Truace
Subscribe

Machine Learning Prediction of Anterior Open-Bite Development After Stabilization Splint Treatment in Temporomandibular Disorder

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 us…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1570· cite & verify
Machine Learning Prediction of Anterior Open-Bite Development After Stabilization Splint Treatment in Temporomandibular Disorder
"09-8018-23" by NavyMedicine is marked with Public Domain Mark 1.0. To view the terms, visit https://creativecommons.org/publicdomain/mark/1.0/.

In brief

Researchers built machine learning models using 35 pre-treatment cephalometric and demographic features from 87 temporomandibular disorder patients. All had been treated with stabilization splints, physical therapy and medication. The models were trained to flag patients whose anterior overbite later decreased in a clinically meaningful way.

The result matters because splint therapy can lead to open-bite development, and an early risk flag could shape how patients are counseled and followed. Discrimination was moderate, the sample was small and retrospective, and the models could not forecast the exact amount of bite change, so clinical use would need further validation.

Main points

  1. In fivefold cross-validation, XGBoost reached an AUC of 0.734, against 0.611 for conventional logistic regression.
  2. With SMOTE oversampling, a random forest reached a comparable AUC of 0.735 and improved sensitivity.
  3. SHAP analysis 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.
  4. Regression models for the exact magnitude of overbite change had near-zero or negative cross-validated R2 values.

The gain

Interpretable machine learning on pre-treatment cephalometric data separated higher-risk from lower-risk TMD patients better than logistic regression, which could support counseling and closer monitoring.

The rundown

The study was a retrospective cohort of 87 patients. The primary outcome was overbite difference, defined as post-treatment minus pre-treatment overbite. Patients were then labeled as Decrease or not for classification.

Classification performance was best with XGBoost, and the random forest with SMOTE matched its AUC while improving sensitivity. SHAP was used to make the models interpretable and to rank which baseline measures drove predictions.

Regression analyses showed that baseline variables alone could not estimate how much the overbite would change. The authors conclude the models support risk stratification, not precise quantitative prediction.

Sources

  1. Peer-reviewedClinical and Experimental Dental Research2026-10-01

The debate