BMI metrics and their association with adiposity, cardiometabolic risk factors, and biomarkers in children and adolescents

Abstract

BACKGROUND:

There are limited data comparing the relative associations of various BMI metrics with adiposity and cardiometabolic risk factors in youth.

OBJECTIVE:

Examine correlations of 7 different BMI metrics with adiposity, cardiometabolic risk factors, and biomarkers (i.e. blood pressure, waist circumference, cholesterol, leptin, insulin, high molecular weight adiponectin, high-sensitivity c-reactive protein (hsCRP)).

METHODS:

This was a cross-sectional analysis of youth in all BMI categories. BMI metrics: BMI z-score (BMIz), extended BMIz (ext. BMIz), BMI percentile (BMIp), percent of the BMI 95th percentile (%BMI_p95), percent of the BMI median (%BMI_p50), triponderal mass index (TMI), and BMI (BMI). Correlations between these BMI metrics and adiposity, visceral adiposity, cardiometabolic risk factors and biomarkers were summarized using Pearson’s correlations.

RESULTS:

Data from 371 children and adolescents ages 8–21 years old were included in our analysis: 52% were female; 20.2% with Class I obesity, 20.5% with Class II, and 14.3% with Class III obesity. BMIp consistently demonstrated lower correlations with adiposity, risk factors, and biomarkers (r = 0.190–0.768) than other BMI metrics. The %BMI_p95 and %BMI_p50 were marginally more strongly correlated with measures of adiposity as compared to other BMI metrics. The ext.BMIz did not meaningfully outperform BMIz.

CONCLUSION:

Out of all the BMI metrics evaluated, %BMI_p95 and %BMI_p50 were the most strongly correlated with measures of adiposity. %BMI_p95 has the benefit of being used currently to define obesity and severe obesity in both clinical and research settings. BMIp consistently had the lowest correlations. Future research should evaluate the longitudinal stability of various BMI metrics and their relative associations with medium to long-term changes in adiposity and cardiometabolic outcomes in the context of intervention trials.

BACKGROUND

Body mass index (BMI) is the most commonly used surrogate measure of adiposity for children and adolescents in research and clinical care, but does not reflect adiposity well in all BMI categories [1, 2]. The rise in prevalence of severe obesity has necessitated a re-evaluation of which weight-for-height metrics are the most accurate as a surrogate of adiposity [3]. In addition to weight-for-height, a number of measures are used clinically and in research to assess risk of clinical outcomes in persons with excess adiposity. For example, indicators of cardiovascular health (such as blood pressure, cholesterol), and biomarkers of cardiovascular health (e.g., leptin, which is associated with increased renal sympathetic nerve activity, and adiponectin which is considered protective against vascular disease and is associated with insulin sensitivity [4, 5]).

Obesity in children is defined as a BMI ≥ 95th percentile (or, equivalently, a z-score ≥1.645) for a child’s sex and age, and severe obesity as a BMI ≥ 120% of the 95th percentile of the CDC growth charts [6, 7]. Although z-scores based on these growth charts are widely used, this metric is inappropriate for BMIs above the 97th percentile (BMIz > 1.88) as this was the highest percentile used in the CDC estimation of the LMS parameters [7, 8]. Extrapolating the CDC LMS parameters to higher BMIs results in z-scores being compressed into a narrow range with values that do not correspond well with the observed data [9, 10]. Further, prior research has shown that BMIz scores do not correlate well with objective measures of adiposity, including in longitudinal assessments [11-13].

