Predicting gingival embrasure risk after invisible orthodontics using multimodal data and machine learning

Area of Science:

• Oral health research
• Biostatistics
• Dental informatics

Background:

• Clear aligner therapy is increasingly popular for orthodontic treatment.
• Gingival embrasure spaces can develop post-treatment, impacting aesthetics and periodontal health.
• Predicting this complication requires robust, data-driven tools.

Purpose of the Study:

• To develop and validate a predictive model for gingival embrasures after clear aligner therapy.
• To identify key risk and protective factors influencing embrasure development.
• To evaluate machine learning algorithms for predictive accuracy.

Main Methods:

• Retrospective analysis of 340 patients undergoing clear aligner therapy.
• Multivariate logistic regression and machine learning (Random Forest, Logistic Regression, SVM) applied to multimodal oral data.
• Model performance assessed using AUC, calibration curves, and SHAP values.

Main Results:

• Multivariate analysis identified bleeding on probing, interproximal bone height, and tooth movement as risk factors.
• Gingival thickness, contact area, papilla height, and bone plate thickness were protective factors.
• The Random Forest model demonstrated superior performance with high AUC values in training and validation sets.

Conclusions:

• A validated risk prediction model for post-clear aligner gingival embrasures was developed using multimodal data.
• The Random Forest algorithm proved optimal for this predictive task.
• The model serves as an objective tool for individualized risk assessment, complementing clinical judgment.