Introduction: Guy’s & St. Thomas’ Urology Department runs a walk-in clinic which provides same-day appointments for their patients and urgent referrals. This service is staffed by junior doctors and advanced nurse practitioners with varying levels of specialty experience. The clinic often overruns due to patient demand and the need for senior input which negatively impacts patient experience and clinic flow. Recently the service has introduced a clinical decision making platform, DemDx, which utilises artificial intelligence (AI) to support clinical diagnosis.
To identify whether AI can improve efficiency in a urology walk-in clinic environment.
To assess if DemDx’s algorithm accurately matches the working diagnosis of clinicians.
To evaluate how DemDx can capture clinical decision making to provide an auditable pathway for quality improvement.
Methods: 254 urology algorithms were developed, based on input from consultant urologists and guidelines. This resulted in 63 end diagnoses and 51 bespoke clinical actions.
Baseline data was taken for the total time required to treat a patient and the number of times a senior was contacted.
The junior clinical team were trained and utilised the interface during their walk-in clinic service.
Anonymised clinical pathways were retrieved from the DemDx server and compared with the patient's electronic medical record (EMR).
The study covered 5 weeks of walk-in clinics with a total of 10 participants.
Results: 20 patients were collected from the DemDx group and 57 patients from the control group.
DemDx correctly diagnosed 14 out of 17 patients (82%) when compared to their EMR diagnosis.
It highlighted 2 pathways that needed to be added to the algorithm (penile fracture and ureteric stent pain). With these excluded the correct diagnoses increases to 93%.
No evidence of disruption or slowing down of the service (Time spent: Control 30[20 - 70] min; DemDX 25 [20 - 33] min; p=0.299 (median [IQR] wilxcox test)).
No difference shown in degree of senior support required (Senior Contacted: Control 10/57 (18%); DemDX 7/20 (40%); p=0.128 (fisher exact test with midp correction)).
100% of pathways and actions completed on the DemDx server were captured and auditable.
Conclusions: These results show potential for AI in assisting diagnosis in urology walk-in services. DemDx accurately mapped clinical presentations into pathways in a transparent and auditable manner. It also highlighted additional diagnostic pathways that needed to be added to the interface. This evidence provides scope for a larger project to evaluate how AI may reduce waiting times in clinics, increase the confidence and competence of clinical staff and improve the overall patient journey.