ECL Cell Tumor and Poorly Differentiated Endocrine Carcinoma of the Stomach: Prognostic Evaluation by Pathological Analysis

Transcription

1 GASTROENTEROLOGY 1999;116: ECL Cell Tumor and Poorly Differentiated Endocrine Carcinoma of the Stomach: Prognostic Evaluation by Pathological Analysis GUIDO RINDI,* CINZIA AZZONI, STEFANO LA ROSA, CATHERINE KLERSY, 6 DONATELLA PAOLOTTI,* SIGRID RAPPEL, MANFRED STOLTE, CARLO CAPELLA, CESARE BORDI, and ENRICO SOLCIA* *Department of Human Pathology, University of Pavia and Istituto di Ricovero e Cura a Carattere Scientifico (IRCCS) Policlinico San Matteo, Pavia, Italy; Institute of Anatomic Pathology, University of Parma, Parma, Italy; Department of Pathology, University of Pavia at Varese, Varese, Italy; 6 Biometric Unit, IRCCS Policlinico San Matteo, Pavia, Italy; and Institute of Pathology, Klinikum Bayreuth and Friedrich-Alexander University, Nürnberg, Germany Background & Aims: Gastric endocrine tumors show a wide spectrum of clinical behavior, and prognostic assessement of individual tumors is difficult. The aims of this work were to identify predictors of tumor malignancy and patient outcome and to provide a rationale for treatment guidelines. Methods: Gastric endocrine tumors (86 enterochromaffin-like cell carcinoids and 16 poorly differentiated carcinomas) were investigated for 15 clinicopathologic variables and for expression of Ki67, P53, and BCL-2 proteins. Data were analyzed by univariate and multivariate statistics for evidence of tumor malignancy and patient survival. Results: Histological grades 2 and 3, size H3 cm, 9 or more mitoses, or H300 Ki67-positive cells per 10 high-power fields identified 26 of 33 (79%) malignant (metastatic or deeply invasive) tumors, and size F1 cm and/or growth restricted to the mucosa characterized 46 of 69 (67%) tumors with benign behavior during a median follow-up of 39 months. Malignancy-predictive models were developed using angioinvasion, size, clinicopathologic type, mitotic index, and Ki67 index. The same variables, in addition to deep gastric wall invasion and histological grade, predicted patient outcome. Conclusions: Criteria for the assessment of malignancy risk and patient outcome were developed for the different tumors, providing a basis for treatment guidelines. Endocrine tumors of the stomach are rare and mostly well-differentiated growths with variable and poorly predictable behavior, from benign to moderately malignant. 1 9 Poorly differentiated endocrine (neuroendocrine) carcinomas (PDECs) of the stomach also have been reported and are particularly aggressive, malignant tumors. 7,10,11 With the exception of a few gastrin and enterochromaffin cell tumors, all well-differentiated endocrine tumors of the stomach are enterochromaffin-like (ECL) cell tumors (carcinoids). 1 7 The ECL cell, the main endocrine cell type of the corpus-fundus mucosa, is known to be highly sensitive to gastrin stimulus and, in turn, to trigger parietal cell acid secretion by releasing histamine. 12 Gastrin, fibroblast growth factor (FGF), and other factors have been shown to have trophic effects on ECL cells, a finding of potential relevance for ECL cell tumorigenesis. 12,13 Recently we described three classes of well-differentiated ECL cell tumors of the stomach on the basis of their background gastric pathology. 7 Hypergastrinemia has been identified as a common pathogenetic trait for the tumors associated either with chronic atrophic gastritis (type I tumors) or with multiple endocrine neoplasia type 1 (MEN-1) and the Zollinger Ellison syndrome (ZES) (type II tumors). Type I and type II tumors are essentially benign or, at most, low-grade malignant neoplasms, causing tumor-related death in no cases in our series and only in occasional cases reported in the literature. 1 7 A third ECL cell tumor class was found in patients free of any specific gastric pathology and in the absence of hypergastrinemia (type III, or sporadic, tumors). As a whole, such tumors have been demonstrated to behave more aggressively, with local or distant metastases in most cases, and caused the death of the patient in a substantial proportion of cases investigated. 7 This classification was conceived to provide simple clinicopathologic criteria useful for treatment of patients and proved effective in some studies. 14,15 However, a few Abbreviations used in this paper: CAG, chronic atrophic gastritis; CI, confidence interval; ECL, enterochromaffin-like; FGF, fibroblast growth factor; HPF, high-power field; HR, hazard ratio; MEN-1, multiple endocrine neoplasia type 1; NSE, neuron-specific enolase; OR, odds ratio; PAS, periodic acid Schiff; PDEC, poorly differentiated endocrine carcinoma; PGP 9.5, protein gene product; R M, explained variation; ROC, receiver operating characteristic; ZES, Zollinger Ellison syndrome by the American Gastroenterological Association /99/$10.00

2 March 1999 PROGNOSIS OF GASTRIC ENDOCRINE TUMORS 533 type I and type II tumors proved malignant, whereas many type III tumors and even 1 of 12 PDEC cases behaved favorably after surgery. 7,15 These observations indicate that the single case may escape the behavior predicted for the tumor class it belongs to and raise the question of whether it is possible to identify behaviorpredicting tumor variables that could be used to evaluate individual tumors and assess their malignant potential. Some variables, such as proliferative markers and angioinvasion, have been identified recently for endocrine tumors of other sites (pancreas, parathyroids, thyroid) The observation of a few cases with coexisting welldifferentiated and poorly differentiated endocrine growths and the reported atypical cases among sporadic carcinoids 7,19 22 suggest the possibility of neoplastic progression from a well-differentiated, histologically benign or low-grade ECL cell tumor to a poorly differentiated, histologically aggressive PDEC. It has been shown that ECL cells of well-differentiated tumors express a common range of endocrine markers, indicating a substantial cell homology between type I, II, and III tumors. On the other hand, most cells composing PDEC retain only abortive signs of endocrine differentiation at the ultrastructural level and share with cells of well-differentiated tumors only the expression of cytosolic markers of neuroendocrine differentiation, like neuron-specific enolase (NSE) and protein gene product (PGP 9.5). 