Unsupervised Machine Learning Identified Distinct Population Clusters Based on Symptoms of Oral Pain, Psychological Distress, and Sleep Problems

OBJECTIVES: The aims of this study were to explore the use of unsupervised machine learning in clustering the population based on reports of oral pain, psychological distress, and sleep problems and to compare demographic and socio-economic characteristics as well as levels of functional domains (work, social, and leisure) between clusters. MATERIALS AND METHODS: In this cross-sectional study, a total of 1613 participants from the National Health and Nutrition Examination Survey in 2017–2018 were analyzed. Five variables, including oral pain, depression, anxiety, sleep apnea, and excessive daytime sleepiness, were selected for cluster analysis using the k-medoids clustering algorithm. The distribution of categorical variables between clusters was assessed using χ(2) test. One-way analysis of variance and Kruskal–Wallis H test were used to compare numerical variables as appropriate. RESULTS: Five distinct clusters were identified: healthy, norm, anxiety, apnea-comorbid, and pain-comorbid. The apnea-comorbid cluster had mean age of 59 years and higher proportion of men. The pain-comorbid cluster had mean age of 56 years and higher proportion of women. Whites constituted a majority of both comorbid clusters. The pain-comorbid cluster demonstrated the least percentage of individuals with college degree, the lowest income, and significant impairment in all functional domains. CONCLUSION: Through the use of unsupervised machine learning, the clusters with comorbidity of oral pain, psychological distress, and sleep problems have emerged. Major characteristics of the comorbid clusters included mean age below 60 years, White, and low levels of education and income. Functional domains were significantly impaired. The comorbid clusters thus call for public health intervention.