Machine Learning Models for Identifying Dental Pain in Adolescents

Area of Science:

• Public Health
• Machine Learning
• Adolescent Health

Background:

• Dental pain is a significant public health concern among adolescents.
• Screening tools are needed to identify at-risk populations for early intervention.

Purpose of the Study:

• To evaluate the effectiveness of machine learning (ML) algorithms in identifying dental pain in Brazilian adolescents.
• To assess the potential of ML for public health screening of dental pain.

Main Methods:

• Utilized data from the Brazilian National Survey of School Health (PeNSE) from 2015 and 2019, including 259,833 adolescents aged 11-18.
• Trained and tested nine ML models on the 2015 dataset and validated externally using the 2019 dataset.
• Evaluated model performance using metrics like Area Under the Curve (AUC) and Recall, alongside fairness assessments.

Main Results:

• The Extra Trees (ET) model demonstrated the best performance, with an AUC of 0.64 and Recall of 0.57 in the test set.
• The model achieved a Recall of 0.57 in the external validation set, indicating a modest ability to identify adolescents with dental pain.
• Fairness analysis revealed disparities in accuracy and recall across different demographic groups (sex, race), with sex, alcohol consumption, and family violence identified as key predictive variables.

Conclusions:

• Machine learning holds promise for identifying dental pain in adolescent populations.
• Current predictive performance and fairness limitations necessitate further research and model refinement for practical public health application.