Detection of Outlying Correlation Coefficients in Multicenter Clinical Trials.

Area of Science:

• Biostatistics
• Clinical Trial Management
• Data Quality Assurance

Background:

• Central statistical monitoring is crucial for identifying data discrepancies across centers in multicentric clinical trials.
• Significant data distribution differences may indicate potential issues like negligence, misconduct, or fraud.
• Existing methods for comparing data distributions vary based on data type and analysis scope (univariate/multivariate).

Purpose of the Study:

• To introduce and evaluate two novel statistical methods for detecting centers with outlying bivariate Pearson correlation coefficients.
• To assess the performance of these methods in identifying centers with unusual correlation patterns.
• To compare the effectiveness of the proposed methods in central statistical monitoring.

Main Methods:

• Development of two distinct methods for detecting outlying bivariate Pearson correlation coefficients across trial centers.
• Method 1: Direct comparison of correlation coefficients between centers.
• Method 2: Conditioning the correlation test on marginal standard deviations for independence from center-specific variability.

Main Results:

• Both proposed methods demonstrated equal performance on simulated data for identifying outlying correlations.
• Application to real-world clinical trial data successfully identified centers with outlying bivariate correlations.
• The two methods showed agreement for centers with average standard deviations but diverged for centers with extreme standard deviations.

Conclusions:

• The presented methods are effective tools for central statistical monitoring in multicentric clinical trials.
• These techniques enhance the ability to detect centers with anomalous bivariate correlations, aiding data quality assessment.
• The methods are versatile and applicable beyond central statistical monitoring to other statistical analysis settings.