Leveraging routine clinical data for dementia risk prediction using machine learning.

Area of Science:

• Gerontology
• Medical Informatics
• Computational Biology

Background:

• Early dementia diagnosis is crucial for interventions that slow progression and reduce costs.
• Longitudinal electronic health record (EHR) data offers a valuable resource for risk prediction.
• Developing accurate predictive models can improve patient outcomes and healthcare efficiency.

Purpose of the Study:

• To develop and evaluate machine learning models for dementia risk prediction using EHR data.
• To identify key clinical features associated with increased dementia risk.
• To assess the utility of routine clinical data for dementia risk stratification.

Main Methods:

• An incidence-based case-control study was conducted using EHR data from UMass Memorial Health (2017-2024).
• Machine learning models, including XGBoost, were developed to predict dementia risk.
• Model performance was evaluated using Area Under the Curve (AUC) and feature importance analysis.

Main Results:

• The study included 5,622 dementia cases and 44,976 controls.
• The XGBoost model achieved an AUC of 0.802, demonstrating good predictive performance.
• Top predictors identified were thyroid-stimulating hormone (TSH), vitamin B12, and HDL cholesterol.

Conclusions:

• Machine learning models integrating comorbid conditions and longitudinal lab data show promise for dementia risk prediction.
• Routinely collected EHR data is a scalable and cost-effective resource for identifying at-risk individuals.
• These findings support the use of predictive analytics in dementia care pathways.