Background
Neuroendocrine tumors (NETs) are slow-growing neoplasms that arise from diverse locations, including the pancreas and gastrointestinal (GI) tract. The prognosis of NETs is heterogeneous and tumor progression has been utilized in most phase-3 clinical trials to refine prognostic stratification. Current knowledge of well-differentiated NET growth kinetics and the relationship to treatment are limited; further understanding may improve the assessment of NET progression, prognosis and ultimately treatment.
Response evaluation criteria in solid tumors (RECIST) is used by oncologists and radiologists, and recognized by regulatory bodies as an assessment of tumor response to therapy, including NETs [
1,
2]. RECIST estimates change in tumor burden using the sum of the longest diameters (SLD) of target lesions over a course of treatment which, together with the appearance of new lesions, is transformed into a categorical variable (complete response [CR], partial response [PR], stable disease [SD], or progressive disease [PD]) [
1,
2].
Although a valuable tool widely used in clinical trials, RECIST is a qualitative variable, overemphasizing tumor shrinkage as a successful response to treatment, and requires improved reproducibility [
3,
4]. Limitations of RECIST extend to accurate assessment of NET disease status, which can be confounded by their slow growth kinetics [
5]. As a result, it can be difficult for practitioners to decide how to adapt treatment. Therefore, a metric that provides a more sensitive measure of tumor growth would prove advantageous in slow-growing tumor assessment, including NETs.
An assessment providing dynamic and quantitative evaluation of tumor kinetics may offer a useful complement to RECIST. Tumor slope has been explored as a prognostic factor in studies on solid tumors [
6‐
10], including well-differentiated NETs [
9,
10]. Tumor growth rate (TGR), which is based on change in tumor volume, has been independently associated with progression-free survival (PFS) using data from phase-1 studies in solid tumors (including upper GI and pancreatic tumors) [
11], and with PFS and overall survival (OS) using data from the phase-3 TARGET study (sorafenib compared with placebo) in metastatic renal cell carcinoma (mRCC) [
12]. More recently, an association of TGR with PFS has been reported in a post hoc analysis from a phase-2 single-arm trial of lanreotide depot/autogel 120 mg for non-functioning intestinal/pancreatic NETs in Japanese patients [
13].
The CLARINET core study demonstrated the antitumor efficacy of lanreotide depot/autogel 120 mg/4 weeks in patients with non-functioning intestinal/pancreatic NETs compared with placebo [
14]. The CLARINET open-label extension confirmed long-term safety and efficacy [
15]. Most participants in the core study had SD at baseline according to RECIST v1.0. In these post hoc analyses, core study tumor growth measurements were evaluated using TGR as a measure of tumor progression and response before and during treatment, and as a prognostic factor for PFS.
Discussion
These post hoc analyses were conducted to evaluate the clinical utility of TGR as a measure of tumor progression, response and prognosis in patients with intestinal/pancreatic NETs receiving lanreotide or placebo, using data from the CLARINET study [
14]. Analyses of the use of TGR to assess tumor progression before treatment (TGR
0) revealed a large proportion of patients’ tumors were actively growing during the pre-treatment period, despite classification as SD according to RECIST v1.0. There was no correlation between Ki-67 at screening and TGR
0, despite an attempt to stratify the data by only including patients with biopsies taken within 1 year of treatment start. Analyses of the use of TGR to assess proliferative activity during treatment (TGR
Tx–0) showed a large proportion of patients had reductions in TGR
Tx–0, which tended to be greater with lanreotide compared with placebo, while most patients were still classified as SD according to RECIST v1.0. TGR
Tx–Tx demonstrated the antitumor efficacy of lanreotide compared with placebo, as early as 12 weeks into treatment, by reduction and subsequent stabilization. Analyses of the use of TGR
0 as a prognostic factor showed that by using a TGR
0 cut-off of 4%/month (found to have optimal association with risk of PD/death by ROC analysis), patients with TGR
0 > 4%/month had a four-fold higher risk of PD/death compared with ≤4%/month, in the overall population and within both treatment groups. TGR
0–12 was also found to be prognostic for PFS in both treatment groups. A multivariate analysis identified five factors independently associated with PFS: lanreotide treatment, progression at baseline, TGR
0, hepatic tumor load (> 25%–≤50 and > 50%), and primary tumor type. Ki-67 at baseline was not identified as a potentially important prognostic factor for PFS, prior therapy (yes) was not prognostic in the presence of other terms and, notably, CgA at baseline, and tumor grade were excluded from the final model.
