Context: In gastric cancer, the significance of perineural invasion remains controversial. Detecting perineural invasion with hematoxylin-eosin staining often leads to misdiagnosis. Labeling nerves by immunohistochemistry greatly assists perineural invasion detection, but it might also be misdiagnosed, because scattered cancer cells are difficult to recognize.
Objective: To reevaluate the significance of perineural invasion in gastric cancer by double immunohistochemical staining that labels both nerves and cancer cells, and to examine agreements on perineural invasion detection between double immunohistochemical staining and single immunochemical staining (to label nerves) or hematoxylin-eosin staining.
Design: We evaluated perineural invasion in 160 cases of gastric cancer with double immunohistochemical staining, single immunochemical staining, and hematoxylin-eosin staining, respectively; then we investigated the prognostic significance of perineural invasion.
Results: Perineural invasion was detected in 65.0% (104 of 160), 38.1% (61 of 160), and 56.9% (91 of 160) of cases with double immunohistochemical staining, hematoxylin-eosin staining, and single immunohistochemical staining, respectively. Agreement was low between double staining and hematoxylin-eosin staining (κ = .34), and most false reports occurred in diffuse gastric cancer. Agreement between single immunochemical staining and double staining was good (κ = .67), but it declined in diffuse gastric cancer (κ = .28). Perineural invasion was closely associated with other clinicopathologic variables. Although perineural invasion-positive patients had a worse outcome than perineural invasion-negative patients, it was not an independent prognostic factor (P = .11; hazard ratio, 0.637; 95% confidence interval, 0.366-1.110).
Conclusions: Double immunohistochemical staining could improve accuracy of perineural invasion detection in gastric cancer, particularly in the diffuse type. Moreover, perineural invasion predicts a poor outcome in gastric cancer, but it cannot provide more information than traditional clinicopathologic variables.