A SAS Program for the 2000 CDC Growth Charts (ages 0 to <20 years)

The purpose of this SAS program is to calculate the percentiles and z-scores (standard deviations) for a child’s sex and age for BMI, weight, height, and head circumference based on the CDC growth charts. Weight-for-height percentiles and z-scores are also calculated. Observations that contain extreme values are flagged as being biologically implausible. These extreme values, however, are not necessarily incorrect.

Although the SAS program can be used to calculate z-scores and percentiles for children up to 20 years of age, the World Health Organization (WHO) growth charts are recommended for children <24 months of age. There are several computer programs available on the WHO and CDC sites that use the WHO growth charts; the latter follows the same steps as does this SAS program for the CDC growth charts.

The SAS program, cdc-source-code.sas [SAS - 8KB] (files are below, in step #1), calculates these z-scores and percentiles for children in your data based on reference data in cdc_ref.sas7bdat. If you’re not using SAS, you can download CDCref_d.csv [CVS -160KB], and create a program based on cdc-source-code.sas [SAS-8KB] to do the calculations.

Instructions for SAS users

Step 1: Download the SAS program (cdc-source-code.sas [SAS -8KB]) and the reference data file (CDCref_d.sas7bdat). Do not alter these files, but move them to a folder (directory) that SAS can access.

For the following example, the files have been saved in c:\sas\growth charts\cdc\data.

Step 2: Create a libname statement in your SAS program to point at the folder location of ‘CDCref_d.sas7bdat’. An example would be:
libname refdir 'c:\sas\growth charts\cdc\data';

Note the SAS code expects this name to be refdir; do not change this name.

Step 3: Set your existing dataset containing height, weight, sex, age and other variables into a temporary dataset, named mydata. Variables in your dataset should be renamed and coded as follows:

Table 1

Variable	Description
agemos	Child's age in months; must be present. The program assumes you know the number of months to the nearest day based on the dates of birth and examination. For example, if a child was born on Oct 1, 2007 and was examined on Nov 15, 2011, the child’s age would be 1506 days or 49.48 months. In everyday usage, this age would be stated as 4 years or as 49 months. However, if 49 months were used as the age of all children who were between 49.0 and <50 months in your data, the estimated z-scores would be slightly too high because, on average, these children would be taller, weigh more, and have a higher BMI than children who are exactly 49.0 months of age. This bias would be greater if only completed years of age were known, and the age of all children between 4 and <5 years was represented as 48 months. If age is known only as the completed number of months (as is data from NHANES 1988-1994 and 1999-2010), consider adding 0.5 so that the maximum error would be 15 days. If age is given as the completed number of years, multiply by 12 and consider adding 6.
sex	Coded as 1 for boys and 2 for girls.
height	Height in cm. This is either standing height (for children who are ≥ 24 months of age or recumbent length (for children < 24 months of age); both are input as height. If standing height was measured for some children less than 24 months of age, you should add 0.8 cm to these values (see page 8 of http://www.cdc.gov/nchs/data/series/sr_11/sr11_246.pdf [PDF-5.4MB]). If recumbent length was measured for some children who are ≥ 24 months of age, subtract 0.8 cm.
weight	Weight (kg)
bmi	BMI (Weight (kg) /Height (m)2). If your data doesn’t contain BMI, the program calculates it. If BMI is present in your data, the program will not overwrite it.
headcir	Head circumference (cm)

Z-scores and percentiles for variables that are not in mydata will be coded as missing (.) in the output dataset (named _cdcdata). Sex (coded as 1 for boys and 2 for girls) and agemos must be in mydata. It’s unlikely that the SAS code will overwrite other variables in your dataset, but you should avoid having variable names that begin with an underscore, such as _bmi.

Step 4: Copy and paste the following line into your SAS program after the line (or lines) in step #3.
%include 'c:\sas\growth charts\cdc\data\CDC-source-code.sas'; run;

If necessary, change this statement to point at the folder containing the downloaded ‘CDC-source-code.sas’ file. This tells your SAS program to run the statements in ‘CDC-source-code.sas’.

Step 5: Submit the %include statement. This will create a dataset, named _cdcdata, which contains all of your original variables along with z-scores, percentiles, and flags for extreme values. The names and descriptions of these new variables in _cdcdata are in Table 2. Additional information on the extreme z-scores is given in a separate section that follows the “Example SAS Code”.

