Clinical research & EBM · Reference
What is the ROC curve and AUC?
A receiver operating characteristic (ROC) curve plots a test’s sensitivity against one minus its specificity across thresholds, and the area under the curve (AUC) summarises how well the test discriminates between two states. AUC ranges from 0.5 (chance) to 1 (perfect).
Sensitivity, specificity and thresholds
A diagnostic test or predictive model usually produces a continuous score, and a threshold decides which scores are classed as positive. Sensitivity is the proportion of true positives correctly identified; specificity is the proportion of true negatives correctly identified. Raising or lowering the threshold trades one against the other: a more permissive threshold catches more positives but produces more false alarms. The ROC curve visualises this trade-off by plotting sensitivity against 1 minus specificity at every possible threshold, so the whole spectrum of operating points is shown at once.
Reading the curve and the AUC
On an ROC plot, the diagonal line from corner to corner represents a test that performs no better than chance. A curve that bows towards the top-left corner indicates better discrimination, because it achieves high sensitivity while keeping the false-positive rate low. The area under the curve condenses the whole curve into one number: 0.5 is chance, 1.0 is perfect discrimination, and intermediate values are sometimes loosely described in bands (for example "acceptable" or "excellent"). The AUC has an intuitive interpretation as the probability that the test ranks a randomly chosen positive case above a randomly chosen negative case.
Uses and cautions
ROC analysis is widely used to evaluate diagnostic tests and to compare predictive models in research, including in machine learning, where it measures a classifier’s ability to separate classes independently of any single threshold. It has limits, however. AUC summarises discrimination but says nothing about calibration — whether predicted probabilities match observed frequencies — and it can be misleading when the two classes are very unbalanced, where a precision–recall curve may be more informative. As with other measures, an AUC should be reported with a confidence interval and interpreted in the context of how the test will be used.
Key facts
At a glance
- ROC plots: Sensitivity versus 1 − specificity
- AUC range: 0.5 (chance) to 1.0 (perfect)
- AUC meaning: Probability a positive outranks a negative
- Better curve: Bows towards the top-left corner
- Measures: Discrimination, not calibration
- Caution: Can mislead with very imbalanced classes
Common questions
FAQ
What does the area under the ROC curve measure?+
The AUC measures how well a test or model discriminates between two states across all thresholds. It ranges from 0.5, meaning no better than chance, to 1.0, meaning perfect separation, and equals the probability that the test ranks a random positive case above a random negative one.
What do sensitivity and specificity mean?+
Sensitivity is the proportion of true positives a test correctly identifies, while specificity is the proportion of true negatives it correctly identifies. The ROC curve shows how the two trade off against each other as the classification threshold changes.
What are the limitations of AUC?+
AUC summarises discrimination but not calibration — whether predicted probabilities match observed rates — and it can be misleading when classes are very imbalanced. In such cases a precision–recall curve may be more informative, and AUC should be reported with a confidence interval.
The step most authors miss
Doing CRediT right? Don’t stop at the statement.
A CRediT statement credits you inside one paper. The recognition CRediT was built for happens when those roles are tied to you, persistently. Sign in with your ORCID — free — and claim your CRediT contributions on casrai.org, the home of the standard. They become a verified, portable part of your identity, not a line that disappears into one PDF.
Free: claim your contributions, then export a journal-ready CRediT statement, schema.org structured data, JATS XML, CSV or BibTeX — and preview your public profile. A membership publishes that profile publicly and verifies the journals you serve.
