Exact Probability Distribution for the ROC Area under Curve.

Area of Science:

• Biostatistics
• Medical Device Development
• Diagnostic Accuracy

Background:

• The Receiver Operating Characteristic (ROC) curve and its Area Under the Curve (AUC) are critical for evaluating in vitro diagnostic (IVD) medical device accuracy.
• Accurate statistical inference for IVD devices relies on the precise probability distribution of the ROC AUC-value.
• Traditional asymptotic approximations for ROC AUC calculations can suffer from imprecision, particularly when correcting for multiple hypothesis testing, impacting IVD device development.

Purpose of the Study:

• To determine the exact probability distribution of the ROC AUC-value for accurate statistical inference.
• To develop a computationally efficient method for obtaining exact critical values and p-values.
• To demonstrate the utility of exact p-values in improving IVD device development.

Main Methods:

• Derivation of the exact probability distribution for the ROC AUC-value.
• Development and application of a geometric interpolation method for faster computation of exact values.
• Illustration of the method using open-access data to compare biomarker performance against a predicate device.

Main Results:

• The exact probability distribution of the ROC AUC-value was determined, allowing for exact critical values and p-values.
• Geometric interpolation significantly increased computational speeds, providing an approximation that is exact in a special case.
• The method demonstrated the superiority of 26 composite biomarkers over a predicate device using real-world data.

Conclusions:

• Obtaining exact p-values is crucial for accurate statistical inference in IVD device evaluation.
• The proposed geometric interpolation method offers a computationally efficient approach to achieve exactness.
• This advancement facilitates more efficient and reliable IVD device development, especially when dealing with multiple hypothesis testing.