Calculating the probability of random sampling for continuous variables in submitted or published randomised...

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Data Integrity

Background:

• Previous analysis of randomized controlled trial (RCT) baseline data using a chi-squared method suggested improbable distributions.
• Subsequent simulations indicated the original chi-squared method was flawed.
• This study addresses the need for accurate statistical methods to assess data reliability in RCTs.

Purpose of the Study:

• To correct and evaluate a chi-squared method for analyzing RCT baseline data.
• To compare the performance of the corrected chi-squared method, ANOVA, and Monte Carlo simulations in assessing random sampling probabilities.
• To re-evaluate the baseline data from Fujii et al. RCTs using improved statistical approaches.

Main Methods:

• Correction of a previously used chi-squared statistical method.
• Application of Analysis of Variance (ANOVA) to assess random sampling.
• Utilization of Monte Carlo simulations for probability analysis.
• Comparison of statistical method performance with precisely and imprecisely reported means.

Main Results:

• The corrected chi-squared and ANOVA methods showed inaccuracies with imprecisely reported means.
• Monte Carlo simulations confirmed significant differences between Fujii et al.'s RCT baseline data and those from other authors (p < 10^-16).
• Monte Carlo analysis identified fewer RCTs with unlikely distributions compared to the original chi-squared method, yet still confirmed highly improbable data in Fujii et al.'s trials.

Conclusions:

• The distribution of baseline data in Fujii et al.'s RCTs is extremely unlikely to have arisen from random sampling.
• Monte Carlo simulations appear to be a suitable screening tool for detecting non-random, potentially unreliable data in RCTs submitted for publication.
• Accurate statistical methods are crucial for ensuring the integrity of data in clinical research.