An LLM-based vetting engine.
Multiple
Multiple
Concluded
This project has been withdrawn. After much work, it was determined that the success of this product hinges on the ability to integrate the solution with the EHR system which at this point in time is not a possibility at GSTT. Referral and vetting projects are better solved with optimisation of the referral pathway than dedicated processing software after the referrals are received.
The RAVE project aims to build a generalisable AI-driven software for performing automated vetting of General Practitioner (GP) referrals to make the secondary care referral vetting process more efficient and saves clinician time.
To realise the project’s intended benefits of cost savings, gains in resource efficiency, and decrease of waiting times for patients, we’ve decided to tackle this issue by using a hybrid model. In this context, we use an LLM to extract meaningful information from referral documents using Natural Language Processing (NLP) and evaluate this information using clinical decision trees to output a suggested referral decision.
The outputs of RAVE are intended to supplement the existing referral vetting process and should be interpreted as advisory information, rather than automatically actionable decisions. This means that clinical teams are always kept in the loop and ultimately have the final decision on how referrals are accepted, redirected, or rejected at the Trust.