P007 Real-world multicentre clinical performance evaluation of an artificial intelligence as a medical device across darker skin tones

Abstract

Patients with darker skin tones have a lower incidence of skin cancer, with < 0.5% of cases melanoma occurring in Black and Asian patients in the UK (Delon C, Brown KF, Payne NWS et al. Differences in cancer incidence by broad ethnic group in England, 2013–2017. Br J Cancer 2022; 126: 1765–73). However, Black and Asian patients are less likely than White patients to have skin cancer diagnosed via the urgent suspected cancer (USC) pathway (48.9% and 60.6%, respectively, vs. 67.3%) and have lower USC referral conversion rates (0.4% vs. 6.6%), suggesting that skin cancer assessment pathways are subject to ethnic disparities in existing models of care. This study reports the performance of a CE class III artificial intelligence as a medical device (AIaMD) to detect melanoma, squamous cell carcinoma (SCC) and basal cell carcinoma (BCC) in patients with darker skin tones. The AIaMD is intended for screening, triage and assessment of nonacral and nonsubungual lesions suspicious for skin cancer and is approved for use across all skin tones. This retrospective, multicentre study included 11 928 lesions in patients with Fitzpatrick skin types V or VI assessed between July 2021 and December 2025. Overall, 32 melanomas (18 invasive), 45 SCCs (including 11 keratoacanthoma) and 100 BCCs were confirmed histologically. Lesions classified as benign by the AIaMD underwent second-read review by a dermatologist. The sensitivities of referral for melanoma (32 of 32), SCC (45 of 45) and BCC (100 of 100) were all 100% (95% confidence interval 97.9–100). All cancers were correctly triaged to urgent dermatologist review. The AIaMD demonstrated high sensitivity for all skin cancers in patients with Fitzpatrick type V or VI skin and may help reduce existing biases in skin cancer referral pathways for patients with darker skin tones to support earlier diagnosis. The authors are employed by the AIaMD provider.