January 2022

Aim: The objective of this study was to evaluate a machine learning model to identify periodontitis patients and their cultivable subgingival microbial features. Methods: We analyzed the composition of the subgingival microbiota in health, gingi- vitis and periodontitis using machine learning. A total of 1026 microbiological records comprising 73 healthy, 205 gingivitis and 748 periodontitis culture samples were ana- lyzed for associated important features and patterns of clustering. Results: The most frequent microorganism was Fusobacterium spp. followed by Prevotel-la intermedia/nigrescens, Porphyromonas gingivalis, Eikenella corrodens, Tannerella forsythia, enteric rods and Aggregatibacter actinomycetemcomitans. The frequency of these microorganisms was higher in periodontitis (p≤0.05). The machine learning algo-rithm efficiently distinguished between health and periodontitis with good performance. Age, P. gingivalis, Fusobacterium spp. and P. intermedia had a high and positive impact on the prediction of periodontitis cases followed by the presence of A. actinomycetem- comitans, T. forsythia and P. micra. There was a discernible increase in the cultivable subgingival microbiota counts from healthy to gingivitis and to periodontitis. Conclusions: Machine learning was able to discriminate between health and periodontitis with good performance. In addition, P. gingivalis, Fusobacterium spp. and P. inter-media were important determinants in the prediction of periodontitis cases.