Risk-factor based Diagnosis for Chronic Periodontitis using Machine Learning Models
1 Department of Clinical Epidemiology and Biostatistics, Faculty of Medicine Ramathibodi Hospital, Mahidol University, Bangkok, Thailand
2 Department of Periodontology, Faculty of Dentistry, Chulalongkorn University, Bangkok, Thailand
3 Center of Excellence in Periodontal Disease and Dental Implant, Chulalongkorn University, Bangkok, Thailand
Abstract
Chronic periodontitis is one of the most common oral diseases worldwide, with a prevalence of 11.2% globally, 15–20% in Asians, and 26% in Thai adults. Symptoms are negligible until it is too late, resulting in tooth loss and reduced quality of life. Diagnosis requires a chairside examination measuring six sites of the gingival sulcus per available tooth — a time- and resource-intensive process. A predictive model to identify the risk of chronic periodontitis could assist in targeted screening. While logistic regression has been commonly applied using demographic and risk behaviour predictors, performance depends on feature selection and engineering. This thesis applies machine learning models — mixed-effects logistic regression (MELR), recurrent neural networks (RNN), and mixed-effects support vector machine (MESVM) — for the diagnosis of chronic periodontitis using the longitudinal Electricity Generating Authority of Thailand (EGAT) cohort. The MELR model (90.5% accuracy) performed better than conventional logistic regression as well as other machine learning models (RNN 70.0%, MESVM 72.7%) even after hyperparameter optimization. Trained models can be applicable for screening in community and public health settings as well as in electronic health record systems.