5,7 So far, no cell marker capable per se of labeling the aggressive tumor behavior has been identified, and little information is available on the molecular events underlying tumor development or progression. Involvement of P53 and BCL-2 oncoproteins has been considered but not extensively investigated on gastric tumors. 23,24 This study assessed the weight of 15 clinicopathologic variables and the expression of three proteins (P53, BCL-2, and Ki67) as predictors of tumor malignancy and patient outcome. Materials and Methods In the last three decades, 258 cases of gastric endocrine tumors were identified in the Departments of Pathology of Bayreuth, Parma, Pavia, and Varese. Of the 258 cases, 217 (84.1%) had been diagnosed as ECL cell tumors, 3 as pyloric gastrin cell tumors, 1 as a pyloric somatostatin cell tumor, 1 as a corpus serotonin cell microtumor, 20 (7.8%) as PDEC, 13 (5.0%) as associations of gastric endocrine tumor and nonendocrine carcinoma in a background of chronic atrophic gastritis (10 ECL cell tumors with 10 adenocarcinomas, 3 of which were early ; 1 PDEC and 1 adenocarcinoma; and 1 ECL tumor and 1 PDEC with 1 choriocarcinoma), and 3 as endocrine tumors not further specified because insufficient material or information was available. Of these, 102 cases representative of the whole series considered as far as variables investigated, 61 with surgical specimens and 41 with endoscopic samples only, were selected for this study on the basis of (1) availability of sufficient tumor tissue and nontumor gastric mucosa, (2) tumor type fitting among those being studied, (3) no other coexisting neoplasia likely to affect patient survival, and (4) available clinical and follow-up information. Of these cases, 69 have been reported previously. 5,7,15 Patients clinicopathologic analysis included age, sex, follow-up in months, disease status, cause of death, presence of hypergastrinemia, type of material (surgical or bioptic) available for study, and evidence for local and/or distant metastases. Tissue samples from tumors and nontumor mucosa were routinely processed and stained with H&E for histology and periodic acid Schiff (PAS)/alcian blue (ph 2.5) for mucins. Extratumoral ECL cell growths and gastritis were evaluated as reported previously. 21,25 Antral gastrin cell hyperplasia was evaluated as described previously 26 in sections immunostained with the avidin-biotin technique according to the method of Hsu et al. 27 Tumor Tissue Analysis Tumor tissue analysis included histological diagnosis, grading, clinicopathologic type, site of development (fundus, corpus, antrum), size, number of growths, level of gastric wall invasion, and presence of angioinvasion, lymphoinvasion, or histologically proven metastases. To facilitate assessment of vascular microinvasion, sections were immunostained for the endothelial marker CD34 (QBEN10, mouse monoclonal; Serotec, Oxford, England, 1:100). Angioinvasion was defined as spaces lined by CD34-immunoreactive endothelial cells, containing red blood cells and showing adhesion of tumor cells to the endothelium. Lymphoinvasion was defined as tumor-filled slender, serpentine channels often crossing the muscularis mucosa or running parallel to it and spaces accompanying arteriolar or venous vessels. Endocrine differentiation of tumor cells was investigated using general endocrine markers such as silver-impregnation methods (Sevier Munger and/or Grimelius) and immunohistochemical tests for chromogranin A and synaptophysin or for the cytosol markers NSE and PGP 9.5 and immunohistochemical tests for the gastrointestinal hormones gastrin, somatostatin, serotonin, and pancreatic polypeptide, as detailed previously. 5,7 Tumors were first classified as (1) differentiated, with diffuse, often intense reactivity for granular as well as cytosolic markers of endocrine differentiation, mild to moderate cellular atypia, and microlobular-trabecular to solid structure; and (2) undifferentiated or poorly differentiated, with substantial reactivity for cytosolic markers; poor, scanty, or absent reactivity for granule markers; severe cellular atypia; and diffuse to poorly defined solid structure. The absence of specific hormone expression in most tumor cells despite reactivity for general endocrine markers (both granular and cytosolic) was regarded as indicative for the diagnosis of ECL cell tumor. In 28 selected cases, the diagnosis was confirmed by ultrastructural analysis, performed as detailed previously. 5,7 Following previously described guidelines

