Summary of findings'. '.

Question	What is the diagnostic accuracy of smartphone applications for detecting cutaneous melanoma in adults?
Participants	Adults with suspicious skin lesions
Prior testing and prevalence	Studies did not report the basis for participant selection. One selected a sample of lesions previously imaged during routine care just before excision of the lesion. The second study evaluated the test on patients who had been referred for further screening of the lesion by a specialist. Prevalence of melanoma was 18% and 35%.
Settings	Secondary care
Target condition(s)	Invasive melanoma and atypical intraepidermal melanocytic variants
Index test	Smartphone applications intended for use by the general public. Lesions not visualised by applications excluded
Reference standard	Histology
Action	If accurate, positive results of smartphone applications will help to highlight lesions of concern to the lay public, promoting earlier diagnosis of melanoma and reducing consultations for benign lesions.
Limitations
Risk of bias	Patient selection methods at high risk of bias due to selective inclusion of lesion types (2/2) and use of a case‐control design (1/2). Test interpretation was blinded to reference standard and pre‐specified for artificial intelligence‐based diagnosis (2/2). Reference standard blinding not described. Timing of index and reference standards not reported. Exclusions due to test failures were not reported (1/2) or their final diagnoses were not described (2/2)
Applicability of evidence to question	High concerns about applicability due to unrepresentative participant samples with high disease prevalences (2/2). Test not applied and interpreted by the intended user of the application (2/2). Reference standard interpretation by experienced histopathologists was not described (1/2).
Total number of studies:2
Detection of melanoma
Quantity of evidence	Number of studies	2	Total participants with test results	332	Total with target condition	86a
Findings	Across the four artificial intelligence‐based applications that classified lesion images (photographs) as either melanomas (one application) or as high risk or 'problematic' lesions (three applications), sensitivities ranged from 7% (95% CI 2% to 16%) to 73% (95% CI 52% to 88%) and specificities from 37% (95% CI 29% to 46%) to 94% (95% CI 87% to 97%). This means that between 27% and 93% of invasive melanoma or atypical intraepidermal melanocytic variants were not picked up as high risk by the automated applications (or as melanomas by one of the four applications). With a prevalence of melanoma ranging between 18% and 37% for these evaluations, the number of melanomas missed was 7 to 55. The single application using store‐and‐forward review of lesion images by a dermatologist had a sensitivity of 98% (95% CI 90% to 100%) and specificity of 30% (95% CI 22% to 40%); the dermatologist missed one melanoma. The number of test failures (lesion images analysed by the applications but classed as unevaluable and excluded by the study authors) ranged from 3 to 31 (or 2% to 18% of lesions analysed). The store‐and‐forward application had one of the highest rates of test failure (15%). At least one melanoma was classed as unevaluable in three of the four application evaluations, the highest number of melanomas excluded by the dermatologist evaluating the store‐and‐forward images (6/60 melanomas assessed).

^aOf the 60 melanomas included in one study, the four applications successfully analysed 54 to 60.