Marginal analysis of multiple outcomes with informative cluster size

Area of Science:

• Dental epidemiology
• Biostatistics
• Public health

Background:

• Periodontal disease surveillance requires understanding links to health and socioeconomic factors.
• Assessing periodontal disease involves multiple tooth-level clinical measurements.
• Variations in tooth count can create informative cluster size, biasing conventional statistical models.

Purpose of the Study:

• To develop a novel statistical method for jointly analyzing multiple correlated binary outcomes in clustered data.
• To address the challenge of informative cluster size in periodontal disease research.
• To improve the accuracy of statistical estimates in dental surveillance studies.

Main Methods:

• Proposed a multivariate outcome cluster-weighted generalized estimating equation (GEE) approach.
• Incorporated cluster-specific weights to account for informative cluster size.
• Compared the proposed method with conventional GEE using data from the Veterans Affairs Dental Longitudinal Study.

Main Results:

• The proposed method demonstrated minimal relative biases in estimates.
• The method achieved excellent coverage probabilities in simulation studies.
• Results indicate improved accuracy compared to conventional GEE for data with informative cluster size.

Conclusions:

• The novel cluster-weighted GEE method effectively analyzes periodontal disease data with informative cluster size.
• This approach offers a more reliable tool for dental surveillance and epidemiological research.
• Accurate statistical modeling is crucial for understanding periodontal disease determinants.