3 534 RINDI ET AL. GASTROENTEROLOGY Vol. 116, No. 3 for the assessment of clinicopathologic type, 7 ECL cell tumors were subclassified as type I tumors when associated with chronic atrophic gastritis (CAG), ECL cell hyperplasiadysplasia, and hypergastrinemia; type II tumors when associated with hypertrophic gastropathy or at least nonatrophic mucosa, ECL cell hyperplasia-dysplasia, and hypergastrinemia, usually in conjunction with MEN-1 ZES; and type III sporadic tumors when not associated with any specific gastric pathology, with normal ECL cell distribution, normal levels of gastrinemia, and no evidence of MEN-1 or ZES. Tumor grading was partly based on previous criteria, 7 with grades 1 and 2 largely corresponding to previous typical and atypical carcinoids, respectively, and grade 3 fully corresponding to neuroendocrine carcinomas. Grade 1 (G1) was further subdivided into grade 1a (G1a), characterized by small, microlobular-trabecular aggregates formed by regularly distributed, often aligned cells (mosaiclike pattern), with regular, monomorphic nuclei, usually inapparent nucleoli, rather abundant, fairly eosinophilic cytoplasm, and almost absent mitoses; and grade 1b (G1b), characterized by significant areas with solid structure, absence of cell alignment, round to spindle cell shape, irregular and moderately polymorphic nuclei of larger size, often with evident nucleoli, and rather few, morphologically typical mitoses. Grade 2 (G2) tumors showed prevalence of solid cellular aggregates and large trabeculae, crowding, and irregular distribution of round to spindle and polyedric tumor cells, fairly large vescicular nuclei with prominent eosinophilic nucleoli, or smaller, hyperchromatic nuclei with irregular chromatin clumps and small nucleoli, considerable mitotic activity, sometimes with atypical mitotic figures, and scant necrosis. Grade 3 (G3) tumors corresponded to PDEC and showed severe histological atypia with solid to diffuse structure and frequent central necrosis. They were composed of tightly packed, small to mid-sized tumor cells showing large, irregular, polymorphic, and hyperchromatic nuclei, with irregular chromatin distribution, evident nucleoli, scant cytoplasm, and frequent, often atypical, mitoses. The proliferative status of tumor cells was evaluated using Ki67 protein antibodies (MIB 1, mouse monoclonal; Immunotech, Marseille, France; 1:50) after microwave antigen retrieval. 28 The Ki67 proliferative index was evaluated in tumor areas showing the highest number of labeled nuclei either by counting the number of positive cells among 2000 tumor cells (percent evaluation) 18 or by counting the number of labeled nuclei per 10 high-power (400 ) microscopic fields ( 10 HPF). In 38 cases evaluated with both methods, no significant difference was found between counting positive cells per 10 HPF and counting per 2000 tumor cells. The count 10 HPF, also known as the Ki67 index, was eventually chosen because we found it more convenient for the practicing pathologist traditionally accustomed to this method for counting mitoses. The tumor mitoses were similarly counted 10 HPF on H&E-stained sections. Transforming marker analysis. Accumulation of P53 protein was evaluated by immunohistochemistry after microwave antigen retrieval using specific antibodies (DO7, mouse monoclonal; Dako, Glostrup, Denmark; 1:100) recognizing an N-terminal sequence of both wild-type and mutated molecule. 28 P53-positive cells were semiquantitatively estimated as percent of tumor cells. The accumulation of BCL-2 oncoprotein was evaluated by immunohistochemistry using a specific antibody (anti-human BCL-2 oncoprotein, mouse monoclonal; Dako; 1:100) after microwave antigen retrieval. 23 BCL-2 accumulation was evaluated in tumor areas showing the highest number of labeled cells by counting positive cells 10 HPF. Statistical Analysis Most variables had a skewed distribution; therefore, median and quartiles were the adopted descriptive statistics. Variables predicting malignancy were identified using logistic regression modeling. Malignancy was defined as evidence of either deep gastric wall invasion (muscularis propria and beyond) or metastases to local lymph nodes or to distant sites. The odds ratio (OR), together with its 95% confidence interval (CI) and the related P value, was calculated for assessment of the relative risk of malignancy of a certain value (category) of the variable with respect to the reference category within the same variable. The adequacy of the model has been checked by calculation of the proportion of benign and malignant cases correctly predicted by the model as well as the area under the receiver operating characteristic (ROC) curve, as a measure of goodness of fit. These statistics range from 0 to 1 (100%); the closer the value to 1, the better the model. 29 Before inclusion in the model, all continuous variables were categorized. Cutoff values were chosen based on previous clinical knowledge. In case of perfect prediction of benign or malignant behavior, cutoff values were adjusted to allow OR calculation. Survival analysis was performed by assuming a proportional hazard for each variable. Univariate predictors of death have been identified by fitting a Cox proportional hazard model. Hazard ratios (HRs) and their 95% CIs were computed for each variable. The predictive ability of the selected Cox model has been assessed by calculation of the explained variation by means of the likelihood-based explained variation (R M ). 30 Multivariate analysis was performed to assess the independent prognostic value of a series of candidate prognostic factors for malignancy and death. Correlation between covariates was assessed to evaluate collinearity between variables to be included in the multivariate model. Two covariates were considered as correlated at R All variables showing a P value of 0.1 by univariate analysis were included in the multivariate model. Stata Statistical Software release 5.0 (College Station, TX) was used for computation. Results Overall Description The 102 patients selected for study had approximately equal sex distribution (55 female, 47 male), with a mean age of 59 years (SD, 13.7); 52% of patients (n 53) were older than 58 years. However, 13 of 16

