Active-control trials with binary data: a comparison of methods for testing superiority or non-inferiority using the...

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Inference

Background:

• Active-control trials are crucial for comparing new treatments against existing ones.
• Assessing treatment effects with binary data using odds ratios is common in clinical research.
• Establishing superiority or non-inferiority requires robust statistical testing methods.

Purpose of the Study:

• To evaluate and compare three asymptotic statistical tests for superiority or non-inferiority in active-control trials.
• To examine the performance of likelihood ratio, Wald, and score tests for binary data analyzed via odds ratios.
• To assess Type I error rates and statistical power across different non-inferiority margins.

Main Methods:

• The study presents three asymptotic tests: likelihood ratio, Wald, and score tests.
• These tests are based on the unconditional binary likelihood for the log-odds ratio.
• Simulations were conducted to compare the tests' performance regarding Type I error and power.

Main Results:

• All three tests demonstrated Type I error rates close to the nominal level in simulations.
• The score test showed slightly elevated Type I error rates with large non-inferiority margins.
• The score test generally yielded the highest statistical power, though differences among tests were not practically significant.

Conclusions:

• Likelihood ratio, Wald, and score tests are viable and implementable methods for active-control trials with binary data.
• The score test is recommended for its generally higher power, while maintaining acceptable Type I error rates.
• These methods and their confidence intervals are readily available in standard statistical software.