 Top of Page

Table 2: Z-Scores, percentiles, and extreme (biologically implausible, BIV) values in output dataset, _cdcdata

Description	Variable				Cutoff for Extreme Z-Scores
	Percentile	Z-score	Modified Z-score to Identify Extreme Values	Flag for Extreme Values	Low z-score (Flag coded as -1)	High z-score (Flag coded as +1)
Weight-for-age for children between 0 and 239 (inclusive) months of age	wapct	waz	_Fwaz	_bivwt	< -5	> 8*
Height-for-age for children between 0 and 239 (inclusive) months of age.	hapct	haz	_Fhaz	_bivht	< -5	>4*
Weight-for-height for children with heights between 45 and 121 cm (this height range approximately covers ages 0 to 6 y)	whpct	whz	_Fwhz	_bivwh	< -4	>8*
BMI-for-age for children between 24 and 239 months of age	bmipct	bmiz	_Fbmiz	_bivbmi	< -4	>8*
Head circumference-for-age for children between 0 and 35 (inclusive) months of age	headcpct	headcz	_Fheadcz	_bivhc	< -5	>5

* Changed in 2016. Additional information is below

Step 6: Examine the new dataset, _cdcdata, with PROC MEANS or some other procedure to verify that the z-scores and other variables have been created. If a variable in Table 1 was not in your original dataset (e.g., head circumference), the output dataset will indicate that all values for the percentiles and z-scores of this variable are missing. If values for other variables are unexpectedly missing, make sure that you’ve renamed and recoded variables as indicated in Table 1 and that your SAS dataset is named mydata. The program should not modify your original data, but will add new variables to your original dataset.

Example SAS code corresponding to steps 2 to 6. You can simply cut and paste these lines into a SAS program, but you’ll need to change the libname and %include statements to point at the folders containing the downloaded files.

libname refdir 'c:\sas\growth charts\cdc\data';
data mydata; set whatever-your-original-dataset-is-named;
%include 'c:\sas\growth charts\cdc\data\CDC-source-code.sas';
proc means data=_cdcdata; run;

Additional Information

Z-scores are calculated as =

Z = [ ((value / M)**L) – 1] / (S * L) ,

in which ‘value’ is the child’s BMI, weight, height, etc. The L, M, and S values are in CDCref_d.sas7bdat and vary according to the child’s sex and age or according to the child’s sex and height. Percentiles are then calculated from the z-scores (for example, a z-score of 1.96 would be equal to the 97.5 percentile). For more information on the LMS method, see http://www.ncbi.nlm.nih.gov/pmc/articles/PMC27365/

 Top of Page

Extreme or Biologically implausible Values

As explained in the Modified z-scores documentation [PDF - 367KB] , the SAS code also calculates modified z-scores that can be used to identify extreme values that may be errors. These modified z-scores are based on extrapolating one-half of the distance between 0 and +2 z-scores to the tails of the distribution. The output from the SAS program contains BIV flag variables that are coded as -1 (modified z-score is extremely low), +1 (modified z-score is extremely high), or 0 (modified z-score is between these 2 cut-points). These BIVs flags (e.g., _bivbmi), along with other variables that are in the output dataset, _cdcdata, are shown in Table 2. A biologically implausible value is not necessarily incorrect, but the value should be further examined, possibly in conjunction with other characteristics of the child.

 Top of Page

2016 Change to BIV cut-points: Rationale

The modified z-scores used for the upper range of valid values was changed in 2016 for a number of the growth chart parameters. Previously, the cut-points for extremely high values were based on recommendations from a 1995 WHO publication (1), but several papers (2–6) have since indicated that these cut-points were probably too restrictive. The WHO cut-points identified many values that were extremely high, but were probably not errors.

On the basis of an analyses of 2- to 18-year-olds in NHANES 1999-2000 through 2011-2012 (3) and 2- to 4-year-olds in CDC’s Pediatric Nutrition Surveillance System (PedNSS) (6), we now suggest that the upper BIV cut points be increased from
(1) +5 to +8 for modified z-scores for weight and BMI, and
(2) +3 to +4 for modified z-scores for height.
These new z-score cut-points roughly correspond to the modified z-scores for the maximum values of the body size measures among 2- to 18-year-olds in NHANES. We are not making changes to the cut-points for the extremely low values of the body size measurements.

If BIV cut-points are used to exclude data, this change would likely affect comparisons of data calculated and cleaned using these new BIV cut-points with data that used the older (WHO 1995) values. The effects of these changes will likely differ across datasets depending upon the true prevalence of extreme values and the accuracy of the recorded data. In an analysis (3) of NHANES 1999-2012 data, for example, as compared with estimates obtained using the WHO 1995 cut-points, the use of the 2016 cut-points increased the prevalence of obesity and extreme obesity (120% of the 95th percentile of BMI-for-age) by about 0.5 percentage points. (Because of the extensive data cleaning in NHANES, published estimates from these surveys do not exclude any of the extremely high values.) In an analysis of PedNSS (6), compared with the WHO 1995 cut-points, the use of the 2016 cut points increased the prevalence of both obesity and extreme obesity by 0.9 percentage points. Because of the relatively low prevalence of extreme obesity among children, particularly pre-school children, a 0.5% to 0.9% increase results in a large proportional change in prevalence.

