AI03 Steps towards safely deploying the world’s first autonomous artificial intelligence as a medical device for skin cancer into a National Health Service teledermatology pathway

Abstract In December 2024, autonomous artificial intelligence as a medical device (AIaMD) was successfully deployed into the urgent suspected skin cancer pathway at an acute National Health Service (NHS) trust. This milestone followed a 2-year partnership between an award-winning teledermatology service and an NHS-awarded AI provider using a UK Conformity Assessed (UKCA) class IIa AIaMD [Thomas L, Hyde C, Mullarkey D et al. Real-world post-deployment performance of a novel machine learning-based digital health technology for skin lesion assessment and suggestions for PMS. Front Med (Lausanne) 2023; 10: 1264846]. Since its launch, this AI-enabled teledermatology service has assessed 10 737 cases. The performance of the AIaMD was monitored using a robust postmarket surveillance strategy, based on recommendations from a White Paper commissioned by NHS England. Prior to transitioning to an autonomous AI pathway, local PMS data showed that the current version of the AIaMD was achieving a sensitivity of 97.3% [95% confidence interval (CI) 93.6–98.8] for all skin cancers, with a benign specificity of 74.0% (95% CI 72.0–75.9). Across 2571 lesions in this dataset, the negative predictive value for correctly ruling out all types of skin cancer was > 99.7% (95% CI 99.2–99.9). Real-world vigilance data from the AIaMD’s deployment nationally included > 135 000 NHS patients, and demonstrated that the AIaMD consistently met or exceeded predetermined sensitivity performance targets set at the upper end of what specialists achieve in the literature. During the partnership, data from appropriately consented patients were utilized to further train and validate the AIaMD, ensuring that AI improvements are driven by real-world patient data that reflect its intended use. Bolstered by the AIaMD’s strong diagnostic performance and validated safety, we leveraged our department’s teledermatology expertise to fully automate our skin cancer pathway. As a vanguard site we were committed to ensuring that the necessary due diligence was completed to facilitate safe and effective implementation. We will provide an overview of the key components involved in this process, including coproductive approaches with stakeholders, and details of our comprehensive PMS. This includes standardization of patient information provided upon discharge, a unique safety net in the form of a text message survey at 6 and 12 months postdischarge, and efforts to reduce health inequities for individuals with darker skin types (by currently routing these patients to a second opinion by a consultant dermatologist). We will present comparative performance data illustrating the evolution of our service from conventional face-to-face clinics, to teledermatology, non­autonomous AI and finally autonomous AI pathways. Finally, we will demonstrate the positive impact this transition has had on increasing capacity within both urgent cancer and routine dermatology at our trust.