4 March 1999 PROGNOSIS OF GASTRIC ENDOCRINE TUMORS 535 Table 1. Tumor Clinicopathologic Type, Histological Pattern and Behavior, and Patient Survival of the 102 Cases Investigated Metastasis b Type Grade n Size a (cm) Mitoses a ( 10 HPF) P53 Ki67 a ( 10 HPF) Lymphoinvasion Angioinvasion Deep wall invasion L D All Tumor death Observation time c Type I G1a (0.5 1) 0 (0-0) 0/ (10 41) 6/33 1/33 0/33 0/33 0/33 0/ G1b ( ) 0 (0 0) 1/ (14 67) 18/25 5/25 1/25 2/25 1/25 3/ G ( ) 0 (0 0) 1/ (12 58) 24/58 6/58 1/58 2/58 1/58 3/ Type II G ( ) 1 (0 2) 0/ (12 39) 9/11 3/11 0/11 2/11 1/11 2/ Type III G ( ) 1.5 (0 3) 5/ (26 114) 9/12 3/12 7/12 5/12 2/12 5/ G2 5 4 ( ) 9 (7 18) 3/ ( ) 5/5 5/5 5/5 4/5 4/5 5/ PDEC G ( ) 44 (29 78) 12/ ( ) 16/16 14/16 16/16 15/16 8/16 16/ a Median (quartile range). b L, local; D, distant. c Person-months. PDECs and 11 of 17 type III ECL cell tumors were in men, and 32 of 58 type I and 7 of 11 type II ECL cell tumors were in women. Tumor Features The main pathological findings of the 102 tumors are outlined in Table 1 and Figures 1 and 2. Ninety-four tumors (92%), including all but 1 of the 86 ECL cell tumors and 9 of 16 PDECs, were in the body-fundus mucosa, and the remaining 8 cases were in the antral mucosa. Single tumors were observed in 50 cases (19 type I, 17 type III, and 14 PDECs), 2 tumors in 5 cases (3 type I and 2 PDECs), and 3 or more tumors in 47 cases (36 type I and 11 type II). Overall, more than 1 tumor was observed in 52 cases; only 7 of these (13%) were malignant, compared with 26 of 50 (52%) solitary tumors. The mean tumor size was 1.76 cm ( 1.87; range, cm); 43 neoplasms (none with evidence of malignancy) were 1 cm, 38 (10 malignant) were cm, and 21 (all malignant) were 3 cm in size. In most cases (73; 71.6%), gastric wall invasion was found to be limited to the mucosa or mucosa and submucosa. In detail, 29 tumors did not extend beyond the mucosa, and 44 did not extend beyond the mucosa and submucosa; 29 tumors were deeply infiltrative, reaching either the muscularis propria (9 cases) or the subserosa (20 cases). Angioinvasion was found in none of the 29 intramucosal, 9 of 44 (20.5%) submucosal, and 22 of 29 (75.9%) deeply invasive tumors. Lymphoinvasion was seen in 1 of 29 (3.4%) intramucosal, 34 of 44 (77.3%) submucosal, and 28 of 29 (96.6%) deeply invasive tumors. Local lymph node metastasis was seen in 28 of 102 cases, all but 1 with lymphoinvasion at the primary site. Distant metastasis was present in 16 cases, all with lymphoinvasion, all but 1 with angioinvasion at the primary site, and 13 with local lymph node metastases. Overall, metastasis was seen in 31 of 102 cases (30.4%). Considering both the variables deep wall invasion and presence of metastases, either to local or distant lymph nodes, 33 of 102 cases (32.4%) were positive for one or both and were classified as malignant. A close relationship was found between deep wall invasion and presence of metastasis: 27 of 29 (93%) deeply invasive cases showed metastasis, compared with 4 of 73 (5.5%) tumors restricted to the mucosa-submucosa (Fisher exact test, P 0.000). When only ECL cell tumors were considered, 11 of 13 (85%) deeply invasive tumors had metastasis, compared with 4 of 73 (5.5%) mucosalsubmucosal (Fisher exact test, P 0.000), i.e., 4 of 44 (9.0%) submucosal and 0 of 29 intramucosal tumors. Grade 1 histology was found in all type I and type II tumors and in 12 of 17 type III tumors. In particular, 38 ECL cell tumors (33 of 58 type I, 3 of 11 type II, and 2 of 17 type III tumors) were G1a. Only 1 of these was malignant, compared with 11 (7 deeply invasive and 9 metastatic, 4 with distant metastases) among the 43 G1b tumors (25 of 58 type I, 8 of 11 type II, and 10 of 17 type III). As shown in Table 1, separation of G1 from G2 tumors among type III ECL cell carcinoids allowed identification of two tumor groups with substantially different outcomes, metastatic patterns, angioinvasion rates, sizes, and mitotic and Ki67 indexes. On the other hand, among type III G1 tumors, no major difference was found between 7 malignant and 5 nonmalignant cases in terms of outcome, size, or mitotic or Ki67 index. The only relevant differences between G2 type III carcinoids and G3 PDECs were longer patient survival and lower mitotic index in G2 type III carcinoids. No mitoses were detected in 56 tumors, including 4 (7%) malignant cases; 1 8 mitoses 10 HPF were found in 28 tumors, 10 of which (36%) were malignant; and 9 or more mitoses were seen in 19 tumors, all of which were malignant. Of the 95 cases investigated for Ki67, 150 labeled cells 10 HPF were found in 20 tumors, all but 1 (95%) malignant. In contrast, only 11 of 75 (15%) tumors with lower Ki67 index proved malignant. Accumulation of P53 was seen in only 21 of 94 (22%) tumors analyzed, 19 of which were malignant and 12 of

5 536 RINDI ET AL. GASTROENTEROLOGY Vol. 116, No. 3 Figure 1. (A and B) Grade 1a ECL cell tumor associated with corpus-fundus atrophic gastritis, showing (A) delicate, trabecular-microlobular aggregates of small, regularly distributed, mildy atypical cells with no mitoses and (B) few Ki67-positive nuclei. (C) Grade 1b ECL cell tumor associated with atrophic gastritis, showing more solid aggregates of cells with moderately hyperchromatic nuclei, often with evident nucleolus. (D F ) Grade 2 sporadic ECL cell tumor with (D) prevalence of large, solid aggregates of irregularly distributed cells, with or without focal preservation of (E) trabecular and (F ) microacinar structure; note focal necrosis (*); fairly large, polyedric nuclei; scattered mitoses (arrows); and (E ) high Ki67 index (magnifications: A, B, C, E, and F, 250 ; D, 100 ; A, C, D, and F, H&E stain; B and E, immunoperoxidase with light hematoxylin counterstain). which were PDECs. P53 was also detected in foci of nonendocrine glandular growth frequently found to be intermingled with PDEC at the mucosal-submucosal level. BCL2 expression was detected in 34 of 94 tumors (36%) for which data were available. The mean number of positive cells observed was HPF (median, 0), with extreme variability (SD, ) and skewing and no significant relationship with malignancy; 9 of 34 BCL2-positive tumors (26%) and 22 of 60 negative cases (37%) were malignant. Follow-up As detailed in Table 1, 20 patients died of tumor disease (median follow-up, 11.5 months; quartiles 5 17).