BIVs vs. Data Errors

These BIVs can be used to flag potentially problematic data points, and the 2016 cut-points were chosen to balance the inclusion of extreme values that are likely to be correct and the exclusion of those that are likely to be incorrect. However, other cut-points can be used and may be more appropriate based on other information specific to your data. If desired, the modified z-scores (3rd column of Table 2) can be used to construct other cut-points for extreme (or biologically implausible) values rather than relying on the BIV flag variables. For example, if you feel that use of the BMI-for-age cut-point of +8 would result in the inclusion of many values that are likely to be errors, you could use F_bmiz > 6 as the definition of a high BMI BIV. This could be re-coded in the output dataset as:

if -5 <= _Fbmiz <= 6 then _bivbmi=0; *plausible;
else if _Fbmiz > 6 then _bivbmi=1; *high BIV;
else if . < _Fbmiz < -5 then _bivbmi= -1; *low BIV;

It would also be possible to use the modified z-scores to identify children who would have been flagged with the older WHO cut-points.

Once a data point has been flagged as a potential problem, other information from the child, if available, could be used to help identify errors and help in the decision to include or exclude the value. For example, if a child with an extremely high BMI also has a high skinfold thickness or arm circumference, the BMI value is more likely to be correct than if the other measure is low. Similarly, in a longitudinal study, one could assess whether a child with an extreme value at 1 time point also has a high value at other examinations. If only weight and height are available at a single examination, one might consider whether a child who has an extremely high weight is also very tall, and if there are other children who weigh nearly as much.

 Top of Page

Defining Severe Obesity and BMI Metrics Among Children with Very High BMIs

The use of the LMS parameters of the CDC growth charts for children with very high BMIs has been shown to result in percentiles that differ from those that are estimated from the data http://www.ajcn.org/content/90/5/1314.full.pdf [PDF-153KB]. Therefore, rather than using the BMI percentiles (and z-scores) to identify and characterize children with severe (or extreme) obesity, it is recommended that these high BMI values be expressed relative to the 95^thpercentile. A BMI value that is 120% of the 95^th percentile in the CDC growth charts is approximately equal to the empirical 99^th percentile. This cut-point (120%) is recommended for the classification of extreme (or severe) obesity among 2- to 19-year-olds [7]. The variables bmiz and bmipct in the SAS output should not be used in the classification of this subset of children with very high BMIs because the highest percentile that was estimated in the CDC growth charts was the 97^th https://www.cdc.gov/nchs/data/series/sr_11/sr11_246.pdf [PDF-5.31MB], https://www.cdc.gov/nchs/data/nhsr/nhsr063.pdf [PDF-87KB].

The drawbacks of expressing very high BMIs as z-scores (or percentiles) have been emphasized by several investigators [8-12]. The SAS code creates 2 variables, bmipct95 and bmidif95, that express a child's BMI relative to the 95^th percentile either as a percentage (bmipct95) or as a difference (bmidif95). These 2 variables are likely to be better measures of adiposity among children who have very high BMIs than are z-scores and percentiles. Bmipct95 can range from below 50 to over 220, and a child with a bmipct95 of 140 would have a BMI that is equal to 1.4 times the 95th percentile. Bmidif95 is the difference (in kg/m²) between the child's BMI and the CDC 95^th percentile for that sex/age. For example, the CDC 95^th percentile for a 20-month-old boy is 18.0 kg/m²; if this boy had a BMI of 21.3kg/m², his bmidif95 would be 3.3 kg/m² (21.3 - 18.0) and his bmipct95 would be 118% (100 × 21.3/18.0). A negative value for bmidif95 (or a bmipct95 < 100) would indicate that the child does not have obesity.  In addition to these 2 variables, the SAS program also outputs the CDC 50th (bmi50) and 95^th (bmi95) percentiles for the child's sex and age.

If a large proportion of children in an analysis have severe obesity (bmipct95 ≥ 120), you should consider expressing all BMIs relative to the 95^th percentile and using bmipct95 or bmidif95 in analyses. The limitations of expressing extremely high BMIs as z-scores apply to both cross-sectional and longitudinal studies, including those that evaluate obesity interventions.

Top of Page