The goal of the proposed research program is to address a large, unmet need for new statistical methodology that specifically addresses the challenges of correlated data from dental research studies. Dental data exhibits multi-level correlation structures, with correlations between teeth and between observations over time on the same patient. Other distinctive features of dental data are the unique spatial correlation patterns between teeth or tooth surfaces, large clusters, informative cluster sizes, censored failure-time responses, and distinct sources of missing responses that need to be handled differently (patient drop-out versus loss of individual teeth). Statistical methodologic research specific to dental data structures is underdeveloped. Although methodology for correlated data has undergone fruitful development, existing methods do not adequately address all the features of multi-level dental data. The proposed research will develop new methods of statistical analysis for such data, and will make these methods readily available for application in practice. The research will focus on inference for generalized linear regression models with multilevel data, and includes the following specific aims: 1) Study properties of the method of Optimal Combined Estimating Equations (OCEE) in comparison with GEE, and extend it to multi-level data and censored failure-time data; 2) Develop the Combined Estimates (CE) method based on combinations of estimators derived from different sources of information in multi-level data, and study its properties in comparison with OCEE and GEE; 3) Characterize the properties of the Empirical Score Test for clustered data, develop a new test procedure that will be valid with a small number of clusters, and compare the new procedure with existing hypothesis testing methods; 4a) Modify the Inverse Cluster-Size-Weighting method of Williamson et al. (2003) so that it is more generally consistent for within-cluster predictors, extend the new method to multi-level data, and apply the CE approach to develop estimators with improved efficiency; 4b) develop new approaches for accommodating missing data that distinguish between missing responses at different levels, and 5) Disseminate the results to statisticians and dental researchers through the analysis of real dental data sets, publications in statistical and dental journals, presentations at statistical and dental conferences, short courses offered to dental researchers, and distribution of computer programs to other researchers for performing data analysis. Through this research, we expect to make a significant impact on the analysis of multi-level dental data.
| Huang, Ying; Leroux, Brian (2011) Informative cluster sizes for subcluster-level covariates and weighted generalized estimating equations. Biometrics 67:843-51 |
| Stoner, J A; Leroux, B G; Puumala, M (2010) Optimal combination of estimating equations in the analysis of multilevel nested correlated data. Stat Med 29:464-73 |
