Marginal analysis of ordinal clustered longitudinal data with informative cluster size

Area of Science:

• Biostatistics
• Dental Research
• Longitudinal Data Analysis

Background:

• Informative cluster size (ICS) is a challenge in longitudinal dental data analysis.
• Existing methods primarily focus on continuous outcomes, not ordinal periodontal disease data.
• Tooth loss can decrease cluster size during studies, complicating analysis.

Purpose of the Study:

• To develop a statistical method for analyzing ordinal clustered longitudinal dental outcomes with informative cluster size (ICS).
• To account for decreasing cluster size due to tooth loss in longitudinal studies.
• To model associations between health covariates (metabolic syndrome, smoking) and periodontal disease progression.

Main Methods:

• Developed longitudinal cluster-weighted generalized estimating equations (CWGEE).
• Employed a proportional odds logistic regression model for ordinal outcomes.
• Estimated within-teeth correlation using two-stage quasi-least squares.
• Validated the method using data from the Department of Veterans Affairs Dental Longitudinal Study and simulations.

Main Results:

• The proposed CWGEE method demonstrated very low bias.
• CWGEE achieved excellent coverage probability in simulations.
• Conventional generalized estimating equations (GEE) without ICS adjustment showed significant bias.

Conclusions:

• Longitudinal CWGEE is a robust method for analyzing ordinal clustered longitudinal dental data with ICS.
• The method effectively handles decreasing cluster sizes due to tooth loss.
• CWGEE provides more reliable results than conventional GEE in the presence of ICS.