Statistical Methods for Comparing Predictive Values in Medical Diagnosis.

Area of Science:

• Medical diagnostics
• Statistical analysis
• Artificial intelligence in healthcare

Background:

• Evaluating binary diagnostic tests requires sensitivity, specificity, and predictive values.
• Comparing predictive values between tests on the same patients poses statistical challenges due to outcome-dependent denominators.
• Accurate comparison is vital for selecting superior diagnostic methods, especially in fields like breast cancer detection.

Purpose of the Study:

• To review existing statistical methods for comparing predictive values of diagnostic tests.
• To propose and evaluate the permutation test as a suitable method for comparing predictive values.
• To demonstrate the application of these methods using a breast cancer diagnostic dataset.

Main Methods:

• Review of statistical techniques for comparing diagnostic test performance metrics.
• Application of the permutation test, a non-parametric approach, for comparing positive and negative predictive values.
• Utilizing a dataset from breast cancer diagnosis involving MRI and mammography/ultrasound combined modalities.

Main Results:

• Existing methods for comparing predictive values present statistical complexities.
• The permutation test offers an intuitive and non-parametric approach to comparing predictive values.
• The study demonstrates the practical application of the permutation test on real-world diagnostic data.

Conclusions:

• The permutation test is a valuable tool for comparing predictive values in diagnostic accuracy studies.
• This method is particularly useful for small sample sizes and complex comparisons.
• The findings aid in selecting more accurate diagnostic strategies for breast cancer and other conditions.