Lexical concept distribution reflects clinical practice.

Area of Science:

• Medical Informatics
• Clinical Documentation
• Health Data Analysis

Background:

• The direct correlation between narrative medical text and actual clinical practice remains unclear.
• Electronic health records (EHRs) offer structured data, but narrative notes contain rich clinical context.
• Validating narrative medical text as a reliable data source is crucial for comprehensive health data analysis.

Purpose of the Study:

• To test the hypothesis that lexical concepts of medication intensification in narrative provider notes correlate with clinical practice.
• To determine if narrative medical documentation reflects real-world clinical decisions.

Main Methods:

• Analysis of over 29,000 medication intensifications from narrative provider notes.
• Comparison with 444,000 electronic medication records for 82 medications (anti-hypertensive, anti-hyperlipidemic, anti-hyperglycemic).
• Calculation of Pearson correlation coefficient between dose increases in notes and therapeutic ranges from EHRs.

Main Results:

• A significant positive correlation (r=0.39, p=0.0003) was found between medication intensification patterns in narrative notes and clinical practice reflected in EHRs.
• Correlations with medication dictionaries (negative controls) were not significant, strengthening the findings.
• This indicates a direct reflection of clinical practice within narrative medical text.

Conclusions:

• Narrative medical documents are a valid and reliable source of clinical data.
• The study provides evidence that the content of provider notes directly mirrors actual clinical practice.
• Lexical analysis of narrative text can be used to infer clinical actions and trends.