Unsupervised phenotyping of the periodontal architecture through high-dimensional clustering of electronic health records: A multicenter study.

backgroundTo identify novel periodontal phenotypes using unsupervised machine learning on a large-scale, multicenter cohort, specifically characterizing disease patterns based on the "periodontal architecture" of localized structural failures (tooth mobility and molar furcation defects) rather than global severity averages alone.

methodsThis cross-sectional study analyzed electronic health records from 15,723 adult patients with periodontitis. A high-dimensional feature vector (D = 72) was constructed for each patient, integrating tooth-specific ordinal grades for mobility and furcation, mean probing depths (PPD), clinical attachment loss (CAL), and systemic health variables. Unsupervised phenotyping was performed using principal component analysis (PCA), t-SNE visualization, and K-means clustering. Cluster validity was assessed via Silhouette Analysis, and phenotypes were compared using ANOVA and Chi-square (X

resultsFour distinct architectural phenotypes were identified: (1) Maintenance/Healthy, representing stability; (2) Anterior-Mobility dominant, defined by high-grade anterior mobility (52.4% prevalence), and the highest diabetes prevalence (12.2%); (3) Molar-Furcation Dominant, a male-dominated group (61%) characterized by advanced posterior furcation defects (93.8% prevalence) despite lower anterior mobility; and (4) Generalized severe, exhibiting global architectural collapse. The phenotypes demonstrated statistically significant separation across all clinical metrics (p < 0.001).

conclusionPeriodontitis manifests as distinct architectural archetypes-specifically "Anterior-Mobility" and "Molar-Furcation" phenotypes-that are often aggregated into a single severity category by traditional staging. These data-driven clusters have unique systemic risk profiles, suggesting that diagnosis and treatment planning should incorporate the specific localization of structural failure. PLAIN LANGUAGE SUMMARY: Severe gum disease (periodontitis) is traditionally classified by overall severity, often grouping different types of tooth damage into the same broad category. To uncover hidden patterns, we used an artificial intelligence technique to analyze the detailed dental and medical records of more than 15,000 patients. Instead of simply grouping patients by how advanced their disease was, the computer identified four distinct patient profiles based on specific patterns of tooth damage and overall health. Interestingly, two of the most severe profiles were fundamentally different. One featured loose front teeth and was strongly linked to diabetes and systemic health issues. The other featured damage between the roots of back teeth (molars) and was primarily driven by local anatomical problems rather than general health. These findings demonstrate that severe gum disease is not a single condition, but rather develops through completely distinct biological pathways. Recognizing these unique patterns allows dental professionals to move beyond a "one-size-fits-all" approach, paving the way for personalized treatments-such as medical screening for patients with loose front teeth and targeted surgical repairs for molar damage.