A predictive model for arthrogenous temporomandibular disorders based on clinical signs and symptoms.

This study aimed to develop and internally evaluate a multivariable statistical model to identify arthrogenous temporomandibular disorders (TMD) using routinely collected clinical data. The model's performance was compared with the Fonseca Anamnestic Index (FAI) alone, using an imaging-based classification as the reference standard. This cross-sectional observational study included 1170 consecutive patients attending their first consultation at a tertiary TMD center between August 2019 and August 2024. Arthrogenous TMD was determined using combined clinical and imaging assessment according to the Dimitroulis classification. Clinical variables, including age, maximum mouth opening (MMO), individual FAI items, and joint-related complaints, were extracted from the EUROTMJ database. Generalized additive models (GAMs) were used to develop predictive models. Performance was assessed using receiver operating characteristic (ROC) curves, area under the curve (AUC), sensitivity, and specificity in training (60%) and test (40%) datasets. The final Fonseca-Dimitroulis (FD-Class) model incorporated age, MMO, selected FAI items (Q2, Q6, Q7), crepitus, and temporomandibular joint (TMJ) locking. The model achieved an AUC of 0.761 in the training dataset and 0.742 in the test dataset, outperforming the FAI alone (AUC = 0.662). This model may support the early identification of arthrogenous TMJ disease and improve decision-making regarding referral for advanced imaging in maxillofacial practice.