The automated risk assessment of lesions by a popular smartphone app appears to be at odds with diagnoses made by a dermatologist, suggesting that many similar skin assessment apps fall short in identifying a patient’s true lesion risk compared with active clinician assessment. This is according to study data in the Journal of the European Academy of Dermatology and Venereology.
Participants from the National Skin Cancer Day, an annual campaign in The Netherlands, were enrolled at multiple academic hospitals across the country (n=125). Prior to blinded rating by a dermatologist, participants indicated ≤2 lesions that caused them concern. These lesions were analyzed by the automated risk assessment of a popular smartphone app (SkinVision app) designed to assess skin lesions. Ratings from the app were compared with the assessment and diagnosis of the clinical dermatologist.
In the cohort, a total of 199 lesions were assessed by both the app and the dermatologist. The app was unable to take a picture of 45% of the lesions. Approximately 67% and 70% of lesions diagnosed by the dermatologist as benign nevi or seborrheic keratoses were categorized as either high risk or medium risk, respectively, by the app. Of 9 basal cell carcinoma cases, the app was unable to perform analysis on 4 atypical nevi and 1 lentigo maligna. There was poor interobserver agreement between the automated risk assessment’s ratings and the diagnoses made by the dermatologist (weighted kappa = 0.02; 95% CI, 0.08-0.12; P =.74).
Limitations of the study included the lack of histopathologic assessment and the app’s inability to perform assessment in nearly half of cases.
“Remarkable in this study was the fact that 45% of the lesions could not be analyzed by the app,” the investigators noted. They agreed that “regulations integrating scientific research and validation in real-life situations are urgently needed to provide consumers with reliable information” regarding smartphone apps and their ability to assess health concerns.
Chung Y, van der Sande AAJ, de Roos KP, et al. Poor agreement between the automated risk assessment of a smartphone application for skin cancer detection and the rating by dermatologists [published online August 19, 2019]. J Eur Acad Dermatol Venereol. doi:10.1111/jdv.15873