|
Abstract:
|
Dental caries is a common oral disease that results in demineralization of the tooth. In caries research, the number of decayed, missing, or filled tooth surfaces (DMFS) are analyzed often. However, missing data due to missing teeth or surfaces (e.g., due to sealants) are common in oral health studies. There is no consensus on how to count these surfaces when coming up with a DMFS score. The problem is that the risk of caries varies dramatically across locations. Often, higher-risk teeth and surfaces are more likely to be missing. We propose a model-based approach for imputing these 'missing' values. Our method exploits the spatial correlation between surfaces and allows for the possibility of informative censoring. This approach should increase efficiency, and in some cases reduce bias. We illustrate the method using data from a study of childhood lead exposure.
|
