Periodontal Disease Classification Affects Associations With Systemic Conditions

Objectives: We tested how periodontal disease classification definitions created using supervised and unsupervised learning systems identify associations with systemic conditions. We compared three periodontal disease classification models: 1) Centers for Disease Control and Prevention & American Academy of Periodontology (CDC/AAP) and the 2) World Workshop Stages and Grades System (WW17), representing supervised learning systems, and the 3) Periodontal Profile Class System (PPC), representing an unsupervised learning system. Supervised learning is a system that uses expert-drawn ‘lines in the sand’ to delineate disease categories. Unsupervised learning is a data-driven method to determine disease categories (reading ‘patterns in the sand’), which produces homogenous groups of people based on a variety of clinical measures.
Methods: Atherosclerosis Risk in Communities study data were used to compare all three models of periodontal disease classification. PPC categories were converted to WW17 Stages and are referred to as PPC-Stages. SAS PROC Logistic was used to create adjusted odds ratios for the following systemic conditions: 1) Intima-media arterial wall thickness (IMT > 1mm), 2) Calcified Arterial Plaque, 3) Diabetes Mellitus, 4) Hypertension, 5) High-Density Lipoprotein cholesterol (HDL < 40), and 6) Obesity (BMI > 30).
Results: Fully adjusted models for IMT, Calcified Arterial Plaque and Diabetes showed no statistically significant odds ratios for periodontal disease classification models based on supervised learning systems. However, there were significant odds ratios for the classification model based on unsupervised learning systems. Fully adjusted models for Hypertension, HDL and Obesity had significant odds ratios for models created with both supervised and unsupervised learning.
Conclusions: PPC-stages appear to be more sensitive in identifying relationships between periodontal disease and systemic conditions. Thus, periodontal definitions created with unsupervised methods allow us to explore in more depth how new disease phenotypes are related to systemic conditions. This understanding may help clinicians make informed treatment decisions.