Changes in BMI z score (BMIz) are widely used in weight control programs and interventions to monitor changes in body fatness, but this metric may not be optimal. We examined the ability of 3 BMI metrics to assess adiposity change among children with a wide range of BMIs.
The sample comprised 343 3-year-old children with serial measurements of BMI and body fatness every 4 months over 4 years. We compared correlations between changes in body fatness, calculated with dual-energy-x-ray absorptiometry, and changes in 3 BMI metrics: BMIz and percentage of the 50th (%50th) and 95th (%95th) percentiles in the CDC growth charts.
About 21% of the participants were Black and 79% were white. Changes in body fatness over 4 years were more strongly associated with changes in %50th and %95th than with changes in BMIz. Correlations with %body fat among all children were r = 0.64 for BMIz versus r = 0.77 to 0.78 for %50th and %95th (P < .001 for differences between the correlations). Stratified analyses showed the difference between the correlations were similar among boys and girls, among white children and Black children, and among children without obesity and those with obesity.
Changes in adiposity among young children are better captured by expressing changes in BMI as a percentage of the 50th or 95th percentiles instead of BMIz change. Using the best BMI metric will allow pediatricians to better assess a child’s change in body fatness over time.
Because of the limitations of BMI z scores for very high BMIs, several alternative BMI metrics have been proposed. However, it is uncertain which metric is best for evaluating BMI change in intervention or when monitoring a child’s BMI over time.
BMI changes expressed relative to the CDC 50th or 95th percentiles are more strongly correlated with adiposity change than BMI z scores. Using the best BMI metric across the entire BMI distribution will allow BMI changes to be more accurately assessed.
About 20% of children and adolescents (ages 2 to 19 years) in the United States have obesity,1 and several interventions and treatment programs have been developed for children with overweight or obesity.2 The focus of these studies is typically BMI change over a relatively short period, such as 1 year. For example, it may be desired to assess changes in BMI (1) over 1 year in an intervention study, (2) during the coronavirus disease 2019 pandemic, and (3) when monitoring children over time.
BMI z scores (BMIz), based on the Centers for Disease Control and Prevention (CDC) growth charts, have frequently been the outcome in these longitudinal studies.2 However, the physiologic outcome of these longitudinal studies is change in body fatness, which is the primary driver of cardiometabolic risk. BMIz has several limitations when examining changes in weight status.3–6 The relation of BMI to BMIz is curvilinear: at high BMIs, BMIz approaches a maximum that differs by sex and age, irrespective of the child’s BMI.7 Figure 1 shows this relationship among 3-, 5-, and 7-year-olds, typical ages of the children in the current study. At relatively low BMIs, a small BMI change can result in a sizeable BMIz change. However, for a child with a high initial BMI, a similar BMI change results in a much smaller BMIz change that varies substantially by sex and age.6 Therefore, changes in BMIz cannot be compared between children with different initial BMI z scores. Although several studies have emphasized the poor performance of BMIz among children with extremely high BMIs8–10 , Figure 1 also illustrates that the BMIz change is also problematic at BMIs below the CDC 95th percentile.
Relation of BMI to BMIz for 3, 5-, and 7-year-old boys and girls. The range of BMI values is based on data from NHANES (1999–2000 through 2017–2018). At age 7 years (84 months), the maximum, possible BMIz is 3.27 SDs among boys and 3.25 SDs among girls (blue line).
Relation of BMI to BMIz for 3, 5-, and 7-year-old boys and girls. The range of BMI values is based on data from NHANES (1999–2000 through 2017–2018). At age 7 years (84 months), the maximum, possible BMIz is 3.27 SDs among boys and 3.25 SDs among girls (blue line).
To overcome the limitations of the CDC BMI z scores, several alternative BMI metrics, such as the percentage of the 50th (%50th) or the 95th percentile (%95th) in the CDC growth charts, have been proposed.3,11 However, the utility of %50th and %95th concerning longitudinal changes in body fatness is not well characterized. A useful BMI metric would be expected to be more strongly associated with changes in body fatness than an inferior metric. We, therefore, examined differences in the relation of changes in body fatness to changes in BMIz, %50th, and %95th among 343 3- to 7-year-old children who had BMIs that were similar to those in the general population.
