Poorly differentiated component in gastric pinch biopsies predicts submucosal invasion

All the cases were selected from EGCs diagnosed on endoscopic pinch biopsies and treated by EMR between January 2002 and November 2007 at Samsung Medical Center, Seoul, Korea. Patients who presented with synchronous GC were excluded.

The first cohort included 60 patients (45 men and 15 women, age range 46–83 years, mean age 65.2 years) who were diagnosed with SMiGC after EMR. None of these patients had endosonographic evidence of lymph node or distant metastases. For the control group, 58 patients (50 men and 8 women, age range 45–84 years, mean age 63.8 years) confirmed to have IMC by EMR during the same period were included.

The endoscopic gross types of the tumors were classified using the five-tier classification system of the Japanese Research Society for Gastric Cancer [10]. The tumor location was classified as upper third, middle third, or lower third of the stomach.

All biopsy specimens were fixed in 10% neutral-buffered formalin, embedded in paraffin, sectioned at 4 μm thickness, and H&E stained. Three to 13 biopsy fragments (mean 5.8 fragments) were available for the SMiGC cases and two to nine fragments (mean 5.1 fragments) for the IMC control cases. For EMR, specimens were pinned to cork mats and fixed in 10% neutral-buffered formalin, and the resection margins were marked by differently colored inks. The fixed specimens were serially sectioned at 2 mm intervals and completely embedded. Three pathologists (Lee SM, Park CK and Kim KM) blinded to the results of the EMRs reviewed all the pinch biopsies and examined them for the presence of specific histologic features described in previous papers on pathologic findings used to distinguish high-grade dysplasia from adenocarcinoma in Barrett’s adenocarcinoma [9, 11, 12] early colonic carcinoma, [13‐15] and gastric carcinoma [16‐21]. These features are WHO histologic grade (well, moderately and poorly differentiated); poorly differentiated component (single or small clusters of poorly differentiated carcinoma cells or signet ring cells in 5% to 50% of the tumor volume) (Figure 1a); intraluminal amorphous and eosinophilic necrotic cell debris (IEND) in at least one neoplastic gland (Figure 1b); wisps of muscularis mucosa entrapped within the neoplastic glands (Figure 1c); papillary architecture (Figure 1d); tumor cribriforming (Figure 1e); and microscopic ulcers (Figure 1f). Disagreement in interpretation was resolved by consensus.

Finally, all EMR specimens were reviewed to confirm the diagnoses (IMC versus SMiGC), and the depth of SM invasion was measured as previously described [22]. Briefly, in cases with distorted or lost muscularis mucosa, immunohistochemical staining with desmin was performed, and the lower aspect of the muscularis mucosa was projected by drawing an imaginary line reaching both ends of the discontinuous muscularis mucosa. This line was used as a baseline, and the distance from this line to the deepest point of infiltration was measured as the depth of SM invasion.

Statistical analyses were performed using the Statistical Package for Social Science version 16.0 software for Windows (SPSS Inc., Chicago, IL, USA). For univariate analysis, the χ-square test and Fisher’s exact test were performed, and for multivariable analysis, each predictive pathologic factor of SM invasion was analyzed by multiple logistic regression analyses. Odds ratios and 95% confidence intervals were calculated to estimate the correlation between SM invasion, clinical factors (age, sex, and tumor location), and morphologic findings, e.g., degree of differentiation, microscopic ulcer, poorly differentiated component, wisps of muscularis mucosa, tumor cribriforming, papillary architecture, desmoplastic reaction, and IEND.

To validate the significant histologic findings in an additional independent cohort, consecutive gastric biopsies diagnosed as adenocarcinoma from January 1 to October 31, 2010 were reviewed by two other pathologists (Joo M and Ahn SM). In the 1497 cases, patients with palliative or multiple cancers, those who were lost to follow-up, or those confirmed as having advanced carcinomas were excluded, leaving a final group of 616 gastric cancers. Among these, the depth of invasion was confined to the mucosa in 423 cases and to the submucosa in 193 cases.

A total of 60 SMiGCs were grossly evaluated by routine endoscopy before EMR. Of these, 34 cases (56.6%) were located in the lower one-third (antrum). The predominant endoscopic findings were combined type IIa and IIc in 28 cases (46.7%), followed by type IIa in seven cases (11.7%) and type IIc in six cases (10%).

Table 1 summarizes the incidence of the various pathologic factors present in the biopsies corresponding to cases eventually diagnosed as SMiGC or as IMC. Poorly differentiated component (p < 0.0001), entrapped wisps of muscularis mucosa (p = 0.0013), cribriforming (p = 0.0084), papillary architecture (p = 0.0074), desmoplastic reaction (p = 0.0157), and IEND (p < 0.0001) were significantly associated with SM invasion. However, patient sex, age, tumor location, histologic grade and microscopic ulcers were not statistically associated with SM invasion.

The specificities of poorly differentiated component, cribriforming, papillary architecture, and IEND for predicting SMiGC in EMR specimens were 94.8%, 16.7%, 23.3%, and 93.1%, respectively; and their sensitivities were 36.7%, 16.7%, 23.3%, and 40%, respectively. Combinations of any two or three were associated with a higher specificity but a lower sensitivity (Table 2).

In multivariate analyses after exclusion of histologic grade, entrapped wisps of muscularis mucosa, and microscopic ulcer, multiple logistic regression analysis corrected by Bonferroni’s method showed that poorly differentiated component [odds ratio (OR) 9.59, p = 0.002], IEND [OR 6.23, p = 0.0126], cribriforming [OR 4.66, p = 0.0318], and papillary architecture [OR 5.52, p = 0.0188] were significantly associated with SMiGC.

In the independent validation cohort, WHO histologic grade (p < 0.0001), poorly differentiated component (p < 0.0001) and embedded wisps of muscularis mucosa (p = 0.0008) were significantly associated with SM invasion. However, cribriforming (p = 0.99), papillary architecture (p = 0.13), desmoplastic reaction (p = 0.09) and IEND (p = 0.58) were not associated with SM invasion (Table 3). In multivariate analyses, poorly differentiated component (p = 0.003) and papillary architecture (p = 0.008) remained significant (Table 4). The diagnostic sensitivity of poorly differentiated component was 48.7% (95% CI, 67.9 ~ 55.8) and specificity was 72.2% (95% CI, 67.9 ~ 76.5). The positive predictive value of poorly differentiated component was 43.5% (95% CI, 36.9 ~ 50.1%) and the negative predictive value was 75.3% (95% CI, 71.1 ~ 79.5).