Discussion
TNM staging represents a simple and accurate instrument for death-risk assessment and patient management at diagnosis, provided, it accurately reflects the biology and natural history of the cancer for which it was designed. Evidence is essential for TNM development, and this is dramatically true for rare "orphan" cancers. We tested two TNM staging systems to determine which was superior in terms of performance when a large series of pancreatic neuroendocrine neoplasms was used. All results were resistant to a series of sensitivity analyses.
Our data indicate that the ENETS TNM is superior in performance to the UICC/AJCC/WHO 2010 TNM and is more accurate. We observed some differences in distribution of sex and age across the ENETS staging system, whereas this was not observed for the UICC/AJCC/WHO 2010 TNM. Cox regression analysis showed that only the ENETS TNM perfectly allocated patients in four risk groups that were statistically significantly different for death risk and almost equally populated, whereas the UICC/AJCC/WHO 2010 TNM compressed the disease into three differently populated classes with most patients being in stage I, in the equally populated stages II and III (which were similar in terms of death risk) and in stage IV. Furthermore, multivariable modeling demonstrated that although both TNM systems' stages resulted in independent predictors of survival, the UICC/AJCC/WHO 2010 TNM stages consistently showed very large 95% confidence intervals per stage, indicating its inconsistent prediction ability. This latter observation deserves one further comment. The fact that the UICC/AJCC/WHO 2010 TNM staging results an independent predictor of survival at multivariable analysis indicates that measuring the cancer extent is per se relevant, as confirmed by a recent single series investigation. This does not necessarily translate into an effective description of cancer course, as demonstrated here. Building a TNM system requires the use of parameters proven as important for treatment-aimed patient stratification. This appears not to be the case for the UICC/AJCC/WHO 2010 TNM, fitting more the aggressive course of ductal adenocarcinoma (smaller size and early invasion of nearby tissues/vessels) rather than the slow malignant pace of neuroendocrine neoplasms (larger size, rare invasion, and late course to metastasis). Indeed, the perfect patient stratification in four risk classes by the ENETS TNM allows for the progressive approach to more aggressive treatment options, as supported by current guidelines.
Ki67 grading is a useful instrument for the management of patients with neuroendocrine tumors. The ENETS/WHO 2010 grade and a two-step proliferation grading by Ki67 were independent predictors of outcome in a multivariable analysis, confirming previously published analysis of single institution series (9,10,28) and strongly supporting the adoption of the recent WHO 2010 classification. Of note, the WHO grading acquires outmost efficacy in the absence of staging information, supporting its use for small biopsy analysis and in the absence of other clinical information, which is a frequent event in daily pathology practice. The recent proposal of 5% Ki67 as cutoff for G1 (12) fits our finding of 4.8% Ki67 at analysis of data under the ROC curve. However, the substantial efficacy overlap of both systems observed in this series does not justify a change in the current WHO grading, which is otherwise meant for all sites of the gut. In view of the large range for G2 grading (3% to <20%), as defined by the ENETS-WHO 2010 grading, this finding suggests 5% as the actual border for the more aggressive disease. This other piece of information may result in more accurate tailoring of therapy to the individual patient.
Potentially curative surgery resulted also in an independent predictor of outcome at multivariable analysis in the present series. This observation strongly supports the adoption of aggressive strategies for curative surgery in the treatment of neuroendocrine neoplasms of the pancreas, as previously emphasized. It may be interpreted consequently that a more aggressive surgical attitude should be pursued even for small size (≤2 cm) lesions, especially in the sporadic setting. Along these lines, the ROC curve analysis for size identified 3 cm as a possible cutoff for more aggressive disease, in part supporting previous observation made on the basis of a limited case series (9,10,12). Caution should be used, however, in interpreting the effect of treatment in an observational study. A higher malignancy of the tumor might in fact hamper the use or success of surgery.
The cutoff number of lymph nodes identified as potentially being associated with adverse survival, as identified by ROC curve analysis, was one. This indicates that a single lymph node metastasis is per se a clinically significant adverse event. However, the ENETS TNM stage IIIB (any T, N1, M0) unexpectedly identified patients with better fare than patients in Stage IIIA (T4, N0, M0). This could be explained by size differences of the two patient groups (N.39 for IIIA and N.156 for IIIB), or by differences in surgical techniques and lymph node sampling, or may simply indicate that T4 is per se a more adverse parameter than N1. Alternatively, given the T definition (see Table 1 ), stage IIIA may well contain a large number of patients not amenable to curative surgery. These observations suggest the need for TNM adjustment as recently proposed. Along these lines and more importantly, when grouped into a single stage III class, the separation in risk group was statistically significant and progressive, confirming the efficacy of the ENETS TNM staging. Thus, a simplification of the ENETS staging system into four single stage classes only is suggested. In contrast, the UICC/AJCC/WHO 2010 TNM stage system was unable to separate intermediate groups, possibly because of the extremely broad T definition of stages IIA and IIB.
Other interesting findings emerged by the descriptive and univariable analysis of this series, specifically, the overall good performance of neuroendocrine neoplasm patients with net separation of insulinomas and the substantial course overlap of nonfunctioning and other non-insulinoma functioning neoplasms. This corresponds with the somehow unique signature of insulinomas as compared with other pancreas neoplasms. The same applies for high-grade G3 neoplasms, which, similar to the few published series, fare the worst and represent a minority of patients (ie, less than 10% of this series). The most typical profile of the patient with sporadic neuroendocrine neoplasm is male and of the 6th decade, whereas younger female are patients with familial (most likely multiple endocrine neoplasia syndrome type 1) disease. Adverse features observed were being male, older age, non-curative surgery, and sporadic nonfunctioning neoplasms, again confirming previously reported data from the registry and single institution series.
This cohort is the largest ever published on neuroendocrine neoplasms of the pancreas listing such detailed information to our knowledge. This series results from the retrospective collection of neoplastic tissues obtained at surgery from 11 European Institutions known as referral centers for oncology, endocrinology, gastroenterology/pancreatology, and pancreas surgery for neuroendocrine neoplasm patients. To reduce the various selection bias expected and to reflect on the daily clinical practice, our strategy was to collect data representing all settings of presentation/cure of this type of patient. Certainly, and this is largely expected, the most obvious and positive effect is the overall elevated survival observed here at 5 and 10 years, likely depending on a mixture of early diagnosis, accurate surgery, tailored medical therapy, and effective follow-up. The combination of such positive features goes along with what is recommended for the cure of rare cancer disease and for neuroendocrine neoplasms specifically. Along these lines, the widespread adoption of common guidelines based on the WHO classification together with the universal health system in European countries may both explain the overall excellent survival reported here.
The major limitation of this study is its retrospective nature, which implies some degree of variation in collecting relevant data. For example, surgical techniques and lymph node sampling may slightly vary among different institutions and surgical teams, potentially resulting in different nodal status definition. Also, the interpretation of Ki67 staining of cancer cells may slightly vary in different pathology laboratories and by the pathologists' expertise, leading to differential proliferation grading assessments. Additionally, variability in treatment at different centers may have influenced patient survival. Collection of relevant data in a large prospective series with uniform protocols for data entry is needed to confirm our findings.
In conclusion, neuroendocrine cancer patient survival depends on multiple factors, certainly one of the most important being therapy. Accurate selection of patients for therapy is needed when the extent of neuroendocrine cancer disease is measured. This can be achieved only by accurate surgical staging. Our data suggest the ENETS TNM staging system as superior and support its use in current clinical practice.