Prognostic factors identification using machine learning in temporomandibular disorder treatment responders
Chollada Chamnanmanoontham1, Thanaphon Tangchoopong2, Jarin Paphangkorakit1
1Department of Oral Biomedical Sciences, Faculty of Dentistry, Khon Kaen University, 40002 Khon Kaen, Thailand.
Background:
Temporomandibular disorders (TMD) are complex chronic conditions that significantly impair quality of life and impose a considerable social and economic burden. Although various treatment strategies have been developed, the prognostic factors influencing therapeutic outcomes remain poorly defined. This study aimed to identify relevant prognostic factors for treatment response using machine learning methods, with patient readiness for discharge serving as the primary outcome.
Methods:
A total of 1050 medical records from patients diagnosed with TMD and treated at the Orofacial Pain and Dental Sleep Medicine clinic, between January 2018 and June 2023, were retrospectively analyzed with a follow-up period of one year. Twenty-six clinical and demographic variables were initially extracted and preprocessed using one-hot encoding. After the removal of highly correlated variables, the final dataset comprised 36 features derived from encoded categorical variables. Seven machine learning algorithms, namely Decision Tree, Random Forest, Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), Gradient Boosting, and Extreme Gradient Boosting (XGBoost), were trained to predict discharge readiness. Model performance was evaluated using accuracy, precision, recall, F1 score, and the receiver operating characteristic-area under the curve (ROC-AUC). Feature importance was assessed using information gain and Shapley Additive exPlanations (SHAP) to interpret model predictions.
Results:
The Random Forest model demonstrated superior performance in predicting readiness for discharge, with an accuracy of 0.7901, precision of 0.8611, recall of 0.8943, F1 score of 0.7901, and ROC-AUC of 0.6442. SHAP analysis identified onset duration, pain severity, anxiety level, presence of neck pain, and body mass index (BMI) as the most influential predictors.
Conclusions:
The identified prognostic factors highlight the multidimensional nature of TMD and support their relevance in guiding patient-specific management strategies. The integration of machine learning approaches may enhance clinical decision-making and contribute to the development of more personalized and effective treatment pathways for TMD.
