BMI-for-age graphs with severe obesity percentile curves: tools for plotting cross-sectional and longitudinal youth BMI data

We categorized weight status according to the 2007 Expert Committee recommendations [12], with the additional category of severe obesity described by Flegal et al. in 2009 [13] and defined in a Scientific Statement of the American Heart Association in 2013 [1]. In addition, we included two distinct classes of severe obesity defined by Skinner and Skelton in 2014 [14]. The resulting six weight classifications are underweight (BMI-for-age <5th percentile), healthy weight (5th to <85th percentile), overweight (85th to <95th percentile), obese class 1 (95th percentile to <120% of the 95th percentile), severe obesity class 2 (120% to <140% of the 95th percentile or BMI 35.0 to <40.0 kg/m²), and severe obesity class 3 (BMI ≥ 140% of the 95th percentile or BMI ≥ 40.0 kg/m²).

The reference population used to determine BMI percentiles and z-scores is based on a large sample of children and adolescents whose height and weight were measured as part of the National Health Examination Surveys (NHES) and the National Health and Nutrition Examination Surveys (NHANES) conducted between 1964 and 1994. NHES and NHANES are part of the National Center for Health Statistics (NCHS) data sets that were used to develop the 2000 CDC growth charts [15]. The NCHS reference data set needed to produce the results output was obtained from the CDC website [11] in two file formats: SAS bat file (cdc_ref.sas7bdat) and an Excel csv file (cdcref_d.csv); both are provided as Additional files with this article. The NCHS reference data used to produce the percentile curves on the BMI-for-age graphs were obtained from tables on the CDC website [16] and are provided as an Additional Excel file (Ref_percentile_curves.xlsx).

Table 1 lists the data inputs required to use the graphing tools. For cross-sectional data sets, each participant ID number must be unique. When participant ID numbers are replicated within a data set, the graphing programs treat the data as longitudinal. Sex is an essential variable because BMI percentiles and z-scores are sex-specific. Age should be computed from date of birth and date of assessment and expressed to one or more decimal places for greatest accuracy. Self-reported age as a whole number may lead to misclassification of weight status, particularly for young children. Height (cm) and weight (kg) must be provided in metric units; height and weight data obtained in English units should be converted to metric units: height (cm) = height (inches) * 2.54; weight (kg) = weight (lbs) / 2.20462. If height and weight data are not available but BMI data are available, then BMI (kg/m²) can be provided instead. The graphing programs utilize height and weight data preferentially to compute BMI, BMI percentiles, and BMI z-scores. In the absence of height and weight values, investigator-provided BMI values will be used to compute BMI percentiles and z-scores. Age, height, weight, and BMI (if provided) should be expressed to the greatest degree of accuracy for which the measurement was obtained; rounding may reduce the accuracy of the computed BMI values.

Legend: Data inputs reflect variable names and formats required in the investigator’s data file. Results output represent the variables generated by the SAS and R graphing programs.

Graphing program files are provided for the statistical analysis software SAS (SAS Institute Inc., Cary, NC) and the statistical computing and graphing environment R [17], as described below. The results output files generated by SAS and R contain identical results; the graphs generated by SAS and R display the same data points and curves and are similar in appearance.

SAS: Investigators who choose to use SAS must have SAS software and the six files listed in Table 2 to generate the results file and BMI-for-age graphs. The first file is the SAS graphing program, two files are SAS macro files, two files are CDC reference data sets, and the sixth file is the investigator’s data set. The first five files are provided as Additional files; these files must be accessible on the investigator’s computer or network and should be placed in the same folder as the investigator’s data file. SAS version 9.4 was used to create the SAS graphing program.

Legend: All files listed in this table, except the Investigator’s Data file, are provided as Additional files with this article.

R: Investigators who choose to use R must have R software and the four files listed in Table 2 to generate the results file and BMI-for-age graphs. The first file is the R graphing program, two files are CDC reference data sets, and the fourth file is the investigator’s data set. The first three files are provided as Additional files; these files must be accessible on the investigator’s computer or network and should be placed in the same folder as the investigator’s data file. We used R version 3.3.0 to create the R graphing program file. R software is available for download free of charge.

