Spatial skew-normal/independent models for nonrandomly missing clustered data.

Area of Science:

• Biostatistics
• Dental Research
• Statistical Modeling

Background:

• Periodontal disease (PD) clinical data, including clinical attachment level (CAL) and tooth presence/absence, are often clustered within subjects.
• Traditional linear mixed models assume normality, which may not hold for PD data exhibiting skewness and heavy tails.
• Spatial correlations between tooth-sites and the informative nature of missing teeth are often overlooked in standard analyses.

Purpose of the Study:

• To develop a unified Bayesian statistical model for analyzing clustered periodontal disease data.
• To incorporate spatial dependencies and non-normal distributions (skewness, heavy tails) in random effects for PD.
• To model both clinical attachment level and tooth presence/absence simultaneously.

Main Methods:

• Development of a shared random effects model within a Bayesian framework.
• Utilizing a spatial skew-normal/independent (S-SNI) distribution for random effects, featuring a conditionally autoregressive (CAR) dependence structure.
• Validation through simulation studies and application to a real-world periodontal disease clinical dataset.

Main Results:

• The proposed S-SNI model effectively captures spatially referenced asymmetric and thick-tailed data structures.
• The unified model provides a significantly improved fit compared to models that do not account for these data features.
• Demonstrated advantages in analyzing complex periodontal disease progression patterns.

Conclusions:

• The Bayesian S-SNI model offers a more robust and accurate approach for analyzing periodontal disease data.
• Accounting for spatial effects and non-normality is crucial for reliable statistical inference in PD studies.
• The methodology provides a valuable tool for understanding periodontal disease progression and its associated factors.