Beyond z-scores, other BMI metrics have been suggested as more accurate surrogates of adiposity, such as percent of the BMI 95th percentile (%BMI_p95), percent of the BMI median (%BMI_p50), triponderal mass index (TMI, weight/height³), and absolute BMI [14, 15]. Additionally, there is a new metric, call the extended BMI z-score, “ext.BMIz.” The ext.BMIz combines the BMIs of children with obesity from the Centers for Disease Control and Prevention (CDC) growth charts (1960–1980) with more recent data (through 2016) from children with obesity [16, 17]. The distribution of these BMIs ≥ 95th percentile was modeled as a half-normal distribution, and the shape parameter was estimated for each sex and 6-month age-group. These values were then smoothed using polynomial regression. This allows for the calculation of percentiles and z-scores among children with obesity, including those with very extreme BMIs. These estimated percentiles and z-scores for children with obesity were then combined with data from children without obesity (BMI < 95th percentile) in the 2000 CDC growth charts.

Data are limited regarding the relative association of the different BMI metrics with objective measures of adiposity and indicators of cardiometabolic risk in youth. Our primary objective was to examine the correlation of multiple BMI metrics with measures of body fat in children and adolescents. Our secondary objective was to examine the correlation of the BMI metrics with cardiometabolic risk factors and biomarkers associated with disease risk in children and adolescents [18].

METHODS

Study design and participants

Data for this cross-sectional study were collected from 371 children and adolescents who participated in a parent study evaluating cardiovascular disease biomarkers in youth (ClinicalTrials.gov Identifier: NCT01508598) [19]. The participants were recruited into this study by flyers and letters mailed from the health system to youth with a qualifying BMI, and through general pediatric clinics in the Minneapolis Metropolitan Area, and the University of Minnesota Masonic Children’s Hospital Pediatric Weight Management Clinic. The inclusion criteria for the parent study included age 8–17 years and BMI in the normal weight range (BMI < 85th percentile), overweight/obesity range (BMI 85th to 1.2 times 95th percentile), and severe obesity range (BMI ≥ 1.2 times 95th percentile). Key relevant exclusion criteria included syndromic obesity and history of bariatric surgery. All parents/guardians and participants provided verbal and written informed consent and assent, respectively. The protocol for the parent study was approved by the University of Minnesota Human Subjects Protection Program.

Measurement of clinical variables, cardiometabolic risk factors, and biomarkers

All testing was performed in the morning after participants had been fasting for a minimum of 12 h. Height and weight were determined with participants wearing light clothes and without shoes using a wall-mounted stadiometer and an electronic scale, respectively. Waist circumference (WC) was measured at the midpoint of the base of the ribcage and the superior iliac crest and the average of three measurements was used. Seated blood pressure was obtained after 5 min of quiet rest, on the right arm using an automatic sphygmomanometer and appropriately fitted cuff. Three measurements were taken and the average of the final two were used. Total body composition, including total percent body fat (BF%) and estimates of abdominal visceral adipose tissue were obtained using dual x-ray absorptiometry (iDXA, General Electric Medical Systems, Madison, WI, USA) and analyzed using its enCore™ software (platform version 16.2) [20].

Blood was drawn for measurement of lipids, glucose, and insulin using standard procedures (analyzed by the Fairview Diagnostics Laboratories, Fairview-University Medical Center, Minneapolis, MN, USA, a CDC certified laboratory). Leptin, C-reactive protein (CRP) (Alpco, Salem, NH), oxidized LDL (oxLDL; Mercodia, Uppsala, Sweden), lipoprotein A (Lp[a]), and high molecular weight (HMW) adiponectin (R&D Systems, Minneapolis, MN), were measured by ELISA in the University of Minnesota Cytokine Reference Laboratory (Clinical Laboratory Improvement Amendments licensed).

BMI metrics

We calculated BMIz, ext.BMIz, BMI percentile (BMIp), %BMI_p95, %BMI_p50, TMI, and absolute BMI. BMI z-scores and percentiles were calculated using the age- and sex-specific values from the CDC growth charts. BMI extended z-scores were calculated using Wei, Parsons, and Odgen’s method to more accurately evaluate extreme BMIs [21]. Of note, for BMI values under the 95th percentile, there is no difference between BMIz and extended BMIz. TMI was calculated by kg/m³. BMI was calculated by kg/m².

