Integrating periodontitis assessment in medical research using computationallyenhanced classification

Integrating periodontitis assessment in medical research using computationally enhanced classification. Abstract Periodontitis is one of the most prevalent non-communicable diseases (NCDs) in adults affecting 64.7-million Americans based on 2009-2012 estimates. Current examination protocols for periodontitis assessment are either inefficient or inaccurate for population-level studies. Full-mouth examination (FME) is considered the gold standard for estimating true periodontitis prevalence, however it is among the most resource- and time-intensive assessment methods in health-research. Despite decades of efforts, oral health researchers have not been able to pragmatize the development of an accurate partial-mouth examination (PME) protocol. Lack of an implementable PME is a major barrier for: 1) identifying community health needs globally, 2) determining public health resource allocation and 3) implementing periodontitis measures in disease association studies; settings where it is impractical or inefficient to utilize FME. Importantly, emerging evidence has implicated periodontal inflammation in the pathogenesis of type 2 diabetes supported by robust pre-clinical causation models and human correlative studies. Nonetheless, definitive data on whether an increased risk for diabetes onset exists in periodontal patients is lacking because current resource and time demanding full-mouth periodontitis examinations hinder periodontitis assessment in adequately powered prospective studies. Therefore, despite the importance of periodontitis-diabetes associations, periodontal measures are often excluded from large medical cohorts due to funding and logistics limitations. The objective in this application is to enable the integration of periodontitis assessment in community and population level surveillance by developing and validating a computationally enhanced PME method for periodontitis assessment with high validity. Conducted by a strong transdisciplinary team with complementary expertise in epidemiology, global health, biostatistics and machine learning, and supported by an extensive FME dataset of over 25,000 participants of the continuous NHANES, the Hispanic Community Health Study (HCHS) and the Oral Infections Glucose Intolerance and Insulin Resistance Study (ORIGINS), this proposal will pursue two specific aims: 1) to computationally enhance the prediction of PME utilizing the novel implementation of machine learning in periodontitis classification, and 2) to assess the performance of the enhanced PME classifier against existing PMEs and “gold standard” FME in investigating the association between periodontitis and glycemic status. The feasibility of the proposed approach is supported by strong preliminary data showing that a Support Vector Machines (SVMs) classifier enhanced the sensitivity of periodontitis prediction from 54% (“naive” counting of diseased sites from a currently used half-reduced definition PME) to 90% (SVM-enhanced disease classification) while maintaining an acceptable false positive rate of 3%. Ultimately, this enhanced PME will be utilized for assembling large populations with periodontitis in a time-cost-effective manner thereby transforming the fields of NCDs epidemiology and global health surveillance.