A risk scoring system for the differentiation between melanoma with regression and regressing nevi

Background: Spontaneous regression of melanomas is relatively common, its prevalence ranging from 10 to 35%. However, regressing nevi can exhibit worrisome feature and simulate melanoma both clinically and dermoscopically. Thus, the presence of regression can represent a confounding factor.

Objective: To investigate the frequency of dermoscopic patterns of "regression" in a series of benign and malignant melanocytic skin lesions, and to design an integrated scoring system. Scoring classifiers are very effective in selecting the significant parameters for discriminating two clinical conditions, thus can rapidly calculate a patient's risk for a given disease.

Methods: We selected a series of 95 regressing melanocytic lesions, including 50 regressing nevi and 45 melanomas with regression. For each lesion, 12 dermoscopic variables (i.e. five types of regression structures, five atypical pigmentation structures, atypical vascular pattern and pink areas) were examined by three expert in dermoscopy (blinded to the histological diagnosis). The dermoscopic evaluation was then combined with patient age, gender, body site and the maximum diameter of lesion. Concordance analysis with Cohen's kappa was performed between the three clinicians. A risk scoring system was designed by the leave-one-out cross-validation procedure to ensure model prediction power.

Results: The predictive score model revealed a sensitivity of 97.8% and a specificity of 75.5% in discriminating nevi and melanomas with regression. Using the score model, the diagnostic performance of the examiners increased by an average of 23.7% in sensitivity and 5.9% in specificity, compared with standard dermoscopic pattern analysis.

Conclusions: We assessed the validity of an integrated risk scoring model as a new methodological approach that could help the dermatologist in the differentiation between melanoma with regression and regressing nevus. Future studies could test the setting up of a score model over an even more complex pool of data obtained from different skin lesions with various diagnostic devices.