Sex influences on periodontal disease and diabetes: A population science approach, with software

Periodontal Disease (PD) continues to remain a major public health burden in the United States. Manifestation and progression of PD are multifactorial, and may vary across sex, with/without additional comorbidities, such as Type-2 Diabetes (T2D), where comorbid subjects are at an elevated risk of compromised oral health. There is an overall paucity of clinically interpretable and nationally representative cross-sectional summaries of numerous risk factors (and their complex interactions) in assessing multi-comorbidity aspects (here, PD and T2D), and precise estimation of associated causal treatments for PD in practice-based settings, factoring in sex influences. Publicly available nationwide survey databases (such as the NHANES), and large oral health databases (such as the HealthPartners®, HP) are important, but somewhat under-utilized resources for such evaluations and practical interpretations, mainly due to several unique statistical and epidemiological complexities, which are often beyond the capabilities of existing standard analytical tools and software packages. Furthermore, how to prioritize patients for oral clinic visits based on their sex determinants, and multi-comorbidity risks continues to remain unresolved. In this project, we address these challenges, and initially propose a stochastically-principled, nationally meaningful, summary risk index (Aim 1) representing cross-sectional PD association from about 11,700 adult dentate subjects, who are part of the NHANES 2009-2014 study, for the 4 target groups: (a) Males with T2D, (b) Males without T2D, (c) Females with T2D, and (d) Females, without T2D. We then refine and validate this derived index, and propose a time-varying PD index (Aim 2) for the four target subgroups, accommodating causality of periodontal treatment effects, via application to the rich, longitudinal, observational HP database of about 25,000 subjects in a practice-based setting, with further model fitting and cross-validation using the Kaiser Permanente Northwest database of about 1,17,000 subjects with similar characteristics. Next, we utilize the time-varying index to construct an optimal policy (Aim 3) for prioritizing highrisk patients for quicker clinic visits. Finally, we produce a free, interactive, web-application tool (Aim 4) via R Shiny, for estimation and computation of the personalized index and recall decisions for any future patient. Our statistically principled, comprehensive, unique index for PD integrating electronic medical records from two large HMOs will be the first of its kind to generate new knowledge in regards to assessing sex influences, in light of T2D. Furthermore, the proposed methodology is readily generalizable to other comorbidities, such as cardiovascular disease, kidney and liver disease, etc. In the longer term, pending rigorous model validation, the derived index has the potential to be integrated into popular chairside software, such as Patterson’s EagleSoft®, thereby facilitating efficient bench to bedside translation.