Overall survival by primary tumor size.
Overall survival by lymph node involvement.
Overall survival by presence of distant metastatic disease.
A, Overall survival by stage for appendiceal carcinoid specific death. Stages I and II overlap and are the same. B, Overall survival by stage for entire database.
Christine S. Landry, Charles Woodall, Charles R. Scoggins, Kelly M. McMasters, Robert C. G. Martin. Analysis of 900 Appendiceal Carcinoid Tumors for a Proposed Predictive Staging System. Arch Surg. 2008;143(7):664–670. doi:10.1001/archsurg.143.7.664
Appendiceal carcinoid tumors (ACTs) are rare, and little is known about the long-term prognosis for these tumors because no staging system exists. Therefore, we sought to investigate prognostic factors associated with ACTs and to create a predictive staging system to accurately estimate prognosis.
In patients with ACTs, TNM staging will accurately predict prognosis.
Retrospective review of 15 983 patients with carcinoid tumors in the Surveillance Epidemiology and End Results (SEER) database from January 1, 1977, to December 31, 2004.
SEER database study.
Nine hundred patients with ACTs (552 females and 348 males; mean age, 47.1 years [age range, 9-89 years]; mean size of the primary tumor, 2.4 cm [range, 0.1-11.5 cm]).
Main Outcome Measure
Clinicopathologic features in patients with ACTs that affect prognosis using a newly created TNM staging system incorporating these parameters.
Lymph node metastasis was found in 137 patients (24%), and distant metastatic disease in 89 patients (10%). Stage-specific survival was statistically significant between stages (P < .001) but not within stages. At multivariate analysis, patient age, primary tumor size, histologic features, lymph node involvement, and distant metastasis were significant factors predicting survival.
Our newly developed TNM staging system accurately predicts prognosis in patients with ACTs. A TNM staging system for ACTs will be helpful not only for physician education about factors that affect the outcome with this disease but also to observe trends in prognosis.
Although carcinoid tumors are the most common malignant lesions of the appendix, appendiceal carcinoid tumors (ACTs) rarely occur.1 This malignant lesion is most commonly discovered during routine appendectomy,2 and little is known about the long-term prognosis. There are 4 histologic subtypes of ACTs: goblet cell, composite cell, enterochromaffin cell, and atypical cell. To our knowledge, no large series has evaluated the significance of these subtypes. No staging system exists for ACTs, and there is controversy about the appropriate surgical management. We sought to investigate prognostic factors associated with ACTs and to create a predictive staging system to accurately estimate prognosis.
The study population was obtained from the National Cancer Institute Surveillance Epidemiology and End Results Program (SEER) database. Begun on January 1, 1973, the nationwide SEER database includes patient demographic data, overall survival, tumor morphologic features, stage at diagnosis, and first course of treatment in a group of patients with cancer. Approximately 26% of the US population, in 18 geographic areas, is represented in the SEER database.2
In searching the SEER database for 1977 to 2004, on May 12, 2007, we identified 15 983 patients with gastrointestinal carcinoid tumors (stomach, colon, rectum, small bowel, and appendix). Nine hundred of these patients had a diagnosis of ACT. Clinicopathologic data evaluation in this patient population included patient age, sex, and race/ethnicity; primary tumor size, histologic features, depth of invasion, and degree of differentiation; lymph node involvement; metastatic disease; and the number of additional malignant lesions.
Patients were excluded from survival analyses if the cause of their death was heart disease; septicemia; atherosclerosis; non-Hodgkin lymphoma; diabetes mellitus; lung or bronchial disease; accident or adverse event; cerebrovascular disease; stomach or duodenal ulcer; pneumonia or influenza; acute myeloid leukemia; chronic liver disease or cirrhosis; brain or other nervous system disease; infectious or parasitic disease; homicide or legal intervention; suicide or self-inflicted injury; aortic aneurysm and dissection; nephritis, nephrosis, or nephrotic syndrome; signs or symptoms of ill-defined conditions; chronic obstructive pulmonary disease; disease of the arteries, arterioles, or capillaries; or disease of the larynx, breast, stomach, prostate gland, uterine cervix, small intestine, urinary bladder, kidney or renal pelvis, or rectum or rectosigmoid junction. Patients in whom the cause of death was miscellaneous malignant tumor, disease of the colon excluding the rectum, or neoplasms in situ or with benign or unknown behavior were considered to have died of ACT if they had only 1 primary malignant lesion. Patients with more than 1 primary malignant lesion were also excluded from survival analysis.