6 March 1999 PROGNOSIS OF GASTRIC ENDOCRINE TUMORS 537 Figure 2. Grade 3 tumor (PDEC) showing (A) solid aggregates of hyperchromatic cells with large necrotic areas; (B) frequent, often atypical, mitoses with high Ki67 index; and (C) P53 immunoreactivity. (D and E ) Partly necrotic neoplastic embolus in a peritumor blood vessel of a grade 3 endocrine carcinoma (D) and a lymph vessel microinvasion in a grade 1b tumor (E) (magnifications: A,63 ; B and C, 250 ; D, 100 ; E, 400 ; A, D, and E, H&E stain; B and C, immunoperoxidase with light hematoxylin counterstain). Fifteen patients died for reasons unrelated to gastric tumor disease (median follow-up, 26 months; quartiles 11 63), and 66 were alive at the time of the last visit, 27 with persistence of tumor disease (median follow-up, 36 months; quartiles 20 60) and 39 without evidence of tumor disease (median follow-up, 66 months; quartiles 30 99). Identification of Predictors of Malignancy Logistic regression univariate analysis showed that of the 13 variables considered (sex, age, antral site vs. other, 2 vs. 1 lesions, lesion size cm vs. 1cm or 2.5 cm vs. 1 cm, histological grade 2 3 vs. 1 or 1b vs. 1a, mitotic index 1 7 vs. 0 or 8 vs. 0, Ki67 index of 150 vs. 150, BCL-2 expression, P53 accumulation, lymphoinvasion, angioinvasion, and clinicopathologic type, one-step change) all but age and BCL-2 expression were statistically significant predictors (P 0.005). Higher relative risks of malignancy were observed for Ki67 index; tumor size, type, and grade; P53 accumulation; mitotic index; and presence of angioinvasion and lymphoinvasion. The adequacy of the model was reflected by the area under ROC curve, which ranged from 0.94 to 0.83 for the variables size, clinicopathologic type, Ki67, mitoses, and angioinvasion. High collinearity was observed among Ki67 index, mitotic index, size, histological grade, and clinicopathologic type. On the basis of these observations, four models of multivariate logistic analysis were developed (Table 2). All models were statistically significant, and their adequacy proved high, with areas under the ROC curve close to 1; in all of them, angioinvasion (but not

7 538 RINDI ET AL. GASTROENTEROLOGY Vol. 116, No. 3 Table 2. Multivariate Logistic Analysis of 102 Gastric Endocrine Tumors Predictors of malignancy Predictors of outcome Variable OR 95% CI P (variable) P (model) ROC HR 95% CI P (variable) P (model) Model A Age ( yr) Sex (M/F) Angioinvasion (yes/no) P53 a ( 1/ 1) Ki67 b ( 150/ 150) Model B Age ( yr) Sex (M/F) Angioinvasion (yes/no) P53 a ( 1/ 1) Mitoses b 1 7/ / Model C Age ( yr) Sex (M/F) Angioinvasion (yes/no) P53 a ( 1/ 1) C-P type c (1-step change) Model D Age ( yr) Sex (M/F) Angioinvasion (yes/no) P53 a ( 1/ 1) Size (cm)( 2/ 2) a Percentage of tumor cells. b 10 HPF. c Clinicopathologic type, step by step from types I to II and III and to PDEC. R M lymphoinvasion) emerged as a statistically significant independent predictor. When angioinvasion was combined with tumor type or size, correct identification of malignant behavior was obtained in 85% and 91% respectively, of all tumor cases and correct identification of benign behavior was obtained in 94% and 94% of cases, respectively. When the analysis was restricted to the 86 ECL cell tumors, all of the above variables remained significant (though less powerful) predictors of malignancy at both univariate and multivariate analysis. In particular, combination of angioinvasion with tumor type or size (ROC 0.91 and 0.94) allowed correct identification of malignant behavior in 80% and 81% of cases, respectively, and of benign behavior in 93% and 94%, respectively. Only when the 5 G2 ECL cell tumors were also dropped did Ki67 lose significant prediction by both univariate and multivariate analyses. However, angioinvasion (in all models of Table 2), clinicopathologic type (as a continuous variable or as type III vs. type I tumors in multivariate model C), size ( 2 cm vs. 2 cm in model D), mitoses (1 8 mitoses 10 HPF vs. no mitoses in model B), grade 1b vs. 1a, and lymphoinvasion (at univariate analysis only) retained significant predictive power. Identification of Predictors of Survival Univariate survival analysis showed that of 16 variables investigated (sex, age, site, number of lesions, size, grade, mitotic index, Ki67 index, BCL-2 expression, P53 accumulation, lymphoinvasion, angioinvasion, clinicopathologic type, gastric wall invasion, local and distant metastasis), all but age and BCL-2 expression were statistically significant predictors (P 0.005). The highest HRs were observed for 7 variables: presence of distant metastases (HR, 32.2), size 2 cm (HR, 33.14), presence of angioinvasion (HR, 67.46), deep wall invasion (HR, 68.26), grade 2 and 3 histology (HR, 79.86), mitotic index 8 (HR, 87.24), and Ki67 index 150 (HR, ). As for the multivariate analysis of malignancy risk (see previous paragraph), four multivariate models, each one involving tumor Ki67 expression, mitoses, clinicopathologic type, or size, were developed to assess the survival HR (Table 2). Similar findings were obtained with two other models involving histological grade or level of

8 March 1999 PROGNOSIS OF GASTRIC ENDOCRINE TUMORS 539 gastric wall invasion. All models were statistically significant, with R M ranging from 0.50 to 0.56; with the exception of model A, in all of them the variable angioinvasion emerged as an independent survival risk factor. The Kaplan Meier curves based on this analysis are shown in Figure 3. Discussion Extensive investigation of 102 gastric endocrine tumors allowed identification of 11 variables (tumor size, site, number, histological grade, mitotic index, Ki67 or P53 protein expression, angioinvasion, lymphoinvasion, clinicopathologic type, and patient s sex) that, upon univariate analysis, proved to be significant predictors of malignancy as defined by metastases and/or deep gastric wall invasion. Mitotic index, Ki67 index, histological grade, size, and clinicopathologic type were found to be strongly correlated to each other, and each was an independent predictor of malignancy in separate models of logistic regression analysis. Angioinvasion was the only factor found to be an independent predictor in all models. The best models predicting tumor behavior were found to be a combination of angioinvasion with clinicopathologic type, mitoses, Ki67 index, and size. When our analysis was restricted to the 86 ECL cell carcinoids, the above factors were found to retain a predictive capacity by univariate analysis and, for angioinvasion, grading, type, size and mitotic index, by multivariate analysis. Separate analysis of the 81 grade 1 ECL cell tumors showed that angioinvasion, tumor type, size, and grade retained a significant malignancy-predictive power. In general, this study confirmed the importance of the clinicopathologic classification of gastric endocrine tumors into three types of ECL cell carcinoids (depending on the associated gastric pathology) and PDECs, formerly called neuroendocrine carcinomas. 7 However, more precise evaluation of individual cases was obtained when the above tumor parameters were also considered. Among G1 tumors, growth restricted to the mucosa (29 cases) and/or tumor size 1 cm (43 cases) predicted benign behavior perfectly. Of the 46 tumors identified by these criteria, only 2 (both involving the submucosa) showed angioinvasion, and none developed metastases or deeply invaded (i.e., beyond the submucosa) the gastric Figure 3. Cumulative proportion of patients surviving (Kaplan Meier curves) according to (A) angioinvasion, (B) tumor size, (C) number of mitoses, and (D) combination of clinicopathologic type with histological grade. The 2 curves of A and B as well as the 3 curves of C differ from each other significantly (P 0.000). In D, a significant difference was found between grade 2 3 and grade 1 patients (P 0.000) as well as between grade 2 and grade 3 (P 0.02); differences between types inside grade 1 were not significant.

