What question did this study set out to answer?

The aim is to identify patients with symptomatic Alzheimer disease using electronic health records data and large language models.

June 20, 2026Open Access

Identification of memory clinic patients diagnosed with alzheimer disease using electronic health records data and large language models

Key Points

The aim is to identify patients with symptomatic Alzheimer disease using electronic health records data and large language models.
Examined electronic health records data including demographics, comorbidities, and clinical notes.
Developed machine learning models for clinical diagnosis prediction of symptomatic Alzheimer disease.
Achieved prediction accuracy one year before clinical diagnosis with AUROC of 0.858.
Prediction models identified symptomatic Alzheimer disease with an AUROC of 0.858 (95% CI: [0.835, 0.881]).
Key features for model accuracy included age, repeating statements, and hypertension.
EHR-derived phenotypes effectively enable the prediction of symptomatic Alzheimer disease diagnosis.

Abstract

Alzheimer disease (AD) is a neurodegenerative disorder marked by gradual decline in memory and thinking. New treatments for early symptomatic AD have increased the need for early and accurate AD diagnosis. This study aimed to identify which patients presenting to a specialty memory clinic were diagnosed with symptomatic AD by examining electronic health record (EHR) data encompassing demographic data, comorbidities, medications, biomarkers, and clinical notes analyzed using large language models. We developed clinical diagnosis prediction machine learning models to predict future diagnosis of symptomatic AD in specialty memory clinic patients presenting with cognitive concerns. The models could accurately predict a diagnosis of symptomatic AD one year prior to clinical diagnosis (AUROC 0.858, 95% CI: 0.835, 0.881). Key features in the model included age, repeating statements, and hypertension. Our findings suggested that models using multifaceted EHR-derived phenotypes could predict diagnosis of symptomatic AD, potentially enabling earlier intervention.

Mark Helpful

Bookmark

Relay

View Full Paper