Multiple characteristics were analyzed to determine the effect on overall survival using the Kaplan-Meier method and χ2 analysis (log-rank test). Variables evaluated included patient age, sex, and race/ethnicity; tumor size, histologic features, depth of invasion, and degree of differentiation; lymph node involvement; and metastatic disease. Multivariate analysis was performed for statistically significant variables using the Cox proportional hazards model.
Statistically significant variables at univariate and multivariate analysis were used to create a staging system for ACTs. T stage was determined by analyzing both size of the primary tumor and depth of invasion. Because depth of invasion was not significant at univariate analysis, only tumor size was used for staging. To determine how to differentiate tumors within the T stage, tumors were classified in 1-cm increments according to size at diagnosis. Survival was compared between each increment using the Kaplan-Meier method and χ2 analysis. Three groups were created within the T stage according to how different increments varied in overall survival. These groups were then compared with each other for overall survival to ensure statistical significance within the T stage. Nodal and distant metastases were also incorporated into the staging system.
Patients in the database were classified into the newly developed staging system. Only patients with a cause of death due to ACTs were included in the staging. Patients who died with more than 1 primary malignant lesion were not included in the staging because this might have affected overall survival. Pairwise comparisons were performed between stages using Kaplan-Meier survival curves and χ2 analysis (log-rank test) to assure statistical significance.
Because carcinoid tumors are often associated with other malignant lesions, we further analyzed the effect on overall survival in patients having multiple cancers. In addition, survival with surgical resection was evaluated. The type of resection and relation to stage at diagnosis were also examined among patients in this database.
Of the 900 patients in the database, 552 were female (61%) and 348 were male (39%). Mean age at diagnosis was 47.1 years (median age, 47 years; age range, 9-89 years) (Table 1).
Univariate and multivariate analyses were performed on the ACT-specific death data set (Table 2). At univariate analysis, female sex, smaller size of primary tumor, greater degree of differentiation, absence of lymph node involvement, absence of distant metastatic disease, and enterochromaffin cell histologic subtype predicted improved survival. At multivariate analysis, only age, tumor size, nodal metastasis, distant metastasis, and histologic subtype remained significant independent predictors of survival (Table 2).
Size of the primary tumor was compared for tumors smaller than 1 cm, 1 cm or larger but smaller than 2 cm, 2 cm or larger but smaller than 3 cm, 3 cm or larger but smaller than 4 cm, 4 cm or larger but smaller than 5 cm, and 5 cm or larger. There was not a significant difference in survival between tumors smaller than 1 cm and tumors 1 cm or larger but smaller than 2 cm (P = .20). Tumors smaller than 2 cm were associated with a better predicted overall survival than were tumors 2 cm or larger but smaller than 3 cm (P = .03). Similarly, tumors smaller than 2 cm were associated with a better predicted survival than were tumors 3 cm or larger but smaller than 4 cm, 4 cm or larger but smaller than 5 cm, and 5 cm or larger (all P < .001). Tumors 2 cm or larger but smaller than 3 cm were associated with a similar prognosis as tumors 3 cm or larger but smaller than 4 cm (P = .08) and tumors 4 cm or larger but smaller than 5 cm (P = .12). However, tumors 2 cm or larger but smaller than 3 cm were associated with a better predicted survival than were tumors 5 cm or larger (P = .03). There was no statistical difference between tumors 3 cm or larger but smaller than 4 cm compared with tumors 4 cm or larger but smaller than 5 cm (P = .66) and tumors 5 cm or larger (P = .94). From this analysis, we divided tumors into 3 categories: T1, smaller than 2 cm; T2, 2 cm or larger but smaller than 3 cm; and T3, 3 cm or larger. Survival was statistically significant between these groups (P < .001) (Figure 1).
Well-differentiated tumors were associated with a significantly better prognosis than moderately or poorly differentiated tumors (P = .02 and P = .001, respectively). However, survival associated with moderately differentiated or poorly differentiated tumors was not significantly different (P = .13). Depth of tumor invasion was not significant at univariate analysis (Table 2). Lymph node involvement and distant metastatic disease did not have a significant effect on prognosis (both P < .001) (Figure 2 and Figure 3).
