Oral Health Parameter-Based Mini-Mental State Examination Indication Using Machine Learning

Among the growing elderly population, the number of people with neurocognitive disease increases, highlighting the need for early diagnosis. Mini-Mental State Examination (MMSE) is one of the tools used to diagnose neurocognitive disease. The existence of a relationship between degraded oral health and decreased MMSE scores is known. Using machine Learning (ML) techniques, the present study aimed to study the potential of using oral health and demographic examination data to indicate the level of MMSE score. Data from a study evaluating oral health over time and data from an ongoing study evaluating the general health in an elderly population were used as inputs to ML models Random Forest (RF), Support Vektor Machine (SVM), and Catboost (CB) for the binary indication of MMSE score 30 or MMSE score 26 or lower was used to find the best classification performance to distinguish between MMSE low and healthy control (HC) groups. The classifiers were trained using the nested cross-validation (nCV) method to mitigate the risk of overfiting. CB and RF achieved the highest classification accuracy of 80%. However, the CB classifier outperformed other classifiers with 76.4 average accuracies over all the nCV combinations. The oral health parameters and demographics used as input to the ML classifiers carry enough information to distinguish between MMSE low and HC groups. This study suggests that oral health examination might provide information that can be used with the help of Machine Learning (ML) to indicate lowered MMSE scores.