The test set included a total of 333 melanocytic lesions belonging to 314 patients. Histopathological examination revealed 120 in situ melanomas (36%), 167 common acquired nevi (50·2%) and 46 Spitz or spitzoid nevi (13·8%) (Table 2). In situ melanoma were more frequently seen in men (55·8%); in older patients [median: 63, interquartile range (IQR): 49·25–72·75] and both in situ melanoma and nevi were mainly located on the back (40% and 41·8%, respectively). In situ melanoma looked flat (100%), heavily pigmented (73·3%) and more than 6 mm in diameter (76·7%).
Dermoscopically, in situ melanomas were typified by an atypical pigment network (69·2% of cases) and regression (38·3%). In nevi, an atypical pigment network was present in 47·9% of cases. The seven–point score (> 1) was suggestive of an atypical melanocytic lesion (97·5% in in situ melanoma and 92·5% in nevi) in both populations.
Confocal examination revealed the presence of epidermal pagetoid cells in 73·3% (88/120) of in situ melanomas and 41·8% (89/213) of nevi: nevi bearing pagetoid cells were mainly compound nevi (44/89, with histological inflammation or regression in a third of cases) and Spitz nevi (26/89); the remaining 19 were junctional nevi (10/89), sclerosing/recurrent nevi (8/89) and one combined nevus. In both in situ melanoma and nevi groups cells were mainly dendritic–shaped and showed a similar distribution, being widespread in the majority of in situ melanoma and nevi typified by epidermal pagetoid spreading [64% (56/88) and 61% (54/89), respectively] (Table S1; see Supplementary Information).
Regarding the main architectural pattern at the dermoepidermal junction, there were no striking differences between the two populations. At the dermoepidermal junction, 30·5% of nevi showed 'dense' nests, whereas in in situ melanomas 'dense' and 'dense and sparse' nests were equally represented (9·2% and 12·5%, respectively); cytological atypia appeared focally located in 36·7% and 31·5% of in situ melanomas and nevi, respectively. However, cytological atypia were widespread in half melanomas (50%).
Spearman's analysis showed that several clinical, dermoscopic and confocal features significantly correlated with the diagnosis of in situ melanoma: age (ρ = 0·422, P < 0·001), female sex (ρ = −0·167, P = 0·002); the dermoscopic finding of an atypical network (ρ = 0·206, P < 0·001), atypical vessels (ρ = −0·115, P = 0·036), irregular dots/globules (ρ = −0·130, P = 0·017) and regression (ρ = 0·191, P < 0·001). The presence of atypical vessels was inversely related to melanoma diagnosis because of the high percentage of this dermoscopic feature in Spitz nevi (11/46) and the presence of few papular lesions with polymorphic vessels among common nevi (12/167), which increases the rate of atypical vessels in benign lesions.
Significant RCM features were: the presence of pagetoid cells (ρ = 0·356, P < 0·001), pagetoid cell shape (ρ = 0·330, P < 0·001) and pagetoid cell location (ρ = 0·347, P < 0·001), clod pattern (ρ = −0·109, P = 0·047) and nest type (ρ = −0·148, P = 0·007), cytological atypia (ρ = 0·383, P < 0·001) and the presence of melanophages (ρ = −0·164, P = 0·003).
The univariate logistic regression confirmed that almost all factors highlighted by Spearman's correlation were good predictors of in situ melanoma diagnosis (Table 3). More specifically, dermoscopic criteria significantly associated with a diagnosis of in situ melanoma were the presence of an atypical pigment network and regression [odds ratio (OR) 2·44, 95% confidence interval (CI) 1·52–3·91 and OR 2·39, 95% CI 1·46–3·92, respectively]. The confocal finding of pagetoid spreading in the epidermis was a strong predictor for a melanoma diagnosis (OR 5·35, 95% CI 3·11–9·20), especially if characterized by roundish and dendritic cells located peripherally or on the entire surface of the lesion. Junctional cytological atypia was the strongest predicting factor for a diagnosis of melanoma, conferring a nine times higher risk of melanoma when widespread (OR 8·98, 95% CI 4·66–17·32).
