Related Experiment Videos
Unconditional exact tests for equivalence or noninferiority for paired binary endpoints
1Division of Biostatistics Research, National Health Research Institutes, Taipei, Taiwan.
Biometrics
|June 21, 2001
Summary
This study introduces an exact test for comparing medical diagnostic procedures, offering a reliable alternative to asymptotic tests for small sample sizes. The new method provides wider confidence intervals, enhancing the assessment of noninferiority and equivalence.
Area of Science:
- Biostatistics
- Medical Diagnostics
- Clinical Trials
Background:
- Comparing diagnostic procedures requires assessing equivalence or noninferiority using correlated proportions.
- Asymptotic tests may lack reliability with small sample sizes, necessitating alternative statistical methods.
Purpose of the Study:
- To propose an unconditional exact test procedure for assessing equivalence or noninferiority between two medical diagnostic procedures.
- To derive unconditional exact confidence intervals for the difference in proportion means.
Main Methods:
- An unconditional exact test procedure is developed using two statistics: a sample-based test statistic and a restricted maximum likelihood estimation (RMLE)-based test statistic.
- The p-value attainment at the null hypothesis boundary and the derivation of unconditional exact confidence intervals are demonstrated.
Main Results:
- The proposed unconditional exact tests provide p-values generally larger than those from asymptotic tests.
- Unconditional exact confidence intervals are typically wider than asymptotic confidence intervals, offering a more conservative assessment.
Conclusions:
- The unconditional exact test procedure is a reliable method for assessing equivalence and noninferiority, especially with small sample sizes.
- Wider confidence intervals from the exact method enhance the robustness of diagnostic procedure comparisons.