Screen Many Predictors with Univariable Fine-Gray Models

Runs [tbl_fine_gray_single()] for each feature and stacks the results, with a multiplicity adjusted p-value. A feature that cannot be fitted (for example zero variance in the rows used) is left out with a warning, so one bad column does not stop the screen.

Usage

tbl_fine_gray_screen(
  data,
  time_col,
  event_col,
  features,
  failcode = 1,
  cencode = 0,
  conf_level = 0.95,
  exponentiate = TRUE,
  p_adjust = "BH",
  ...
)

Arguments

data

A data frame, one row per subject.

time_col

Follow-up time.

event_col

Event type, as in [tbl_cumulative_incidence()]: a censoring code, then one code per cause.

features

Character vector of predictor columns.

failcode

Code of the cause of interest (default 1).

cencode

Code for censoring (default 0).

conf_level

Confidence level.

exponentiate

Report the subdistribution hazard ratio (default); `FALSE` gives the coefficient on the log scale.

p_adjust

Method for `stats::p.adjust()` (default `"BH"`), applied across all rows.

...

Passed to `cmprsk::crr()` (for example `maxiter`).

Value

The stacked [tbl_fine_gray_single()] rows plus `p.adj`.

See also

[tbl_fine_gray_single()], [tbl_fine_gray_multi()]

Examples

if (requireNamespace("cmprsk", quietly = TRUE)) {
  set.seed(1)
  d <- data.frame(time = rexp(150, 0.1), type = sample(0:2, 150, TRUE, prob = c(0.3, 0.4, 0.3)),
                  age = rnorm(150, 60, 8), bmi = rnorm(150, 27, 4), arm = rep(c("A", "B"), 75))
  tbl_fine_gray_screen(d, "time", "type", c("age", "bmi", "arm"))
}
#> # A tibble: 3 × 11
#>   feature term  estimate std.error statistic p.value conf.low conf.high     n
#>   <chr>   <chr>    <dbl>     <dbl>     <dbl>   <dbl>    <dbl>     <dbl> <int>
#> 1 age     age      1.00     0.0144     0.341   0.733    0.977      1.03   150
#> 2 bmi     bmi      0.973    0.0277    -0.991   0.322    0.921      1.03   150
#> 3 arm     armB     0.827    0.236     -0.805   0.421    0.521      1.31   150
#> # ℹ 2 more variables: n_event <int>, p.adj <dbl>