Predicting Longitudinal Caries Trajectories from Childhood to Early Adulthood

Area of Science:

• Oral Health
• Data Science
• Public Health

Background:

• Traditional methods for analyzing caries progression lack predictive power.
• Machine learning (ML) has not been previously applied to predict caries trajectories from childhood to early adulthood.

Purpose of the Study:

• To use unsupervised ML for clustering longitudinal caries data into distinct trajectory groups.
• To employ supervised ML to predict trajectory group membership using behavioral, dietary, fluoride, and sociodemographic variables.

Main Methods:

• Longitudinal data from 560 participants of the Iowa Fluoride Study were analyzed.
• Unsupervised K-means clustering identified three caries trajectory groups: low, medium, and high.
• Supervised ML models, including extreme gradient boosting, predicted trajectory group membership.

Main Results:

• Three distinct caries trajectories were identified: low (70.5%), medium (21.1%), and high (8.4%).
• Extreme gradient boosting achieved 85.9% accuracy and a Brier score of 0.21 in predicting trajectory membership.
• Key predictors included sex, socioeconomic status, water fluoride concentration, and intake of sugar-sweetened beverages and juice.

Conclusions:

• This study pioneers the use of combined ML models to accurately predict caries trajectories from childhood to adulthood.
• ML-driven prediction of caries trajectories can facilitate early identification of at-risk individuals.
• Findings support the development of targeted, age-appropriate preventive interventions for dental caries.