUC < 0·7 for less accurate, 0·7 ≤ AUC < 0·9 for moderately accurate and AUC ≥ 0·9 for highly accurate(Reference Greiner, Pfeiffer and Smith25). The present study was able to detect an AUC of 0·6 with 90 % power and a significance level of 0·05 in all sub-group analysis. In addition, the performance of the capacity to identify elevated BP, dyslipidaemia and combined CVRF is presented as the OR of having these outcomes for every increase in one unit of the anthropometric indicator (1·0 sd increase in BMIZ and WCZ and 0·1 increase in WHtR). These were derived from a logistic regression model adjusted for age and sex when appropriate.
Differences in AUC were analysed with the command ‘roccomp’ in Stata, using algorithm suggested by DeLong(Reference Delong, Delong and Clarke-Pearson26) to evaluate the equality of AUC between WHtR, BMIZ and WCZ to identify CVD risk factors, and the differences by sex or age group. The significant difference was identified by the χ 2 test with P <0·04. To facilitate comparisons between different age groups, children aged 6 to 9 years were defined as children and children aged 10 to 18 years were defined as adolescents(Reference Sawyer, Azzopardi and Wickremarathne27). Due to the relatively small number of children in age- and sex-specific sub-groups with CVRF in the sample with blood lipid data, all analyses by sex were conducted for all health outcomes, and analyses by age were conducted for elevated BP only. The receiver-operating characteristic regression analysis with univariate logistic regression models were used to calculate the Youden index from sensitivity and specificity at each cut-off. The optimal cut-off for anthropometric indicators was defined based on the maximum Youden index (J value = sensitivity + specificity – 1)(Reference Youden28).
The probability of having the outcome at a certain cut-off was calculated from the logit model of anthropometric indicators (as dichotomised variables) and health outcomes. The OR were calculated from logistic model of anthropometric indicators and health outcomes at each specific anthropometric cut-off, adjusted for age and sex and no adjustments. Odds and 95 % CI were reported with the statistical significance being considered P < 0·04 using Bonferroni correction for six random tests.
Results
In total, 10 949 subjects were included in the analyses, 50·6 % were male and mean age was 10·7 (sd 3·4) years (range 6–18 years). Analyses that considered blood lipids were based on the sub-sample of 1009 children (48·2 % male) who had blood lipid data. The prevalence of elevated BP, dyslipidaemia and the combined CVRF was 26·5, 49·3 and 12·2 %, respectively (Table 1). All anthropometric indicators, except WCZ and BP profiles, were different between boys and girls. The prevalence of abnormal lipids was not different between boys and girls. Elevated BP, overweight and obesity were more common among boys than girls (31·2 v. 21·6 %, 48·2 v. 33·1 % and 26·8 v. 10·0 %, respectively). Dyslipidaemia and combined CVRF were common in children than adolescents (59·5 v. 42·1 % and 16·7 v. 9·0 %, respectively)
Table 1. Characteristics of school-aged children and adolescents in Ho Chi Minh City, Vietnam (2014–2015) (Mean values with their standard errors; numbers and percentages)
WC, waist circumference; WHtR, waist-to-height ratio; SBP, systolic blood pressure; DBP, diastolic blood pressure; CVRF, cardiovascular risk factor.
* P <0·001 for the difference of mean or prevalence between boys and girls.
† P <0·04 for the difference of the prevalence between children and adolescents.
The performance of anthropometric indicators to identify elevated BP, dyslipidaemia and combined CVRF is presented in Table 2 and Figs. 1–3. Overall, all anthropometric indicators had capacity in identifying elevated BP, dyslipidaemia and combined CVRF. For the identification of elevated BP, WCZ (AUC 0·64) performed better (P < 0·04) than BMIZ (AUC 0·63) and WHtR (AUC 0·61). All indicators performed better for identifying elevated BP in adolescents (AUC 0·64–0·66) than in children (AUC 0·61–0·62) (P < 0·001). Similarly, the OR of having elevated BP in adolescents ranges from 1·69 to 2·45, whereas these ratios in children range from 1·33 to 1·97 (Table 2). For the identification of dyslipidaemia, BMIZ (AUC 0·64) performed similarly to WHtR (AUC 0·66) but performed better than WCZ (AUC 0·62) (P < 0·04) (Table 2). For identification of children and adolescents who had at least three CVRF, all indicators had a moderately accurate capacity (AUC range 0·72–0·74). The OR for having dyslipidaemia associated with a unit increase in anthropometric indicators ranged from 1·35 to 2·24, while these OR for combined CVRF ranged from 2·06 to 4·52 (Table 2).