Significant variables at multivariate analysis were incorporated into a staging system consistent with the American Joint Committee on Cancer TNM staging system. The T classification was categorized according to tumor size: smaller than 2 cm, 2 cm to 3 cm, or 3 cm or larger; the N classification was classified as 0 for no lymph node metastasis and 1 for nodal metastasis; and the M classification was categorized as 0 for the absence of distant metastasis and 1 for the presence of distant metastasis (Table 3 and Table 4). Individual stages were created by comparing all combinations of T, N, and M status using the Kaplan-Meier method and χ2 analysis via pairwise comparisons.
Tumors in patients who died of known carcinoid disease were staged with the newly developed staging system. One hundred twelve tumors were classified as stage I, 40 as stage II, 66 as stage III, and 55 as stage IV. Survival between stages I and II in the database could not be analyzed because no patients died of stage I or II disease. There was a statistically significant difference between stages II and III (5 deaths [7.5%]; P = .05) and between stages III and IV (25 deaths [45%]; P < .001). Figure 4A shows the survival curve with the new staging system using the modified database (P < .001). Likewise, a survival curve for the entire database is shown in Figure 4B (P < .001). For the entire database, 116 tumors were classified as stage I, 45 as stage II, 79 as stage III, and 89 as stage IV. When stage I (4 deaths [4.3%]) and stage II (5 deaths [11%]) were compared in the entire database, there was no statistical difference in overall survival (P = 1.00). There were 19 deaths (24%) in the group with stage III disease and 59 deaths (66%) in the group with stage IV disease. For disease-specific survival, 5-year survival was 100% for stages I and II, 78% for stage III, and 32% for stage IV. The 5-year overall survival rates for the entire database, including patients with noncarcinoid causes of death and additional malignant lesions, were 93% for stage I, 78% for stage II, 53% for stage III, and 22% for stage IV (Table 5).
Patients who underwent surgical resection tended to have better survival than patients who did not undergo surgery (P = .01). Excisional biopsy was performed in 20 patients, local excision in 30, hemicolectomy in 204, partial colectomy in 188, colectomy and adjacent organ resection in 40, total colectomy in 6; and surgical resection not otherwise specified in 235. No surgery was performed in 8 patients (Table 6). Of the patients who did not undergo surgery, 2 patients had distant metastases, and the tumor stage was unknown in the remaining 6 patients.
From our analysis, the SEER registry reported that 5.63% of the carcinoid tumors detected from January 1, 1977, to December 31, 2004, were of appendiceal origin. In 2003, Modlin et al3 analyzed the SEER database from 1973 to 1999 and reported a 4.77% incidence of ACTs among all patients with carcinoid tumors. This was a sharp decline from previous tumor registries such as the End Results Group, which reported a 44% incidence from 1950 to 1969, and the Third National Cancer Survey, which reported a 35% incidence from 1969 to 1971.3 The lower incidence in the SEER database may be because only tumors that were considered malignant were reported after 1986.4 Whether the actual incidence of ACTs is decreasing is unknown, especially inasmuch as carcinoid tumors are not staged and many institutions do not consider them malignant.
Analysis of the SEER database demonstrated the mean size of the primary tumor to be 2.4 cm at diagnosis. This database likely does not represent the average patient with carcinoid tumors because many studies have reported that most ACTs are smaller than 2 cm at diagnosis.5- 10 The predominant reason why the SEER database likely does not represent the general population is that most tumors are diagnosed in the community during routine appendectomy and are, therefore, not reported to the SEER registry.
Primary tumor size was a statistically significant factor in overall survival at both univariate and multivariate analyses (Table 2). We analyzed primary tumor size in terms of survival in 1-cm increments but found no difference in survival between tumors smaller than 1 cm and tumors ranging from 1 to 2 cm. We further analyzed the effect of size on overall survival and found statistically worsened survival as tumor size increases. Other studies in the literature confirm the effect of larger size and worsening prognosis.11- 13
We analyzed the effect of depth of tumor invasion on overall prognosis but found no correlation between depth of invasion and overall survival. However, most of the patients in the SEER database had tumor invasion beyond the muscularis propria; only 17% of tumors did not invade beyond the muscularis propria. Data in the literature about depth of invasion of ACTs is limited. However, some investigators have reported a correlation of enlarging tumor size with mesoappendiceal involvement.14,15
Goblet cell carcinoid tumors account for most of the tumors with histologic information in the SEER database. The next most common subtypes were composite cell tumors, followed by enterochromaffin cell and atypical ACTs. Overall, enterochromaffin cell tumors comprised the most favorable histologic subtype. Multiple institutions have reported similar prognostic effects of histology for ACTs. For example, the goblet cell subtype is thought by some to have a higher potential for malignancy than other carcinoid tumors,16- 18 and other studies report that the goblet cell variant may have varying potential for malignancy.19- 21 Likewise, composite cell tumors may be associated with a worse prognosis because of features they share with adenocarcinomas.22
Because of the rarity of ACTs, most studies on this malignant lesion include few patients. As a result, little is known about the true malignant potential of these tumors. In addition, treatment of ACT is unclear, especially in its advanced stages.
