DemDx AI-driven triage platform development, stage 1
Research type
Research Study
Full title
DemDx AI-driven Ophthalmology triage platform development - Stage 1: data collection and training of Machine Learning (ML) models
IRAS ID
290843
Contact name
Anika Kadchha
Contact email
Sponsor organisation
Moorfields Eye Hospital NHS Foundation Trust
Duration of Study in the UK
1 years, 8 months, 13 days
Research summary
Introduction: Demand for eye services is expected to grow at a rate of 5% annually, in step with an older demographic. The Royal College of Ophthalmology predicts an additional 1.56 million outpatient attendances per year for Glaucoma and Age Related Macular Degeneration, in 15 years. Artificial intelligence techniques, including machine learning (ML) algorithms, could support AHPs in improving the triage process's accuracy and speed. ML algorithms used in general emergency departments have shown that they might be better than clinicians at predicting the need for admission and more sensitive and specific than normal emergency room protocols. This study is a continuation of a previous study where ML algorithms were trained and evaluated on approximately 2,584,008 NHS letters from Moorfields Eye Hospital. Our ML model reached 79% as F1 score using the most informative features.
Objectives: This project aims to train and test ML algorithms to be integrated into Dem Dx’s novel Ophthalmologic Triage Platform.
Methods: This will be a prospective observational study in the Adult A&E of Moorfields Eye Hospital, City Road, with no patient recruitment necessary. Clinical presentation of cases triaged will be input in a data collection form by the clinical team (Ophthalmology Consultants, Senior Registrars, and Nurses) during the consultations, using SNOMED keywords. This will be combined with patient data, including clinical history, examinations, investigations results, final diagnosis and onward referrals from the PAS system, and then used to build ML models to predict onward triage outcomes, including referrals and investigations requests, based on the potential diagnosis. An initial accuracy of 80% for the most common Ophthalmology diagnosis is aimed using 15,000 samples from clinical consultations, expected to be collected in the first 6-months. The ML algorithms will be updated on an ongoing basis until the project's end.
REC name
London - Riverside Research Ethics Committee
REC reference
21/LO/0294
Date of REC Opinion
17 May 2021
REC opinion
Further Information Favourable Opinion