Baseline-Anchored Longitudinal Composite Index

Builds a composite score for every subject and time point. Each variable is standardised with the **baseline** mean and SD (at `baseline_time`, across subjects), multiplied by its polarity (`alpha`, +1 if a higher value means worse/more, -1 if lower does), and the score is the mean of the available standardised variables: `index = mean_i( alpha_i * (x_i - mean0_i) / sd0_i )`. At baseline each variable has mean 0, so the index has mean 0 at baseline and a change reads as standard deviations of the baseline distribution. For one time point (cross-sectional data) use [compute_composite_index()].

Usage

compute_longitudinal_composite_index(
  data,
  variables,
  index_map = NULL,
  id_col = "subject_id",
  time_col = "time",
  baseline_time = 0,
  index_name = "composite_index",
  min_vars = 1L
)

Arguments

data

A data frame in long format.

variables

Numeric columns that contribute.

index_map

Optional data frame with columns `variable` and `alpha` (+1 or -1); variables not listed get +1.

id_col, time_col

Subject and time columns.

baseline_time

Time that defines the baseline rows.

index_name

Name of the new column.

min_vars

Smallest number of non-missing standardised variables a row needs to get a score (default 1); otherwise `NA`.

Value

`data` as a tibble with the index column added. The attribute `baseline` has one row per variable (`variable`, `alpha`, `mean0`, `sd0`, `n0`). A variable whose baseline SD is 0 or undefined (fewer than two baseline values) is left out with a warning.

See also

[compute_composite_index()]

Examples

set.seed(1)
d <- expand.grid(id = 1:20, time = c(0, 5, 15))
d$a <- rnorm(60, 10 + d$time / 5, 2)
d$b <- rnorm(60, 50 - d$time, 5)
out <- compute_longitudinal_composite_index(d, c("a", "b"), data.frame(variable = "b", alpha = -1),
                                            id_col = "id")
tapply(out$composite_index, out$time, mean)
#>             0             5            15 
#> -2.733789e-16  6.333390e-01  2.215371e+00