Table 2. Performance of anthropometric indicators to identify elevated blood pressure, dyslipidaemia, and combined cardiovascular risk factors in school-aged children and adolescents in Ho Chi Minh City, Vietnam (Areas under the curve or odds ratios and 95 % confidence intervals)
BMIZ, BMI z-score; WCZ, waist circumference z-score; WHtR, waist-to-height ratio.
* OR for 1·0 unit increase (1 sd for BMIZ and WCZ and 0·1 for WHtR) using logistic regression adjusted for age and sex when it is appropriate in survey analysis.
† P <0·04 for the difference between WCZ and WHtR.
‡ P < 0·05 for the difference in AUC between BMIZ and WCZ.
§ P < 0·001 for the difference between boys and girls.
‖ P < 0·001 for the difference between children and adolescents.
¶ P <0·04 for the difference between BMIZ and WHtR.
** P < 0·04 for the difference between children and adolescents.
Fig. 1. AUC of anthropometric indicators to identify elevated blood pressure in school-aged children in Ho Chi Minh City, Vietnam. ROC, receiver-operating characteristic; BMIZ, BMI z-score; WHtR, waist-to-height ratio; WCZ, waist circumference z-score.
Fig. 2. AUC of anthropometric indicators to identify dyslipidaemia in school-aged children in Ho Chi Minh City, Vietnam. ROC, receiver-operating characteristic; BMIZ, BMI z-score; WHtR, waist-to-height ratio; WCZ, waist circumference z-score.
Fig. 3. AUC of anthropometric indicators to identify at least three cardiovascular risk factors in school-aged children in Ho Chi Minh City, Vietnam. ROC, receiver-operating characteristic; BMIZ, BMI z-score; WHtR, waist-to-height ratio; WCZ, waist circumference z-score.
Optimal cut-offs for each anthropometric indicator are presented in Tables 3 and 4. The optimal cut-offs for BMIZ and WCZ to identify elevated BP (Table 3) were 0·4 and 0·06 sd, whereas these cut-offs to identify dyslipidaemia or combined CVRF (Table 4) were 1·01 and 1·10 sd for BMIZ and 0·45 and 0·62 sd for WCZ. In contrast, optimal WHtR cut-offs for all CVRF considered ranged from 0·47 to 0·50 (Tables 3 and 4). The sensitivity and specificity for each of the optimal anthropometric cut-offs were higher than 50·0 %, except for the optimal BMIZ cut-off (sensitivity 47·2 %) and optimal WCZ cut-off (sensitivity 39·7 %) to identify dyslipidaemia, and some sex-specific sub-groups. All optimal cut-offs for identification of the combined CVRF in children and adolescents provided a high specificity (>70·0 %) (Table 4).
Table 3. Optimal cut-offs of anthropometric indicators to identify elevated blood pressure in school-aged children and adolescents in Ho Chi Minh City, Vietnam
BMIZ, BMI z-score; WCZ, waist circumference z-score; WHtR, waist-to-height ratio.
J value (Youden index) = sensitivity + specificity – 1.
Table 4. Optimal cut-offs of anthropometric indicators to identify dyslipidaemia and combined cardiovascular risk factors in school-aged children and adolescents in Ho Chi Minh City, Vietnam
BMIZ, BMI z-score; WCZ, waist circumference z-score; WHtR, waist-to-height ratio.