Methods
As previously summarized,12 data for this analysis were based on a study of preschool children that focused on changes in body fatness. The study recruited 372 3-year-olds in 2001 and 2002 in the Cincinnati area. Eligibility criteria were (1) full-term (≥ 37 weeks’) gestation, (2) no chronic health conditions influencing growth and development, and (3) children from households in which parents were either both Black or both White. Other race and ethnicity groups were not included because of the small numbers of Asian (1.6%), Hispanic (1.3%), and ‘other’ or multiracial (2.6%) race-ethnicities in the Cincinnati area during study enrollment.
Serial measurements of body composition, weight, and height were scheduled every 4 months from ages 3 to 7 years,13 but some children were lost to follow-up. Further, because of scheduling conflicts, some children were first examined at 2 years, and the last visit occurred at 8 years. Children were lightly dressed in shorts and pants with an elastic waist and t-shirts. The Cincinnati Children’s Hospital Medical Center institutional review board approved the study, informed parental written consent was obtained, and CDC determined that this research was exempt from institutional review board review.
BMI Metrics
Height was measured using a wall-mounted stadiometer and weight with a digital scale; BMI was calculated as kg/m2. Sex- and age-standardized BMI z scores (BMIz) were based on the CDC growth charts.14,15 In addition to BMIz, we examined 2 other metrics that account for differences in BMI by sex and age: (1) BMI expressed as a percentage of the CDC 50th percentile (%50th) and (2) as a percentage of the CDC 95th percentile (%95th).3,11,16
Body Fatness
As previously described,17 dual-energy x-ray absorptiometry (DXA) was used to calculate fat mass. The child’s entire body was scanned in the array-fan beam mode with a Hologic 4500 instrument (Hologic, Bedford, MA) and Hologic Pediatric software. Printed scans were examined by trained personnel, and children with substantial movement in the limbs or trunk were excluded. Because there is a decrease in head size relative to body size with age among young children,13 the skull was excluded from all DXA measurements. Sedation was not used.
Of the 372 enrolled children, 368 had 1 or more valid DXA scans. We further limited our analyses to the 343 children with at least 2 DXA scans conducted 9 or more months apart. The mean number of examinations per child was 12. We express body fatness as %body fat (fat mass ÷ weight). However, this metric is influenced by both the amount of fat mass and the amount of fat-free mass.18,19 Therefore, we also examined the relation of changes in the BMI metrics to changes in 2 other measures of body fatness: fat mass index (fat mass ÷ height2 ) and height-adjusted (with linear regression) fat mass.18 Because the various body fatness metrics showed similar results, only those for %body fat are shown.
Statistical Methods
We used R 4.1.220 for all analyses. In addition to presenting descriptive characteristics of the sample, we also examined changes in BMI and body fatness between the first and last examination with age using mixed-effects models to account for the correlated, repeated measurements from each child21,22 .
We then focused on the Pearson correlations between changes in the 3 BMI metrics to changes in body fatness. Changes were calculated using data from each child’s first and last exams; the mean interval was 3.9 years. Differences between the magnitudes of the correlated correlation coefficients were assessed as described by Meng et al.23,24 This test, for example, evaluated whether there was a statistically significant difference in the relation of changes in %body fat to changes in %50th versus changes in BMIz. Multiple regression was also used to determine if the prediction (R2) of changes in %body fat could be improved by adding baseline levels of each BMI metric to models that included changes in the specified BMI metric.
We also examined correlations over an ∼1-year interval by stratifying on the year of age and examining changes between the first visit at each year.
Results
About 79% of the children were White and 21% were Black. Across all examinations, the mean BMIz was 0.48 SDs, and 10.5% of the BMIs obtained throughout the study were greater than or equal to the CDC 95th percentile.
Table 1 shows mean levels of various characteristics at the first and last visits by sex. The mean age at the first visit was 3.4 years and was 7.2 years at the final visit; the mean interval between the first and last visits was 3.9 years (range: 0.9–4.5 years). About 7% of children had obesity at the initial examination, but none had severe obesity. Over the ∼4 years of follow-up, levels of BMI and most BMI metrics increased; however, mean levels of %95th and %body fat decreased slightly (P < .05 for all changes with age).
