Restoration's Longevity in Endodontically Treated Teeth: A Machine Learning Survival Analysis From Randomised...

Area of Science:

• Dental research
• Biostatistics
• Machine learning in healthcare

Background:

• Endodontically treated teeth (ETT) restorations require accurate prognostic models.
• Estimating the longevity of ETT restorations is crucial for patient outcomes and treatment planning.

Purpose of the Study:

• To develop and evaluate machine learning (ML) survival models for predicting the success and survival rates of restorations in ETT.
• To identify key predictors of restoration longevity in ETT.

Main Methods:

• Consolidated data from four controlled clinical trials (424 patients, 618 restorations, up to 17 years follow-up).
• Evaluated Gradient Boosting Survival, Random Survival Forests, and Survival Support Vector Machine models.
• Utilized 10-fold cross-validation and hyperopt for hyperparameter tuning. Assessed performance using AUC, C-index, IPCW C-index, and Brier score.

Main Results:

• Gradient Boosting Survival model demonstrated superior performance for survival rate prediction (AUC=0.83).
• Random Survival Forest model showed higher accuracy for success rate prediction (AUC=0.73).
• Patient age, tooth type, and dentist experience were identified as significant predictors. Fairness analysis indicated performance disparities across sexes and countries.

Conclusions:

• Machine learning models exhibit high predictive performance for ETT restoration longevity, particularly for survival rates.
• ML offers a promising framework for data-driven evaluation of success and survival outcomes in endodontic treatments.