9 540 RINDI ET AL. GASTROENTEROLOGY Vol. 116, No. 3 wall during a median follow-up of 39 months (2648 person-months). This indolent behavior is in keeping with previous findings and most reports from the literature and supports a conservative approach for the treatment of nonangioinvasive, grade 1 ECL cell carcinoids 1 cm in size, especially when of type I, i.e., associated with diffuse corpus-fundus (type A) CAG and hypergastrinemia. 1 3,7,14,15,32 However, when they are 1 cm or more in size or angioinvasive, all tumors should undergo investigation for possible deep wall invasion, 31,32 a pattern we found to be closely related with metastases and highly predictive of patient outcome. To suppress hypergastrinemia and thus to prevent gastrin-promoted tumor progression or recurrence, antrectomy has been proposed in CAG-associated cases Favorable results have been reported in some, although not all, patients. An octreotide suppression test has been proposed recently to test whether tumor cells retain sensitivity to gastrin, thus predicting a beneficial outcome of antrectomy. 36 Helicobacter pylori infection, a potential cause of gastric gland atrophy 37,38 and ECL cell proliferation, 13 has been found in approximately 20% of CAG associated with ECL cell carcinoids, often with substantial amounts of surviving parietal cells. 21 In some cases of H. pylori positive atrophic gastritis, bacterial eradication is followed by considerable restoration of acidopeptic glands and acid secretion with reduction of gastrinemia. 39,40 Therefore, before antrectomy is considered, bacterial eradication should be obtained in any H. pylori positive patient with CAG-associated carcinoidosis. Although the prognosis of most patients with MEN ZES and type II gastric ECL cell tumors depends on the behavior of associated pancreatic or duodenal gastrinomas more than on the behavior of gastric tumors, one ECL cell tumor caused the death of a patient in our series and a few other malignant cases have been observed (R. T. Jensen, personal communication). 22 In such patients, careful search for associated pancreatic, duodenal, parathyroid, or other tumors and family investigation for the MEN-1 gene mutation 10 are mandatory. For type III (sporadic) ECL cell tumors, a surgical approach is recommended, 31,41 even when they show grade 1 histology, if they are 1 cm or more in size or invade at least the submucosa. Indeed, 7 of 11 tumors showing this pattern proved malignant in our series, including 5 with metastases. An important new finding of this study was the identification of five cases of sporadic type III ECL cell tumors showing grade 2 histology which, at least in this limited series, were all malignant. Such tumors differed from the remaining ECL cell carcinoids in that they had greater size, mitotic index, and Ki67 index; more frequent angioinvasion and distant metastasis; and shorter survival. However, the median survival of G2 ECL cell tumors (17 months; quartiles 5 17) was longer than that of PDECs (8 months; quartiles 4 12). Sporadic malignant carcinoids with clinicopathologic patterns and outcomes similar to those of the present G2 cases, including cases with atypical carcinoid syndrome, 20 have been reported in the literature, 42,43 although they were not clearly separated from PDECs. Grade 2 tumors may represent a distinct category of tumors with actual malignancy intermediate between those of PDECs and of grade 1 ECL cell tumors. Grade 3 histology allowed for ease of separation of the 16 PDEC cases from the 86 ECL cell carcinoids. This separation, also supported by differences in metastatic, invasive, and proliferative patterns, may fit in with recent genetic evidence showing that intestinal PDECs share loss of heterozygosity for the p53, APC, and DCC genes with ordinary, nonendocrine adenocarcinomas but not with carcinoids. 22 Such genetic differences would support a separate origin of at least most PDECs from pluripotent epithelial stem cells and of carcinoids from differentiated endocrine cells. 22 Our finding of extensive P53 protein accumulation in PDECs, also involving residual foci of glandular growth, may fit with this interpretation. However, the occasional observation of carcinoid-pdec tumors either as intimately admixed growths in type A CAG 21 or as concomitant, separate lesions with common genetic alterations in MEN-1 syndrome 22 and our detection of P53 protein in 50% of type III carcinoids investigated suggest that progression of some ECL cell tumors to PDECs might also occur. The poor outcome of PDECs supports previous studies recommendation of aggressive surgery coupled with chemotherapy for such high-grade malignancies. 11 Similar treatments are probably also worthy of consideration for grade 2 carcinoids 44 and have been applied with some success to metastatic tumors associated with the carcinoid syndrome. 41 In conclusion, in our series histological grade 2 or 3, size of 3 cm or more, mitotic index of 9 or more, and/or Ki67 index of 300 or more (all perfect predictors of malignancy) characterized 26 of 33 (77%) malignant gastric endocrine tumors, whereas size below 1 cm and/or growth restricted to the mucosa (both perfect predictors of benign behavior) identified 46 of 69 (67%) tumors showing no actual malignancy at the time of diagnosis as well as benign behavior during follow-up. The malignancy risk of the remaining tumors could be assessed on the basis of angioinvasion, clinicopathologic type III, grade 1b, 1 8 mitotic index, and size from 1 to 2.9 cm. The same parameters, as well as histological grade and

