Key points

Introduction

The field of neuroendocrine oncology has grown significantly over the past decade. Over that time, substantial collaborative efforts have allowed the successful conduct of multiple randomized controlled trials that have changed both the clinical practice of oncology and the scientific practice of conducting future studies. These studies provide ample opportunity for learning in study design and execution, and our intent in this article is to consolidate the lessons learned and offer direction as the field continues to advance.

We would submit that the key principles for neuroendocrine tumor (NET) clinical trials moving forward are the selection of homogeneous patient populations, assessment of standardized criteria for progression and response by real-time centralized review, and rigorous study design. These principles arise from a history of rigorous clinical investigation that has evolved together with improvements in technology that enable us to conduct ever more sophisticated investigations. Similarly, we believe that these issues are central to interpretation of any given study, and should be reviewed when considering the results of any clinical trial.

Introduction

The field of neuroendocrine oncology has grown significantly over the past decade. Over that time, substantial collaborative efforts have allowed the successful conduct of multiple randomized controlled trials that have changed both the clinical practice of oncology and the scientific practice of conducting future studies. These studies provide ample opportunity for learning in study design and execution, and our intent in this article is to consolidate the lessons learned and offer direction as the field continues to advance.

We would submit that the key principles for neuroendocrine tumor (NET) clinical trials moving forward are the selection of homogeneous patient populations, assessment of standardized criteria for progression and response by real-time centralized review, and rigorous study design. These principles arise from a history of rigorous clinical investigation that has evolved together with improvements in technology that enable us to conduct ever more sophisticated investigations. Similarly, we believe that these issues are central to interpretation of any given study, and should be reviewed when considering the results of any clinical trial.

Historical overview

The earliest clinical trials for patients with NET evaluated conventional chemotherapy, and highlight many of the salient issues of clinical trial design in this patient population. These issues include grouping tumors by primary site and clinical aggressiveness, selection of response criteria, and assessment of those criteria.

One of the first studies tested streptozocin in 52 patients with metastatic pancreatic NETs (pNETs) in the 1970s. This relatively large study for the era was conducted after an initial evaluation of the drug in 4 patients with pNET and 4 patients with extrapancreatic NET (carcinoid) revealed 1 response in a patient with pNET and no responses in the carcinoid patients. This later study required the collaboration of 50 investigators to accrue 52 patients, and is notable for its inclusion of exclusively patients with pNET. Following their accrual, patients received standardized doses of the therapy and were followed for response. Response criteria were strictly defined to incorporate both improvements in hormone secretion and tumor volume assessed by physical examination of the assessing investigator. By its nature, this study was uncontrolled, but given the lack of alternative therapies, evidence of relevant activity established streptozocin as the standard therapy for advanced pNET.

Subsequently, 2 randomized studies of approximately 100 patients each were conducted by the Eastern Cooperative Oncology Group (ECOG) to develop streptozocin-based chemotherapy further. The first demonstrated superiority of 5-fluorouracil combined with streptozocin over streptozocin alone in a population of patients with pNET, although it continued to use a composite endpoint of biochemical and measureable response, with approximately one-third of patients eligible for classification of response based on biochemical parameters, and an unknown proportion eligible based on physical examination. Secondary endpoints of progression-free survival (PFS) and overall survival were not statistically different between the 2 arms. The second study evaluated 3 regimens: streptozocin with 5-fluorouracil, streptozocin with doxorubicin, and single-agent chlorozotocin, with doxorubicin/streptozocin demonstrating superiority. Nearly half of all patients in that study could be classified as responders based on biochemical criteria. However, both PFS and overall survival were significantly improved with the combination ( P <.005 for both endpoints and both comparators). These studies established the standard therapy for pNETs until 2011, although notably, later evaluation in the modern era of cross-sectional imaging would suggest that the radiographic response rate of pNETs to streptozocin-based doublet chemotherapy is actually less than 10%. Importantly, these studies highlight some of the key study design issues that continue to arise in the field. Multi-institutional cooperation was required to achieve even modest accrual of a homogeneous group of patients and intermediate endpoints, such as objective radiographic response rate, were used due to feasibility. Also of note, given the challenges of patient accrual, the time lapse between each of these studies was approximately 10 years.

