Outcomes

Key findings

Several key findings emerged from the project:

1. Most research in the area of autonomous systems for patient triage focuses on the final stage (examination and assessment) of the triage process, with the primary focus on admission identification.
2. To date, the validation, deployment and clinical trial of autonomous systems for patient triage is very limited.
3. Autonomous and AI systems for patient triage tend to rely on complete data being available. However, in many real-world scenarios, only partial data can be collected. Therefore, data samples that lack some of the patient features are representative for what the such systems will encounter after deployment, and should not be excluded from their training and validation.
4. Our semi-structured interviews with healthcare professionals within the NHS revealed both support for future adoption of DAISY-like solutions and a series of concerns and expectations that, in the participants’ view, should be addressed prior to their deployment.
5. The DAISY benefits indicated by the healthcare professionals in our interviews range from reduction in patient wait times in the emergency departments nad assistance in the streamlining of the triage process to support in calling for appropriate diagnostics and for further patient examination, and identification of those unwell and requiring more immediate and urgent attention.
6. Trust was identified as a critical factor influencing the utility and wide-spread  adoption of the system.  Certain participants expressed hesitancy in uptake until such time as credibility, reliability, and regulatory approval of the system and the underlying algorithm could be demonstrated.
7. The study participants were also concerned that DAISY might not be reliably or securely implemented resulting in reputational vulnerability to both the clinician and the system itself. Additionally, explainability and transparency were recognised as important factors. Participants wanted to know the basis upon which determinations and underlying assumptions are made in the diagnosis and the reasons for doing so.
8. Last but not least, the idea that the practice of medicine is both an art and a science was identified by the participants. DAISY’s inability to practise medicine as an art is viewed as a potential shortcoming of the system.  
9. Overall, the study revealed the expectation of DAISY as a support, back-up or second resource in A&E triage, which requires careful  integration into the existing triage process.
10. Setting up the DAISY testbed revealed technical challenges around the integration of the vital parameter measurement devices (e.g., for measuring temperature, oxygen level, and blood pressure) - produced by different manufacturers, and having different specifications and level of documentation, etc., these are challenging to integrate into an operational end-to-end system.
11. Asking patients to use even simple devices to measure their vital parameters is error prone, e.g., hair in the way can introduce errors in measuring forehead temperature, and improper use of blood pressure monitors leads to inaccurate measurements.
12. Real-world data sets collected in the hospital are difficult to curate for use in the training and testing of AI and autonomous systems, due to omissions, inaccuracies and final diagnosis (i.e., after additional medical tests were carried out on the patient) being recorded instead of the A&E triage assessment needed (as the ‘ground truth’) for training and testing DAISY.
13. Securing the ethical approval for each stage of the project involving human participation (even for interview and questionnaire-based studies) required significant effort and time.
14. Both the technical work and studies conducted as part of the DAISY project confirm the important role that trustworthy autonomous systems can play in streamlining key stages of the A&E triage process.

Outputs

Impact