Partial AUC measures the area under the ROC curve over a chosen false-positive range, for when only part of the curve is clinically relevant — screening at FPR ≤ 10%, for instance.
The full AUC averages discrimination over every possible threshold; in many clinical settings only one region of the ROC matters. A screening test must keep FPR very low (so it doesn't flood the system with false positives); a confirmatory test must keep FNR very low. Partial AUC restricts the integration to the relevant FPR range and reports the area within it.
Raw partial AUC depends on the chosen range, making cross-test comparison hard. McClish's correction rescales it back to a 0–1 metric for comparable reporting.
Continuous predictor + binary outcome + FPR range [low, high].
Partial AUC raw + McClish-corrected + full AUC + 95% bootstrap CI (B = 200).
Report the McClish-corrected version when comparing across studies that used different FPR ranges; report raw partial AUC when the range is fixed by convention.