Descriptive Characteristics at the First and Last Visits for the 343 Children With at Least 2 DXA Scans
| Boys (n = 180) | Girls (n = 163) | |||||
|---|---|---|---|---|---|---|
| Initial Exam | Final Exam | Annual Changeb | Initial Exam | Final Exam | Annual Changeb | |
| Age (y) | 3.4±0.3a | 7.2±0.7 | — | 3.4±0.3 | 7.2±0.6 | — |
| % Black | 17.2 | 17.2 | — | 24.5 | 24.5 | — |
| BMI | 16.3±1.2 | 16.9±2.4 | 0.18±0.04 | 16.2±1.4 | 17.2±2.7 | 0.30±0.05 |
| BMIz | 0.32±1 | 0.49±1 | 0.06±0.01 | 0.36±1 | 0.55±1 | 0.07±0.01 |
| %50th | 103.1±8 | 108.7±16 | 1.7±0.2 | 104.2±9 | 110.9±17 | 2.0±0.3 |
| %95th | 90.5±7 | 87.6±12 | −0.7±0.2 | 89.3±8 | 86.6±13 | −0.6±0.2 |
| % Obesity | 7.8 (4.7–12.6) | 11.1 (7.3–16.5) | 5.5 (2.9–10.2) | 15.3 (10.6–21.7) | ||
| %Body fat | 26.0±5 | 22.8±6 | −1.0±0.1 | 30.6±6 | 28.1±7 | −0.8±0.1 |
| Boys (n = 180) | Girls (n = 163) | |||||
|---|---|---|---|---|---|---|
| Initial Exam | Final Exam | Annual Changeb | Initial Exam | Final Exam | Annual Changeb | |
| Age (y) | 3.4±0.3a | 7.2±0.7 | — | 3.4±0.3 | 7.2±0.6 | — |
| % Black | 17.2 | 17.2 | — | 24.5 | 24.5 | — |
| BMI | 16.3±1.2 | 16.9±2.4 | 0.18±0.04 | 16.2±1.4 | 17.2±2.7 | 0.30±0.05 |
| BMIz | 0.32±1 | 0.49±1 | 0.06±0.01 | 0.36±1 | 0.55±1 | 0.07±0.01 |
| %50th | 103.1±8 | 108.7±16 | 1.7±0.2 | 104.2±9 | 110.9±17 | 2.0±0.3 |
| %95th | 90.5±7 | 87.6±12 | −0.7±0.2 | 89.3±8 | 86.6±13 | −0.6±0.2 |
| % Obesity | 7.8 (4.7–12.6) | 11.1 (7.3–16.5) | 5.5 (2.9–10.2) | 15.3 (10.6–21.7) | ||
| %Body fat | 26.0±5 | 22.8±6 | −1.0±0.1 | 30.6±6 | 28.1±7 | −0.8±0.1 |
Values are mean ± SD or prevalence (95% confidence interval).
Estimated change in characteristic per year as estimated in mixed-effects models using all of the data from each child. All changes with age were statistically significant at the 0.05 level.
—, not applicable.
Supplemental Figure 4 shows the cross-sectional associations, within each sex and race group, between the BMI metrics and %body fat at the examination closest to age 5.3 years (the mean age across all examinations). In general, %body fat was most strongly associated with %95th and %50th, and least strongly correlated with BMIz. These differences were seen among boys and girls, and among White children and Black children.
Figure 2 shows that between the first and last visit, change in %body fat was more strongly correlated with changes in %50th and %95th than with changes in BMIz. The weaker association with BMIz was evident in each race and sex group, and differences between the correlations for BMIz and the metrics on the basis of percentages were statistically significant at the 0.0001 level. Multiple regression analyses (not shown) indicated that the R2s of models predicting %body fat change from changes in BMIz increased substantially (ΔR2 = 0.15) if initial BMIz was included as an additional predictor. In contrast, adding baseline levels of %50th or %95th to models already containing changes in these metrics did not improve the prediction of %body fat change.
Correlations between changes in the BMI metrics and change in %body fat between the first and last visit.
Correlations between changes in the BMI metrics and change in %body fat between the first and last visit.
Additional analyses, stratified by obesity status at the initial examination, showed that the differences among the BMI metrics were similar among children with or without obesity. For example, among the 23 children with obesity at baseline, changes in body fatness tended to be more strongly correlated with changes in %50th and %95th (r = 0.77–0.79) than with BMIz change (r = 0.71). Among the 320 children with an initial BMI < 95th percentile, correlations were r = 0.77 to 0.79 for changes in %50th and %95th versus r = 0.67 for changes in BMIz.
