Skip to contents

Per-Subject Summaries of Repeated Measurements

Source: R/transform_longitudinal_features.R
transform_longitudinal_features.Rd

Collapses each subject's repeated values of a feature into one number, using one or more transformations that are often tried as predictors of an outcome: change from baseline, within-subject variability, or the largest swing.

Usage

transform_longitudinal_features(
  data,
  id_col,
  time_col,
  features,
  methods = c("pct_change_baseline", "variance", "sd", "max_signed_diff"),
  format = c("long", "wide")
)

Arguments

data

A data frame in long format (one row per subject and time).

id_col, time_col

Names of the subject and numeric time columns.

features

Character vector of numeric feature columns.

methods

One or more of `"pct_change_baseline"`, `"variance"`, `"sd"`, `"max_signed_diff"`.

format

`"long"` (default; one row per subject, feature and method) or `"wide"` (one row per subject, columns named `<feature>__<method>`).

Value

A tibble. Long format: the id column (named as `id_col`), `feature`, `method`, `value`, `n_obs`. Wide format: the id column and the value columns.

Definitions

Within a subject and feature, missing values are dropped and the rest are ordered by `time_col` (ties in time keep the data order). `n_obs` is the number of values used. * `"pct_change_baseline"`: `100 * (last - first) / abs(first)`, with `first` the value at the earliest time (the baseline) and `last` the value at the latest time; the absolute value keeps the sign meaning "went up" or "went down" when the baseline is negative. `NA` if the baseline is 0 or there is only one value. * `"variance"` and `"sd"`: the sample variance and standard deviation (divisor `n - 1`) of all values; `NA` with fewer than two values. * `"max_signed_diff"`: `max - min` of the values, **signed by time order**: positive when the (first) maximum comes after the (first) minimum, negative when it comes before, and 0 when all values are equal. `NA` with fewer than two values.

See also

[batch_cox_single()], [plot_feature_transform_heatmap()]

Examples

long <- data.frame(id = rep(1:3, each = 3), time = rep(c(0, 5, 15), 3),
                   hr = c(80, 90, 70, 60, 61, 62, 100, 90, 80))
transform_longitudinal_features(long, "id", "time", "hr")
#> # A tibble: 12 × 5
#>       id feature method               value n_obs
#>    <int> <chr>   <chr>                <dbl> <int>
#>  1     1 hr      pct_change_baseline -12.5      3
#>  2     1 hr      variance            100        3
#>  3     1 hr      sd                   10        3
#>  4     1 hr      max_signed_diff     -20        3
#>  5     2 hr      pct_change_baseline   3.33     3
#>  6     2 hr      variance              1        3
#>  7     2 hr      sd                    1        3
#>  8     2 hr      max_signed_diff       2        3
#>  9     3 hr      pct_change_baseline -20        3
#> 10     3 hr      variance            100        3
#> 11     3 hr      sd                   10        3
#> 12     3 hr      max_signed_diff     -20        3