#4 Overcoming Limitations of AUC in Evaluating Model Performance

Music City Center 

The Area Under the Receiver Operating Characteristic Curve (AUC) is one of the most widely used metrics for assessing the discriminatory ability of risk prediction models. It possesses several desirable properties: it is a proper scoring rule and remains independent of event rates, making it a valuable tool in model evaluation. However, its limitations are often overlooked. This presentation will focus on key challenges associated with AUC, including pitfalls in using it for external validation and how study design choices can significantly influence its value. We aim to provide a more nuanced perspective on AUC's role in model assessment and guide best practices for its application in clinical research.

AUC

c-index

validation

two-class classification