Statistical analysis

Descriptive statistics were summarized by mean and standard deviation (SD) for continuous variables and frequency with percent for categorical variables. Analyses are presented for the full sample (primary focus of this report) as well as by BMI categories: (1) Normal and Overweight (BMI < 95th percentile for age and sex); (2) Class I Obesity (BMI between 1.0 and 1.2 times the 95th percentile for age and sex); (3) Class II Obesity and greater (BMI greater than 1.2 times the 95th percentile for age and sex). Associations between BMI metrics and measures of adiposity and cardiometabolic risk factors and biomarkers were summarized using Pearson’s correlations and evaluated using Meng’s test, which is a test for correlated correlation coefficients [22]. Local polynomial curves were used to display the non-linear association between the BMI metrics and BF%, both overall and by sex.

Systolic blood pressure (SBP) was standardized by age and sex to CDC percentiles. Triglyceride/high density lipoprotein ratio (TG/HDL) and WC were standardized by age and sex using the 2011–2012 and 2013–2014 data cycles of the National Health and Nutrition Examination Survey (NHANES) [23]. Standardized values were obtained by fitting a linear regression model using the NHANES data with age and sex as covariates, and using the fitted model to calculate and record the residuals from our dataset. All graphs and analyses were conducting using R version 3.5.3 (R Foundation, Vienna, Austria) [24].

Participants

The sample consisted of 371 children and adolescents ages 8–21 years old (52% female, mean age 13 years) (Table 1). Most participants were White (77%), 10% were Black, and 11% Latino. Over one-third had a BMI in the normal weight category (38.5%) while 6.5% were in the overweight category, 20.2% had Class I obesity, 20.5% had Class II Obesity, and 14.3% had Class III obesity. The median BMI was 25.2.

Table 1.

Demographics and BMI metrics overall and by body mass index (BMI) category.

n (%)	Overall N = 371	Normal weight and overweighta N = 167, 45%	Class I obesityb N = 75, 20%	Class II and III obesityc N = 129, 35%
Age, mean (SD)	13 (3)	13 (3)	12 (3)	14 (3)
Male	179 (48)	91 (55)	36 (48)	52 (40)
Race, Ethnicity
Asian	6 (1.6)	2 (1.2)	2 (2.7)	2 (1.6)
Black	37 (10)	15 (9)	6 (8)	16 (12)
American Indian	3 (1)	0 (0)	0 (0)	3 (2)
Mixed	32 (9)	8 (5)	8 (11)	16 (12)
White	287 (77)	139 (83)	57 (76)	91 (71)
Other	5 (1)	2 (1)	2 (3)	1 (1)
Latino	42 (11)	9 (5)	11 (15)	22 (17)
BMI Metrics
BMIz, mean (SD)	1.3 (1.2)	0.1 (0.8)	2.0 (0.2)	2.5 (0.2)
Ext.BMIz, mean (SD)	1.4 (1.4)	0.1 (0.8)	1.9 (0.1)	2.8 (0.6)
BMIp, mean (SD)	78 (28)	54 (26)	97 (1.1)	99 (0.4)
%BMIp95, mean (SD)	105 (31)	76 (9)	110 (6)	141 (18)
%BMIp50, mean (SD)	144 (43)	104 (13)	150 (10)	193 (25)
TMI, mean (SD)	17 (5)	12 (2)	18 (2)	23 (3)
BMI mean (SD)	27 (9)	19 (3)	27 (3)	37 (7)

^aBMI < 95th percentile for age and sex.

^bBMI ≥ 95th percentile and <1.2 × 95th percentile for age and sex.

^cBMI ≥ 1.2 × 95th percentile for age and sex.