Anthropometric cut-offs for BP varied by age and sex. To identify elevated BP, younger boys had a higher cut-off than older boys for all anthropometric indicators (Table 3). In addition, the optimal WCZ cut-offs for boys to identify elevated BP were lower than the mean WC of the population (−0·12 sd in children and −0·22 sd in adolescent). To identify dyslipidaemia and combined CVRF, BMIZ cut-offs for boys were higher than those for girls, and this finding was similar in younger age groups to identify elevated BP. The optimal cut-offs for WHtR had a smaller variation by age group and sex (0·44–0·51) compared with other indicators WCZ (−0·22 to 0·62 sd) and BMIZ (0·40–1·96 sd) (Tables 3 and 4).
The probability of having CVRF above the anthropometric cut-offs, as well as OR for those above v. below the cut-off, is presented in Table 5, comparing optimal cut-offs identified from our survey v. those commonly used. The differences in probability of having CVRF between the commonly recommended >1·0 sd BMIZ cut-off and the optimal BMIZ cut-off were small for all the three CVD risk factor outcomes considered (0·281 v. 0·244, 0·528 v. 0·529, 0·118 v. 0·125, respectively) (Table 5). This was also the case for the different WHtR cut-offs. The optimal BMIZ cut-offs and the recommended BMIZ cut-off of 1·0 sd also showed comparable OR, sensitivity and specificity. At a BMIZ cut-off of 2·0 sd, the OR were higher than those for the optimal or 1·0 sd cut-offs, but the 2·0 sd cut-off had lower sensitivity (28·6 %) than the 1·0 sd cut-off (54·4 %). At the WHtR cut-off of 0·5, the OR for having elevated BP, dyslipidaemia or the combined CVRF were comparable to those for the BMIZ cut-off of 1·0 sd. However, the OR of having elevated BP at the optimal WHtR cut-off of 0·48 was higher (OR 3·74; 95 % CI 2·88, 4·85) compared with the 0·50 WHtR cut-off (OR 2·29; 95 % CI 2·01, 2·67).
Table 5. Probability of having risk factor and the OR of each risk factor for suggested and optimal BMI z-score (BMIZ) and waist-to-height ratio (WHtR) cut-offs (Odds ratios and 95 % confidence intervals; percentages)
* Optimal BMIZ cut-offs to identify elevated blood pressure, dyslipidaemia and combined cardiovascular risk factors were 0·4, 1·01 and 1·10 sd, respectively.
† Optimal WHtR cut-offs to identify elevated blood pressure, dyslipidaemia and combined cardiovascular risk factors were 0·48, 0·47 and 0·5, respectively.
‡ P value of goodness of fit for logistics models adjusted for age and sex. Crude OR was generated from the logistic model of health outcome and anthropometric indicators, adjusted OR was the logistic model adjusted for age and sex, and all models were analysed using ‘svy’ prefix.
The relationship between the anthropometric variables considered as continuous variables and the probability of having CVRF is illustrated in Figs. 4–6. Based on the shape of the probability curve, the probability of having elevated BP or dyslipidaemia did not change markedly (approaching linearity) with each increase in BMIZ, WCZ and WHtR cut-offs (Figs. 4 and 5). In contrast, the probability of having CVRF had a greater variation (approaching a u-curve) with each increase in BMIZ, WCZ and WHtR cut-offs that are central in the distribution and lie closer to the optimal cut-points (Fig. 6).
Fig. 4. Probability of having elevated blood pressure among children and adolescents by recommended and optimal anthropometric cut-offs. (a) By BMI z-score (BMIZ) cut-offs. (b) By waist circumference z-score (WCZ) cut-offs. (c) By waist-to-height ratio (WHtR) cut-offs. Recommended cut-offs: BMIZ (1·0 sd), WCZ (0·5 sd), WHtR (0·5). Optimal cut-offs: BMIZ (0·4 sd), WCZ (0·06 sd), WHtR (0·48).
