Avoiding Blunders When Analyzing Correlated Data, Clustered Data, or Repeated Measures

Area of Science:

• Rheumatology
• Biostatistics
• Epidemiology

Background:

• Rheumatology research frequently involves correlated and clustered patient data.
• Improper analysis of such data, by treating observations as independent, can lead to flawed statistical inference.
• This study addresses common analytical errors in rheumatoid arthritis (RA) research.

Purpose of the Study:

• To compare statistical inference from generalized linear models (GLM) versus mixed-effects models when analyzing correlated rheumatology data.
• To highlight the impact of ignoring correlation on effect size estimation and statistical significance.
• To emphasize the importance of modeling correlation in rheumatology research.

Main Methods:

• Utilized a subset of data from 633 rheumatoid arthritis patients (Raheel et al., 2017).
• Analyzed binary RA flare and continuous swollen joint counts as outcomes.
• Compared GLM with generalized linear mixed models and generalized estimating equations to account for data correlation.

Main Results:

• Regression coefficients (β) were similar across methods.
• Standard errors increased significantly when correlation was incorporated into the models.
• Ignoring correlation led to underestimated standard errors, narrower confidence intervals, and inflated Type I error rates.

Conclusions:

• Failure to account for correlation in rheumatology data analysis can result in overestimated effect sizes and misleading P values.
• Accurate statistical inference requires appropriate modeling of correlated and clustered observations.
• Properly accounting for correlation is crucial for reliable findings in rheumatoid arthritis studies.