Differences between metrics

Differences across all BMI metrics were statistically significant within each measure of adiposity, cardiometabolic risk factor, and cardiometabolic biomarker assessed. The magnitude of correlations between BMI metrics and cardiometabolic risk factors varied widely across BMI categories (Fig. 1; Supplemental Table 1). Correlations between BMI metrics and cardiometabolic biomarkers tended to be weaker than correlations between BMI metrics and cardiometabolic risk factors. Differences across all BMI metrics were statistically significant within each cardiometabolic risk factor. Strengths of correlations between BMI metrics and cardiometabolic biomarkers varied widely across BMI categories (Fig. 1; Supplemental Table 2). Differences across all BMI metrics were statistically significant within each cardiometabolic biomarker. Visual representation of the different metrics and their correlation with percent body fat in the overall sample is represented in Fig. 2, and with all of the outcomes is represented in Fig. 3.

Fig. 1. This is a scatter plot showing the coefficient of correlation on the Y axis and the different outcomes measured on the X axis.

Each point in the graph is a correlation coefficient for the 6 different BMI metrics (they are represented by a corresponding letter and color). The left upper and lower panels represent participants with a BMI in the normal weight or overweight categories; the middle upper and lower panels represent participants with a BMI in the Class I obesity category; the right upper and lower panels represent participants with a BMI in the Class II category and higher.

Fig. 2. This is a scatter plot with a best fit line by Lowess smoother [30].

All six panels are showing correlation with one outcome, percent body fat, on the Y axis, and the 7 different BMI metrics. The blue line is male participants. The pink line is female participants. The black line is both sexes combined. These plots include participants in all BMI categories.

Fig. 3. This is a line graph showing the coefficient of correlation (vertical axis) for each BMI metric (horizontal axis), and each different line is a different outcome.

This figure includes all the participants, in the BMI categories. BMI body mass index, BMIz BMI z-score, ext.BMIz BMI extended z-score, BMIp BMI percentile, %BMI_p95 BMI percent of the 95th Percentile, BMI_p50 BMI percent of the median, TMI triponderal mass index, BMI body mass index, SBP systolic blood pressure, HDL high density lipoprotein, hsCRP high-sensitivity C-reactive protein, oxLDL oxidized low-density lipoprotein, Lp(a) lipoprotein(a), HMW high molecular weight.

Adiposity results by BMI Metric

Below we highlight the 6 BMI metrics and notable correlations with measures of adiposity. All results can be seen in the tables and figures. Table 2 and Fig. 3 show the correlation coefficients for the overall sample; Supplemental Table 1 and Fig. 1 show them by BMI category. Figure 2 shows the BMI metrics and percent body fat.

Table 2.

Association between BMI metrics and adiposity, cardiometabolic risk factors, and biomarkers, as measured by Pearson’s correlation coefficient, among all participants (all BMI categories).

	BMIz	ext.BMIz	BMIp	%BMIp95	%BMIp50	TMI	BMI	p valuea
Adiposity
Percent body fat	0.847	0.855	0.768	0.859	0.870	0.868	0.810	<0.001
Percent visceral fat	0.726	0.752	0.601	0.814	0.813	0.790	0.776	<0.001
Cardiometabolic risk factors
SBP percentile	0.473	0.482	0.434	0.475	0.475	0.477	0.422	<0.001
TG/HDL ratiob	0.522	0.512	0.461	0.512	0.499	0.454	0.432	<0.001
WCb	0.875	0.907	0.756	0.941	0.936	0.889	0.900	<0.001
Cardiometabolic biomarkers
Leptin	0.711	0.759	0.605	0.807	0.831	0.837	0.820	<0.001
Insulin	0.599	0.641	0.492	0.689	0.693	0.654	0.693	<0.001
hsCRP	0.401	0.426	0.358	0.448	0.463	0.489	0.481	<0.001
oxLDL	0.275	0.258	0.264	0.250	0.256	0.261	0.250	<0.001
Lp(a)	0.182	0.178	0.190	0.156	0.164	0.179	0.193	<0.001
HMW adiponectin	−0.493	−0.483	−0.468	−0.470	−0.471	−0.450	−0.484	<0.001

BMI body mass index, BMIz BMI z-score, ext.BMIz extended BMI z-score, BMIp BMI percentile, %BMI_p95 BMI percent of the 95th Percentile, BMI_p50 BMI percent of the median, TMI triponderal mass index, BMI body mass index, SBP systolic blood pressure, TG triglyceride, HDL high density lipoprotein, WC waist circumference, hsCRP high-sensitivity C-reactive protein, oxLDL oxidized low-density lipoprotein, Lp(a) lipoprotein(a), HMW high molecular weight.

