Association between periodontitis and diabetes

Background + Aims

• Periodontitis shares important risk factors with cardiometabolic disease, particularly smoking and elevated blood glucose.
• Information routinely collected in medical settings could therefore help identify people with undiagnosed periodontal disease and support referral for dental assessment.
• This study aimed to:
  • Develop and internally validate a practical model for identifying existing moderate-to-severe periodontitis in US adults.

Materials + Methods

• Cross-sectional analysis of 3,017 adults aged over 30 years, with more than 14 teeth, from NHANES 2011–2012.
• Periodontal examinations measured probing depth and clinical attachment loss at six sites per tooth. Participants were classified using CDC/AAP criteria.
• Candidate predictors included demographic characteristics, smoking, body mass index, blood pressure, cholesterol and HbA1c, which reflects average blood glucose over approximately two to three months.
• Researchers used multivariable logistic regression to select a model balancing simplicity and predictive performance.
• Internal validation used bootstrapping – resampling the original dataset to assess how reliably the model performed.

Results

• 37.1% of participants had moderate periodontitis and 13.2% had severe periodontitis.
• Current smoking was associated with substantially higher odds of moderate-to-severe disease: OR 2.91 (95% CI 2.23–3.80).
• HbA1c ≥5.7% was also associated with higher odds: OR 1.29 (95% CI 1.07–1.57).
• The final five-variable model included age, gender, ethnicity, HbA1c and smoking, achieving an area under the ROC curve (AUC) of 0.773.
  • AUC describes how well a model distinguishes people with and without the target condition. 0.5 indicates chance performance and 1.0 perfect discrimination.
• Sensitivity was 70.0%: approximately 7 in 10 participants with moderate-to-severe periodontitis were identified.
• Specificity was 67.6%: approximately two thirds without moderate-to-severe disease were correctly classified.
• The larger 11-variable model achieved an AUC of 0.801, offering relatively modest additional discrimination

Limitations