Fig. 5. Probability of having dyslipidaemia among children and adolescents by recommended and optimal anthropometric cut-offs. (a) By BMI z-score (BMIZ) cut-offs. (b) By waist circumference z-score (WCZ) cut-offs. (c) By waist-to-height ratio (WHtR) cut-offs. Recommended cut-offs: BMIZ (1·0 sd), WCZ (0·5 sd), WHtR (0·5). Optimal cut-offs: BMIZ (1·01 sd), WCZ (0·49 sd), WHtR (0·47).
Fig. 6. Probability of having combined cardiovascular risk factors among children and adolescents by recommended and optimal anthropometric cut-offs. (a) By BMI z-score (BMIZ) cut-offs. (b) By waist circumference z-score (WCZ) cut-offs. (c) By waist-to-height ratio (WHtR) cut-offs. Recommended cut-offs: BMIZ (1·0 sd), WCZ (0·5 sd), WHtR (0·5). Optimal cut-offs: BMIZ (1·1 sd), WCZ (0·62 sd), WHtR (0·5).
Discussion
Our results showed that one in three, one in two, and more than one in ten children in urban Vietnam had elevated BP, dyslipidaemia and at least two CVRF, respectively. BMIZ, WCZ and WHtR had capacity to identify Vietnamese children and adolescents who have elevated BP, dyslipidaemia or at least three CVRF. Overall, all anthropometric indicators had moderate and comparable capacity to identify at least three CVRF. The indicator that performed best for identification of elevated BP was WCZ, though BMIZ performed almost equally well. BMIZ and WHtR performed relatively better than WCZ for identification of dyslipidaemia.
Overall, our AUC values were comparable with those reported in a meta-analysis of children and adolescents from thirty studies which included four studies from China (in which AUC 0·64–0·68 for hypertension, 0·67–0·7 for dyslipidaemia and >0·7 for at least three morbidities)(Reference Lo, Wong and Khalechelvam10). The performance of anthropometric indicators in our study, when evaluated as AUC achieved to identify elevated BP, was lower than reported to identify hypertension in recent studies in Malaysian and Chinese children (AUC around 0·60 in our study v. AUC around 0·70 in the other two studies)(Reference Cheah, Chang and Hazmi29,Reference Lu, Luo and Xie30) .
Elevated blood pressure
According to the WHO, raised BP is a leading risk factor non-communicable diseases, including CVD, in Vietnam(7). The prevalence of hypertension in Vietnamese adults in 2015 was 23·1 % (males) and 14·9 % (females). There are no known national BP data for children and adolescents. The prevalence of elevated BP in our survey was 29·3 % (adolescents), whereas in a study of 693 in high schoolchildren in HCMC in 2007 showed that this prevalence was 21·5 %(Reference Nguyen, Tang and Kelly31). Our result indicated that the prevalence of elevated BP in adolescents was higher than in children. Also, similar to other studies, the prevalence of elevated BP was higher in boys than in girls(Reference Nguyen, Tang and Kelly31,Reference Dong, Wang and Wang32) , particularly in adolescents.
Regarding the capacity to identify elevated BP, we found that BMIZ and WCZ performed better than WHtR (P < 0·04) in predicting elevated BP. This finding is consistent with recent studies in children (7–15 years old) in Tianjin, China(Reference Lu, Luo and Xie30), and adolescents (12–17 years old) in Sarawak, Malaysia(Reference Cheah, Chang and Hazmi29), and with a previous meta-analysis of studies in children and adolescents(Reference Ma, Wang and Liu33). However, in another study of children (6–10 years old) in Guangzhou, China, Liang (2015) found that WC was the best indicator to identify elevated BP(Reference Jj and Chen Yj34). In contrast, a study from the Chinese National Survey on Student’s Constitution and Health in children (7–17 years old) suggested that BMIZ had a better performance to identify elevated BP compared with WCZ, WHtR z-score, waist-to-hip ratio z-score and skinfold thickness z-score(Reference Dong, Wang and Wang32). Collectively, the current evidence indicates that WHtR is less useful than BMIZ and WCZ for identification of elevated BP.
