Machine Learning to Build Prediction for Unmet Dental Care Needs

Objectives: This study aimed to identify factors predictive of unmet dental care needs using machine learning.
Methods: The 2016 Medical Expenditure Panel Survey data were used in this study. The sample was weighted to represent the entire population of the United States. Machine learning methods such as Random Forest, ExtraTrees Classifier, Decision Tree Classifier, and Gradient Boosting Classifier were used to identify variables that are predictive of unmet dental care needs based on relative variable importance. The results were examined using chi-square test and independent samples t-test across demographic groups.
Results: A total of 34,655 participants and 237 variables were included in this study. Different tree classifiers consistently identified 14 top predictors of unmet dental care needs. Age was the most important variable followed by person’s total income, census region of residence, inability to get necessary medical care and family size. The risk prediction model for unmet dental care needs reached an accuracy of 82.6%.
Conclusions: This study demonstrated that social determinants of health are strong contributors to unmet dental care needs. The use of machine learning tree classifiers provides the ability to process hundreds of variables at once and enables discovery top predictors that were not previously found using traditional statistics. This study can aid members of the dental public health (both providers and policy-makers) to develop cost-efficient practices and interventions for high-risk patients having unmet dental care needs.