shows the breakdown of gender and age, WHO melanoma subtype, mutation frequencies of BRAF
, and anatomic site of the primary tumor of the cohort analyzed in this study and the previous study (Viros et al. 2008
) for comparison. With the exception of a greater median tumor thickness (2.0 mm vs. 1.3 mm) and a greater number of NMs (23.6% vs. 5.3%) in the present cohort, the compositions were very similar. shows the median weighted kappa scores as a measure of interobserver agreement ranging from 0.35 to 0.69, corresponding to fair (scores between 0.2 and 0.4) and to substantial (scores between 0.6 and 0.8) agreement. The best agreement was obtained with the criteria pigmentation, scatter of intraepidermal melanocytes, and solar elastosis, which showed better inter-observer agreement than the established WHO melanoma subtype.
Clinical characteristics of patients and melanomas by mutation status of BRAF and NRAS in both cohorts.
Interobserver agreement on morphologic features and WHO melanoma subtype
As shown in and Table S2
mutant primaries had increased upward scatter and nest formation of intraepidermal melanocytes, consisted of larger and rounder, more heavily pigmented melanocytes, displayed a thickened epidermis of their radial growth phase areas, were better circumscribed, and had less solar elastosis of their surrounding skin than primaries that did not have BRAF
mutations. The features showed a linear association with BRAF
mutation status, i.e. the strength of the association increased monotonously with higher (or lower) scores. Patients with BRAF
-mutant primaries were also significantly younger than patients whose melanomas did not have BRAF
mutations were further significantly more common in melanomas on the trunk compared to those in acral location. The associations between the morphologic and clinical features are similar in direction as in the previous study by Viros et al. (Viros et al. 2008
) but the strength of the association, as expressed by the odds ratios, varies. In particular, the associations for scatter and nesting are weaker in the present study.
Association of histomorphologic features,anatomic site, and clinical characteristics and BRAF mutation status
When applying the binary decision tree determined in the previous study in order to predict BRAF
mutation status, prediction could be made with 60.3% accuracy (52.0% sensitivity, 72.2 % specificity, p=0.0004, calculated by Fisher's Exact test), compared to a probability of 49.7% to predict BRAF
mutation by chance (). The original decision tree determined in the prior study uses scatter, nesting, and pigmentation as decision nodes for predicting BRAF
status. Because the current study established that the feature “nesting” used in this tree had comparatively lower kappa scores, we tested alternative trees composed of features with higher reproducibility but comparable associations with BRAF
status. The resulting binary tree that best predicted BRAF
mutation status had cell shape at the entry node followed by nodes using pigmentation and solar elastosis (Figure S1
). This classification tree created by single binary classification tree classifier was too complex to be practicable. Therefore, smaller alternative trees were created taking into account the information obtained in the binary classification tree classifier as well as inter-rater agreement and odds ratios of association of characteristics and BRAF
mutation status ().
Prediction algorithms of BRAF mutation status using only morphologic variables
As scoring of solar elastosis can be affected by variation in the hematoxylin eosin staining protocol, we excluded those 90 slides that were not processed at UCSF. This increased the prediction accuracy from 61.9% (sensitivity 79.0%, specificity 44.9%, p<0.0001) to 65.5% (sensitivity 80.2%, specificity 51.2%, p<0.0001), using the prediction algorithm that included solar elastosis as the entry node (, Table S1
Independent of BRAF
mutation status, NRAS
mutations were found to be significantly associated with low or absent scatter and better circumscription (p=0.026 and p<0.03 respectively, Table S2
). However, because the frequency of NRAS
mutations was lower and NRAS
status could not be established in 32 cases that were wild type for BRAF
, our power to identify features associated with NRAS
mutations was limited.
In the prior study, we took advantage of the association between BRAF
and age and used age as a proxy for BRAF
status to analyze possible associations with survival and pattern of metastasis in a large melanoma series from a registry, for which no mutation status was available (Viros et al. 2008
). In that study, an age cutoff of 55 years was determined in the study cohort to best predict BRAF
mutation status. Patients in the registry whose melanomas were predicted to be BRAF
mutant because they were younger than 55 years had a significantly better survival and were more prone to relapse in the regional lymph node basin than patients aged 55 years or older. The same association was seen in the current cohort (Table S3 and Figure S2a
) (Mehta & N. R. Patel 1986
). The available follow-up information in the current study allowed us to test more directly, whether these associations reflect any effect of BRAF
mutation status on survival and pattern of metastasis. Supporting the prior studies, we found that BRAF
-mutant primaries were more likely to first metastasize to regional lymph nodes while melanomas without BRAF
mutations were more likely to show non-nodal metastases (logistic regression, p=0.046; Fisher's Exact test, p=0.069) (). By contrast, when we tested the role of BRAF
mutations on overall and disease-free survival directly, no significant associations were found (p-values p=0.58 and p=0.39, respectively) (Figure S2b+c
Primaries with BRAF mutations are more likely to metastasize to the regional node.