Classifying Delayed Dental Care Using Machine Learning: A National Health Interview Survey Analysis

Delays in dental care worsen oral disease and mirror broader inequities in health care access and use.

To estimate the 12-month prevalence of delayed dental care among US adults, to characterise disparities across demographic and socioeconomic groups and to evaluate the utility of machine learning models for identifying individuals at elevated risk of delay.

This study used cross-sectional analysis of the 2023 National Health Interview Survey (NHIS) Sample Adult file (N = 54,927). Survey weights were applied to obtain nationally representative estimates. Descriptive statistics were stratified by age, sex, education, dental insurance coverage and race/ethnicity. In parallel, supervised machine learning classifiers were developed to classify delayed dental care as a binary outcome. Class imbalance was addressed using oversampling applied to training data only. Model performance was evaluated using accuracy, precision, recall, F1-score and area under the receiver operating characteristic curve (AUC), and model interpretability was assessed using SHapley Additive exPlanations (SHAP).

Overall, 14.5% of adults reported delaying dental care in the prior year. Delay was higher among adults aged 35-50 (18.3%) and 51-64 (17.2%) and lower among those 65 or older (11.0%). Racial/ethnic differences were evident: Black/African American adults (19.1%) and individuals reporting multiple races (21.1%) had higher prevalence of delayed care than did White adults (14.2%). Adults without dental insurance were more likely to report delays, underscoring the role of coverage. An educational gradient was observed, with higher delays among those with lower attainment. Among the evaluated classifiers, the Light Gradient Boosting Machine (LightGBM) demonstrated the strongest overall performance (accuracy = 84.87%). The SHAP analysis identified education, income-to-poverty ratio and insurance status as the most influential predictors.

Delayed dental care affects a substantial share of US adults and disproportionately impacts groups defined by age, race/ethnicity, education and insurance status. Interpretable machine learning models can complement traditional survey analyses by supporting population-level risk stratification. Policies that expand dental coverage, reduce financial barriers and target outreach to high-risk populations may mitigate inequities. Given the cross-sectional design, findings should be interpreted as predictive rather than causal. Future research should incorporate contextual determinants (eg, geographic access and provider availability) and longitudinal data to refine population targeting and evaluate intervention impact.