The optimal BMIZ cut-off in children was 1·88 sd, whereas the optimal cut-off in adolescents was 0·45 sd. This contradiction may reflect the difference in the relative relationship between BMIZ and elevated BP. In adolescents, the prevalence of elevated BP and the OR of having elevated BP were higher than in children (29·3 v. 22·5 % and 1·69 v. 1·33 %, respectively). Due to the higher prevalence of elevated BP in adolescents, the prediction of elevated BP in the total population was influenced by the relative relationship between BMIZ and elevated BP in adolescents. This may explain the low optimal BMIZ cut-off in total population (0·4 sd) compared with the optimal cut-offs for children (1·98 sd).
Adolescents had an increased risk of elevated BP even when their BMIZ was within the healthy range. Thus, screening for elevated BP in adolescents should not be limited to children with overweight or obesity according to the current definition (WHO 2006). Given the high prevalence of hypertension in Vietnamese adults (18·9 %)(Reference Tran, Truong and Tran35), and the evidence for an increased risk of developing CVD in adulthood for children with elevated BP(Reference Hao, Wang and Treiber36), it may be beneficial to consider the lower BMIZ cut-off of 0·4 sd for the detection of Vietnamese children and adolescents likely to have elevated BP. However, the suggestion of BMIZ of 0·4 sd cut-off to identify elevated BP may not be practical in the context of the relatively high prevalence of children with overweight and obesity at the BMIZ cut-off of 1·0 sd. It would likely lead to a large burden on the health system due to increased management of children at risk of elevated BP. At the BMIZ cut-off of 1·0 sd, children are even more likely at risk of elevated BP. Measuring BP is cheap and non-invasive technique, so BP screening could be recommended in children and adolescents(Reference Flynn, Kaelber and Baker-Smith22) by using an electronic sphygmomanometer.
Dyslipidaemia and combined cardiovascular risk
Dyslipidaemia is common among school-aged children in HCMC, particularly in those aged 6–9 years. The prevalence of dyslipidaemia in our survey (49·3 %) was higher than in school-aged children (6–18 years old) in Bejing, China, in 2014 (28·9 %)(Reference Ding, Cheng and Yan37). Dyslipidaemia is strongly associated with BMI, breakfast skipping(Reference Kim, Song and Park38), screen time and lack of physical activity(Reference Zheng, Chen and Zhao39). Although we were not able to take these covariates into consideration, our prediction models were adjusted for age and sex. In addition to the higher prevalence of dyslipidaemia in children compared with adolescents, the prevalence of combined CVRF also showed a similar trend (16·7 v. 9·0 %). When considering the prevalence of dyslipidaemia and combined CVRF in the context of the nutrition transition, there are indications that children in urban Vietnam are moving to stage four of the nutrition transition process(Reference Popkin40). Thus, the prevention of CVRF should be prioritised from an early age.
For the identification of dyslipidaemia and combined CVRF, our results indicated that although BMIZ and WHtR had higher capacity than WCZ to detect dyslipidaemia, all anthropometric indicators had a comparable capacity to identify combined CVRF. This is in agreement with the finding from Lo and colleagues’ meta-analysis(Reference Lo, Wong and Khalechelvam10), with a recent study of Korean adolescents (10–19 years old) (AUC for BMI percentile, WC percentile, WHtR 0·68–0·69)(Reference Choi, Hur and Kang41), and with a cohort study of Chilean children(Reference Vasquez, Corvalan and Uauy42). Therefore, it is highly likely that all three anthropometric indicators are useful for the identification of Vietnamese children and adolescents with dyslipidaemia and combined CVRF.
Utility of BMI
BMIZ had moderate capacity to identify combined CVRF but was less accurate for identification of elevated BP and dyslipidaemia. The optimal BMIZ cut-off of 0·4 sd had the higher capacity to detect children with elevated BP than using BMIZ cut-offs of 1·0 and 2·0 sd, which are the criteria recommended by WHO for identification of overweight and obesity in children(Reference de Onis, Onyango and Borghi15). However, the optimal cut-offs for identification of dyslipidaemia and combined CVRF were close to 1·0 sd and therefore more consistent with the recommendation from WHO to use 1·0 sd (BMI 25 kg/m2 at 19 years old) as a cut-off for overweight(Reference de Onis, Onyango and Borghi15). Given that the sensitivity and specificity of the 1·0 sd BMIZ cut-off to identify combined CVRF were acceptable and similar to the optimal cut-off of 1·10 sd, continued promotion of the use of the 1·0 sd BMIZ cut-off would help focus on the increased risk of CVRF in addition to the identification of overweight in school-aged children and adolescents.
