An international team of researchers reports in the journal Caries Research that artificial intelligence (AI) models can predict dental caries presence and severity across different age groups using data from the 2023 Brazilian Oral Health Survey, a national epidemiological study coordinated by the Brazilian Ministry of Health. The analysis points toward identifying the population clusters most affected by cavities through broader social determinants of health rather than individual clinical diagnostics alone, an approach the authors say could support targeted public health programs.
The study's senior author is Fernando Neves Hugo, DDS, MSc, PhD, chair of the Department of Epidemiology and Health Promotion at New York University College of Dentistry. "The emergence of AI offers a transformative opportunity for public health surveillance and risk stratification, including addressing the global burden of cavities," Dr. Hugo said. "These are powerful tools that could be helpful for targeting resources and delivering preventive care to specific groups that will benefit the most."
The survey collected demographic information, including age, sex, and race, along with socioeconomic and behavioral factors associated with oral health. The analysis focused on three age groups recommended by the World Health Organization: 6,698 12-year-old children, 8,041 adolescents aged 15 to 19, and 8,987 adults aged 35 to 44, according to the study. On the socioeconomic side, the data collected included wealth, household density, private dental insurance status, and water source; on the behavioral side, how long it had been since the participant's last dental visit. Participants also underwent clinical examinations to assess the presence of untreated caries and disease severity.
Conventional statistical analysis of the data showed oral health in the population emerging from a complicated mix of socioeconomic, structural, and behavioral influences. Of all the factors examined, wealth and household density carried the most predictive weight for cavity risk. Across all age groups, participants classified as non-White had a higher caries burden, while private dental insurance and recent dental visits were associated with lower caries prevalence. Water access was also associated with caries prevalence: 62.5 percent of 12-year-olds without access to piped water had untreated caries, compared with 35.5 percent of those with access to piped water.
When the team applied several machine learning algorithms to the dataset, the models demonstrated varying predictive performance across age groups, with stronger discrimination of caries presence among adults. The researchers also evaluated the relative contributions of socioeconomic, behavioral, and clinical factors. The study reports that an explainability analysis using SHAP values, a method that ranks how much each variable contributes to a model's predictions, confirmed that socio-structural factors outweighed individual clinical markers across all age groups, with wealth score and household density emerging as the primary structural predictors of caries risk.
"Our machine learning approach acts as a high-resolution sensor that identifies individuals whose profiles signal a heightened risk that would otherwise be invisible within community averages," Dr. Hugo said.
The authors acknowledge several limitations, including the study's cross-sectional design, which prevents causal conclusions, its focus on urban populations, and the lack of external model validation. The researchers also note that the findings may not translate directly to other countries, where differences in dental care access and socioeconomic conditions could influence caries risk. Looking ahead, they aim to repeat the AI-enabled method using country-specific factors to build prevention strategies tailored to each population. As a next step, Dr. Hugo and his colleagues are preparing a parallel analysis using US data from the National Health and Nutrition Examination Survey.
Figures and Images
