Parametric and nonparametric methods for confidence intervals and sample size planning for win probability in...

Area of Science:

• Medical Statistics
• Clinical Trials Methodology

Background:

• Likert scale data are widely used in medical research, particularly in randomized trials.
• Statistical analysis often involves raw scores or rank information.
• Quantifying treatment effects can be done by comparing scores between treatment and control groups.

Purpose of the Study:

• To derive confidence intervals and sample size formulas for win probability in parallel-group randomized trials using Likert data.
• To compare the performance of parametric and nonparametric statistical methods for analyzing Likert scale data.
• To provide guidance on the preferable analytical approach for Likert data in clinical trials.

Main Methods:

• Development of asymptotic parametric and nonparametric confidence intervals for win probability.
• Inclusion of analyses for studies with only follow-up scores and those with baseline and follow-up measurements.
• Performance assessment through simulation studies using real-world Likert item and scale data.

Main Results:

• Parametric methods showed poor performance regarding bias, interval coverage, and precision, even without baseline adjustment.
• Nonparametric approaches demonstrated excellent performance for both unadjusted and adjusted win probabilities.
• Simulations confirmed the superiority of nonparametric methods in various performance metrics.

Conclusions:

• Nonparametric methods are recommended as the preferable approach for two-group randomization trials with Likert data.
• The study provides practical insights and SAS code for implementing nonparametric analyses.
• These findings enhance the reliability of statistical inference in medical research utilizing Likert scale measurements.