The development of octreotide, an octopeptide somatostatin analogue, was similar in its style and limitations. Following separately reported assessments in 25 carcinoid patients and 22 patients with pNET, consistent evidence of biochemical control was observed, and it became the standard therapy for patients with NETs of all primary sites for control of hormonal syndromes. Octreotide’s development was prescient in its separation of NETs by primary site, but in turn hindered by small sample sizes. No evidence of tumor regression was observed, and no claim of improved PFS or overall survival was made in these early studies. It was not until the P rospective, R andomized Study on the Effect of O ctreotide LAR in the Control of Tumor Growth in Patients With Metastatic Neuroendocrine Mid gut Tumors (PROMID) study that there was evidence from 85 randomized patients that a somatostatin analogue could delay time to progression in patients with midgut NETs. Most patients had minimal liver involvement, and there was no formal assessment of whether patients had progression before enrollment, although half of the enrolled patients were less than 4.3 months from diagnosis. Notably, the PROMID study used standard radiographic criteria to evaluate response and progression, and incorporated blinded central radiology review to determine progression.

Therefore, in considering the critical studies between 1970 and 2010, a steady march of progress is apparent. Themes of attempting to achieve patient population homogeneity through primary site stratification and selection, as well as collaborative adherence to standardized response criteria run throughout. Meanwhile, the evolution of the radiology field allowed for the use of standardized imaging criteria for response and blinded centralized radiology review.

Patient selection

The primary issue in selecting patients with NET for clinical trials is ensuring the relevant homogeneity of the population under study. This is a challenging goal in NETs, as a homogeneous population of patients is required to address a focused research question, but defining eligible patients too narrowly could easily render accrual too slow to answer the question while it is still relevant to clinical practice. A homogeneous population may represent a selected group that is felt to be most likely to benefit from a given therapy, but may also represent a group with uniform prognosis, such that therapeutic impact can be more readily detected. Stratified randomization can allow for a heterogeneous population of patients with distinct prognoses to be balanced with respect to the primary endpoint. This strategy can permit larger studies to be completed without a completely homogeneous population, but allows limited interpretation of subgroup analyses, as the subgroups need not be balanced with respect to the stratification variables.

We have learned much about the biologically relevant heterogeneity of patients with NET in the past several years, as novel agents have been investigated in heterogeneous populations of patients with NET. In phase II studies of everolimus, sunitinib, and pazopanib, clear differences in response rates between pNETs and extrapancreatic NETs were observed, suggesting that different primary sites have distinct biology and should be studied separately. The exact biological reasons for these differential responses remain obscure, but the introduction of next-generation sequencing to the neuroendocrine field has demonstrated clear genomic differences between well-differentiated pancreatic, small bowel, and pulmonary NETs. Although frequent lesions in MEN1, DAXX, ATRX, and the mammalian target of rapamycin pathway were observed in pNETs, only occasional alterations of CDKN1B have been observed in small bowel NETs, whereas ARID1A, EIF1AX, and MEN1 are most commonly altered in pulmonary NETs. However, it remains unclear whether further splitting these tumors by molecular phenotype will be required to achieve biologically relevant homogeneity, and whether these genetic alterations beget similar pathophysiology in the context of different primary sites and histologies.

At present, our thinking is that separating NETs for clinical study by primary site is an appropriate approach when feasible. This is especially important in single-arm studies in which the comparison is against a historical control. In randomized studies, stratification may also be appropriate when primary site is prognostic, but there is reason to believe treatment may benefit a wider study population. In our view, primary site and histology are relevant proxies for unmeasurable biological variables, and as such are useful variables for achieving biologically relevant patient homogeneity in study design and execution.

The strategy of separating NETs by primary site is further supported by recent work in our group about the predictive value of phase II studies. In that work, we conclude that drugs tested in different tumors have measurably distinct odds of eventual approval, in turn suggesting that different phase II designs may be required to reduce the frequency of advancing drugs to phase III studies based on false-positive phase II results. Although it remains to be demonstrated whether NETs of varied primary sites have distinct odds of therapeutic efficacy, the differences observed in the activity of sunitinib and streptozocin argue that pNETs should be considered separately from other NETs, as the prior probability of success is likely to be different.

In addition to grouping patients by primary site, the importance of progressive disease on study entry also has become apparent. Studying this population of patients serves 3 purposes. First, it studies the population of patients in greatest need of therapy, as indolent disease can often be observed without additional therapy in many patients. Second, it studies a more homogeneous population by excluding that fraction of patients who will have stable disease on study regardless of therapy. Finally, by selecting patients with more aggressive disease, it increases the event rate, thereby shortening the necessary follow-up time and accrual requirements. These properties allow for the more rapid completion of studies asking the most relevant questions about the population most in need of therapy. As a result, the field has moved toward including only patients with progressive disease within the 12 months before study enrollment, using that entry criterion in the pivotal studies of everolimus and sunitinib. In these studies, which included only patients with progressive pNETs, the median PFS was reassuringly similar, at 4.6 and 5.5 months, respectively.