Nonparametric and semiparametric group sequential methods for comparing accuracy of diagnostic tests

Area of Science:

• Biostatistics
• Medical Diagnostics
• Clinical Trial Design

Background:

• Traditional comparative diagnostic accuracy studies use fixed sample designs.
• Ethical and efficiency concerns necessitate periodic data monitoring in medical studies, especially with expensive or risky procedures.
• Limited research exists on sequential sampling for comparative receiver operating characteristic (ROC) studies.

Purpose of the Study:

• To propose a nonparametric group sequential design for comparing ROC curves.
• To address the need for adaptive sampling in diagnostic accuracy evaluations.
• To improve the ethical and statistical efficiency of comparative diagnostic studies.

Main Methods:

• Developed a nonparametric group sequential design adapting weighted area under the ROC curve statistics.
• Implemented the nonparametric approach for sequential ROC curve comparison in nonsmall-cell lung cancer screening.
• Described a semiparametric sequential method using proportional hazard models.

Main Results:

• The nonparametric approach demonstrated robustness to model misspecification.
• Simulation studies indicated excellent finite-sample performance for the proposed nonparametric method.
• Comparative analysis showed favorable statistical properties against semiparametric and parametric alternatives.

Conclusions:

• The proposed nonparametric group sequential design offers a robust and efficient method for comparing diagnostic test accuracy.
• This approach is suitable for sequential monitoring in medical studies, enhancing ethical considerations and resource utilization.
• The nonparametric method provides reliable performance even with potential model misspecifications in ROC analysis.