Self-reported periodontitis: A multiethnic community-based validation study.

OBJECTIVE

Self-reported periodontal measures offer a practical alternative to clinical examinations in large-scale studies, but few validation efforts exist in Asia, where periodontitis is highly prevalent. Moreover, the diagnostic performance of individual self-reported measures and the relevant sociodemographic factors differs across populations. This study aims to clinically validate self-reported periodontal measures in a multiethnic community-based population and to develop several predictive models that combine sociodemographic characteristics and self-reported periodontal measures.

METHODS

We analysed cross-sectional data from 426 diabetes-free participants in Singapore who completed the Centers for Disease Control and Prevention/American Academy of Periodontology (CDC-AAP) self-reported periodontal questionnaire and underwent full-mouth periodontal examinations. Periodontitis status was defined using the 2012 CDC-AAP case definitions. Multivariable logistic regressions and area under the curve (AUC) analyses evaluated predictive models for periodontitis. Additionally, we performed an exploratory analysis, training and testing five machine learning models to predict periodontitis.

RESULTS

Participants had a mean age of 48.9±9.9 years, 55.6% were female, and 16.4%, 42.5%, and 18.1% had mild, moderate, and severe periodontitis, respectively. A multivariable model incorporating one self-reported question (loose teeth), age, and ethnicity demonstrated good discrimination for severe periodontitis (AUC=0.76; sensitivity/specificity=0.60/0.80). While the machine learning models achieved similar AUC (0.67-0.76), they tended to be highly specific (0.99-0.75) but had much lower sensitivity (0.12-0.63).

CONCLUSION

Selected self-reported periodontitis questions were useful for detecting severe periodontitis in this population. While machine learning models integrated sociodemographic factors with all 8 self-reported questions, larger and more diverse datasets are needed to enhance model robustness and generalisability across international populations.

CLINICAL SIGNIFICANCE: Self-reported periodontal measures validated in a multiethnic population demonstrated good predictive value for severe periodontitis when combining self-report of loose teeth with demographic variables age, gender, and ethnicity. The application of machine learning models further enhanced diagnostic performance, highlighting potential for scalable, data-driven periodontal screening and surveillance across diverse populations.