Urgent skin cancer referrals are rising, and autonomous AI tools have been proposed as a solution to the significant pressure on dermatology services. Using publicly available cancer waiting time data from 24 NHS trusts, this real-world interrupted time series and meta-analysis showed no consistent pooled improvement in Faster Diagnosis Standard breaches (the proportion of patients referred with suspected cancer waiting over 28 days from referral to diagnosis) following deployment of an AI-assisted lesion triage system, with marked heterogeneity ranging from benefit to deterioration across trusts. These findings highlight that the impact of diagnostic AI is highly context-dependent and underscore the need for robust local evaluation and cost-benefit assessment before widespread adoption.
Carson et al. (Thu,) studied this question.