What is Assign AI?
Assign AI suggests which patients may need input from a clinician and assigns their requests to a dedicated inbox. From this inbox, receptionists can more quickly assign requests to clinicians or clinicians may work directly from it to save time, thus reducing workload and improving patient safety.
Assign AI can be switched on and off from the Unassigned inbox. When it is enabled for the first time, a new ‘Clinical’ group inbox is created. Any new requests that are then submitted that Assign AI thinks require review by a clinician are automatically assigned there. It does not assign requests already in the Unassigned inbox to the Clinical inbox if they need clinical input.
Assign AI's suggestions are meant to assist, not replace clinical judgment. You should therefore still triage and reply to all Patchs requests ASAP.
What are the benefits of Assign AI?
Without Assign AI, initial triage decisions are made by receptionists. The advantages of Patchs doing this instead are that it:
- Improves patient safety by speeding up triage decisions, helping requests that need clinician input get reviewed quicker
- Reduces workload by suggesting triage decisions
- Improves the accuracy of the initial triage, because it's learned how to triage from GPs
- Reduces the clinical risk placed on receptionists when triaging, by providing suggestions
How accurate is Assign AI?
The most important thing for Assign AI is its ability to identify patients that need input from a clinician so they get seen sooner. The best way to do this is to measure the percentage of these requests it gets right, which is called the sensitivity. The current version of Assign AI has a sensitivity of 91%. Nothing is 100% accurate, even human GPs. It is unknown how well human GPs perform the same task as Assign AI so we can compare them, but we are conducting research to find out.
Further detail on the performance of all AIs is given in the Clinical Safety Case Report.
What should I do if I disagree with a suggestion from Assign AI?
If Assign AI thinks a request needs input from a clinician when it doesn’t (false positive)
- If you are a receptionist, you should not assign it to a clinician, and deal with it yourself.
- If you are a clinician, you should not select 'Me' when answering the 'Who is most appropriate to resolve this?' question when making a triage decision.
If Assign AI misses a request that needs input from a clinician (false negative)
- If you are a receptionist, you should assign the request to a clinician.
- If you are a clinician, you should select 'Me' when answering the 'Who is most appropriate to resolve this?' question when making a triage decision.
Because Patchs is constantly learning, this will help it correct its mistakes in future.