In urban Vietnam, given the high prevalence of dyslipidaemia and CVRF, particularly in children (6–9 years old), these risk factors should be detected early to prevent the future morbidity. However, currently, universal screening for CVD in children and adolescents is not widely recommended and practised(Reference Bloetzer, Bovet and Suris43). Thus, based on the performance of BMIZ, this indicator with a suggested cut-off of 1·0 sd could be promoted as a screening tool to identify CVRF. Children with a BMIZ cut-off more than 1·0 sd would be at 5 times higher odds of having at least three CVRF than those with BMIZ 1·0 sd or less.
Utility of waist circumference z-scores
Regarding the utility of WCZ, our results showed that the optimal cut-off was 0·06 sd for the prediction of elevated BP, 0·49 sd for the prediction of dyslipidaemia and 0·62 sd for the prediction of combined CVRF. The sensitivity of this WCZ cut-off for identification of dyslipidaemia was low (39·7 %), and age- and sex-specific WC cut-offs were not available. Thus, this indicator should not yet be considered as a screening tool for CVRF in Vietnamese children, until more evidence is gathered. WCZ was the best indicator to identify elevated BP, particularly in boys. Notably, boys were identified with high risk of elevated pressure even at cut-offs that were lower than the mean WC of the current population. This is likely due to the currently, on average, significantly higher WC in Vietnamese boys and also reflects the fact that abdominal obesity was highly associated with elevated BP in Vietnamese children, particularly in boys. Given the sensitivity of WC in determining elevated BP in children, the establishment of age- and sex-specific WC cut-offs will enhance the identification of risk screening in Vietnamese children.
Utility of waist-to-height ratio
Regarding the utility of WHtR, results from our study suggest that children with high WHtR had approximately 6-fold increased odds of having combined CVRF compared with those with lower WHtR. Sensitivity for the WHtR optimal cut-offs was low but acceptable (>50·0 %) but lower for the 0·5 cut-off. A study in Korean children also reported a very low sensitivity (27·8 %) for the WHtR 0·5 cut-off in the identification of at least two cardiometabolic risk factors(Reference Choi, Hur and Kang41). The evidence thus suggests that the use of WHtR 0·44 for boys and WHtR 0·43 for girls, instead of the universal 0·5 cut-off, would be optimal for screening of CVD risk factors in Korean children(Reference Choi, Hur and Kang41). However, in our study, the optimal cut-off to identify combined CVRF was 0·5, which is the same as the suggested WHtR cut-off. In addition, the AUC (0·74), sensitivity (63·7 %) and specificity (78·5 %) were acceptable. From a practical point of view, the message ‘keep your waist a half of your height’ is easy to communicate to the community(Reference Graves, Garnett and Cowell44). Thus, a WHtR cut-off of 0·5 could be recommended as a screening tool for the CVRF in urban children in Vietnam.
A preference for discrete anthropometric indicators or continuous variables for the purpose of identifying young people who are likely to carry CVD risk factors was also considered. The results suggest that there were small differences in the probability of having a risk factor between the optimal cut-offs and the commonly suggested cut-offs and that the probability of carrying risk factors changed at a more or less constant rate in particular for elevated BP and dyslipidaemia. Although the use of specific cut-offs might result in misclassification of diseases(Reference Perkins and Schisterman45), from a practical point of view, the use of complex prediction models using continuous variables may not be feasible in some low-resource settings, particularly in low- and low–middle-income countries. This could lead to delays in screening individuals with a high risk of disease in low-resource settings. Therefore, discrete cut-offs rather than continuous variables to identify CVRF in children would appear to be more useful in low-resource settings such as Vietnam. However, there still remains a need for ongoing validation of the optimal discrete anthropometric cut-offs for identification of Asian children at high risk of cardiovascular risk factors(Reference Bloetzer, Bovet and Suris43). Given the benefit of intervening early in life to reduce the risk of CVD in adulthood, interventions that promote healthy eating and physical activity should be prioritised rather than waiting for the validation of cut-offs for risk(Reference Gillman46).
