| Study Tier | Sample Size (Smallest Group) | Statistical Reliability |
|---|---|---|
| Exploratory Pilot | 15–20 | High risk of over-optimism. ROC curves may look great but won’t replicate in a new cohort. |
| Standard Discovery | 30–40 | AUC values begin to stabilize and the margin of error (Confidence Interval) narrows significantly. |
| Robust Validation | 80–100+ | Necessary for clinical-grade biomarkers. Allows for complex machine learning without overfitting. |