^aDetermined by Meng’s test.

^bStandardized for age and sex using NHANES data.

BMIz

In the overall sample, BMIz had the third lowest correlation with percent body fat (r = 0.847) and second lowers with percent visceral fat (r = 0.726).

Novel extended BMI z-score (ext.BMIz)

In the overall sample, ext.BMIz had the 4th lowest correlation with percent body fat (0.855) and 3rd lowers with visceral fat (r = 0.752).

BMIp

In the overall sample, BMIp had the lowest correlation with percent body fat (r = 0.768) and with percent visceral fat (r = 0.601).

%BMI_p95

In the overall sample, %BMI_p95 had the 3rd highest correlation with percent body fat (r = 0.859) and the highest correlation with percent visceral fat (r = 0.814).

%BMI_p50

In the overall sample, %BMI_p50 had the highest correlation with percent body fat (r = 0.870) and the second highest correlation with percent visceral fat (r = 0.813), though nearly the same as % BMI_p95 (r = 0.814) and TMI (r = 0.868).

TMI

In the overall sample, TMI had the second highest correlation with percent body fat (r = 0.868) and the third highest correlation with percent visceral fat (r = 0.790).

BMI

In the overall sample, BMI had the second lowest correlation with percent body fat (r = 0.810) and the 4th highest correlation with percent visceral fat (r = 0.776).

Below is a brief overview of results of the secondary objective, which was to examine the correlation of BMI metrics with cardiometabolic risk factors and biomarkers:

Overview of correlations with cardiometabolic risk factors

Waist Circumference: Correlations ranged from 0.756 (BMIp) to 0.941 (%BMI_p95).

TG/HDL Ratio: Correlations ranged from 0.432 (BMI) to 0.522 (BMIz).

SBP percentile: Correlations ranged from 0.422 (BMI) to 0.477 (TMI).

Overview of cardiometabolic biomarkers

Leptin: Correlations ranged from 0.605 (BMIp) to 0.837 (TMI).

Insulin: Correlations ranged from 0.492 (BMIp) to 0.693 (BMI and %BMI_p50).

hsCRP: Correlations ranged from 0.358 (BMIp) to 0.489 (TMI). oxLDL: Correlations ranged from 0.250 (BMI and %BMI_p95) to 0.275 (oxLDL).

Lp(a): Correlations ranged from 0.156 (%BMI_p95) to 0.190 (BMIp). HMW Adiponectin: Correlations ranged from −0.450 (TMI) to −0.493 (BMIz).

DISCUSSION

Our goal in this study was to systematically compare the strength of correlation of multiple BMI metrics, including the novel ext. BMIz, with numerous measures of adiposity and cardiometabolic risk factors and biomarkers in youth across a wide range of BMI values and BMI categories. The main finding of this study was that all of the BMI metrics, with the exception of BMIp, performed relatively similarly in terms of the strength of association with adiposity, cardiometabolic risk factors and biomarkers taken together. However, %BMI_p95 and %BMI_p50 had the highest correlation coefficients with measures of adiposity (total body and visceral fat). Given that visceral adipocytes secrete a number of inflammatory and coagulopathic markers [25-27], we were interested in which BMI metric best predicted overall and visceral adiposity. We found that %BMI_p95 best correlated percent visceral fat, closely followed by %BMI_p50, and that BMIp had the lowest correlation with percent visceral adiposity. Notably, all BMI metrics tended to demonstrate low correlations with many of the risk factors and biomarkers assessed, which is consistent with previous research [13].

