Machine learning-based analysis of behavioral and dental plaque microbiome features associated with dental caries in 5-year-old children from urban and rural Yunnan

Abstract

This study aimed to integrate questionnaire-based oral health behavior data with dental plaque microbiome profiles to identify behavioral and microbiological factors associated with dental caries among 5-year-old children from urban and rural areas of Yunnan Province. It also evaluated the performance of these features in machine learning classification models. This study included children from Fengqing County (rural) and Hongta District (urban) in Yunnan Province, who were categorized into a caries-free (CF) group and a dental caries (DC) group based on their caries status. Dietary habits, oral hygiene practices, and regional background data were collected using a questionnaire. Dental plaque samples were collected simultaneously for 16 S rRNA gene sequencing. Based on the relative abundances of the microbial community at the phylum and genus levels, differences in microbial profiles were compared between children with different caries statuses and across regions. Random forest, least absolute shrinkage and selection operator (LASSO), and other machine learning models were employed to identify key microbial features and evaluate the performance of these features, together with relevant behavioral variables, in classifying caries status. Questionnaire analysis revealed significant differences between the DC and CF groups in location, frequency of dessert consumption, and nighttime post-brushing sugar intake (NPSI). By location, Fengqing County and Hongta District differed significantly in NPSI and use of fluoridated toothpaste. Plaque microbiome analysis revealed significant phylum-level differences between the DC and CF groups, particularly in Bacteroidetes, Fusobacteria, and Proteobacteria. A phylum-level diagnostic model identified Fusobacteria as a key discriminative variable (area under the curve [AUC] = 0.737). LASSO analysis identified three genera— Capnocytophaga , Haemophilus , and Comamonas —with Capnocytophaga contributing to diagnostic performance (AUC = 0.720). Incorporating location, dessert consumption, and NPSI further improved model performance (AUC = 1.000). Behavioral factors, regional background, and dental plaque microbial characteristics are jointly associated with the risk of dental caries in 5-year-old children, suggesting that regional differences may partially reflect underlying socioeconomic conditions. However, as these findings are correlational, further validation is required.