Improving the Accuracy of Artificial Intelligence Triage in Primary Care
Trial statusRecruiting
Trial phaseNot applicable
Trial typeInterventional
Biological sexAll
AgeNot listed
SponsorUniversity of Manchester
About this trial
WHY ARE WE DOING THIS? When patients contact their GP practice, the first step is to work out what kind of help they need and how quickly it's needed. This is called 'triage' and is important for patient safety.
Artificial Intelligence (AI) can help make triage faster. While AI is already being used in the NHS, we don't know how accurate it is or if it treats all patients fairly.
WHAT WILL WE DO?
We will collect anonymised data from patients that use an AI triage system called Patchs in GP practices in England. The project will last four years. We will analyse the data in four steps:
1. Look at data from GP practices using Patchs without AI triage to see how they currently triage patients and what problems they face. 2. Use data from GP practices using Patchs (both with AI on and off) to make the AI triage more accurate. 3. Check data from GP practices using Patchs with AI triage off to measure how well the updated AI system works. 4. Give the improved AI triage system to GP practices already using AI.
At each step, we will check whether patients from different backgrounds are treated fairly.
HOW WILL WE ANALYSE THE DATA? We will use statistical methods to compare the triage decisions made by the AI with those made by clinical staff. This analysis will also be used to check that the AI works fairly for patients from different backgrounds.
WHAT DIFFERENCE WILL WE MAKE? Our research will show the problems with triage and explain how an improved AI system could help patients get the care they need more quickly.
Eligibility criteria
This trial accepts healthy volunteersQualifiers
GP practices using the Patchs system
Disqualifiers
N/A
Trial design
Design model
Parallel
Treatments tested in this trial
AI triage
Other interventionPatchs AI automates parts of the triage and workflow process for GP practices when any patient uses the Patchs online consultation system. It is intended to assist, not replace human decision-making. Patchs AI aims to reduce GP practice workload by minimising manual tasks and improve patient safety by helping patients receive appropriate care sooner. Patchs AI uses information about the patient and their online consultation to suggest an urgency, clinical topic, staff role, and mode to conduct the consultation. Based on these suggestions it can provide patients with relevant health information, ask questions to elicit further information, and/or advise them to contact alternative care providers. Suggestions from Patchs AI can be accepted or rejected by GP practice staff and patients, which is used to monitor its safety and re-train the system.
No AI triage
Other interventionUse of the Patchs system with no AI triage i.e. manual triage only
Treatment groups
226,821 Participants
are divided into 2 treatment groupsGroup A: AI triageExperimental treatment 1 intervention
Group B: No AI triageActive comparator 1 intervention
Trial outcomes
Primary outcomes
1
F1 score
F1 score of each AI triage module (harmonic mean of its precision and recall)
Time frame
From enrollment to the end of the study, anticipated to be 4 yearsSponsors and contacts
Click on the lead sponsor to view all of their trials.
University of Manchester
Lead sponsor
University of Cambridge
Collaborator
Source ClinicalTrials.gov