Evaluate continuous markers with ROC curves

roc() measures how well one or more continuous markers discriminate a binary outcome. It reports the area under the ROC curve (AUC) with a confidence interval and, by default, the Youden-optimal cutpoint with its sensitivity, specificity, and predictive values. With two or more markers, their AUCs are compared pairwise with the DeLong test. For a test that is already binary use diag_test(). Requires the pROC package.

Usage

roc(
  data,
  marker,
  outcome,
  positive = NULL,
  direction = c("auto", ">", "<"),
  cutpoint = c("youden", "none"),
  conf.level = 0.95,
  d = 2,
  percent = TRUE
)

Arguments

data: A data frame.
marker: One or more numeric marker columns: a bare name, c() of names, or a tidyselect expression.
outcome: The binary outcome column: a bare name or string.
positive: The level of outcome that means "disease present". If NULL, common labels such as "Yes", "1", or "Positive" are detected, falling back to the last level.
direction: String. Which marker values indicate disease: "<" if higher values do, ">" if lower values do, or "auto" to let pROC choose by comparing the group medians.
cutpoint: String. "youden" reports the cutpoint that maximises sensitivity + specificity - 1; "none" reports the AUC only.
conf.level: Number between 0 and 1. Confidence level for AUC intervals.
d: Integer. Decimal places for displayed estimates.
percent: Logical. If TRUE, show the cutpoint's sensitivity, specificity, and predictive values as percentages. The AUC and the cutpoint itself are always shown as decimals.

Value

A simtab_result of class simtab_roc. Print it to see the AUC table and comparisons, convert it with as.data.frame(), or save it with export_docx(), export_pptx(), or export_xlsx(). Unrounded results are stored in $data.

Details

Statistical methods

ROC curves and AUCs are computed with pROC (Robin et al., 2011), using DeLong intervals for the AUC and the paired DeLong test to compare markers measured on the same patients (DeLong et al., 1988). A cutpoint chosen from the same data is optimistic: its sensitivity and specificity will usually be lower in new patients (Ewald, 2006), and a note says so. The AUC describes discrimination only, not calibration.

Missing data

Rows with a missing outcome or marker value are dropped, across all markers at once, so every marker is evaluated on the same patients. A message reports how many rows were removed. Each marker needs both outcome classes among the remaining rows.

Modifying the result

Use fmt() to change decimals or percentages and ggplot2::autoplot() to draw the ROC curves. The curve coordinates and DeLong comparisons are stored in $data.

References

Robin, X., Turck, N., Hainard, A., et al. (2011). pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics, 12, 77. doi:10.1186/1471-2105-12-77 .

DeLong, E. R., DeLong, D. M., & Clarke-Pearson, D. L. (1988). Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics, 44(3), 837–845. doi:10.2307/2531595 .

Ewald, B. (2006). Post hoc choice of cut points introduced bias to diagnostic research. Journal of Clinical Epidemiology, 59(8), 798–801. doi:10.1016/j.jclinepi.2005.11.025 .

Examples

if (requireNamespace("pROC", quietly = TRUE)) {
  substudy <- subset(epitabl, diagnostic_substudy == "Yes")

  # AUC and Youden cutpoint for point-of-care troponin
  roc(substudy, poc_hstn_value, adjudicated_acs)

  # Compare two markers with the paired DeLong test
  fit <- roc(substudy, c(poc_hstn_value, systolic_bp), adjudicated_acs,
             positive = "Yes")
  fit

  # AUC only, no data-driven cutpoint
  roc(substudy, poc_hstn_value, adjudicated_acs, cutpoint = "none")
}
#> Removed 21 observation(s) with missing values (2.3%).
#> Auto-detected outcome positive level: 'Yes'
#> Removed 21 observation(s) with missing values (2.3%).
#> Removed 21 observation(s) with missing values (2.3%).
#> Auto-detected outcome positive level: 'Yes'
#> 
#> ROC Curve Analysis
#> ==================
#> Outcome: adjudicated_acs (positive = 'Yes')
#> CI method: delong | Direction: auto
#> 
#>          Marker       AUC (95% CI) Direction Cutpoint Sensitivity Specificity
#>  poc_hstn_value 0.90 (0.89 - 0.92)         <                                 
#>  PPV NPV
#>         
#> ℹ Methodological guidance
#>   2 additional methodological notes hidden. Run simtablr_guidance("teaching")
#>   separately before printing to show all advice.