10 March 1999 PROGNOSIS OF GASTRIC ENDOCRINE TUMORS 541 level of gastric wall invasion, also predicted patient s outcome. Thus, parallel analysis of tumor growth and patient s survival allowed (1) confirmation of the benign nature and favorable outcome of most gastrin-promoted, CAG- or ZES-associated (type I or II) ECL cell carcinoids; (2) separation of a subset of sporadic (type III) ECL cell carcinoids characterized by grade 2 histology and showing an outcome intermediate between that of grade 1, benign or low-risk tumors, and that of grade 3, highly malignant cancers; and (3) development of a system of malignancy and outcome prediction that improves the capacity to assess the prognosis of individual tumors and may provide a basis for diagnostic and treatment guidelines of the different tumor forms. 45 References 1. Carney JA, Go VLW, Fairbanks VF, Moore SB, Alport EC, Nora FE. The syndrome of gastric argyrophil carcinoid tumors and nonantral gastric atrophy. Ann Intern Med 1983;99: Borch K, Renvall H, Liedberg G. Gastric endocrine cell hyperplasia and carcinoid tumors in pernicious anemia. Gastroenterology 1985;88: Muller J, Kirchner T, Muller-Hermelink HK. Gastric endocrine cell hyperplasia and carcinoid tumours in atrophic gastritis type A. Am J Surg Pathol 1987;11: Solcia E, Capella C, Fiocca R, Rindi G, Rosai J. Gastric argyrophil carcinoidosis in patients with Zollinger Ellison syndrome due to type 1 multiple endocrine neoplasia. A newly recognized association. Am J Surg Pathol 1990;14: Bordi C, Yu JY, Baggi MT, Davoli C, Pilato FP, Baruzzi G, Gardini A, Zamboni G, Franzin G, Papotti M, Bussolati G. Gastric carcinoids and their precursor lesions. A histological and immunohistochemical study of 23 cases. Cancer 1991;67: Lehy T, Cadiot G, Mignon M, Ruszniewski P, Bonfils S. Influence of multiple endocrine neoplasia, type 1, on gastric endocrine cells in patients with the Zollinger Ellison syndrome. Gut 1992;33: Rindi G, Luinetti O, Cornaggia M, Capella C, Solcia E. Three subtypes of gastric argyrophil carcinoids and the gastric neuroendocrine carcinoma: a clinicopathological study. Gastroenterology 1993;104: Thomas RM, Baybick JH, Elsayed AM, Sobin LH. Gastric carcinoids. An immunohistochemical and clinicopathological study of 104 patients. Cancer 1994;73: Debelenko LV, Emmert-Buck MR, Zhuang Z, Epshteyn E, Moskaluk CA, Jensen RT, Liotta LA, Lubensky IA. The multiple endocrine neoplasia type I gene locus is involved in the pathogenesis of type II gastric carcinoids. Gastroenterology 1997;113: Chejfec G, Gould V. Malignant gastric neuroendocrinomas. Hum Pathol 1977;8: Matsui K, Kitagawa M, Miwa A, Kuroda Y, Tsuji M. Small cell carcinoma of the stomach: a clinicopathological study of 17 cases. Am J Gastroenterol 1991;86: Håkanson R, Tielemans Y, Chen D, Anderson K, Ryberg B, Mattsson H, Sundler F. The biology and pathobiology of ECL cells. Yale J Biol Med 1992;65: Kidd M, Miu K, Tang LH, Perez-Perez GI, Blaser MJ, Sandor A, Modlin IM. Helicobacter pylori lipopolysaccharide stimulates histamine release and DNA synthesis in rat enterochromaffin-like cells. Gastroenterology 1997;1113: Rappel S, Altendorf-Hofmann A, Stolte M. Prognosis of gastric carcinoid tumours. Digestion 1995;56: Rindi G, Bordi C, Rappel S, La Rosa S, Stolte M, Solcia E. Gastric carcinoids and neuroendocrine carcinomas: pathogenesis, pathology and behavior. World J Surg 1996;20: Rosai J, Carcangiu ML, DeLellis RA. Tumors of the thyroid gland. In: Rosai J, ed. Atlas of tumor pathology. 3rd series, fascicle 5. Washington, DC: Armed Forces Institute of Pathology, Abbona GC, Papotti M, Gasparri G, Bussolati G. Proliferative activity in parathyroid tumors as detected by Ki67 immunostaining. Hum Pathol 1995;26: LaRosa S, Sessa F, Capella C, Riva C, Leone BE, Klersy C, Rindi G, Solcia E. Prognostic criteria in nonfunctioning pancreatic endocrine tumours. Virchows Arch 1996;429: Sweeney EC, McDonnell L. Atypical gastric carcinoids. Histopathology 1980;4: Ahlman H, Dahlstrom A, Enerback L, Granerus G, Nilsson O, Persson S, Tissell LE. Two cases of gastric carcinoids: diagnostic and therapeutic aspects. World J Surg 1988;12: Solcia E, Rindi G, Fiocca R, Villani L, Buffa R, Ambrosiani L, Capella C. Distinct patterns of chronic gastritis associated with carcinoid and cancer and their role in tumorigenesis. Yale J Biol Med 1993;65: Bordi C, Falchetti A, Azzoni C, D Adda T, Canavese G, Guariglia A, Santini D, Tomassetti P, Brandi ML. Aggressive forms of gastric neuroendocrine tumors in multiple endocrine neoplasia type I. Am J Surg Pathol 1997;21: Azzoni C, Doglioni C, Viale G, DelleFave G, De Boni M, Caruana P, Ferraro G, Bordi C. Involvement of BCL-2 oncoprotein in the development of enterochromaffin-like cell gastric carcinoids. Am J Surg Pathol 1996;20: Vortmeyer AO, Lubensky IA, Merino MJ, Wang CY, Pham T, Furth EE, Zhuang Z. Concordance of genetic alterations in poorly differentiated colorectal carcinomas and associated adenocarcinomas. J Natl Cancer Inst 1997;89: Solcia E, Bordi C, Creutzfeld W, Dayal Y, Dayan AD, Falkmer S, Grimelius L, Havu