Estimation of receiver operating characteristic curve when case and control require different transformations for...

Area of Science:

• Biostatistics
• Medical Diagnostics
• Biomarker Analysis

Background:

• Receiver operating characteristic (ROC) curves are vital for assessing diagnostic biomarker accuracy.
• Standard methods often apply a common Box-Cox transformation to biomarker data from cases and controls.
• Biomarker data from diseased and non-diseased groups may require different transformations for normal approximation.

Purpose of the Study:

• To address limitations of existing ROC curve estimation methods when biomarker data require heterogeneous transformations.
• To propose a novel method for estimating ROC curves and their area under the curve (AUC) with different transformations.
• To evaluate the performance of the proposed method against existing parametric and nonparametric approaches.

Main Methods:

• Developed a new statistical method for ROC curve and AUC estimation under heterogeneous Box-Cox transformations.
• Compared the proposed method with nonparametric estimators and estimators using a common Box-Cox transformation.
• Utilized HIV infection data from the National Health and Nutrition Examination Survey (NHANES) for empirical validation.

Main Results:

• Established that existing methods using a common Box-Cox transformation are biased when data require different transformations.
• Demonstrated that the proposed method provides valid and less biased estimates of the ROC curve and AUC.
• The new method shows improved performance compared to common transformation and nonparametric approaches.

Conclusions:

• Heterogeneous transformations in biomarker data necessitate specialized methods for accurate ROC analysis.
• The proposed method effectively handles differing data transformations, improving diagnostic biomarker evaluation.
• This approach enhances the reliability of ROC curve and AUC estimation in biostatistical and clinical research.