O01 Real-world effectiveness of artificial intelligence-assisted lesion triage on cancer waiting times

Abstract

There is an urgent need to improve delivery of NHS waiting time targets for cancer care. Dermatology is under particular pressure with a rapidly rising suspected skin cancer referral burden. Artificial intelligence (AI) is proposed as a solution, with SkinAnalytics Deep Ensemble for Recognition of Malignancy (DERM) AI diagnostic tool classified as a CE class III medical device for use in skin cancer triage and assessment. Implementation of this tool is claimed to reduce the need for up to 95% of face-to-face reviews. However, to date, analysis of the impact of DERM on cancer waiting times in England has not been reported. Our aim was to evaluate the real-world impact of DERM using publicly available cancer waiting time data. The breach percentage of cancer waiting times Faster Diagnosis Standard by month for NHS trusts adopting DERM was obtained and presented graphically. Interrupted time series analysis with random effects meta-­analysis was undertaken to investigate individual NHS trusts and the overall effect. Overall, 24 UK NHS trusts (1196 trust-months) that adopted DERM in skin cancer pathways between April 2021 and June 2025 were examined. The results showed substantial variation, with five (21%) showing significant improvement in postimplementation trends, compared with four (17%) significantly deteriorating. Meta-analysis revealed no significant pooled effect (pooled slope change +0.06 percentage points per month (95% confidence interval −0.36 to 0.48, P = 0.78), with high heterogeneity (I2 = 84%, Q = 140, P < 0.001). Baseline breach rates correlated weakly with level and slope changes (r = −0.28, P = 0.18 and r = 0.10, P = 0.64, respectively). Real-world analysis of cancer wait breaches shows marked variation in effectiveness of DERM implementation in NHS trusts, ranging from marked improvement to notable deterioration. The current absence of significant pooled effects suggests careful cost–benefit analysis is required to assess the value of these tools and the requirement to measure metrics beyond algorithm accuracy at a local NHS trust level when considering adoption.