Clinical decision support model for tooth extraction therapy derived from electronic dental records

Area of Science:

• Dental Informatics
• Artificial Intelligence in Healthcare
• Prosthodontics

Background:

• Tooth extraction is a critical first step in prosthodontic treatment plans.
• Accurate decision-making for tooth extraction requires considering multiple patient factors.
• Clinical decision support (CDS) models may aid dentists in this crucial choice.

Purpose of the Study:

• To develop a CDS model for predicting tooth extraction therapy.
• To utilize electronic dental records (EDRs) for model construction.
• To evaluate the model's performance in clinical scenarios.

Main Methods:

• A retrospective study analyzed 4135 deidentified EDRs from 3559 patients.
• Knowledge-based algorithms structured raw EDR data for feature extraction.
• Recursive feature elimination filtered redundant features.
• Tooth extraction prediction was modeled using binary and triple classification with five machine learning algorithms.
• The Extreme Gradient Boost (XGBoost) algorithm was evaluated.

Main Results:

• The triple classification model achieved a higher F1 score (0.856) than the binary model (0.847).
• XGBoost demonstrated superior performance among the five algorithms.
• XGBoost achieved high accuracy (0.962 binary, 0.924 triple), precision (0.865 binary, 0.879 triple), and recall (0.830 binary, 0.836 triple).
• The CDS model's performance surpassed that of two verifying prosthodontists.

Conclusions:

• The developed CDS model demonstrates high efficacy in supporting tooth extraction therapy decisions.
• The model leverages electronic dental records for data-driven clinical insights.
• AI-powered decision support systems show promise in improving dental treatment planning.