Abstract:Early detection of individuals with a high risk of dementia is crucial for prompt intervention and clinical care. This study aims to identify high-risk groups for developing dementia by predicting the outcome of the Mini-Mental State Examination (MMSE), using historical data collected from community-based primary care services. To mitigate the effect of inter-individual variability and enhance the accuracy of the prediction, we implemented a multi-stage method powered by supervised and unsupervised machine learning methods. Firstly, we preprocessed the original data by imputing missing values and using a wrapper-based feature selection algorithm to pick significant features, resulting in ten variables out of 567 being selected for further modeling. Secondly, we optimized hierarchical clustering to partition the unlabeled data into groups by their similarities, and then applied supervised machine learning models to build subgroup-specific prediction models for the identified groups. The results demonstrate that the proposed subgroup-specific prediction models generated from the multi-stage method achieved satisfactory performance in predicting the outcome classes of dementia risk. This study highlights the potential of incorporating unsupervised and supervised learning models to predict high-risk cases of dementia early and facilitate better clinical decision-making.

Predicting Dementia Risk for Elderly Community Dwellers in Primary Care Services Using Subgroup-specific Prediction Models