Multivariate Regression Analysis
In the logistic forward stepwise multivariate regression model (Table 3), the effect of each variable, significantly associated with histological diagnosis in the univariate analysis was adjusted for the effect of the remainder variables in the model.
The dermoscopic observation of an atypical network and regression were shown to be independent predicting factors for in situ melanoma diagnosis (OR 3·44, 95% CI 1·70–6·97 and OR 4·17, 95% CI 1·93–9·00, respectively). On RCM, the presence of epidermal pagetoid spread confers a 2·8 times higher risk of melanoma (OR 2·83, 95% CI 1·32–6·04) and the presence of junctional cytological atypia was confirmed as the strongest RCM predicting factor (OR 3·39, 95% CI 1·38–8·30 when focal and OR 8·44, 95% CI 3·21–22·16 when widespread).
The confocal observation of dense nests and melanophages is not frequent in in situ melanoma compared with nevi (OR 0·30, 95% CI 0·12–0·77 and OR 0·44, 95% CI 0·22–0·88, respectively) (Figure 1).
Dermoscopic and confocal predictors in the diagnosis of in situ melanoma. Dermoscopic features independently associated with in situ melanoma diagnosis were atypical pigment network (a) and regression (b) (20 × original magnification). Upon reflectance confocal microscopy, the findings of epidermal pagetoid spread (arrows) (c) and junctional cytological atypia with single melanocytes (arrows) surrounding dermal papillae (asterisks) confer a higher risk for malignancy. Whereas, the confocal findings of dense nests (arrowheads) (e) and melanophages (circles) (f) are not associated with in situ melanoma diagnosis (scale bars, 250 μm).
In order to exploit the improvement in diagnostic accuracy provided by RCM, a multistep diagnostic process considering both the dermoscopic and confocal criteria from the multivariate analysis was developed (Figure 2). A score for predicting in situ melanoma was developed assigning +1 and −1 points, respectively, to each of the dermoscopic or confocal criterion statistically positively and negatively associated with in situ melanoma in the multivariate analysis. Cytological atypia was awarded a value of +1 when focally represented and +2 when widespread. The presence of dense nests was inversely associated with in situ melanoma (OR 0·300) and was used in the score.
Multistep diagnostic algorithm to identify in situ melanoma using specific dermoscopic and confocal features. An overall score of two or more must lead to surgical excision. RCM, reflectance confocal microscopy.
The dermoscopic finding of an atypical network and/or regression adds one point each to the score.
From RCM, the presence of pagetoid infiltration adds one more point to the score. The finding of cytological atypia scores one if it is 'focal' and two if 'widespread'; observation of dense melanocytic nests and melanophages subtracts one point each from the final score. Examples are given in Figures 3 and 4.
A dermoscopically inconspicuous in situ melanoma (a; 20 × original magnification) showing, on reflectance confocal microscopy examination, dendritic pagetoid cells (arrows) covering the entire surface area (b) and atypical cells (arrows) surrounding dermal papillae (asterisks) at the dermoepidermal junction (c), which is characterized by an irregular ringed pattern (scale bars, 250 μm). Pagetoid spread and widespread junctional atypia confer a score of +3 using the algorithm and will lead to the removal of the lesion.
A dermoscopically atypical lesion characterized by an atypical pigment network (a; 20 × original magnification). The confocal examination showed a regular epidermis, devoid of pagetoid spreading (b) and a regular ringed junctional pattern without cytological atypia (c) (scale bars 500 μm). Using the algorithm, we obtain a score of +1, not enough to lead to surgical removal. Histology revealed a compound nevus.
An overall score of two or higher signifies melanoma with a sensitivity of 92·5% and a specificity of 61% (AUC = 0·827), and must therefore lead to surgical removal of that lesion.
The validation set (including a total of 100 lesions with 50 in situ melanoma) confirmed the good performance of the multistep scoring, with 92% sensitivity and 58% specificity levels (AUC = 0·819). In addition, the algorithm reached a high sensitivity (94·9%) on the panel of 59 early invasive melanomas.
The British Journal of Dermatology. 2018;179(1):163-172. © 2018 Blackwell Publishing