We created a staging system using a large population of patients with ACTs. The SEER database was advantageous because it includes a large number of patients with advanced disease. Because these tumors commonly are found as synchronous disease with other primary tumors, we excluded these patients from the analysis to determine the strongest overall factors that influence prognosis. From our analysis, we found that the most important factors in prognosis include primary tumor, nodal involvement, and distant metastasis. Depth of tumor invasion was not a statistically significant predictor of survival and, therefore, was not included in the staging system.
We first created a staging system with 4 stages. Because enlarging primary tumor size and lymph node involvement are strong predictors of survival at univariate and multivariate analyses, stages I and II should be statistically different. Inasmuch as none of the patients with stage I or II tumors died, the distinction between stages I and II may be of little clinical significance. When patients were classified into the staging system from the entire database (overall, not disease-specific survival), there was a differentiation of curves between stage I and stage II tumors. This indicates that there likely is a difference between stage I and II tumors, especially in patients with additional malignant lesions. Longer follow-up and more patient data are needed to determine whether there is a true difference between stage I and II disease.
Because there was no difference between stage I and II disease, we combined these stages into 1 stage and performed an analysis using the Kaplan-Meier method between 3 stages. There was a significant difference between these stages (P≤.001). Prognosis for patients classified into 3 stages can accurately be predicted. However, because there is a difference in the overall database between stages I and II, we recommend differentiating patients with stage I or stage II disease until more definitive data are available.
Most sources report that ACTs larger than 2 cm merit right-sided hemicolectomy.5,23- 25 However, Moertel et al9 from the Mayo Clinic reported that simple appendectomy may be adequate for surgical resection of tumors larger than 2 cm in elderly patients or patients at high operative risk, given the low incidence of recurrence of these tumors. That study10 recommended right-sided hemicolectomy for tumors larger than 2 cm in younger patients and for tumors with vascular involvement and invasion of the mesoappendix. Similarly, results of a study of 48 patients with ACTs by Bamboat and Berger25 from Massachusetts General Hospital suggested simple appendectomy for tumors larger than 2 cm because of the indolent nature of these tumors and their low rate of recurrence. Only 5 patients in this analysis had tumors larger than 2 cm. One of these patients underwent simple appendectomy, and the other 4 underwent right-sided hemicolectomy. Because these studies are based on small numbers of patients, the true malignant potential of these tumors may not have been fully recognized.
Patients in the SEER database underwent a variety of surgical procedures at all stages of the disease. Clearly, patients with advanced disease had shorter overall survival. However, our study did not include enough patients to determine which operative procedure provides the best overall survival at specific stages of disease. Likewise, tumor stage was unknown in most of the patients who did not undergo surgery. Furthermore, there has been some discussion in the literature as to whether the pathologic diagnosis should dictate the type of surgical resection. Some institutions have recommended more aggressive resection of goblet cell carcinoid tumors, whereas other institutions have reported varying degrees of aggressiveness in treating these tumors.18,23 More information is needed to determine the true effect of histologic features on overall prognosis.
Many limitations are associated with use of the SEER database. First, the true incidence of ACTs is unknown because most such tumors found incidentally during emergent appendectomy are not likely to be reported to the SEER registry. Second, the type of surgical resection is not entirely clear for all cases. Appendectomy is not reported as a method of surgical resection in the database. Third, the cause of death in patients registered in the database is also unclear. We assumed that cause of death due to “colon excluding rectum” was because of ACTs. This does include patients who may have died of other conditions, such as diverticulitis. Likewise, patients who may have died of pathologic conditions related to the cancer, such as sepsis or small-bowel obstruction, were excluded from our survival analysis. Fourth, we did not have information on patient comorbidities, lymphovascular invasion, and actual location of the tumor within the appendix. However, even with these limitations, to our knowledge, our study is the largest review of ACTs and demonstrates the effect that a prospective staging system will have on risk stratification, type of treatment, and follow-up. In addition, these limitations can be rectified by instituting a prospective staging system that will enable determination of the true incidence of this tumor type.
