Extracting Critical Information from Unstructured Clinicians' Notes Data to Identify Dementia Severity Using a...

Area of Science:

• Medical Informatics
• Clinical Data Analysis
• Natural Language Processing in Healthcare

Background:

• Alzheimer disease and related dementias (ADRD) severity is crucial for patient care but often missing in structured electronic health record (EHR) data.
• Severity information is frequently embedded within unstructured clinical notes, limiting its accessibility for clinicians.

Purpose of the Study:

• To evaluate the feasibility of using keyword and rule-based matching to extract ADRD severity from EHRs.
• To assess potential biases associated with this data extraction method.

Main Methods:

• Utilized EHR data from 2014-2019 for patients with ADRD diagnoses.
• Developed a rule-based algorithm in Python to identify ADRD severity using cognitive test scores and explicit severity terms from clinical notes.
• Assessed algorithm performance using accuracy, specificity, sensitivity, and F1-score.

Main Results:

• Over 56% of patients lacked documented ADRD severity.
• The algorithm achieved high performance metrics (>90% accuracy, specificity, sensitivity) for detecting severity information.
• Identified demographic differences between patients with documented mild versus advanced ADRD.

Conclusions:

• A rule-based algorithm can effectively extract ADRD severity from unstructured EHR data, demonstrating feasibility.
• Potential biases in documentation practices across healthcare systems must be considered when interpreting results.