Medical Artificial Intelligence · uk
Triaging Low-Risk Skin Lesions First: UK Autonomous AI Pilot Frees Up Dermatology Consultation Time
A 16-month evaluation at two UK hospitals found that around one-quarter to one-third of patients could complete their suspected skin cancer referral pathway following AI assessment. The specialist time saved offers the potential to expand services, but six missed cancers also underscore the need for ongoing follow-up to support autonomous decision-making.
Referrals for suspected skin cancer often require a specialist to determine whether a lesion is dangerous. When large numbers of benign lesions enter the same urgent care pathway, limited consultation time becomes harder to allocate. A real-world evaluation at two UK National Health Service (NHS) hospitals tested the use of artificial intelligence to directly triage some low-risk cases. The results showed that around one-quarter to one-third of patients could complete their referral pathway without further physician review, freeing up time for specialist services.
Presented at the 2026 European Academy of Dermatology and Venereology (EADV) annual congress, the study tracked operations at two hospital sites over 16 months. It covered 8,391 patients, representing 94% of local urgent referrals for suspected skin cancer; 86% of patients consented to autonomous decision-making. The study used Skin Analytics’ DERM system, a CE-marked Class III AI medical device, with autonomous triage introduced only after initial validation.
DERM analyzes smartphone photographs of lesions and dermoscopic images to classify risk. Cases assessed as benign can be discharged directly from the pathway, with patients receiving guidance on monitoring subsequent changes in their lesions. Higher-risk cases and those meeting exclusion criteria are referred to a teledermatologist for review, with outpatient appointments or surgery arranged when necessary. Patients unsuitable for image-based triage continue to receive face-to-face care. The evaluation therefore examined an entire care pathway, rather than just an image recognition model.
After excluded cases were removed, the actual autonomous discharge rates at the two hospitals were 31% and 25%, respectively. Teledermatologists discharged a further 24% and 25% of patients, respectively. The study estimated that, compared with conventional face-to-face care, approximately 2,851 hours of physician time could be saved in total, equivalent to an approximately 62% increase in consultation capacity. Assuming 20 minutes per consultation, this translates to more than 8,500 outpatient appointments. This is potential capacity calculated from time savings and cannot be interpreted directly as the hospitals having already completed that many additional consultations.
Care arrangements also changed: compared with the standard teledermatology pathway, the proportion requiring routine follow-up fell from 27% to 12%. The biopsy rate was also lower than the 43% reported for conventional face-to-face care; materials released by EADV listed it as 27%, while HealthDay reported rates ranging from 26% to 29% across the autonomous pathways. However, these pathway comparisons alone are insufficient to establish that all avoided biopsies were unnecessary procedures, and they have yet to answer whether patients received treatment sooner as a result.
The safety data need to be considered separately. A national dataset that included the two study sites showed that the system’s sensitivity exceeded 98% for each of invasive melanoma, squamous cell carcinoma, and basal cell carcinoma, with a specificity of 72.1%. These performance figures were not calculated solely from the 8,391 patients in this evaluation. Post-market surveillance nevertheless identified six cancers in patients who had been incorrectly discharged, including five basal cell carcinomas and one melanoma in situ. No adverse outcomes were identified during the available follow-up period, but this does not mean that missed diagnoses carry no risk. It also highlights the importance of ongoing monitoring and mechanisms for patients to return for care.
The findings have so far been released as a conference poster and still need to be replicated in larger populations and different healthcare settings to determine whether time savings and diagnostic safety can both be achieved. The study equipment was funded by Chelsea & Westminster Hospital NHS Foundation Trust, and the study also received partial support from La Roche-Posay. Lead author Lucy Thomas also serves as a consultant to Skin Analytics and a member of its clinical advisory board. These relationships should be considered alongside the results. For autonomous medical AI, obtaining medical device status is only the starting point; the real test is whether errors can be continually identified and corrected once it enters routine care.