Rounding, but not randomization method, non-normality, or correlation, affected baseline P-value distributions in...

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Analysis

Background:

• Assessing P-value distributions for baseline variables in randomized controlled trials (RCTs) is crucial for trial integrity.
• Potential limitations include randomization methods, variable normality, correlations, and P-value calculation from rounded data.

Purpose of the Study:

• To investigate factors limiting the comparison of observed versus expected P-value distributions for baseline continuous variables in RCTs.
• To evaluate the impact of randomization, normality, correlation, and data rounding on P-value distributions and AUC-CDF.

Main Methods:

• Analysis of P-value distributions and AUC-CDF in 13 RCTs and simulations.
• Simulations incorporated eight randomization methods, varying variable normality and correlation strength.
• P-values were calculated from both precise and rounded summary statistics.

Main Results:

• P-value distributions and AUC-CDF were consistent with theoretical uniform distributions across various randomization methods and non-normal variables.
• Stratification and minimization randomization produced fewer P-values <0.10 than expected.
• Most baseline variable correlations were weak (-0.2 to 0.2).
• Increasing correlation strength did not affect P-value distribution uniformity.
• P-value distributions derived from rounded summary statistics deviated from uniformity, though empirical distributions could be generated.

Conclusions:

• Randomization methods, variable non-normality, and correlation strength have minimal impact on baseline P-value distribution or AUC-CDF.
• P-values calculated from rounded summary statistics exhibit non-uniform distributions, posing a potential limitation.