Related Experiment Videos
Modeling multivariate binary responses with multiple levels of nesting based on alternating logistic regressions: an
1Department of Obstetrics, Gynecology and Reproductive Sciences, University of Medicine and Dentistry of New Jersey, New Brunswick, NJ 08901, USA. cande.ananth@umdnj.edu
Journal of Dental Research
|September 24, 2004
Summary
Alternating logistic regressions (ALR) effectively model clustered dental data, like caries aggregation. This method efficiently analyzes complex dependencies in nested data, offering a valuable alternative to generalized estimating equations.
Area of Science:
- Dental Research
- Biostatistics
- Epidemiology
Background:
- Clustered binary responses are frequent in dental research, requiring analysis of both risk and aggregation.
- Second-order generalized estimating equations (GEE2) is a standard but computationally intensive method for such data.
- Alternating logistic regressions (ALR) offers a computationally efficient alternative, particularly for large clusters.
Purpose of the Study:
- To illustrate the application of ALR for analyzing clustered binary responses in dental research.
- To model caries aggregation within a dataset exhibiting three levels of nesting: surfaces, regions, and jaws within subjects.
- To evaluate the dependence structure and risk of caries aggregation using ALR.
Main Methods:
- Applied alternating logistic regressions (ALR) to a dental dataset with nested structures (tooth surfaces, interproximal regions, jaws, subjects).
- Modeled both marginal response probabilities (risk) and the dependence structure (aggregation) between clustered responses.
- Calculated odds ratios (OR) to quantify within-region and between-region aggregation of caries.
Main Results:
- Caries lesions demonstrated strong aggregation within subjects, indicating a spatially distributed risk.
- The minimum within-interproximal (IP)-region odds ratio was 2.25 (95% CI: 1.15, 4.41).
- Within-IP-region odds ratios consistently exceeded between-IP-region odds ratios, confirming localized aggregation.
Conclusions:
- Alternating logistic regressions (ALR) provide a convenient and effective method for explicitly modeling dependence structures in clustered dental data.
- ALR is a computationally efficient alternative to GEE2 for analyzing complex nested data, especially with large clusters.
- The findings suggest ALR's applicability to various dental research problems involving clustered or nested binary outcomes.