We found that no BMI metric was clearly superior across most outcomes and BMI categories, but %BMI_p95 tended to perform consistently well, including across different BMI categories and outcomes assessed. This is in line with prior work using NHANES data showing that metrics referencing the 95th percentile were most closely correlated with measures of body fatness [3, 11, 28]. While identifying a single BMI metric that clearly performed the best in our data is challenging, we found that BMIp consistently performed the worst. This is supported by previous literature suggesting that BMIp does not accurately reflect adiposity in children, particularly above the 99th percentile as a wide range of adiposity is compressed between the 99th and 100th percentiles [11, 29]. When analyzing BMI status in children, one should use %BMI_p95 for children with a BMI greater than the 95th percentile.

Unlike prior research, we did not find clear evidence that BMIz was less correlated with adiposity in children compared to other BMI metrics [11, 28, 29]. This may be because our sample had fewer extremely high BMI’s than other samples. We also analyzed a new extended BMIz score, ext.BMIz, in our data. As expected, the ext.BMIz performed the same as the traditional BMIz in lower BMI categories because it is designed to be the same as the BMIz under the 95th percentile. In participants with Class II obesity or greater, the ext.BMIz was better correlated with BF%, TG/HDL ratio, leptin, HMW adiponectin, and hsCRP, and similarly correlated with percent visceral fat, compared to the more traditional BMIz metric.

The relevance and impact of our findings are underscored by the fact that the distribution of BMI is rapidly shifting in the US and throughout the world toward higher BMI values across all ages [29]. From a clinical and research perspective, it is important to identify which BMI metric(s) most closely reflects body fatness since measuring adiposity using DXA, computed tomography, or magnetic resonance imaging is often impractical. Clinically, % BMI_p95 has the advantage of already being used to define classes of obesity in youth [3]. Many electronic medical record platforms currently contain an algorithm to automatically calculate BMI values relative to the 95th percentile. Using %BMI_p95 also has the advantage of families and clinicians being able to conceptualize the 95th percentile as a treatment target. For example, if a child has a BMI at 1.2 times the 95th percentile, a reasonable clinical goal could be striving to achieve a BMI below 1.0 times the 95th percentile (i.e., a BMI no longer in the obesity category), which is perhaps easier to understand than achieving a BMI relative to the median. Additionally, because BMI changes across the different ages of childhood, absolute BMI numbers are not good goals or reference points for patients and families. Another benefit of %BMI_p95 is that it performs similarly on both the CDC and World Health Organization (WHO) growth curves [13].

Strengths of our study include the relatively large sample size, wide range of BMI and body fat values, and the inclusion of total percent fat, visceral fat, cardiometabolic risk factors and biomarkers. Limitations include the fact that our study population was primarily white and that the data were cross-sectional in nature. Additionally, our sample did not have children under the age of 8 years, and these results might not apply to children less than 8 years old. By using correlations to measure performance, we recognize some additional limitations. Correlation is affected by the amount of variability in the data, and correlations are a unit-less summary measure that can be difficult to interpret. In general a higher correlation is better, and we evaluated if correlations were statistically significant or not, but there is not a clear meaningful magnitude of difference. There was also differing variances among the metrics.

CONCLUSION

In conclusion, there was no single BMI metric that consistently and strongly outperformed others when considering all of the adiposity and cardiometabolic variables assessed, but %BMI_p95 and %BMI_p50 were numerically the most highly correlated with measures of adiposity (total body and visceral fat). The %BMI_p95 has the added benefit of currently being used to define obesity and severe obesity and can be used conceptually by researchers and clinicians as a target goal for treatment (i.e. reducing %BMI_p95 to one or below, which represents moving from the obesity category to the overweight category). BMIp consistently had the lowest correlations with percent body fat, visceral adiposity, and other cardiometabolic risk factors and biomarkers. Future research should evaluate how well these BMI metrics perform as outcome measures in the context of interventional trials as well as the strength of tracking with change in adiposity, cardiometabolic risk factors, and biomarkers.