Summary of findings'. 'Summary of findings table.

Question	What is the diagnostic accuracy of dermoscopy, in comparison to visual inspection, for the detection of cutaneous invasive melanoma and atypical intraepidermal melanocytic variantsin adults?
Population	Adults with lesions suspicious for melanoma, including: those with limited prior testing (presenting in primary, community or private dermatology settings), and referred populations (presenting in secondary care or specialist skin cancer clinics).
Index test	Dermoscopy with or without the use of any established algorithms or checklist to aid diagnosis, including: in‐person evaluations (face‐to‐face diagnosis), and image‐based evaluations (diagnosis based on assessment of a clinical image).
Comparator test	Visual inspection
Target condition	Cutaneous invasive melanoma and atypical intraepidermal melanocytic variants
Reference standard	Histology with or without follow‐up to confirm absence of malignancy in benign‐appearing lesions
Action	If accurate, positive results ensure melanoma lesions are not missed but are appropriately excised (or referred), and those with negative results can be safely reassured and discharged.
	Number of studies		Total lesions			Total melanomas
Quantity of evidence	104		42,788			5700
Limitations
Risk of bias (in‐person; image‐based)	Potential risk for participant selection from use of case‐control type design (19 image‐based), inappropriate exclusion criteria (8; 25), or lack of detail (17; 27). All dermoscopy interpretation was blinded to reference standard diagnosis. Dermoscopy thresholds were clearly pre‐specified (25; 50). Low risk for reference standard (29; 63); high risk from use of expert diagnosis or > 20% of benign lesions with no histology (5; 11). Blinding of reference standard to clinical diagnosis reported only in one image‐based evaluation. High risk for participant flow (15; 26), due to differential verification (6; 15), and exclusions following recruitment (10; 16). Timing of tests was not mentioned in 23 (18).
Applicability of evidence to question (in‐person; image‐based)	Participants restricted to those with melanocytic lesions only (10; 35), or other narrowly defined groups (5 image‐based), or to those with histopathology results (29; 57), and included multiple lesions per participant (8 in‐person). High concern for dermoscopy (16; 57), with no description of diagnostic thresholds (8; 25), or reporting of average or consensus diagnoses (9; 35). Dermoscopic image interpretation blinded to clinical images (51 image‐based). Little information given concerning the expertise of the histopathologist (28; 50).
Findings
We included 104 study publications (providing data for 103 cohorts of lesions). We separated a priori 83 publications providing 86 datasets for evaluation of the primary target condition into in‐person (n = 26), and image‐based (n = 60), evaluations. Subsequent analysis confirmed differences in accuracy according to the different approaches to diagnosis (P < 0.0001). Analyses of studies by degree of prior testing revealed no obvious effect on accuracy; the study publications provided insufficient relevant information, and the majority of studies were apparently conducted in referred populations, which hampered our analyses. The findings presented are based on results for all studies regardless of position on the clinical pathway. Sensitivities at fixed specificities and specificities at fixed sensitivities are given for illustrative purposes only and should not be taken as indicative of actual test performance.