Figure 3 shows the correlations between changes in the BMI metrics and changes in %body fat over approximately 1-year intervals, stratified by sex and age (x-axis). Generally, the correlations between the BMI metrics and body fatness were similar to those in Figure 2: change in %body fat was more strongly correlated with changes in %50th and %95th than with changes in BMIz. However, the magnitudes of these differences were smaller at age 3 years than at older ages.
Sex- and age-stratified correlations between the 1-year changes in the BMI metrics and body fatness.
Sex- and age-stratified correlations between the 1-year changes in the BMI metrics and body fatness.
Discussion
Our results, based on serial measurements of body fatness from ages 3 to 7 years among children with BMI levels comparable to those in the general population, indicate that there are substantial differences in the ability of various BMI metrics to assess adiposity change in young children. Changes in %body fat are more strongly associated with changes in BMI expressed as a percentage of the 50th or 95th percentile of the CDC growth charts than with BMIz change. Although our results are based on data collected approximately 20 years ago, there is little reason to think that the relation between changes in body fatness and BMI has changed over this period.
BMIz helps characterize a child’s BMI among boys and girls of different ages, but several studies have emphasized that BMIz does not accurately characterize the BMIs of children with severe obesity8–10,25 . Other investigators3–6 , without data on body fatness, have suggested that because the magnitude of BMIz change is associated with the initial BMIz value, BMI z scores are not optimal for analyzing changes in adiposity. This arises because the LMS method for constructing the CDC growth charts,26,27 results in a curvilinear relation between BMI and BMIz (Fig 1). For example, a 1-unit change in BMI is associated with a much larger change in BMIz among thin children than among heavy children. In contrast to the curvilinear associations with BMIz, BMI is linearly related to %50th and %95th, so changes in these metrics do not depend on initial BMI (Fig 4). A 25% increase in %50th, for example, is associated with the same BMI change regardless of the initial BMI. Regression models that predicted change in %body fat also confirmed that BMIz change is challenging to interpret without knowing a child’s initial BMIz.
Relation of BMI to %50th and %95th for 5-year-old (60 months) boys and girls. The range of BMI values is based on 5-year-olds in NHANES (1999–2000 through 2017–2018). Age 5 years was chosen because it approximates the mean age of children across all examinations.
Relation of BMI to %50th and %95th for 5-year-old (60 months) boys and girls. The range of BMI values is based on 5-year-olds in NHANES (1999–2000 through 2017–2018). Age 5 years was chosen because it approximates the mean age of children across all examinations.
Our results showing that changes in body fatness are more strongly associated with changes in %50th and %95th than with changes in BMIz among children confirm those reported for older children. For example, a 2-year study of 8- to 10-year-olds (N = 557), with a mean BMIz of 0.6 SDs, showed that changes in body fatness were more strongly associated with changes in BMI and %50th (on a logarithmic scale) than with changes in BMIz.28
Other investigators have examined body fat and BMI changes, but most have focused on children with obesity. One study29 of 59 5- to 17-year-olds with obesity compared changes in body fatness to changes in BMI, BMIz, and BMI expressed as a percentage of the cut points for a BMI of 25 kg/m2 (age 18 years) in the International Obesity Task Force (IOTF)30 . Over 2 years, change in %body fat was more strongly associated with changes in BMI and %IOTF-25 than with changes in BMIz (r = 0.68–0.70 vs 0.57). IOTF-25 corresponds to the CDC growth charts’ 86th to 87th percentile. Another study of 339 2- to 12-year-olds, 80% of whom had obesity and 34% of whom had severe obesity, found that changes in BMI (untransformed), %50th, and %95th over 1 year were more strongly correlated with changes in %body fat (R2s of 0.53–0.55) than were changes in BMIz (R2 = 0.38).31 However, this study assessed body fatness using bioelectrical impedance, a less accurate technique than DXA32 . Other data, such as the tracking of BMIz over time, have also shown that BMIz is an inferior BMI metric among children with severe obesity.8–10
The current study examined 3 BMI metrics, but others have been proposed. These include BMI distance from the median, adjusted %distance from the median, and %distance from the median expressed on a logarithmic scale.11 Because the intercorrelations between changes in %50th and %95th with these other BMI metrics in the current study were r≥0.98, we focused only on %50th and %95th. These strong intercorrelations indicate that similar results would likely be obtained in analyses using any metric that expresses BMI relative to the CDC 50th or 95th percentile. Further, additional analyses found similar differences between changes in the BMI metrics and body fatness if the latter was assessed using fat mass (kg), fat mass index (fat mass ÷ height2 ), or height-adjusted fat mass rather than %body fat.
