Breaking barriers: a statistical and machine learning-based hybrid system for predicting dementia.

Area of Science:

• Computational neuroscience
• Medical informatics
• Machine learning in healthcare

Background:

• Dementia affects millions globally, necessitating early detection for effective intervention.
• Current diagnostic methods (clinical exams, cognitive tests) are time-consuming and expensive.
• There is a need for noninvasive, efficient tools for early dementia prediction.

Purpose of the Study:

• To develop and validate a noninvasive hybrid diagnostic system for early dementia prediction.
• To utilize patient electronic health records (EHRs) for dementia risk assessment.
• To improve upon existing diagnostic approaches through machine learning integration.

Main Methods:

• A hybrid diagnostic system combining statistical feature selection (F-score) and ensemble machine learning (ML) was developed.
• The system employed an ensemble voting classifier integrating decision tree, naive Bayes, logistic regression, support vector machines, and random forest models.
• Performance was evaluated using cross-validation and metrics including accuracy, sensitivity, specificity, ROC curve, and Matthew's Correlation Coefficient (MCC) on the SNAC dataset (n=43040).

Main Results:

• The proposed hybrid diagnostic system achieved a high accuracy of 98.25%.
• Excellent performance was also demonstrated with sensitivity of 97.44%, specificity of 95.744%, and MCC of 0.7535.
• The system significantly outperformed baseline ML models and prior feature selection techniques.

Conclusions:

• The developed hybrid ML system provides a highly accurate and efficient noninvasive method for early dementia prediction.
• Leveraging EHRs and advanced ML techniques offers a promising avenue for improving dementia diagnostics.
• This approach has the potential to facilitate timely preventive strategies, reducing the burden of dementia.