Beyond Dictation: Generative Al as a Scalable Solution for Patient-Centred Letter Writing in Oculoplastic Surgery
Author: Antonia Vieira Vender
Base Hospital / Institution: Moorfields Eye Hospital
ePoster presentation
Abstract ID: 25-369
Purpose
Robust, patient-specific documentation is integral to oculoplastic practice: it sustains continuity of care, limits medicolegal liability, and preserves patient confidence. Escalating clinical throughput and administrative demands, however, increasingly constrain the time available for meticulous letter preparation.Our aim was to determine whether a large-language model (LLM), prompted with specialty-tailored instructions, can generate oculoplastic correspondence of comparable or superior quality and readability to that produced by clinicians or medical secretaries.
Methods
We retrospectively analysed all outpatient letters generated for consecutive attendees at a tertiary oculoplastic clinic between 1 January and 31 March 2025 (n = 216). Letters were stratified by author— clinician (n = 96), medical secretary (n = 72), or LLM (n = 48) —and independently evaluated using a modified Ensuring Quality Information for Patients (EQIP) instrument and the Flesch Reading Ease Score (FRES).
Results
EQIP scores differed significantly across groups (p < 0.01). Clinician-authored letters scored lower than both secretary-generated and LLM-generated letters, reflecting frequent reliance on jargon, abbreviations, and acronyms. The LLM achieved a higher mean EQIP score than medical secretaries (p = 0.03), primarily by explicitly articulating treatment risks, benefits, natural history, and alternatives to suggested management plan. FRES analysis similarly identified lower readability in clinician letters (p < 0.01). No difference was found in FRESH between LLM and secretaries (p = 0.08).
Conclusion
A bespoke-prompted LLM can rapidly translate clinical notes into structured, patient-centred correspondence, outperforming traditional medical secretarial workflows in content quality and matching their readability. With rigorous governance, domain-specific fine-tuning, and systematic validation, generative Al constitutes a credible adjunct for enhancing precision, efficiency, and clarity of communication in high-volume oculoplastic services.
Additional Authors
| First name | Last name | Base Hospital / Institution |
|---|---|---|
| Victoria | Ngai | Northwick Park Hospital |
| Basil | Suresh | Watford General Hospital |
| Piers | Daniel | Moorfields Eye Hospital |
| Mohsan | Malik | Moorfields Eye Hospital |
| Claire | Daniel | Moorfields Eye Hospital |