Our analysis of ACTs is consistent with multiple studies in which the most important factors affecting overall prognosis include size of the primary tumor, involvement of lymph nodes, and the presence of distant metastases. Other important factors that contribute to overall survival include histologic subtype, degree of differentiation, and number of additional malignant lesions. Depth of tumor invasion was not found to be a predictor of survival.
The implementation of a staging system is imperative to learn more about the prognosis as well as the appropriate treatment for ACTs. Our newly developed staging system accurately predicts disease prognosis. However, more data and longer follow-up are needed to determine whether there is a difference in survival between stage I and stage II tumors. A TNM staging system for ACTs will be helpful not only for physician education about factors that affect outcome for this disease but also to follow trends in prognosis.
Correspondence: Robert C. G. Martin II, MD, Division of Surgical Oncology, Department of Surgery, James Graham Brown Cancer Center, University of Louisville School of Medicine, 315 E Broadway, Room 313, Louisville, KY 40202 (Robert.firstname.lastname@example.org).
Accepted for Publication: January 14, 2008.
Author Contributions:Study concept and design: Martin. Acquisition of data: Landry. Analysis and interpretation of data: Woodall, Scoggins, and McMasters. Drafting of the manuscript: Landry and Martin. Critical revision of the manuscript for important intellectual content: Woodall, Scoggins, and McMasters. Statistical analysis: Landry. Administrative, technical, and material support: Landry, Woodall, Scoggins, McMasters, and Martin. Study supervision: Scoggins and Martin.
Financial Disclosure: None reported.
Previous Presentation: This paper was presented at the 115th Scientific Session of the Western Surgical Association; November 6, 2007; Colorado Springs, Colorado; and is published after peer review and revision. The discussions that follow this article are based on the originally submitted manuscript and not the revised manuscript.
David M. Nagorney, MD, Rochester, Minnesota: Dr Landry and her colleagues have constructed a TMN staging system for appendiceal carcinoids, orphan cancers of the GI [gastrointestinal] tract. Their proposed system showed clear survival stratification by stage. The TMN system for these cancers, if validated, would be useful because the AJCC/UICC [American Joint Committee on Cancer/International Union Against Cancer] systems are not only the most widely used but also are needed for these uncommon cancers currently lacking any reliable staging system to assess and to compare outcomes. Whether their proposal will be considered propitious is yet to unfold. Although I'm optimistic about its chances, I'm not without reservations.
The key points that I gleaned from the study were: (1) 94% were goblet cell or composite histology or adenocarcinoids, (2) less than 40% were well-differentiated cancers, (3) average size was 2.4 cm, (4) 25% were nodal metastases and 10% were distant metastases, and (5) 20% of patients harbored other cancers. The study size was excellent, and the use of the SEER database, though imperfect, was appropriate. Ask yourself, Are these classic carcinoids? I don't think so, and I think that the histopathologic heterogeneity of carcinoids may limit wide use of the TNM system. I have a few questions.
Data reported to SEER are based on reporting overt cancers, and, for carcinoids, experts still argue whether only some or all of appendiceal carcinoids are, in fact, cancer. Few argue that goblet cells or composites are not malignant. If all carcinoids are cancer, how would that affect the staging system? Is there some way to cull classic carcinoids (enterochromaffin type) to import into your model?
Given the impact of carcinoid dedifferentiation and histopathology, should this TNM system be used selectively or for all appendiceal carcinoids?
Finally, although this study was not designed to address surgical treatment, recommendations in the paper were hedged. Given the limited morbidity and mortality of a right hemicolectomy and the expected stage for the carcinoid histopathology presented, can one offer anything less for the treatment of the primary tumor?
Dr Martin: Data reported to the SEER database do demonstrate somewhat of an individual institutional bias with regard to the definition of what is reported as a cancer. For example, an 8-cm ACT without nodal metastases would not be labeled a cancer by most pathologists. This is controversial and represents a limitation today of the staging and reporting. We have not captured and stratified these patients to truly define the biology of these tumors' behavior.
I do believe histopathologic classification should be used in the staging system. However, this data set is limited because a large majority of patients do not have this recorded. With instituting a staging system, we would be able to capture and then revise once the histopathology data are more robust.
As a cancer liaison for the Commission on Cancer and having the burden to do quality assurance on a number of different disease stagings, I find the majority of the required revisions occur with the most complex staging systems: breast and melanoma. For that reason, we started simple, with the plan that we would expand and revise once we captured a greater number of patients with this diagnosis.
We are not in a position to make any type of surgical recommendations from this SEER data set.