Tests for equivalence or non-inferiority for paired binary data.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Medical Device Evaluation

Background:

• Therapeutic equivalence and non-inferiority assessments are crucial in medical diagnostics.
• Comparing paired binary endpoints often relies on confidence intervals for response rate differences.
• Existing methods may have limitations in accurately controlling statistical errors.

Purpose of the Study:

• To investigate and compare two asymptotic test statistics for assessing equivalence or non-inferiority.
• To evaluate the performance of Wald-type and RMLE-based tests for paired binary endpoints.
• To determine sample size requirements for establishing therapeutic equivalence.

Main Methods:

• Derivation of sample size and power functions for two asymptotic tests.
• Computation of type I error and power using exact probabilities.
• Comparison of a Wald-type (sample-based) test with a restricted maximum likelihood estimation (RMLE)-based test.

Main Results:

• The RMLE-based test demonstrates superior control over type I error compared to the sample-based test.
• A minimal sample size of 120 is required for establishing equivalence with a 0.15 symmetric limit.
• The RMLE-based test without continuity correction performs well at the boundary point.

Conclusions:

• The RMLE-based test is recommended for assessing therapeutic equivalence with paired binary endpoints due to its robust error control.
• Accurate sample size calculation is essential for reliable equivalence studies.
• The findings provide practical guidance for clinical trial design and data analysis.