Reproducibility of Her2/neu scoring in gastric cancer and assessment of the 10% cut-off rule

Abstract

The application of Trastuzumab on gastric cancer patients is based on Her2/neu immunostaining. The testing method relies on visual estimation of both membranous staining intensity, and positive tumor ratio with respect to a 10% cutoff. We evaluated the effect of inter- and intraobserver variations of both factors on therapeutic decision, especially if the positive tumor ratio hovers around the 10% cutoff. Ten pathologists scored 12 Her2/neu immunohistologically stained whole sections of gastric cancer. Applying the common rules for Her2/neu testing for gastric cancer, they separately noted the strongest identifiable staining intensity and the corresponding positive tumor ratio. Scoring was done repeatedly using the microscope, plain virtual microscopy, and virtual microscopy with a manual outline drawing function. Agreements on the strongest identified staining intensities were moderate. Overall concordance correlation coefficients of positive tumor ratios ranged from 0.55 to 0.81. Reproducibility was not improved by virtual microscopy. Pathologists have a good ability to estimate ratios of clearly demarcated areas, but gradients in staining intensities hinder reproducible visual demarcation of positive tumor areas. When hovering around the 10% positive tumor ratio cutoff there is a risk of misinterpretation of the staining results. This could lead to a denial of Trastuzumab therapy. Assessment of Her2/neu expression should be carried out by experienced pathologists because they can more reproducibly rate membranous staining intensities. The low reproducibility of positive tumor ratio is inherent in the testing method and cannot be improved by virtual microscopy. Therefore, we propose to reconsider the 10% cut-off limit.

Keywords: Cut-off value; Her2/neu; gastric cancer; virtual microscopy; visual perception.

Figure 1

Overview of all ratings, grouped by the rating method (microscopy, virtual microscopy, and assisted virtual microscopy.) Each diagram shows 10 bars, corresponding to the ratings of one specimen by 10 pathologists, using the indicated method. Ratings are shown separately for positive tumor ratio (green bars ≥10%, and red bars <10%), and staining intensity (blue bars). Below the diagrams, mean and standard deviation of positive tumor ratio of the 10 ratings are indicated. The top row shows overview images of the 12 scanned slides.

Figure 2

Examples of the outlines of tumor areas (pale blue) and positive tumor areas (pale red) that were manually drawn by three pathologists. They illustrate the variations in assessment of positive tumor areas, and also of total tumor area. Full data are shown in Table S2B.

Figure 3

Comparison of positive tumor ratios between methods. Each data point represents two positive tumor ratios of the same case, rated by the same pathologist, using two methods. Solid diagonal lines mark complete agreement. Dotted lines denote 10% cutoff and isolate the discordant ratings in the upper left and lower right quadrant. (A) compares the microscopic method with virtual microscopy, and (B) compares virtual microscopy with assisted virtual microscopy. (C) How well the pathologists were able to estimate the positive tumor ratios of the outline sketches they had drawn, compared with their calculated ratios (Pearson's correlation coefficient was 0.974, P < 0.001.) Linear regression (dashed line) shows that positive tumor ratios were systematically overestimated.