Reference Files for Parameters by Age

As growth rates and patterns differ by age and sex, reference values differ by these two characteristics.

For the various measures, we use the WHO reference data for children <2 years of age (<730.5 days) and CDC growth charts for children >=2 years of age (>=730.5 days). This is in accord with the CDC recommendation. The reference data for both charts include the LMS parameters, in which L is the Box-Cox transformation for normality, M is the median, and S is the coefficient of variation.

So, the first step is combining the metadata for each measure and ensuring sex and age are shown consistently. We separate WHO and CDC metadata for weight-based derivations due to (1) the WHO adjustment (restricted application of the LMS method) and (2) the CDC extended method for high BMIs.

Additionally, as the CDC growth charts only offer monthly age intervals, on the advice of a retired CDC expert in this area, we added interpolation code in our template to have a record for each day of age. When converting the CDC months into days (30.4375 days in a month), we rounded to the nearest whole number of days. Relying on the monthly charts will result in less accurate calculations, particularly for weight and height values among children who are 2 years of age. If you would prefer to rely solely on the monthly chart here, then you could remove the rounding and the call to the derive_interp_records() function.

We do this pre-processing of the metadata using the following code for BMI as an example:

data(WHO_bmi_for_age_boys)
data(WHO_bmi_for_age_girls)
data(cdc_bmiage)

who_bmi_for_age <- who_bmi_for_age_boys %>%
  mutate(SEX = "M") %>%
  bind_rows(who_bmi_for_age_girls %>%
    mutate(SEX = "F")) %>%
  # Keep patients < 2 yrs old
  filter(Day < 730.5) %>%
  rename(AGE = Day) %>%
  # AGEU is added in metadata, required for derive_params_growth_age()
  mutate(AGEU = "DAYS") %>%
  arrange(AGE, SEX)

cdc_bmi_for_age <- cdc_bmiage %>%
  mutate(
    SEX = case_when(
      SEX == 1 ~ "M",
      SEX == 2 ~ "F",
      TRUE ~ NA_character_
    ),
    # Ensure first that Age unit is "DAYS"
    AGE = round(AGE * 30.4375),
    AGEU = "DAYS"
  ) %>%
  # Interpolate the AGE by SEX so that we get CDC metadata by day instead of
  # month in the same way as WHO metadata
  derive_interp_records(
    by_vars = exprs(SEX),
    parameter = "BMI"
  ) %>%
  # Keep patients >= 2 yrs till 20 yrs - Remove duplicates for 730 Days old which
  # must come from WHO metadata only
  filter(AGE >= 730.5 & AGE <= 7305) %>%
  arrange(AGE, SEX)

Here is how the first records of the WHO metadata for BMI now look:

Similarly, for the CDC metadata:

For BMI in the CDC metadata, a dispersion parameter (Sigma) is used to calculate BMI percentiles and z-scores above the sex- and age-specific 95th percentile (P95).

The above example only shows the parameter BMI, but this follows other parameters like weight and height. We can combine WHO and CDC metadata for height as a WHO adjustment is unnecessary. Note that the head circumference parameter metadata comes only from the WHO reference data, as CDC provides no equivalent for children > 2 years of age.

Reference Files for Weight by Length/Height

WHO provides additional reference data for weight-for-length (recumbent length), instead of age. Again, we combined the metadata for sex to create a single reference file.

There are also weight-by-height files available, but upon advice from a retired CDC expert in this area, we chose only to include the weight-for-length files, as once children can stand, BMI by age is a more appropriate measure to use. However, in our template, we give the metadata values variable a generic name, HEIGHT_LENGTH, so that any user could equally choose to pass weight-for-height files if preferred.

We do this using the following code:

data(who_wt_for_lgth_boys)
data(who_wt_for_lgth_girls)

who_wt_for_lgth <- who_wt_for_lgth_boys %>%
  mutate(SEX = "M") %>%
  bind_rows(who_wt_for_lgth_girls %>%
    mutate(SEX = "F")) %>%
  mutate(HEIGHT_LENGTHU = "cm") %>%
  rename(HEIGHT_LENGTH = Length)

Here is how the metadata now looks for a body length of 65cm:

Derived Variables for Parameters by Age

A variable for current analysis age could be achieved using the following code (you might first need to derive BRTHDT if not already available in ADSL, including any partial date imputation rules if those are possible in your study data collection rules):

# Calculate Current Analysis Age AAGECUR and unit AAGECURU
advs <- advs %>%
  derive_vars_duration(
    new_var = AAGECUR,
    new_var_unit = AAGECURU,
    start_date = BRTHDT,
    end_date = ADT
  )

Here is how these age variables look:

