Comprehensive Summary
This study presents a new device, namely autonomous artificial intelligence as a medical device (AIaMD) into a real-world urgent skin-cancer referral pathway, which is the first time such a system has been used independently in clinical practice. Since launching, this device has assessed over 10,700 cases and data has shown that it achieved a 97.3% sensitivity for detecting skin cancer and a negative predictive value over 99.7%. Due to rigorous post-market surveillance and workflow safeguards, the authors argue this deployment demonstrates that AI can be safely and effectively used to triage skin lesions- potentially reducing unnecessary referrals and speeding up care for patients who need it.
Outcomes and Implications
Although the AI system achieved high sensitivity in studies, its implementation still carries risks and limitations. A substantial fraction of benign lesions may be flagged as suspicious, potentially leading to unnecessary referrals or biopsies rather than reducing workload as intended. There is limited evidence about how well it performs across diverse patient populations: data remain sparse especially for people with darker skin tones, raising concerns about diagnostic disparities. Additionally, non-cancer or non-urgent skin conditions may be deprioritized or delayed if routed out of urgent pathways incorrectly.