Statistical Validation of Unsupervised Clustering for Adolescent TMD: A Cross-Sectional Study.

Abstract

Objective: This study employs unsupervised clustering to identify Temporomandibular disorders (TMD) phenotypes in adolescents, aiming to identify distinct clusters based on biopsychosocial features. It compares these clusters with conventional TMD classifications to assess if this method offers enhanced insights into TMD diagnosis.

Methods: Data from 662 adolescent patients with TMD were analyzed using unsupervised clustering and classified into four groups based on DC/TMD Axis I: 1 (disc displacement), 2 (joint pain), 3 (muscle pain), and 4 (combined joint and muscle pain). Patient-reported outcomes were measured with instruments including the Brief Pain Inventory, the Pain Catastrophizing Scale, the Symptom Checklist-90-Revised, and the Pittsburgh Sleep Quality Index. Statistical analyses validated the clusters against conventional classifications.

Results: Three distinct clusters were identified: High Impact (n = 70), Mild Symptoms (n = 423), and High Catastrophizing (n = 169), each displaying unique patterns in pain severity, pain catastrophizing, psychological distress, and sleep disturbances. Multinomial logistic regression of conventional TMD classifications revealed that only pain severity significantly differentiated the subcategories among these biopsychosocial factors.

Conclusions: The findings underscore the variability in TMD presentations among adolescents and suggest that integrating phenotyping into the conventional diagnostic approach could significantly improve diagnostic accuracy and treatment outcomes, facilitating better management of high-risk adolescent patients.