Periodontal disease risk prediction via machine learning

Recent efforts in the field of periodontology have focused on examining the biological basis of susceptibility to disease, customizing disease prevention and treatment strategies based on the principles of precision dentistry. In the treatment of individuals with periodontal disease, it is important to determine the individual periodontal risk to develop treatment protocols for each patient rather than the "average treatment" approach. Therefore, the aim of this study is to predict the periodontal disease risk of individuals and to divide individuals into periodontal disease risk categories. In this cross-sectional study, demographic, clinical and radiological examinations of 1057 subjects were performed. Participants were diagnosed according to the new periodontal disease classification. Accordingly, a total of 21 categoric independed variables obtained from patients were imported to the Data Loading Window (DLW) which is generated for this study. K-means cluster analysis, one of an unsupervised machine learning, was used to determine the periodontal risk groups. The results of the cluster algorithm was validated by discriminant functions, silhouette analyzes and akaike's information criterion (AIC). As a result of k-means cluster analysis, study participants were divided into three different periodontal risk categories. These categories were defined as low risk (n=462), medium risk (n=336) and high risk (n=259). It was determined that 96.4% of individuals with uncontrolled diabetes were in the high-risk group. While individuals with stage 4 periodontitis were not in the low risk group, it was 5.4% in the medium risk group and 94.6% in the high risk group. A computer-based and user friendly periodontal risk prediction tool (PRPT) was presented for clinical practice. It is assumed that physicians can calculate their patients periodontal disease risk via PRPT and manage their patients periodontal treatment planning.