Materials and Methods
We retrospectively collected surgical excision specimens of 119 granular cell tumours of the skin diagnosed in seven centres (Hôpital Saint-Louis, Hôpitaux Universitaires de Strasbourg, Hôpital La Pitié-Salpêtrière, Hôpital Cochin-Tarnier, Centre Hospitalier Intercommunal de Créteil, Laboratoire de Pathologie Cutanée de la Roquette, Cabinet de Dermatopathologie Mathurin-Moreau) between January 2001 and December 2011. Five H&E step sections and one orcein staining were performed in all cases.
First, malignancy criteria from Fanburg-Smith et al were assessed in order to classify tumours as benign, atypical or malignant. Other histopathological features were then assessed: peripheral growth pattern (well limited or invasive), depth of the tumour (Breslow thickness), largest diameter of the tumour, perineural invasion, presence of intra-tumoural large nerve (>0.2 mm diameter), presence of (so-called) pustulo-ovoid bodies, presence and distribution of inflammatory infiltrate, arrector pili muscle invasion, and vascular invasion. Vascular invasion was classified as 'vascular wall infiltration' (infiltration of the subendothelial layers of the vessel without obliteration), 'vascular obliteration' (infiltration of the vessel wall with absence of central lumen; fully plugged vessel) or 'intravascular permeation' (free-floating tumour cells in the vascular lumen). When present, vascular invasion was studied further with immunohistochemical staining for S100 protein (rabbit polyclonal, Dako, Glostrup, Denmark), CD34 (clone QBend10, Beckman Coulter, Brea, California, USA) and smooth muscle actin (clone 1A4, Dako), performed on serial sections with an automated indirect immunoperoxidase technique (Benchmark XT; Ventana, Tucson, Arizona, USA).
Location of the tumour (head and neck, trunk, genital area, upper limbs, lower limbs, palms and soles) and age at diagnosis were collected. Follow-up data were available in 25 patients.
Slides were reviewed collectively by five pathologists or dermatopathologists (MB, BC, FLP, LP and FP), and only non-equivocal findings were considered. Data analysis was performed using R V.2.15.2 statistical software (R-Foundation for Statistical Computing, Vienna, Austria). Multiple tumours in the same patient typically had similar features, and were not assumed to be independent. Therefore, only the first tumour in each patient (114 tumours) was included in the statistical analysis. Differences between groups were assessed using Wilcoxon signed-rank tests (quantitative variables) and Fisher's exact test (qualitative variables). Probabilities were corrected for multiple comparisons using the Bonferroni method. A p value ≤0.05 was considered statistically significant. For qualitative variables, 95% CI were estimated with the prop.test function in the built-in R stats package.