AI Helps Predict Who’s Most at Risk of Losing Their Teeth
Machine learning reveals how age, income, and dental habits shape oral health outcomes
Tooth loss is often seen as an unavoidable part of aging—but new research suggests that artificial intelligence (AI) could help dentists spot those most at risk long before the damage is done.
In a study published in PLOS ONE, researchers from Harvard University, the University of São Paulo, and the University of Otago used machine learning algorithms to predict tooth loss among American adults. Surprisingly, the study found that socioeconomic factors—like education level, income, and access to regular dental care—were just as important as traditional dental indicators in forecasting who would lose their teeth.
Why Tooth Loss Still Matters
While the overall rate of tooth loss has declined in recent decades, it remains a major concern for older adults and low-income populations. Losing teeth can affect not only a person’s ability to eat and speak but also their self-esteem and general health.
“Identifying who is most at risk allows us to intervene earlier and prevent tooth loss, rather than waiting until extraction becomes the only option,” said lead author Dr. Hawazin Elani of the Harvard School of Dental Medicine.
Turning Big Data into Dental Insight
The research team analyzed data from nearly 12,000 adults who took part in the U.S. National Health and Nutrition Examination Survey (NHANES) between 2011 and 2014.
They developed and compared five types of machine learning models—including logistic regression, random forests, and neural networks—to predict three key outcomes:
Complete tooth loss (edentulism)
Having fewer than 21 teeth (non-functional dentition)
Missing any teeth at all
Each model was trained using data such as participants’ age, education, employment, medical history, and dental care habits, and then tested on new, unseen data.
The most accurate model—Extreme Gradient Boosting Trees (XGBoost)—achieved an impressive 88.7% accuracy in predicting complete tooth loss.
Beyond Cavities and Gum Disease
One of the study’s most striking findings was that models using socioeconomic and health data outperformed those using only dental indicators such as decayed teeth or periodontal disease.
This means that knowing a patient’s education level, income, and access to routine dental care can sometimes be a stronger predictor of future tooth loss than their current dental exam.
Age, education, and routine dental visits emerged as the top three predictors, followed by income, employment status, and home ownership. Chronic diseases—like arthritis, diabetes, and hypertension—also contributed to risk.
Implications for Prevention and Policy
The study highlights how AI could help dentists and policymakers better target prevention programs. For clinicians, machine learning tools could flag high-risk patients for closer monitoring or early intervention. For public health officials, the findings underscore the need to address social inequalities in oral health access.
“This research shows that tooth loss isn’t just about dental hygiene—it’s also about social and economic conditions,” said co-author Dr. Ichiro Kawachi, a public health expert at Harvard. “Machine learning can help us see these patterns more clearly and design fairer health strategies.”
The Next Step: Long-Term Prediction
Although the results are promising, the authors note that more work is needed. The current study used cross-sectional data, meaning it provides a snapshot in time rather than following people over years. Future research using longitudinal data could help validate these AI models and make them even more accurate at predicting tooth loss over time.
Still, the study represents a milestone in applying data science to dentistry. By merging public health and AI, researchers are one step closer to preventing one of the most visible and impactful signs of poor oral health.
Reference
Elani HW, Batista AFM, Thomson WM, Kawachi I, Chiavegatto Filho ADP. Predictors of tooth loss: A machine learning approach. PLOS ONE 16(6): e0252873 (2021). DOI: 10.1371/journal.pone.0252873