Machine learning-based approach for predicting the extraction time of a mandibular third molar.

Area of Science:

• Oral and Maxillofacial Surgery
• Machine Learning in Healthcare
• Surgical Outcomes Prediction

Background:

• Predicting surgical duration for mandibular third molar extraction is difficult due to complex interactions between patient, tooth, and operator factors.
• Accurate time prediction is crucial for surgical planning and resource management.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting mandibular third molar extraction surgical time.
• To identify key predictors of surgical duration using patient demographics, radiographic features, and surgeon variables.

Main Methods:

• Retrospective review of 713 mandibular third molar extractions.
• Development of machine learning models using a no-code platform (Prediction One).
• Reframing prediction as binary classification at clinically relevant time thresholds (≥31, ≥46, ≥61 minutes) and benchmarking against traditional models.

Main Results:

• The model predicting durations ≥31 minutes achieved the best performance (accuracy 0.75, AUC 0.80), outperforming comparator models.
• Surgeon experience and specialist certification were the most significant predictors of surgical time.
• Root-related radiographic morphology was also a strong contributor, while Pell and Gregory classification showed limited predictive value.

Conclusions:

• Threshold-based prediction of mandibular third molar extraction time is feasible and more effective than continuous-time prediction.
• Surgeon experience and specific radiographic features are key determinants of surgical duration.
• The study identified factors with substantial and minimal impact on predicting surgical time, aiding in clinical practice.