Machine Learning Prediction of Anterior Open‐Bite Development After Stabilization Splint Treatment in Temporomandibular Disorder
- 1Department of Oral and Maxillofacial Radiology, Dental and Life Science Institute, School of Dentistry Pusan National University Yangsan Republic of Korea
- 2Department of Oral Medicine Pusan National University Hospital Busan Republic of Korea
- 3Dental Research Institute Pusan National University Dental Hospital Yangsan Republic of Korea
Abstract
Objectives To develop interpretable machine learning (ML) models for risk stratification of clinically meaningful anterior overbite decrease after stabilization splint therapy in temporomandibular disorder (TMD) patients, and to compare ML performance with conventional logistic regression. Materials and Methods This retrospective cohort included 87 TMD patients treated with stabilization splints, physical therapy, and pharmacologic therapy. Thirty‐five pre‐treatment cephalometric and demographic features were used. The primary outcome was overbite difference (OD = post‐treatment minus pre‐treatment overbite). For classification, patients were labeled as Decrease (OD < −0.5 mm; n = 37) versus Non‐Decrease (OD ≥ −0.5 mm; n = 50), combining No Change and Increase due to severe imbalance in the Increase subgroup. Six classifiers (logistic regression, random forest, extreme gradient boosting (XGBoost), Light Gradient Boosting Machine, support vector machine, k‐nearest neighbors) were evaluated using stratified fivefold cross‐validation and leave‐one‐out cross‐validation. Class imbalance was handled using class‐weighting and the Synthetic Minority Over‐sampling Technique (SMOTE) applied within the cross‐validation loop. For regression, nine algorithms were tested to predict continuous OD. Model explainability was assessed using SHapley Additive exPlanations (SHAP). Results In fivefold cross‐validation, XGBoost achieved the best discrimination (AUC = 0.734), outperforming logistic regression (AUC = 0.611). With SMOTE, the random forest achieved a comparable AUC (0.735) with improved sensitivity. Across regression models, cross‐validated R2 values were near‐zero or negative, indicating limited ability to predict the exact magnitude of overbite change from baseline variables alone. SHAP ranked pre‐treatment overbite, ramus height, upper occlusal plane to SN angle, sella‐to‐condylion distance, and condylar positional indices among the most influential predictors for classification. Conclusion Interpretable ML enables improved risk stratification for clinically meaningful anterior overbite decrease after stabilization splint therapy in TMD patients, whereas predicting the exact magnitude of overbite change remains limited using pre‐treatment cephalometrics alone. These models may support individualized counseling and closer monitoring for patients at higher risk of open‐bite development during splint therapy.
Related
- Short-Term Clinical Outcomes of Ready-Made Soft Splints Versus Digitally Fabricated, Customized Hard Stabilization Splints in Temporomandibular Disorders: An Exploratory Randomized Study
- Photobiomodulation versus stabilization splint, both combined with self-management, for acute temporomandibular myalgia: 48-week outcomes of a partially randomized controlled study
- Effects of stabilization splint combined with manual therapy on condylar morphology and skeletal parameters in patients with temporomandibular disorders
- Development and internal validation of a machine learning prediction model for postoperative hemorrhage following modified radical mastectomy for breast cancer
- Machine learning prediction model for distant metastasis at diagnosis in pediatric neuroblastoma: development, interpretation, and temporal validation
BibTeX
@article{hwang2026machine,
title = {{Machine Learning Prediction of Anterior Open‐Bite Development After Stabilization Splint Treatment in Temporomandibular Disorder}},
author = {Jae Joon Hwang and Hye‐Mi Jeon and Hye‐Min Ju and Sung‐Hee Jeong and Yong‐Woo Ahn and Soo‐Min Ok},
journal = {Clinical and Experimental Dental Research},
year = {2026},
month = oct,
doi = {10.1002/cre2.70447},
url = {https://doi.org/10.1002/cre2.70447},
note = {CC BY},
}
RIS
TY - JOUR
TI - Machine Learning Prediction of Anterior Open‐Bite Development After Stabilization Splint Treatment in Temporomandibular Disorder
AU - Jae Joon Hwang
AU - Hye‐Mi Jeon
AU - Hye‐Min Ju
AU - Sung‐Hee Jeong
AU - Yong‐Woo Ahn
AU - Soo‐Min Ok
PY - 2026
DA - 2026/10/30
JO - Clinical and Experimental Dental Research
DO - 10.1002/cre2.70447
UR - https://doi.org/10.1002/cre2.70447
AB - Objectives To develop interpretable machine learning (ML) models for risk stratification of clinically meaningful anterior overbite decrease after stabilization splint therapy in temporomandibular disorder (TMD) patients, and to compare ML performance with conventional logistic regression. Materials and Methods This retrospective cohort included 87 TMD patients treated with stabilization splints, physical therapy, and pharmacologic therapy. Thirty‐five pre‐treatment cephalometric and demographic features were used. The primary outcome was overbite difference (OD = post‐treatment minus pre‐treatment overbite). For classification, patients were labeled as Decrease (OD < −0.5 mm; n = 37) versus Non‐Decrease (OD ≥ −0.5 mm; n = 50), combining No Change and Increase due to severe imbalance in the Increase subgroup. Six classifiers (logistic regression, random forest, extreme gradient boosting (XGBoost), Light Gradient Boosting Machine, support vector machine, k‐nearest neighbors) were evaluated using stratified fivefold cross‐validation and leave‐one‐out cross‐validation. Class imbalance was handled using class‐weighting and the Synthetic Minority Over‐sampling Technique (SMOTE) applied within the cross‐validation loop. For regression, nine algorithms were tested to predict continuous OD. Model explainability was assessed using SHapley Additive exPlanations (SHAP). Results In fivefold cross‐validation, XGBoost achieved the best discrimination (AUC = 0.734), outperforming logistic regression (AUC = 0.611). With SMOTE, the random forest achieved a comparable AUC (0.735) with improved sensitivity. Across regression models, cross‐validated R2 values were near‐zero or negative, indicating limited ability to predict the exact magnitude of overbite change from baseline variables alone. SHAP ranked pre‐treatment overbite, ramus height, upper occlusal plane to SN angle, sella‐to‐condylion distance, and condylar positional indices among the most influential predictors for classification. Conclusion Interpretable ML enables improved risk stratification for clinically meaningful anterior overbite decrease after stabilization splint therapy in TMD patients, whereas predicting the exact magnitude of overbite change remains limited using pre‐treatment cephalometrics alone. These models may support individualized counseling and closer monitoring for patients at higher risk of open‐bite development during splint therapy.
SN - 2057-4347
ER -