Kappa statistic for clustered matched-pair data

Area of Science:

• Biostatistics
• Statistical Methods

Background:

• The kappa statistic assesses agreement in matched-pair data.
• Existing methods struggle with clustered data and within-cluster correlation.

Purpose of the Study:

• Propose a nonparametric variance estimator for the kappa statistic in clustered matched-pair data.
• Evaluate its performance and compare it to existing methods.

Main Methods:

• Utilized the delta method and sampling techniques.
• Developed a nonparametric variance estimator, avoiding distributional assumptions.
• Conducted extensive Monte Carlo simulations.

Main Results:

• The proposed estimator provides consistent estimation for the kappa statistic.
• It performs well with a moderate to large number of clusters (K ≥50).
• Outperforms estimators ignoring intra-cluster correlation, especially when correlation (ρ) is ≥0.3.

Conclusions:

• The proposed variance estimator is reliable for clustered matched-pair data.
• It offers improved accuracy in coverage probability with increasing intra-cluster correlation.
• Demonstrated practical utility through real-world data analysis.