Another BMI metric, extended BMIz, has also been proposed as an alternative to BMIz33 . The metric shows a fairly linear relation with BMI among children with obesity, but its values are identical to BMIz for children without obesity. Further, longitudinal correlations between changes in BMIz and changes in extended BMIz in the current study were >0.99, and changes in body fatness were almost identically correlated with changes in BMIz and extended BMIz. This similarity may be because of the low prevalence of severe obesity among the enrolled children.
It has been suggested that until consensus is reached on which BMI metric is optimal for assessing change, researchers may wish to present the results of analyses on the basis of several BMI metrics.25 However, it is possible that focusing on changes in %50th may be preferred in many situations. For example, BMI expressed relative to the 50th percentile of the CDC growth charts is easily understood, and the 50th percentile was more accurately estimated than the 95th percentile15 (see Figs 60 and 62 of Kuczmarski et al). Further, %50th can be adjusted for changes in the dispersion of BMI values with age in studies with a long follow-up or a wide range of ages at the initial examination.11 The dispersion of BMI values in the CDC growth charts increased by about twofold between the ages of 3.5 years and 18 years.8
Several limitations of the current study should be considered. First, analyses were based on a convenience sample enrolled in 2001 and 2002, but BMI levels and the prevalence of obesity were close to those among similarly aged children in NHANES 2001 to 2006. The mean BMI of children in the current study and NHANES34 during this time period was 16.5 kg/m2. Although data for the current study were collected about 20 years ago, there is little reason to think that the association between changes in body fatness and BMI has changed during this period. Second, although our results are based on 3- to 7-year-olds, our findings concerning changes in BMI metric and body fatness are similar to those found among 5- to 17-year-olds,29 2- to 12-year-olds,31 and 8- to 10-year-olds,28 suggesting they are likely generalizable to children of various ages. Another limitation of the current study is the lack of data on Hispanic and Asian children; this was based on the low prevalences of these groups at the time of study recruitment. Therefore, it is possible that our results cannot be generalized to Hispanic, Asian, or other children. It was also not possible to examine the performance of the various BMI metrics among children with severe obesity, but stratified analyses showed that %50th and %95th were superior metrics among children with obesity and those without obesity.
Conclusions
Although assessment of BMI is commonplace, body fat is not typically assessed in medical practice or research studies because of time limitations, lack of equipment, and cost. However, excess body fat is a better marker of metabolic health than BMI. Thus, using the best BMI metric to assess change in body fatness among children is essential for clinical management and interpreting the results of weight prevention programs, clinical trials, and monitoring of children over time. Our results indicate that BMIz is not the best BMI metric in studies focusing on changes in body fatness or when monitoring children over time. This is true for both children with obesity and those without obesity. Therefore, rather than focusing on change in BMIz, it would be preferable to use change in %50th or %95th.
Dr Freedman conceptualized the analysis, analyzed the data, and drafted and revised the manuscript; Drs Woo and Daniels conceptualized the study, was involved in collecting the data, and critically reviewed the manuscript; and all authors have approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
The Cincinnati Children’s Hospital Medical Center institutional review board approved the study. The Centers for Disease Control and Prevention has determined that this research is exempt from IRB review.
The findings and conclusions of this report are those of the authors and do not necessarily represent the Centers for Disease Control and Prevention’s official position.
FUNDING: Supported by grant R01HL064022 from the National Heart, Lung, and Blood Institute. Funded by the National Institutes of Health (NIH).
CONFLICT OF INTEREST DISCLOSURES: The authors have no conflicts of interest relevant to this article to disclose.
- BMIz
BMI z score
- CDC
Centers for Disease Control and Prevention
- DXA
dual-energy x-ray absorptiometry
- %50th
BMI expressed as a percentage of the 50th percentile of the CDC growth charts
- %95th
BMI expressed as a percentage of the 95th percentile of the CDC growth charts
- IOTF
International Obesity Task Force
- NHANES
National Health and Nutrition Examination Survey
Comments