The results output file is an Excel csv file (BMI_Results.csv) that contains the data input variables provided by the investigator plus the output variables indicated in Table 1. BMI (kg/m²) is computed as weight (in kg) divided by height (in meters) squared. Sex- and age-specific BMI percentiles and BMI z-scores are determined based on the reference population. BMI as a percent of the 95th percentile (BMI_95) is an important metric for identifying severe obesity. Weight status is determined based on BMI percentile and BMI as a percent of the 95th percentile. Each weight status category is mutually exclusive, with obesity classes 1, 2, and 3 being distinct and all weight status categories summing to 100%. Therefore, if investigators wish to determine the prevalence of obesity in their sample, they must add the three classes of obesity. Likewise, to determine the prevalence of severe obesity, it is necessary to add obesity classes 2 and 3.

Table 3 lists the file names and file formats for the BMI-for-age graphs generated by SAS and R. These graphs can be output as Adobe Illustrator Encapsulated PostScript (.eps) vector graphics files and/or as Adobe Acrobat Portable Document Format (.pdf) files. Figs. 1 and 2 are provided as examples of cross-sectional and longitudinal graphs generated using the SAS and R program files. These graphs are provided for illustrative purposes only; the data contained in them were drawn from a series of published [6, 18, 19] and unpublished studies that were approved by the Washington University in St. Louis Institutional Review Board.

Our BMI-for-age graphs are designed to mimic the CDC’s 2000 sex-specific BMI-for-age growth charts for males and females [15, 20], with age on the x-axis, BMI (kg/m²) on the y-axis, and several standard percentile curves (i.e., 5th, 50th, 85th, and 95th) displayed on each graph. Four distinct features of these new BMI-for-age graphs are an expanded BMI range to accommodate youth with severe obesity, the inclusion of the two severe obesity percentile curves (i.e., 120% of the 95th percentile to signify the cut point for obese class 2 and 140% of the 95th percentile as the cut point for obese class 3), the ability to plot multiple children and adolescents on each graph, and the ability to generate cross-sectional and longitudinal graphs.

Figure 1a and b depict cross-sectional data for 3900 females and 4000 males, respectively, aged 2 to <20 years, whose height and weight were measured. Each symbol represents a single child or adolescent, with his/her age apparent from the x-axis, BMI indicated on the y-axis, and weight status reflected by the position relative to the percentile curves. Points above the top two percentile curves on each graph reflect severe obesity (i.e., obese classes 2 and 3). These scatterplots enable one to view the extent of obesity and severe obesity quickly and easily.

Figure 2a and b depict longitudinal data for 30 females and 22 males whose height and weight were tracked for up to four years. Each set of circles with a connecting line represents one child; each circle represents one measurement time point. The two informative aspects of these graphs are: 1) the BMI-for-age percentile at each measurement time point represents the child’s weight status at that time point and 2) the slope of each child’s line relative to the percentile curves represents an increase, no change, or a decrease in BMI-for-age percentile over time. A steep upward slope, as is evident for some participants, signifies a rapid (and potentially undesirable) increase in BMI.

Investigators and other individuals who use these graphing tools may modify the SAS and R program files to accommodate the BMI range of their data (i.e., setting the y-axis range lower or higher than 60 kg/m²), the input data file format (e.g., .xlsx, .csv, .bat), and graph preferences (e.g., symbol style, size, and color; line color and thickness; and font style and size). Also, investigators can insert additional code into the beginning of the SAS and R program files to compute age (from date of birth and date of assessment) and to convert height and weight from English to metric units. Other modifications that may be desirable include inserting additional percentile curves on the BMI-for-age graphs (e.g., 3rd, 10th, 25th 75th, 90th, 97th, 99th).

Strengths of these graphing tools are their novelty, the inclusion of severe obesity percentile curves, the ability to plot thousands of youth on each graph, and the flexibility to plot cross-sectional or longitudinal data. A limitation is the need for either SAS or R software and reference data sets.