Robust nonparametric methods with variable selection for clustered dental data

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Most adults in the US are affected by tooth loss due to periodontal disease or dental caries. Prevention of tooth loss is achieved through professional dental treatment and personal oral health self-care by maintaining the natural dentition in a state of comfort and function. In order to develop appropriate dental treatment planning, dental health care professionals must understand the most effective biological, socio-demographic, behavioral and other medical factors that can affect tooth loss as determined by periodontal disease or dental caries status. Currently, however, there is no consensus concerning the most important factors that may influence dental disease, nor the optimal statistical methods for identifying these factors. There is a need for variable selection methods in robust statistical models for periodontal disease and dental caries outcomes that accommodate the clustered nature of these data (i.e. multiple outcomes from each subject). Goals: The proposed study will develop fixed effects (covariates) and random effects selection techniques for multivariate dental data with robust modeling of the latent random effects induced by clustering and will apply these methods to available databases recording dental health status to advance knowledge about factors associated with tooth loss. Subjects: The statistical methods will be evaluated on a dataset of 300 dentate subjects who were enrolled in the Gullah African-American (AA) Diabetics Study as part of the SC COBRE for Oral Health. For generalizability, the methods will be investigated on national data collected as part of NHANES (1999- 2004). Available data and study design: Periodontal status (determined by pocket depth and clinical attachment level), caries status (determined by tooth level DMFS index), other relevant biological/medical status, smoking, behavioral (brushing and flossing), demographic (poverty status) and other parameters have been collected at the Medical University of South Carolina. The Gullah AA subjects represent an interesting population with minimal genetic admixture whose dental health status remains vastly unknown. NHANES data are publicly available. Significance: The new statistical methods will advance public health by providing dental researchers enhanced knowledge about the nature of the associations between covariates and dental health and, more broadly, by enabling researchers to better target risk assessment and prevention strategies, thereby improving health status. PUBLIC HEALTH RELEVANCE: There is a lack of consensus among dental hygenists to select the most important covariables that might influence tooth loss as determined by caries and periodontal disease. Our proposed robust statistical methods will address this issue with specific applications to explore the dental health status of Gullah-speaking African- Americans, as well as national data collected as part of NHANES (1999-2004). Our methods will have a profound impact on overall public health and the long term goal is to provide dental researchers (as well as other health scientists) a better understanding about repeated and longitudinal dental (health) data so as to prevent and control disease and improve dental health.