Medical Technology · uk
AI triage for skin cancer frees up clinic time, while UK real-world study also reveals missed-diagnosis risks
Two UK hospitals incorporated autonomous AI into their suspected skin cancer referral pathways, estimating savings of 2,851 hours of physician time over 16 months. The prospect of increased capacity is promising, but six missed cases are a reminder that safety monitoring must continue throughout the system’s operation.
Referrals for suspected skin cancer often require specialist assessment, only for the lesions to turn out to be benign. With clinic time limited, appropriately triaging lower-risk cases has become an important issue for healthcare services. A study at two UK hospitals tested whether artificial intelligence could directly take on some triage decisions and estimated how much physician time it could free up.
The prospective real-world study, presented at the 2026 European Academy of Dermatology and Venereology (EADV) annual meeting, included 8,391 patients, representing 94% of urgent suspected skin cancer referrals at the two hospitals; overall, 86% of patients consented to autonomous AI decision-making. Following initial validation, the system was put into use to analyze clinical photographs and dermoscopic images captured with smartphones. Cases classified as benign could have their referral episode closed directly, while higher-risk cases were referred for review by a teledermatologist.
After ineligible cases were excluded, AI directly triaged 31% and 25% of patients at the two hospitals, respectively, without physician review. The researchers estimated that, compared with traditional face-to-face consultations, this pathway saved 2,851 hours of physician time over 16 months. Assuming 20 minutes per consultation, this would provide capacity for more than 8,500 additional clinic visits. However, this represents potential capacity calculated from time savings, rather than an actual increase in the number of visits.
The change in the pathway was also reflected in subsequent management: compared with standard teledermatology services, the proportion requiring routine follow-up fell from 27% to 12%; the biopsy rate was 27%, lower than the 43% seen with traditional face-to-face care. These two comparisons used different comparator pathways and cannot be combined to describe a before-and-after change in the same group of patients. Whether reduced interventions lead to better long-term outcomes also remains to be established.
The safety data came from a national dataset that included these two hospitals and cannot be directly treated as having the same scope as the deployment study of 8,391 patients. Monitoring showed sensitivity above 98% for invasive melanoma, squamous cell carcinoma, and basal cell carcinoma, with specificity of 72.1%. High sensitivity means that most cancers can be identified, but it does not guarantee that no lesion will be missed.
Post-market surveillance nevertheless identified six cases in which referrals had been incorrectly closed, including five basal cell carcinomas and one melanoma in situ. The study report stated that no adverse outcomes had been identified during the available follow-up period; this does not mean that the long-term effects of delayed diagnosis have been ruled out. The missed cases also highlight the need for ongoing follow-up and mechanisms to identify problematic cases again when using autonomous triage.
The current findings were presented at a conference, and the two reports did not provide the full number of patients in the national safety dataset, confidence intervals for each cancer type, or detailed follow-up durations. The available information is also insufficient to assess performance across skin tones and different healthcare settings. The reports stated that the device used bears a CE mark and is a Class III medical device, but obtaining the mark cannot replace safety evaluation after deployment.
Dermatology Times also disclosed that the device used in the study was funded by Chelsea & Westminster Hospital NHS Foundation Trust and that the study received partial funding from La Roche-Posay. Lead author Lucy Thomas also serves as an independent consultant and clinical advisory board member for Skin Analytics. These relationships need to be considered when interpreting the results. For hospitals, the next priority is to establish whether the time saved can translate into more timely care, while keeping the risk of missed diagnoses under monitoring that allows cases to be traced and improvements to be made.