Composite likelihood inference for ordinal periodontal data with replicated spatial patterns.

Area of Science:

• Biostatistics
• Spatial Statistics
• Biomedical Data Analysis

Background:

• Analyzing spatial ordinal data across multiple subjects is challenging due to limited statistical methods.
• Existing approaches often fail with replicated spatial data, common in disease mapping.
• Periodontal disease (PD) status data presents a typical scenario for this analytical gap.

Purpose of the Study:

• To propose a novel multisubject spatial ordinal model incorporating geostatistical structures.
• To develop computationally scalable parameter estimation and diagnostic methods.
• To address limitations in analyzing complex spatial ordinal datasets in biomedical research.

Main Methods:

• A latent variable representation is used within a regression framework.
• Maximum composite likelihood estimation is employed for computational scalability.
• Generalized surrogate residuals are developed for model diagnostics.

Main Results:

• The proposed model effectively handles multisubject spatial ordinal data.
• Maximum composite likelihood method demonstrates asymptotic properties for parameter estimates.
• Simulation studies confirm sound finite sample performance of the methods.

Conclusions:

• The developed methodology provides a robust framework for analyzing multisubject spatial ordinal data.
• The approach is applicable to complex biomedical datasets, such as periodontal disease.
• A companion R package (clordr) facilitates practical implementation.