Building on a T90-supported training experience working jointly with scientists in Dentistry and Biostatistics, this project aims to develop new techniques for modeling dental outcome data by drawing on modern data-science techniques for spatially structured data, specifically accounting for both adjacent-tooth and cross-mouth patterns of association. An investigation into alternative possible mechanisms for oral-health consequences of methamphetamine use serves as a motivating example. A recurring narrative in both the clinical literature and media reports has been the frequent presentation of rampant dental disease in habitual methamphetamine (meth) users, a condition described colloquially as ?meth mouth?. The proposed mechanisms for the increased dental disease are the ?contaminant theory?, where corrosive elements are released when meth is smoked inducing acid-mediated erosion of tooth enamel, and a lack of proper oral health behaviors (OHBs) over extended periods of time. Because both theories explain basic features of available data, a more refined data-science framework is needed to help distinguish these alternatives, achieved here by characterizing in greater detail the patterns of association that would be expected under the respective theories. Tests of alternative models can be based on available data from a sample of 571 meth users. Oral-health outcomes were collected following protocols used in the population-based National Health and Nutrition Examination Survey (NHANES), with additional data also collected on meth-use history and other attributes known to be linked to dental disease. Evidence from preliminary investigations of dental caries data points not only to strong spatial patterns of association between adjacent teeth but also to substantial association between corresponding teeth on opposite sides of the mouth. Conditionally autoregressive statistical models for spatial data are in widespread use, but methods for accommodating multiple spatial neighbor relationships have only recently been conceptualized, giving rise to an interdisciplinary research opportunity for statistical-methods development applicable to dental-outcome data. To address missing teeth, the project will also seek to build on techniques in the literature on statistical analysis with incomplete data, in particular with an approach accounting for alternative mechanisms that could explain why teeth are missing. Consistent with the goals of the NIDCR 2014-2019 Strategic Plan, it is anticipated that the proposed methodologies will be applicable to a wide variety of oral-health (and other health-science) research settings.
New data-science models that account for spatial relationships among oral-health outcomes and the generative process giving rise to missing teeth will strengthen the ability of researchers to answer scientific questions about the patterns and underlying mechanisms of dental disease. An example from a study of the oral-health consequences of methamphetamine use illustrates the potential for sophisticated data-analysis strategies to provide insights for early detection of disease and for needed treatment. More generally, it is anticipated that by more efficiently extracting information from available data, refined analysis frameworks will facilitate personalized oral-health interventions and will yield better public-health outcomes.
