Almost Clinical: Linguistic properties of synthetic electronic health records

TL;DR

This paper analyzes the linguistic properties of synthetic electronic health records (EHRs) for mental health by generating a large, controlled corpus across four clinical genres using templated prompts and varying input demographics and clinical features. Employing Systemic Functional Linguistics, it examines agency, modality, and information flow to quantify how large language models (LLMs) grammatically construct medical authority and patient agency. The study finds that while LLMs can produce coherent, clinically plausible texts, they exhibit register shifts, insufficient clinical specificity, and notable inaccuracies in procedures or medications, with biases evident in ethnicity-linked drug mentions. These findings highlight both the potential and the limitations of synthetic EHRs for research and development, emphasizing the need for careful fine-tuning, bias mitigation, and ethically responsible use. The work provides a framework for genre-sensitive linguistic analysis of synthetic clinical data and offers guidance toward safer, more faithful AI-assisted healthcare documentation.

Abstract

This study evaluates the linguistic and clinical suitability of synthetic electronic health records (EHRs) in the field of mental health. First, we describe the rationale and the methodology for creating the synthetic corpus. Second, we assess agency, modality, and information flow across four clinical genres (Assessments, Correspondence, Referrals and Care plans) to understand how LLMs grammatically construct medical authority and patient agency through linguistic choices. While LLMs produce coherent, terminology-appropriate texts that approximate clinical practice, systematic divergences remain, including registerial shifts, insufficient clinical specificity, and inaccuracies in medication use and diagnostic procedures.

Figures (1)

• Figure 1: The network for modality annotation