Test	In‐person visual inspection alone versus visual inspection plus dermoscopy: any algorithm or threshold
Data analysed	Visual inspection				13 datasets; 6740 lesions; 459 cases
	Dermoscopy				26 datasets; 23,169 lesions; 1664 cases
Resultsa	Sensitivity (95% CI) %		Fixed specificity		Fixed sensitivity		Specificity (95% CI) %
Visual inspection	76% (66 to 85)		80%		80%		75% (57 to 87)
Dermoscopy	92% (87 to 95)						95% (90 to 98)
Numbers applied to a hypothetical cohort of 1000 lesionsb
	TP	FN	FP	TN	TP	FN	FP	TN
At a prevalence of 5%	VI: 38 D: 46 ↑ 8	VI: 12 D: 4 ↓ 8	190	760	40	10	VI: 238 D: 47 ↓191	VI: 713 D: 904 ↑191
At a prevalence of 12%	VI: 91 D: 110 ↑19	VI: 29 D: 10 ↓ 19	176	704	96	24	VI: 220 D: 44 ↓176	VI: 660 D: 836 ↑176
At a prevalence of 21%	VI: 160 D: 193 ↑ 33	VI: 50 D: 17 ↓ 33	158	632	168	42	VI: 198 D: 40 ↓158	VI: 5935 D: 750 ↑158
Test:	Image‐based visual inspection alone versus visual inspection plus dermoscopy: any algorithm or threshold
Data analysed	Visual inspection				11 datasets; 1740 lesions; 305 cases
	Dermoscopy				60 datasets; 13475 lesions; 2851 cases
Results	Sensitivity (95% CI) %		Fixed specificity		Fixed sensitivity		Specificity (95% CI) %
Visual inspection	47% (34 to 59)		80%		80%		42% (28 to 58)
Dermoscopy	81% (76 to 86)						82% (75 to 87)
Numbers applied to a hypothetical cohort of 1000 lesionsc
	TP	FN	FP	TN	TP	FN	FP	TN
At a prevalence of 18%	VI: 85 D: 146 ↑ 61	VI: 95 D: 34 ↓ 61	164	656	144	36	VI: 476 D: 148 ↓328	VI: 344 D: 672 ↑328
At a prevalence of 24%	VI: 113 D: 194 ↑81	VI: 127 D: 46 ↓ 81	152	608	192	48	VI: 441 D: 137 ↓304	VI: 319 D: 623 ↑304
At a prevalence of 39%	VI: 183 D: 316 ↑ 133	VI: 207 D: 74 ↓ 133	122	488	312	78	VI: 354 D: 110 ↓244	VI: 256 D: 500 ↑244
Test	Results according to algorithm used to assist dermoscopy interpretation
	Datasets	Lesions; cases	Sensitivity (95% CIs) %	Specificity (95% CI) %	Numbers in a cohort of 1000 lesionsd
					TP	FN	FP	TN
In‐person					At median prevalence of 12%
No algorithm	8	4704; 849	88% (75 to 95)	87% (80 to 92)	106	14	114	766
Pattern analysis	6	4307; 296	92% (87 to 95)	92% (68 to 98)	110	10	70	810
ABCD at > 5.45 (or likely)	5	1438; 160	81% (62 to 92)	92% (82 to 97)	97	235	70	810
Image‐based					At median prevalence of 24%
No algorithm	24	4498; 941	76% (70 to 82)	79% (71 to 85)	182	58	61	699
Pattern analysis	20	4621; 989	83% (76 to 88)	87% (80 to 92)	199	41	99	661
ABCD at > 5.45	7	2471; 406	81% (60 to 92)	81% (69 to 89)	194	46	144	616
7PCL at ≥ 3	11	3408; 798	80% (63 to 91)	67% (51 to 80)	192	48	251	509
3PCL	7	1505; 363	74% (61 to 85)	60% (42 to 76)	178	62	304	456
3PCL: three‐point checklist; 7PCL: seven‐point checklist; ABCD(E): asymmetry, border, colour, differential structures (enlargement); CI: confidence interval; D: dermoscopy; FN: false‐negative; FP: false‐positive; TN: true‐negative; TP: true‐positive; VI: visual inspection

^aNumbers for a hypothetical cohort of 1000 lesions are presented for two illustrative examples of points on the SROC curves: firstly for the sensitivities of tests at fixed specificities of 80%; and secondly for the specificities of tests at fixed sensitivities of 80%.
 ^bNumbers estimated at 25th, 50th (median), and 75% percentiles of invasive melanoma or atypical intraepidermal melanocytic variants prevalence observed across 26 datasets reporting in‐person evaluations of dermoscopy added to visual inspection.
 ^cNumbers estimated at 25th, 50th (median), and 75% percentiles of invasive melanoma or atypical intraepidermal melanocytic variants prevalence observed across 60 datasets reporting diagnosis using dermoscopic images
 ^dNumbers estimated at median prevalence (50th percentile), of invasive melanoma or atypical intraepidermal melanocytic variants observed across 26 datasets reporting in‐person evaluations of dermoscopy added to visual inspection and then for 60 datasets reporting diagnosis using dermoscopic images