Skip to contents

Summarise the diagnostic performance of a numeric marker

Source: R/roc.R
roc_summary.Rd

Computes the area under the receiver operating characteristic curve (AUC) with a confidence interval and the optimal operating point of a numeric marker against a binary outcome. A tibble with one row is returned, so the result can be combined across markers or subgroups.

Usage

roc_summary(
  marker,
  outcome,
  positive = NULL,
  direction = "auto",
  best_method = c("youden", "closest.topleft"),
  ci_level = 0.95,
  power = FALSE,
  alternative = c("one.sided", "two.sided")
)

Arguments

marker

Numeric vector of marker values.

outcome

Binary outcome. Logical vectors are coded TRUE as the event; numeric 0/1 vectors use 1 as the event. Any other coding requires positive to name the event level.

positive

Value of outcome that represents the event.

direction

Direction of the comparison passed on to pROC::roc(). "auto" lets pROC choose the direction from the data; "<" and ">" fix it explicitly.

best_method

Method used to choose the optimal operating point, "youden" (default) or "closest.topleft".

ci_level

Confidence level for the AUC interval.

power

If TRUE, add the power of the AUC test as reported by pROC::power.roc.test(). The value may be NA when power cannot be estimated for the sample.

alternative

Alternative hypothesis used for the power estimate.

Value

A tibble with one row: sample size, number of events, AUC and its confidence interval, the optimal threshold, and the sensitivity, specificity, positive and negative predictive values, and accuracy at that threshold.

See also

odds_ratio() for the odds ratio at the optimal threshold.

Examples

set.seed(1)
marker <- c(rnorm(40, 0), rnorm(40, 1))
outcome <- rep(c(0, 1), each = 40)
roc_summary(marker, outcome)
#> # A tibble: 1 × 12
#>       n n_event   auc auc_ci_lower auc_ci_upper direction threshold sensitivity
#>   <int>   <int> <dbl>        <dbl>        <dbl> <chr>         <dbl>       <dbl>
#> 1    80      40 0.789        0.692        0.887 <             0.828        0.65
#> # ℹ 4 more variables: specificity <dbl>, ppv <dbl>, npv <dbl>, accuracy <dbl>