Matched-pair noninferiority trials using rate ratio: a comparison of current methods and sample size refinement

Area of Science:

• Biostatistics
• Diagnostic Test Evaluation
• Clinical Trial Design

Background:

• Noninferiority testing is crucial for evaluating new diagnostic tests against established reference standards.
• Assessing the ratio of sensitivity and/or specificity is key for determining if a new test is not unacceptably worse than a reference.

Purpose of the Study:

• To review and compare methods for establishing noninferiority between diagnostic tests based on sensitivity and specificity ratios.
• To evaluate the performance of different statistical tests and estimation methods for noninferiority.

Main Methods:

• Review of one-sided noninferiority tests: logarithmic transformation and Fieller-type tests.
• Comparison of implementation methods: sample-based, constrained least-squares estimation, and constrained maximum likelihood estimation.
• Empirical study to evaluate type I error rates and power of different methods.

Main Results:

• Constrained maximum likelihood estimation (CMLE) methods consistently provide better control of type I error rates.
• Approximate sample size formulae associated with CMLE are asymptotically valid, yielding powers close to the desired level.
• CMLE methods demonstrated superior performance in the empirical evaluation.

Conclusions:

• Constrained maximum likelihood estimation is recommended for establishing noninferiority of diagnostic tests due to its robust error rate control.
• The associated sample size formulae offer reliable planning for studies comparing test performance.
• Methods based on CMLE are practical and effective, as shown in a clinical laboratory example.