A Deep Neural Network-Based Method for Prediction of Dementia Using Big Data.

Area of Science:

• Gerontology
• Public Health
• Artificial Intelligence in Medicine

Background:

• The aging Korean population faces a growing financial burden due to rising dementia rates.
• Timely dementia recognition and intervention are crucial for effective medical treatment.
• Accessible health behavior and medical service usage data can aid in dementia prescreening.

Purpose of the Study:

• To develop and evaluate a deep neural network (DNN) model for predicting dementia using accessible health data.
• To assess the efficacy of a DNN model combined with scaled Principal Component Analysis (PCA) for dementia prediction.

Main Methods:

• Utilized data from 7031 subjects aged over 65 from the Korea National Health and Nutrition Examination Survey (KNHANES) (2001, 2005).
• Applied Principal Component Analysis (PCA) and min/max scaling for data preprocessing and feature extraction.
• Compared the proposed DNN/scaled PCA methodology against five established machine learning algorithms.

Main Results:

• The proposed DNN/scaled PCA methodology achieved an Area Under the Curve (AUC) of 85.5%.
• This performance surpassed that of the other machine learning algorithms evaluated.
• The model demonstrates significant potential for early dementia detection.

Conclusions:

• The developed deep learning model offers a promising approach for early dementia prescreening.
• This method leverages readily available health behavior and medical service usage data.
• The tool can assist both patients and healthcare providers in managing dementia risks.