BackgroundPeriodontal disease is a prevalent chronic inflammatory condition associated with systemic health complications. Early identification of high-risk individuals is crucial for targeted preventive and therapeutic interventions. In this study, we aimed to develop and evaluate machine learning models using routine blood tests to predict periodontal disease risk. This will help to develop accessible and cost-effective screening tools for early detection in a non-dental setting.MethodsThis study utilized data from the 2013-2014 National Health and Nutrition Examination Survey (NHANES), including full-mouth periodontal examinations, demographic variables, and routine blood tests (complete blood count [CBC], lipid profile, liver and kidney function tests). Periodontitis was defined as a binary outcome based on attachment loss and probing depth at interproximal sites. Individuals meeting any of these criteria were classified as having periodontitis. Seven machine learning models were developed and evaluated using precision, recall, and accuracy metrics.ResultsThe Random Forest Classifier achieved the highest performance, scoring 0.91 in precision, recall, and accuracy for predicting periodontitis. Key predictive features included smoking status, age, education level, gamma glutamyl transferase, albumin, blood urea nitrogen, glucose, phosphorus, creatinine, basophil count, and total calcium levels.ConclusionThis study underscores the promise of machine learning, especially the Random Forest Classifier, in predicting periodontal disease risk using routine blood tests and demographics. The model accurately identified periodontitis cases, with key features revealing the complex relationship between systemic health and periodontal disease. This suggests the potential for developing machine learning-based screening tools for early periodontal disease detection in non-dental settings.