N. Histopathological classification of nonantral gastric endocrine growths in man. Digestion 1988;41: Annibale B, Bonamico M, Rindi G, Villani L, Ferrante E, Vania A, Solcia E, Delle Fave G. Antral gastrin cell hyperfunction in children. Gastroenterology 1991;101: Hsu SM, Raine L, Fanger H. Use of avidin-biotin-peroxidase complex (ABC) in immunoperoxidase technique. J Histochem Cytochem 1981;29: Cattoretti G, Pileri S, Parravicini C, Becker MHG, Poggi S, Bifulco C, Göran K, D Amato L, Sabattini E, Feudale E, Reynolds F, Gereds J, Rilke F. Antigen unmasking on formalin-fixed paraffinembedded tissue sections. J Pathol 1993;171: Tilford MJ, Robertson PK, Fiser DH. Using 1fit and 1roc to evaluate mortality prediction models. Stata Tech Bull 1995;28: Schemper M, Stare J. Explained variation in survival analysis. Stat Med 1996;15: Wänberg B, Grimelius L, Granerus G, Conradi N, Jahnsson S, Ahlman H. The role of gastric resection in the management of multicentric argyrophil gastric carcinoids. Surgery 1990;108: Wiedenmann B, Jensen RT, Mignon M, Modlin CJ, Skogseid B, Doherty G, Öberg K. Preoperative diagnosis and surgical management of neuroendocrine gastroenteropancreatic tumors: general recommendations by a consensus workshop. World J Surg 1998;22: Richards AT, Hinder RA, Harrison AC. Gastric carcinoid tumors associated with hypergastrinaemia and pernicious anaemia. Regression of tumors by antrectomy. S Afr Med J 1987;72: Eckhauser FE, Lloyd RV, Thompson NW, Raper SE, Vinik AI. Antrectomy for multicentric, argyrophil gastric carcinoids. A preliminary report. Surgery 1988;104:

11 542 RINDI ET AL. GASTROENTEROLOGY Vol. 116, No Hirschowitz J, Griffith H, Pellegrin D, Cummings OW. Rapid regression of ECL-cell gastric carcinoids in pernicious anemia after antrectomy. Gastroenterology 1992;102: Higham AD, Dimaline R, Varro A, Attwood S, Armstrong G, Dockray GJ, Thompson DG. Octreotide suppression test predicts beneficial outcome from antrectomy in a patient with gastric carcinoid tumor. Gastroenterology 1998;114: Kuipers EJ, Uyterlinde AM, Pena AS, Pals G, Nelis GF, Festen HPM, Meuwissen SGM. Long term sequelae of Helicobacter pylori gastritis. Lancet 1995;343: Eissele R, Brunner G, Simon B, Solcia E, Arnold R. Gastric mucosa during treatment with lansoprazole: Helicobacter pylor is a risk factor for argyrophil cell hyperplasia. Gastroenterology 1997;112: Tucci A, Biasco G, Paparo GF. Effect of eradication of Helicobacter pylori in patients with fundic atrophic gastritis. N Engl J Med 1997;336: Annibale B, Marignani M, Azzoni C, D Ambra G, Caruana P, D Adda T, Delle Fave G, Bordi C. Atrophic body gastritis: distinct features associated with Helicobacter pylori infection. Helicobacter 1997; 2: Ahlman H, Wängberg B, Nilsson O, Grimelius L, Granerus G, Modlin IM, Stenqvist O, Schersten T. Aspects of diagnosis and treatment of the foregut carcinoid syndrome. Scand J Gastroenterol 1992;27: Mendelsohn G, De La Monte S, Dunn JL, Yardley JH. Gastric endocrine tumors, endocrine cell hyperplasia, and associated intestinal metaplasia. Cancer 1987;60: Granberg D, Wilander E, Stridsberg M, Granerus G, Skogseid B, Öberg K. Clinical symptoms, hormone profiles, treatment, and prognosis in patient with gastric carcinoids. Gut 1998;43: Johnson LA, Lavin P, Moertel CG, Weiland L, Dayal Y, Doos WG, Geller SA, Cooper HS, Nime F, Massé S, Simson IW, Sumner H, Fölsch E, Engstrom P. Carcinoids: the association of histological growth pattern and survival. Cancer 1983;51: Modlin IM, Tang LH. Approaches to the diagnosis of gut neuroendocrine tumors: the last word (today). Gastroenterology 1997;112: Received June 26, Accepted November 17, Address requests for reprints to: Enrico Solcia, M.D., Department of Human Pathology, via Forlanini 16, Pavia, Italy. Fax: (39) ; grindi@ipv36.unipv.it. Supported by grants from the Italian Ministry of Health (to IRCCS Policlinico San Matteo), from Ministero dell Universita e Ricerca Scientifica e Tecnologica (MURST) and University of Pavia (to G.R. and E.S.), and from MURST, University of Parma, and Associazione Italiana per la Ricerca sul Cancro (to C.B.). S.L.R. is supported by an Anna Villa Rusconi Foundation Fellowship. Oddi of the sphincter of Oddi Ruggero Oddi ( ) was born in Perugia, Italy, where he began his medical studies, then transferred to the university at Bologna. His doctoral thesis was based on a meticulous investigation of the sphincter that monitors the egress of bile and pancreatic juice into the duodenum. In fact, it was Francis Glisson who first described the sphincter in However, Glisson postulated the main purpose of the sphincter as preventing entrance of intestinal contents into the common bile duct, whereas Oddi s elucidation of its function and dysfunction was more precise and accurate. Therefore, the eponym in Oddi s favor is deserved. Sadly, in later life Oddi became embroiled in unseemly controversies and on more than a few occasions was brought to trial in Italy on charges of abusive commerce of medical products. A broken man, he resorted to morphine for surcease and died in disgrace. We prefer to remember the Oddi of his younger days when his research led to notable advances in our knowledge of biliary and pancreatic secretion. Contributed by WILLIAM S. HAUBRICH, M.D. Scripps Clinic and Research Foundation, La Jolla, California