Start Here
Not sure which test to use? The Diagnostic Testing Wizard asks a few questions and gives you a direct recommendation.
Open the Diagnostic Testing Wizard →Diagnostic Testing Hub
Evaluating how well a test or biomarker distinguishes patients with a condition from those without it.
Related Calculators
ROC Curve & AUCEvaluates a continuous diagnostic test's ability to discriminate between patients with and without a condition across all possible thresholds.Diagnostic AccuracyComputes sensitivity, specificity, PPV, and NPV from a 2x2 diagnostic classification table (disease present/absent vs. test positive/negative).Youden IndexIdentifies the optimal ROC cutoff that maximizes Sensitivity + Specificity − 1.DeLong ROC ComparisonStatistically compares two AUCs measured in the same patients (e.g. a new biomarker vs. a standard test).Decision Curve AnalysisMeasures whether using a prediction model to make clinical decisions produces net benefit compared to treating everyone or no one.
Related Articles
Medical Research Examples
ROC Curve & AUC: Serum procalcitonin evaluated for bacterial vs. non-infectious SIRS (n=50 each): AUC=0.87 (good); optimal cutoff by Youden J=2.1 ng/mL gives Sensitivity=82%, Specificity=80%.
Diagnostic Accuracy: A rapid antigen test evaluated against PCR in 200 patients (45 TP, 5 FN, 10 FP, 140 TN): Sensitivity=90%, Specificity=93%, PPV=82%, NPV=97%.
Common Mistakes
- Reporting only AUC without sensitivity, specificity, and the chosen cutoff value.
- Choosing a cutoff based on the training data without any adjustment for optimism or overfitting.
- Interpreting PPV/NPV from a study sample as if they apply directly to a population with a very different disease prevalence.
- Failing to report the 95% CI for AUC, sensitivity, or specificity.
- Treating a small pilot AUC estimate as precise, without acknowledging the wide confidence interval.
- Not following STARD reporting guidelines for diagnostic accuracy studies.
Reporting Recommendations
Journals expect more than a p-value. For each test, report the effect estimate and its 95% confidence interval alongside the p-value:
ROC Curve & AUCAUC, 95% CI; sensitivity and specificity at the optimal cutoff (Youden Index); PPV, NPV
Diagnostic AccuracySensitivity, specificity, PPV, NPV -- each with a 95% CI -- plus overall accuracy
Youden IndexYouden J statistic, optimal cutoff value, sensitivity and specificity at that cutoff
DeLong ROC ComparisonΔAUC, z statistic, p-value
Decision Curve AnalysisNet benefit at each clinically relevant probability threshold, compared against treat-all and treat-none reference lines
Recommended Learning Order
- Confidence Interval fundamentals — understand how precision is quantified for any diagnostic estimate
- Diagnostic Accuracy (Sensitivity/Specificity/PPV/NPV) — start with a test that already has a fixed positive/negative cutoff
- ROC Curve & AUC — move to a continuous biomarker and evaluate every possible cutoff
- Youden Index — pick a single optimal cutoff from the ROC curve
- DeLong ROC Comparison — statistically compare two tests measured in the same patients
- Decision Curve Analysis — most advanced: assess real clinical utility, not just discrimination
Frequently Compared Tests
ROC/AUC vs. Diagnostic Accuracy (Sensitivity/Specificity)
Use ROC/AUC when your biomarker is continuous and you need to find the best cutoff; use the Diagnostic Accuracy calculator directly once a test is already dichotomized into positive/negative.
Frequently Asked Questions
When do I use ROC/AUC vs. simple sensitivity and specificity?
Use ROC/AUC when your test result is continuous and you need to find the best cutoff or compare overall discrimination. Use the Diagnostic Accuracy calculator once your test is already dichotomized into positive/negative.
What AUC value counts as a 'good' test?
0.5 = no better than chance, 0.7-0.8 = acceptable, 0.8-0.9 = excellent, greater than 0.9 = outstanding.
Do sensitivity and specificity change with disease prevalence?
No -- sensitivity and specificity are intrinsic properties of the test. PPV and NPV, however, do change with prevalence.
How do I choose the optimal cutoff on an ROC curve?
The Youden Index (Sensitivity + Specificity − 1) identifies the threshold that maximizes combined correct classification, though clinical context (cost of false negatives vs. false positives) may justify a different cutoff.
What is STARD and why does it matter?
STARD (Standards for Reporting of Diagnostic Accuracy Studies) is the reporting guideline most medical journals require for diagnostic accuracy studies.
Can I compare two diagnostic tests directly?
Yes -- if both were measured in the same patients, use the DeLong test to statistically compare their AUCs.
What does Decision Curve Analysis add beyond AUC?
It measures whether using the model to guide treatment decisions provides real clinical benefit (net benefit) across a range of probability thresholds -- a model can have good discrimination (AUC) but poor real-world decision-making value.