Derived Variables for Weight by Length/Height

Similar to the above, a variable for current analysis length/height could be achieved using the following code:

# Derive Current HEIGHT/LENGTH at each time point Temporary variable
advs <- advs %>%
  derive_vars_merged(
    dataset_add = advs,
    by_vars = c(get_admiral_option("subject_keys"), exprs(AVISIT)),
    filter_add = PARAMCD == "HEIGHT" & toupper(VSSTRESU) == "CM",
    new_vars = exprs(HGTTMP = AVAL, HGTTMPU = VSSTRESU)
  )

Derived Parameters by Age

Parameters for BMI-for-age z-scores and percentiles could be achieved using the derive_params_growth_age() function as follows:

## Derive Anthropometric indicators (Z-Scores/Percentiles-for-Age) based on Standard Growth Charts
## For BMI by Age
advs <- advs %>%
  slice_derivation(
    derivation = derive_params_growth_age,
    args = params(
      sex = SEX,
      age = AAGECUR,
      age_unit = AAGECURU,
      parameter = VSTESTCD == "BMI",
      analysis_var = AVAL,
      set_values_to_sds = exprs(
        PARAMCD = "BMISDS",
        PARAM = "BMI-for-age z-score"
      ),
      set_values_to_pctl = exprs(
        PARAMCD = "BMIPCTL",
        PARAM = "BMI-for-age percentile"
      )
    ),
    derivation_slice(
      filter = AAGECUR < 730.5,
      args = params(
        who_correction = TRUE,
        meta_criteria = who_bmi_for_age
      )
    ),
    derivation_slice(
      filter = AAGECUR >= 730.5,
      args = params(
        bmi_cdc_correction = TRUE,
        meta_criteria = cdc_bmi_for_age
      )
    )
  )

If only a z-score or percentile were needed, you leave out one of the set_values_to_ arguments.

Once again, it is essential to remember the importance of the metadata being supplied to the above function via the meta_criteria argument. Our default case from this package uses WHO reference data for children < 2 years of age and CDC reference data for children >= 2 years of age. As the user, you can use other (e.g., International Obesity Task Force) reference data files created with the LMS method.

For BMI only, you’ll notice from the code above that we have set bmi_cdc_correction = TRUE. This is because we used the CDC extended percentiles (> 95th percentile) to monitor high BMI values here. If you left this argument as default FALSE value, only the 2000 CDC Growth Chart for BMI would be used from the CDC metadata. In 2022, the CDC released extended percentiles (and z-scores) for BMIs > 95th percentile because the original CDC method did not accurately characterize very high BMIs.

For all weight-based parameters (including BMI), you’ll also see in the code above that we have set who_correction = TRUE. This comes from the WHO Child Growth Standards Guidelines (pages 301-304). This correction is made because the WHO used the LMS method only for weight-based measures for which the calculated z-scores were between -3 and +3; this correction allows for the calculation of more extreme z-scores. If you are not using the WHO metadata (or WHO metadata for length or head circumference), you could leave who_correction = FALSE, and only the usual LMS method would be applied. WHO recommends using this correction for all parameters that involve weight (including BMI).

Here is how these newly derived parameters look:

For height parameters by age, it should be noted that the WHO growth chart we use for children <2 years of age refers to body length, whereas the CDC growth chart for children >=2 years of age refers to height. This is explained thoroughly in the next section.

Derived Weight by Length/Height

The following code is for Weight by Length/Height z-score and percentile for children < 2 years of age, which assumes the “HEIGHT” parameter is always collected as recumbent length.

You’ll notice the derive_params_growth_height() function used is very similar to that used above, but now passing in height or length instead of age.

advs <- advs %>%
  restrict_derivation(
    derivation = derive_params_growth_height,
    args = params(
      sex = SEX,
      height = HGTTMP,
      height_unit = HGTTMPU,
      meta_criteria = who_wt_for_lgth,
      parameter = VSTESTCD == "WEIGHT",
      analysis_var = AVAL,
      who_correction = TRUE,
      set_values_to_sds = exprs(
        PARAMCD = "WGTHSDS",
        PARAM = "Weight-for-length/height Z-Score"
      ),
      set_values_to_pctl = exprs(
        PARAMCD = "WGTHPCTL",
        PARAM = "Weight-for-length/height Percentile"
      )
    ),
    filter = AAGECUR < 730.5
  )

We consulted with a pediatrician with extensive experience working with length and height data, who advised that analysts rarely know whether standing height or recumbent length was measured. It depends mainly on the age at which the child can stand (on average, two years) and the preference of the measurer.

If you have strict CRF (case report form) guidelines that standing height (rather than body length) was measured, you could add 0.7 cm to these values to approximate the child’s body length.