I recently heard an account of the difficulties a clinician faced when drafting a letter relating to a patient who had changed their gender. It was clear from the manner in which the clinician spoke about the matter that it had caused considerable professional concern, and that the reviewing process consumed a significant amount of valuable clinical time.

Potential scenarios that may contribute to the accidental inclusion of incorrect pronouns include:

Clinical management systems generate pre-populated draft letters. This can be problematic if the system hasn't been updated with the patient's current information, including their gender identity

Administrative staff are not aware of a patient's recent gender transition. This can lead to them using outdated pronouns in their responses to correspondence

The patient's transition occurred after the initial correspondence was received. In such cases, subsequent replies might inadvertently use outdated pronouns if the transition is not reflected in the patient's records.

When considering the issue, it occurred to me that current, advanced large language models (LLMs) could be well-suited to checking and correcting anonymised versions of such correspondence.

To test this idea, I utilised Gemini (version ‘1.5 Flash') to request appropriate gender rendering of a draft letter which contained both male and female pronouns. The LLM proved highly efficient in identifying and correcting these errors, reporting the changes made.

It is crucial to stress that if using a publicly accessible LLM, there is a risk that the prompt (input) text might be analysed. However, if the LLM is integrated with a word-processing package such as Google Docs or MS Word-365 and the account security meets the requirements of the Data Security and Protection Toolkit (DSPT), it seems less likely that any inadvertent inclusion of patient data in inputs for such text analysis would pose a data breach risk.