To the Editor: We thank Dr Parsons1 for his comment and would like to clarify our decision to include and prefer MelanA/MART-1 in our diagnostic algorithm to distinguish spindle cell from desmoplastic melanoma.2

The summary of our literature meta-review that Dr Parsons refers to tabulates the MelanA/MART-1 and HMB45 labeling fraction in spindle cell and desmoplastic melanoma derived from 67 different studies. The apparent difference of marker fractions in desmoplastic melanoma (8.8 vs 18.8%; P<0.0001; n=54 studies) derives in large parts from studies (n=37) where only one of the two biomarkers was assessed. The key limitation of this approach to summarize data for the assessment of discriminatory value is, that the percentage of positive cases is based on a theoretical summary of separate studies on different cases rather than a direct comparison of two markers in the same samples. In other words, we do not know the MelanA/MART-1 status in the ∼300 more desmoplastic melanomas that have only been stained for HMB45 (but not for MelanA/MART-1; ie, discrepant analysis). To assess and compare the discriminatory value of these two biomarkers, it is crucial to restrict the comparison to studies where samples have been assessed with both markers (Figure 1a). When restricting the tabulation to those nine studies that allow analysis of both markers on a case-by-case basis (Figure 1b),2, 3, 4, 5, 6, 7, 8, 9, 10, 11 there is no significant difference in the labeling fraction of HMB45 and MelanA/MART-1 in desmoplastic melanoma (P=0.79).

Figure 1
figure 1

Assessment of discriminatory value of MelanA/MART-1 and HMB45 in desmoplastic melanoma. (a) Screenshot of the 37 desmoplastic melanoma studies used to compose the overall labeling fraction of MelanA and HMB45. Shading (gray and black) is visualizing those 15 studies that assessed both markers and black indicates that only 9 studies allow comparison of staining pattern of both markers on a case-by-case basis (see b). For a legible and more detailed tabulation see Supplementary Table 7 in our original report.2 (b) Summary of studies and staining results on a case-by-case basis demonstrate no significant difference in MelanA- and HMB45-labeling fractions (P-value from Fisher’s exact test). (c) Staining results by marker and histotype in our original study cohort (pale)2 and 35 new cases (black squares). Top three rows demonstrate staining pattern to reach or confirm the diagnosis of melanoma, whereas bottom two rows demonstrate staining results of HMB45 and MelanA. When specifically comparing the discriminatory value of HMB45 and MelanA in spindle cell vs desmoplastic melanoma, there are twice as many misclassifications with HMB45 when compared to MelanA: in our initial cohort (triangles) as well as in the additionally examined cases of spindle cell and desmoplastic melanoma (asterisks).

Full size image

Our reason to tabulate all available studies (listed separately in Supplementary Tables 6–8)2 into one combined table (Supplementary Table 9)2 was to assess individual labeling fractions of markers that allow reaching or confirming the diagnosis of melanoma in this specific setting. Briefly, we found that markers otherwise reliable in the setting of conventional melanoma (among these HMB45 and MelanA/MART-1) are with 8.8 and 18.8%, respectively, not as useful for this task in desmoplastic melanoma. In contrast, our meta-review revealed S100, SOX10 and p75 as reliable markers for the positive diagnostic identification of melanoma in the setting of spindle cell as well as desmoplastic melanoma. Once the diagnosis ‘melanoma’ is reached or confirmed, our algorithm can help in subclassification of spindle cell vs desmoplastic melanoma, when appropriate. Rarely, if ever, studies report staining results on a case-by-case basis and it is difficult to assess technical and diagnostic variability between reports. Therefore we based our algorithm strictly on our own analysis of test performance on cases derived from two different institutions.2 When specifically assessing the discriminatory value of the two biomarkers in spindle cell and desmoplastic melanoma, we found eight misclassifications using HMB45, whereas MelanA showed only four misclassifications in our original series (Figure 1c, triangles). In the meantime, we had the opportunity to expand our study cohort (Figure 1c, black squares). Again, with respect to discriminatory value, we found four additional misclassifications using HMB45, whereas MelanA showed two additional misclassifications (Figure 1c, asterisks). These findings in new independent samples validate our prior analysis. Despite these test-performance considerations and the ‘exclusionary’ appearance of our diagnostic algorithm, a biomarker status should not replace careful histomorphological assessment and clinicopathological correlation.

In sum, when assessing the discriminatory value of two biomarkers, it is—in our opinion—not sufficient to summarize literature data of marker fractions. It is rather necessary to compare test performance in a combined assessment of both markers in the same samples.