Sensitivity, specificity, and diagnostic accuracy of three dermoscopic algorithmic methods in the diagnosis of doubtful melanocytic lesions: the importance of light brown structureless areas in differentiating atypical melanocytic nevi from thin melanomas

Abstract

Background: Over the past decade numerous epiluminescence microscopy (ELM) criteria and algorithmic methods have been developed to improve the diagnosis of cutaneous melanocytic lesions.

Objective: Our purpose was to compare the sensitivity, specificity, and diagnostic accuracy of 3 algorithmic methods (pattern analysis, ABCD rule of dermoscopy, and the 7-point checklist) on a series of highly atypical melanocytic lesions. We also determined the diagnostic value of distinct ELM structures by evaluating their frequency in these lesions.

Methods: A total of 198 consecutive atypical macular melanocytic lesions were studied. ELM assessment was based on the presence or absence of 23 dermoscopic features. Two ELM-experienced dermatologists classified each lesion as benign or malignant using the pattern analysis, the ABCD rule of dermoscopy, and the 7-point checklist method. After surgical excision, 102 lesions were histologically diagnosed as Clark's nevi and 96 as thin melanomas (TMs) (mean tumor thickness, 0.3 mm). ELM and histologic diagnoses were then compared to assess the sensitivity, specificity, and diagnostic accuracy as well as positive and negative predictive values (PPV and NPV, respectively) for TMs of the 3 algorithmic methods. Univariate and multivariate analyses were performed to determine which ELM criteria were most strongly associated with TM.

Results: Of the melanocytic lesions studied, 82.3% were correctly diagnosed by using pattern analysis (85.4% sensitivity, 79.4% specificity, 79.6% PPV, and 70.8% diagnostic accuracy), compared with correct diagnosis of 79.3% (84.4% sensitivity, 74.5% specificity, 75.7% PPV, and 67.8% diagnostic accuracy) and 71.2% (78.1% sensitivity, 64.7% specificity, 67.6% PPV, and 57.7% diagnostic accuracy) with the ABCD and the 7-point checklist methods, respectively. The 7-point checklist yielded the highest number of false-negative results (21.8%) with respect to the ABCD rule (15.6%) and pattern analysis (14.6%). Univariate analysis showed that an atypical pigment network, a pigment network with sharp margins, irregular nonuniform brown globules, a nonuniform pigment distribution, homogeneous areas, and light brown structureless areas were the most sensitive and specific ELM features for TM. A backward stepwise logistic regression analysis revealed that the criterion with the strongest TM association was light brown structureless areas (odds ratio = 27.9; 95% confidence interval, 8.6-90.9).

Limitations: The presence and value of light brown structureless areas should also be investigated in clinically nonatypical macular melanocytic lesions.

Conclusion: The pattern analysis method showed the highest sensitivity, specificity, and diagnostic accuracy for TM. Light brown structureless areas were both a statistically significant discriminator and the most reliable predictor of TM (PPV = 93.8%, positive likelihood ratio = 16). Therefore the use of this previously underestimated ELM criterion may not only improve diagnostic performance of equivocal macular melanocytic lesions but also significantly decrease the rate of false-negative results obtained with the 7-point checklist method.