Statistics Glossary → Partial AUC
Partial AUCconcept
The area under the ROC curve computed over a restricted range of specificity (or sensitivity) only, for example where specificity exceeds 0.90.
The full AUC sums up a test's discriminating ability over all possible thresholds in a single number. In practice, though, some threshold regions are never used: if a screening test must not produce many false positives, only the high-specificity region matters. The partial AUC (pAUC) computes the area under the ROC curve only within this relevant range. Two tests with similar full AUCs can differ clearly in the region of interest if their curves cross.
The raw partial AUC depends on the width of the range and is hard to read intuitively; if the range is specificity 0.90-1, its largest possible value is 0.10. The McClish correction rescales it so that a random test scores 0.5 and a perfect test 1. The limits of the range should be chosen in advance on clinical or practical grounds, not after looking at the data; with narrow ranges the estimate is more uncertain.
An everyday example
Two cancer screening tests have almost identical full AUCs of 0.85 and 0.84. But to avoid sending healthy people for unnecessary biopsies, only thresholds with specificity above 0.95 can be used. In this region the first test's partial AUC (McClish-corrected) is 0.78 and the second's 0.62. At the thresholds that matter in practice, the first test is clearly better.
In MerQur
In the ROC Curve analysis, the 'Partial AUC (specificity >=)' option gives the area under the ROC curve for specificity in [lower limit, 1] with the McClish correction; 0 switches it off.