Predicting Alzheimer's Disease Diagnosis, a Decade or more Years before Onset using the Electronic Health Record and...

Area of Science:

• Computational neuroscience
• Medical informatics
• Epidemiology

Background:

• Early detection of Alzheimer's disease (AD) is crucial for intervention.
• Electronic Health Records (EHRs) offer a potential data source for predicting AD.
• Machine learning (ML) methods can analyze complex EHR data for predictive modeling.

Purpose of the Study:

• To develop and validate an EHR-based machine learning model for predicting Alzheimer's disease (AD) onset.
• To identify key predictive features from EHRs for AD risk up to 10 years prior to diagnosis.
• To establish a novel tool for early AD detection and potential intervention.

Main Methods:

• Utilized a large dataset of 19,473 AD cases and 111,922 controls from EHRs.
• Trained a random forest model using 2,499 features from records spanning over 10 years pre-diagnosis.
• Employed 5-fold cross-validation and a 75/25% train/test split for model evaluation.

Main Results:

• The model achieved an Area Under the ROC Curve (AUC) of 0.80, indicating good predictive performance.
• Key predictors identified include age, sex, race, ethnicity, BMI, and comorbidities like cardiovascular and inflammatory diseases.
• Other significant factors included sleep and mood disorders, trauma, seizures, and specific medications/disorders.

Conclusions:

• This study presents the first EHR-based model capable of predicting Alzheimer's disease (AD) 10 years before clinical onset.
• The model demonstrates the potential of leveraging routinely collected EHR data for early AD risk assessment.
• Findings can inform the development of preventative strategies and early intervention programs for Alzheimer's disease.