Prognostic value of epidermal growth factor receptors in gastric cancer: a survival analysis by Weibull model incorporating long-term survivors

Tissue samples were fixed in buffered 4 % formalin, embedded in paraffin, and used for TMA construction as described [22]. A slide with a representative tumor was selected, and an area of the tumor was circled on the slide. Using TMA technology (Beecher Instruments, Silver Spring, MD, USA), the area of interest in the donor block was cored twice with a needle 1.0 mm in diameter and the core transferred to a recipient paraffin block.

TMA sections were stained with primary antibodies: HER1 [H11 mouse monoclonal (Dako, Carpinteria, CA, USA); dilution, 1:100], HER2 [A0485 rabbit polyclonal (Dako); dilution, 1:1500], HER3 [RB-9211 rabbit polyclonal (Nterminal; Neomarkers, Fremont, CA, USA); dilution, 1:100], and HER4 [RB-9045 rabbit polyclonal (C-terminal; Neomarkers); dilution, 1:300]. A standard peroxidase-conjugated streptavidin–biotin method was used to detect the staining reaction (LSAB+; Dako). External positive control tissues included samples of placental tissues positive for the antibodies studied. For negative controls, primary antibodies were omitted and phosphate-buffered saline was substituted. Staining was evaluated by light microscopy and interpreted by a pathologist who was blind to the clinical information. Membrane staining was evaluated for HER1, HER2, HER3, and HER4. Staining of HER1, HER3, and HER4 was classified into four categories (0 no staining, 1+ light staining, 2+ moderate staining, 3+ strong staining) according to established criteria [15, 23]. The recommendations of the consensus panel for HER2 in gastric cancer were used for the classification of HER2 [24, 25]. Cytoplasm staining was evaluated for HER1, HER3, and HER4, and also classified into four categories (0 no staining, 1+ light staining, 2+ moderate staining, 3+ strong staining) according to the criteria used in a similar study [15]. The sections classified as 0 and 1+ were considered to be negative and those classified as 2+ and 3+ were considered to be positive, for both membrane and cytoplasm expression.

Correlations between receptor expression and clinicopathological characteristics were calculated by the chi-square test. Kappa coefficient and McNemar test were utilized to assess concordance among receptor expression. The study sample contained a large number of patients with long-term survival (Fig. 1), a distribution that permitted the use of the Weibull model with a mixture model incorporating long-term survivors for survival analysis [26‐28]. Figures 2 and 3 demonstrate the adequacy of this parametric model to the nonparametric Kaplan–Meier model, which permits its use. The regression model incorporating long-term survivors with Weibull distribution was also used for the study of the prognostic variables. Because the proportionality assumption of the Cox model was not held, the Cox regression model was not an appropriate choice for analyzing the present data. p values < 0.05 were considered to indicate statistical significance. The analyses were performed using the R software.

A total of 198 samples were available for HER1 analysis: 9 % with positive membrane staining (4 % with a 2+ score and 5 % with a 3+ score) and 45 % with positive cytoplasm staining. There was no correlation between HER1 and age, Laurén histological classification, tumor depth, nodal metastases, or TNM stage.

Thirty-four of the 201 samples (17 %) showed positive membrane HER2 staining (11 % with a 2+ score and 6 % with a 3+ score). HER2 positivity was correlated with more advanced age (p = 0.011) and with the Laurén’s intestinal type (p = 0.001). There was no correlation between HER2 expression and tumor depth, nodal metastases, or TNM stage.

Of the 200 samples available for HER3, only 1 showed membrane reactivity. However, 62 % showed cytoplasm positivity, which was related to more advanced age (p = 0.008) and to Laurén’s intestinal type (p < 0.001). There was no correlation between cytoplasmic HER3 expression and tumor depth, nodal metastases, or TNM stage. The correlation among membrane HER3 expression, clinicopathological characteristics, and overall survival was not evaluated because only one patient showed membrane positivity.

Twenty-nine of the 199 samples available (15 %) showed positive membrane HER4 staining, with a 2+ score in 10 % and a 3+ score in 5 %; 24 % of these showed positive cytoplasm staining. There was no correlation between HER4 expression and age, Laurén histological classification, tumor depth, nodal metastases, or TNM stage. Examples of positive immunohistochemistry for HER1, HER2, HER3, and HER4 are shown in Fig. 4.

There was a concordance among expression of the EGFR family receptors, with the exception of membrane HER1 and HER4, and membrane HER2 and HER4 (Table 2).

Receptor expression did not differ according to disease stage (Table 3).

The exploration of the Weibull model incorporating long-term survivors allowed the estimation of cure fraction, estimated at 51 % [95 % confidence interval (CI), 0.42–0.59]. It was also possible to evaluate the instantaneous risk, which demonstrated an increasing risk of death with time, but was more impressive at 4, 12, and 17 months after surgical treatment in the sample studied (Fig. 5).

The regression model incorporating long-term survivors with Weibull distribution adjusted for age, surgical radicality, type of surgery, Laurén histological classification, adjuvant treatment, TNM stage, and cell receptors revealed that TNM stage was the only variable with a prognostic influence (Fig. 6).