Skip to contents

Simulate pediatric body weight and height from NHANES

Source: R/nhanes_peds.R
sample_nhanes.Rd

Simulates a virtual pediatric population with realistic body weight and/or standing height by weighted resampling of children in the National Health and Nutrition Examination Survey (NHANES), see nhanes_peds, optionally smoothed with random noise.

Usage

sample_nhanes(
  n = 500,
  ages = 2:17,
  sex = c("Male", "Female"),
  vars = c("WT", "HT"),
  cycles = NULL,
  method = c("resample", "smooth"),
  bandwidth_factor = 1,
  seed = NULL,
  data = pmxTools::nhanes_peds
)

Arguments

n

Number of children to simulate per age x sex stratum.

ages

Ages in whole years (2 to 17 for nhanes_peds).

sex

"Male", "Female" or both.

vars

Measures to simulate: "WT" (body weight, kg), "HT" (standing height, cm) or both.

cycles

NHANES releases to use (values of data$CYCLE); NULL uses all of them.

method

"resample" (default) returns the donor children's recorded values; "smooth" adds kernel noise to them.

bandwidth_factor

Multiplier for the kernel bandwidth when method = "smooth"; values above 1 smooth more. Ignored, with a warning if supplied, when method = "resample".

seed

Optional random seed. The previous random number generator state is restored afterwards.

data

Reference data with the columns of nhanes_peds.

Value

A tibble with one row per simulated child and columns ID, AGE, SEXN (1 = male, 2 = female), SEX, the requested vars, SOURCE_CYCLE and SOURCE_SEQN (the NHANES release and sequence number of the donor child). Attributes: "kernels", a tibble of per-stratum diagnostics (N_NHANES donors, N_EFF effective sample size, H_WT and/or H_HT bandwidth SDs on the log scale, 0 when resampling, and RHO, the weighted log weight-height correlation, when both measures are simulated); "cycles", "vars" and "method", as used.

Details

Each age (whole years) x sex stratum is simulated separately:

  1. Donors are the stratum's children with every requested measure in vars. A child missing height is therefore still used when only weight is requested.

  2. Each donor's sampling probability is its MEC exam weight, normalised within its NHANES release, times an equal share per release, so that no release dominates because of a larger total weight. The equal share is a modelling choice, not an official pooled NHANES weight.

  3. n donors are drawn with these probabilities. With method = "resample" (the default) each simulated child gets its donor's recorded values, so every value is one that was measured.

  4. With method = "smooth", Gaussian noise is added to the donors' values on the log scale, so values stay positive and continuous rather than repeating the reference values. The noise covariance is $$H = (b \cdot 0.9)^2 \, n_{eff}^{-2/(d+4)} \, S R S$$ where b is bandwidth_factor, d is the number of vars, \(n_{eff}\) is Kish's effective sample size, S holds each log measure's robust scale \(\min(SD, IQR/1.349)\) and R is the weighted correlation matrix of the log measures. With one measure this is the robust Silverman rule \(h = 0.9 \, \sigma \, n_{eff}^{-1/5}\); with both, the correlated noise preserves the weight-height relationship.

Use compare_nhanes() and plot_nhanes() to check the simulated population against the reference.

See also

nhanes_peds, compare_nhanes(), plot_nhanes()

Examples

sim <- sample_nhanes(n = 100, ages = c(2, 8, 14), seed = 20261008)
head(sim)
#> # A tibble: 6 × 8
#>      ID   AGE  SEXN SEX      WT    HT SOURCE_CYCLE SOURCE_SEQN
#>   <int> <int> <int> <chr> <dbl> <dbl> <chr>              <dbl>
#> 1     1     2     1 Male   13.7  90.7 2013-14            81515
#> 2     2     2     1 Male   13.4  91.6 2015-16            90822
#> 3     3     2     1 Male   13.5  87   2021-23           136484
#> 4     4     2     1 Male   15.3  93.5 2015-16            84344
#> 5     5     2     1 Male   11.7  83.1 2013-14            73710
#> 6     6     2     1 Male   12.6  87.1 2015-16            85690

# Smoothed: continuous values around the reference children
smoothed <- sample_nhanes(n = 100, ages = c(2, 8, 14), method = "smooth",
                          seed = 20261008)
attr(smoothed, "kernels")
#> # A tibble: 6 × 7
#>     AGE SEX    N_NHANES N_EFF   H_WT   H_HT   RHO
#>   <int> <chr>     <int> <dbl>  <dbl>  <dbl> <dbl>
#> 1     2 Male        366  263. 0.0429 0.0171 0.735
#> 2     2 Female      381  238. 0.0474 0.0168 0.752
#> 3     8 Male        408  275. 0.0835 0.0153 0.703
#> 4     8 Female      364  245. 0.0909 0.0176 0.690
#> 5    14 Male        327  208. 0.104  0.0188 0.637
#> 6    14 Female      333  204. 0.0814 0.0137 0.435

# Weight only, using every child with a measured weight
wt <- sample_nhanes(n = 100, ages = 10, vars = "WT", seed = 1)
summary(wt$WT)
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
#>   22.60   33.17   39.25   42.21   49.67  101.80