Biomarkers.

Area of Science:

• Computational Medicine
• Geriatric Medicine
• Data Science

Background:

• Early dementia diagnosis is critical for timely interventions and improved patient outcomes.
• Electronic health records (EHRs) provide valuable longitudinal data for identifying dementia risk factors.
• This study leverages machine learning and EHR data to predict dementia risk.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the risk of dementia.
• To utilize longitudinal EHR data from a large hospital system for dementia risk prediction.
• To assess the impact of comorbidities and lab test results on dementia risk prediction.

Main Methods:

• Utilized EHR data from UMass Memorial Medical Center (2017-2024).
• Identified dementia cases using ICD-10 codes, excluding patients under 65.
• Employed logistic regression, random forest, and XGBoost models with 21 features, including comorbidities and lab tests, assessed via 5-fold cross-validation.

Main Results:

• The study included 30,162 participants (mean age 80 ± 12 years, 69.7% women).
• XGBoost model achieved the highest predictive performance with an AUC of 0.82 using comorbidities alone.
• Incorporating lab test information improved performance, with XGBoost reaching an AUC of 0.83.

Conclusions:

• Developed machine learning models capable of predicting dementia risk using EHR data.
• Findings underscore the potential of EHRs for early dementia detection.
• Further validation in diverse healthcare systems is recommended.