The present study has a number of strengths. The results were derived from a large and representative sample of children and adolescents in HCMC, Vietnam and included blood lipid data from approximately one-thousand subjects. Blood samples can be difficult to collect in school-aged children especially in lower to middle-income countries. The present study does, however, have some limitations. It is a cross-sectional study, and thus, the results cannot provide strong evidence of the predictive capacity of anthropometric indicators to identify elevated BP, dyslipidaemia and combined CVRF – longitudinal data would provide stronger evidence. Although a second reading of abnormal BP reduces measurement error, one measurement of BP cannot eliminate the possibility of ‘white-coat’ hypertension in the elevated BP group. However, regardless of the cause of hypertension, this group is still at higher risk of CVD and mortality than the normotensive BP group(Reference Briasoulis, Androulakis and Palla47,Reference Huang, Huang and Mai48) . The pubertal stage of children in the present study was unavailable and therefore could not be analysed in regard to its association with anthropometric indicators and health outcomes. In addition, data on dietary intake and physical activity were unavailable, so prediction models could not be adjusted for these factors. Finally, for the development of predictive criteria, the population for tool development should not be the same as the population used for tool validation. However, due to the relatively smaller number of observations in the lipid profile data, we were not able to do this. Thus, it is necessary to validate the capacity of optimal cut-offs identified by our analyses in another sample of children and adolescents in Vietnam and to examine to what extent these cut-offs provide the same predictive capacity in different groups of Vietnamese children.
In conclusion, the results indicated that all anthropometric indicators had power to discriminate the elevated BP, dyslipidaemia and combined CVRF in school-aged children and adolescents in HCMC, Vietnam. Discrete anthropometric indicators such as a BMIZ cut-off of 1·0 sd and the WHtR cut-off of 0·5 sd could be considered as a screening tool for CVRF in this setting, the additional utility of continuous variables appeared minimal. However, further research in other study populations should be done to confirm the validity of these cut-offs in other settings. Consensus regarding the optimal anthropometric cut-offs to identify children at high risk of CVD in Asian populations should eventually be established.
Acknowledgements
The authors would like to thank Dr Tran Thi Minh Hanh, Dr Pham Ngoc Oanh, Dr Vu Quynh Hoa, Dr Phan Nguyen Thanh Binh and staff at the HCMC Nutrition Centre for conducting the survey of nutritional status of school-aged children in HCMC in the school year 2014−2015. The authors would especially like to thank Dr Do Thi Ngoc Diep, the former Director of HCMC Nutrition Centre, for permission to use this dataset for the present analyses. The authors also acknowledge the technical support from Dr Signe Bruun at the Hans Christian Andersen Children’s Hospital, Denmark and her colleges for helping to apply elevated BP criteria in our study, based on the new Clinical Practice Guideline for Screening and Management of High Blood Pressure in Children and Adolescents.
T. M. T. M. and J. C. v. d. P. formulated the research questions and scope of the manuscript, T. M. H. T. designed the survey of Nutritional Status Among School-Aged Children, T. M. T. M. carried out the analyses, J. C. v. d. P. contributed to the analysis plan, L. J. assisted with the statistical analyses, T. M. T. M. wrote the first draft of manuscript, J. C. v. d. P., D. G., L. J., Q. C. T. and T. M. H. T. contributed to interpretation of the result. All authors contributed to writing the final draft of manuscript.
This work was supported by the Australian Government Research Training Program, and QUT HDR Tuition Fee Scholarship to T. M. T. M. for the programme Doctor of Philosophy at Queensland University of Technology, Brisbane, Australia.
The authors declare that there are no conflicts of interest.
