Diagnostic accuracy: sensitivity, specificity, ROC

Quantifies how well a new test identifies patients with the condition, compared with a reference standard.

Clinical example

A rapid blood test for myocardial infarction compared with the final clinical diagnosis in patients who presented with chest pain.

When to use it

You compare a new test with a reference standard in the same patients.
The test result is yes or no, or a continuous value.
You want sensitivity and specificity and, for a continuous test, the ROC curve.

Assumptions to check

Appropriate reference standard

The reference standard should be as close to the truth as practical, and applied to all patients regardless of the test result.

Representative spectrum of patients

Accuracy depends on who is tested. Include patients like those in whom the test will be used.

What to report

The 2x2 table against the reference standard
Sensitivity and specificity with 95% CI
AUC with 95% CI for a continuous test
The threshold used, and whether it was set in advance

Common mistakes

Report the effect, not just the p-value

Give the estimate (difference, ratio) with its 95% confidence interval and the exact p-value.

In your software

# R
library(pROC)
roc_obj <- roc(d$reference, d$test_value)
plot(roc_obj)
ci.auc(roc_obj)

SPSS

ROC test_value BY reference (1)
  /PLOT=CURVE(REFERENCE)
  /PRINT=SE COORDINATES
  /CRITERIA=CUTOFF(INCLUDE) TESTPOS(LARGE) DISTRIBUTION(FREE) CI(95).