Background: This retrospective cohort study developed an artificial intelligence (AI) model to predict incident dementia and evaluated its predictive performance using a validation cohort. The study participants were 7384 older adults (age ≥ 75 years) who underwent regional dental checkup in Gifu Prefecture, Japan, in 2018 and 2020. Methods: The National Database of Health Insurance of Japan (NDB) was integrated with dental checkup data, and the participants were randomly divided into two datasets: training (n = 5169) and validation (n = 2215). A data analytics tool was utilized to create the AI model with training data in 2018 and data on the presence or absence of dementia development in 2020. Results: The AI model trained solely on NDB data showed a sensitivity of 0.73 and specificity of 0.91 in predicting the presence or absence of dementia development after 2 years. By contrast, the AI model trained on NDB and dental checkup data showed a sensitivity of 0.75 and specificity of 0.95, indicating improvement in both metrics. Conclusions: Combining different sets of data, such as NDB and dental checkup data, for training may be useful for improving the accuracy of AI models to predict dementia development.
Iwai et al. (Thu,) studied this question.