Nonparametric statistical inference method for partial areas under receiver operating characteristic curves, with...

Area of Science:

• Genomic research
• Statistical inference
• Biostatistics

Background:

• The ROC50 index is widely used in genomic research for evaluating classifier performance.
• Current statistical inference methods for the ROC50 index are limited, hindering its application.
• There is a need for accessible statistical tools for ROC50 index analysis.

Purpose of the Study:

• To develop a novel statistical method for inference on the ROC50 index.
• To derive a simple and direct variance calculation formula for partial areas under the ROC curve.
• To provide confidence interval and test statistic formulas based on asymptotic normality.

Main Methods:

• Review of statistical methods for partial areas under ROC curves.
• Derivation of a nonparametric variance calculation formula for partial areas.
• Application of trimmed U-statistics theory to establish asymptotic normality.
• Comparison with existing bi-normal approaches.

Main Results:

• A simple and direct variance calculation formula for partial areas under the ROC curve was derived.
• The partial area under the ROC curve was shown to have an asymptotic normal distribution.
• Formulas for confidence intervals and test statistics were provided.
• The method was validated through simulation studies and applied to a large genomic dataset.

Conclusions:

• The developed nonparametric method offers a practical approach for statistical inference on the ROC50 index.
• The findings facilitate the use of the ROC50 index in genomic research by providing robust statistical tools.
• The method's validity is supported by theoretical underpinnings and empirical validation.