The findings from the present analyses accord with other similar analyses. The advantage of TGR as a rapid measure of tumor response has been reported previously; TGR identified antitumor efficacy during treatment at the first tumor evaluation in phase-1 and -3 studies of mRCC [
11,
12,
16]. The utility of TGR in non-functioning intestinal/pancreatic NETs was first described in a recent single-arm phase-2 trial of 32 Japanese patients [
13]. A numerical reduction in TGR was identified within 12 weeks of initiating lanreotide treatment, despite 65% of patients being classified as PD at baseline (RECIST v1.1), and was sustained from pre-treatment to last-value [
13]. Our analyses expand upon this study in four ways: firstly, they reveal that TGR can detect early differences between lanreotide and placebo groups, despite the slow growth kinetics of NETs; secondly, they demonstrate sustained lanreotide antitumor activity compared with placebo using the large dataset from CLARINET; thirdly they provide additional evidence that although lanreotide reduced TGR versus placebo overall, tumor growth was ongoing in many patients, including those classified as having SD; and finally, in many cases a PD event was the result of ongoing tumor growth, rather than the result of a sudden increase in TGR. Our findings regarding the utility of TGR during treatment (TGR
Tx) for monitoring tumor response and the durability of drug effect also resonate with other analyses from phase-2 and -3 trials in other tumor types [
6,
12,
17,
18]. TGR
0 measurements were suggestive of actively growing tumors and, as seen with other tumor types, TGR
0 did not reflect RECIST status during pre-treatment and early treatment evaluation [
11,
12,
19]. Others have highlighted the inadequacies of RECIST as a measure of tumor response because it involves condensing information into four categories that are defined before treatment, regardless of growth kinetics. Additionally, RECIST may not always be relevant for slow-growing tumors or treatments that stabilize growth, and any observed pseudoprogression, even if uncommon in NETs, may be misrepresented as PD by RECIST [
3,
19,
20]. Our findings that a TGR
0 cut-off of 4%/month and TGR
0–12 are prognostic for PFS are consistent with others who identified an association between growth kinetics and clinical outcomes. In a large retrospective analysis of 20 phase-1 mRCC trials, a 9% decrease in progression hazard was observed with every 10% decrease in TGR (reference compared with experimental period) [
11], and TGR was associated with PFS regardless of treatment (sorafenib or placebo) in a retrospective analysis from the phase-3 TARGET trial in mRCC [
12]. A new response metric for glioblastoma incorporating a linear model of radial tumor expansion computed at first post-radiation scan was prognostic for PFS, and agreed with more complex anatomic and spherical equivalents [
21,
22]. In addition, response outcomes utilizing growth dynamics have also been previously shown to be prognostic for OS in renal and prostate cancer [
12,
17,
18,
21‐
23]. The TGR
0 cut-off of 4%/month could, therefore, be of value within routine clinical practice by providing a prognosis for PFS before treatment start, thus allowing clinicians to make earlier decisions regarding future treatment. TGR
Tx-0 would also be potentially useful in the clinic to identify patients who are or are not benefitting early in the course of their treatment (e.g. TGR
12–0). TGR
0 (≤/> 4%/month), if confirmed to be prognostic for other therapies, and TGR
Tx-0 would be of particular importance when planning and monitoring more toxic treatments, such as peptide receptor radionuclide therapy and chemotherapy, ensuring these are continued only for as long as is necessary.
The lack of correlation between Ki-67 and TGR
0 in this analysis was surprising. However, known difficulties in assessment of the Ki-67 index (intra- and intertumoral staining heterogeneity and counting methods, for example) may, in part, account for this [
24,
25]. Ki-67 was not identified as a potentially important prognostic factor for PFS from the exploratory multivariate analysis, despite the known importance of Ki-67 as a prognostic marker in NETs [
26]. Nevertheless, our findings accord with a previous exploratory analysis of prognostic factors using data from the CLARINET study [
27]. Despite being of potential interest in the univariate setting, CgA levels at baseline, tumor grade, BMI, and sex were excluded from the final multivariate model presented here. Reassuringly, this resonates with previous analyses of prognostic factors for PFS [
13,
27], suggesting CgA levels and tumor grade are less robust prognostic factors than hepatic tumor load and TGR
0 in patients with NETs.
This study was not without limitations. There are inherent limitations in post hoc analyses that potentially limit their interpretation. Any confounding due to anisotropy was not accounted for, as target lesions were assumed to be spherical, although this does tend to be the case for liver metastases. Target lesions followed for TGR may be slow growing, and all lesions within a patient were assumed to be similar; therefore, TGR for an individual may not be representative of their overall tumor targets. Inaccuracies in TGR may be introduced by errors in SLD measurements and, as tumor growth was assumed to be exponential, deviations from this growth pattern; in addition, for TGRTx-0, lesions used for the tumor assessments during the screening period did not have to be the same lesions as those assessed during the treatment period. In concordance with the SLD calculation used in RECIST v1.0, non-target and new lesions were not considered in TGR calculations. The Ki-67 and TGR0 correlation analysis and exploratory analysis of the prognostic value of Ki-67 were limited due to a number of reasons. Firstly, many tumor biopsies were collected several years before the start of treatment, although additional analyses (restricted to the subgroup in whom biopsies were taken in the year before treatment initiation) provided similar results to the overall population. Secondly, Ki-67 data were either unreliably quantified or missing for 41 patients who were enrolled into the CLARINET study based on mitotic index. Thirdly, a number of patients had Ki-67 values recorded as < 1% (N = 15) or < 2% (N = 29), which restricted the way Ki-67 data could be handled; and few patients (N = 17) with Ki-67 values > 5% were included. The ROC area-under-curve implied that TGR0 4%/month cut-off was not strongly deterministic of PFS. Further validation will be required to determine whether this cut-off is relevant in other study populations. Thirteen patients with progression based on non-target or new lesions were excluded from the prognostic value analysis of the TGR0 4%/month cut-off. However, as a similar number of patients were excluded in each treatment group, it is unlikely to have affected the results.
These limitations notwithstanding, our findings suggest TGR has potential clinical utility as a novel metric for proliferative activity, particularly in future studies of somatostatin analogs or novel targeted therapies in NETs in which subtle changes in tumor growth are expected but may not be identified using RECIST. Further advantages of TGR utility include its potential in individualizing patient treatment, with the possibility that therapy can be adjusted based on a more precise analysis of tumor kinetics.
Acknowledgments
The authors thank all patients involved in the study, as well as their caregivers, care team, investigators and research staff in participating institutions. The authors thank Tom Vizard, PhD and Germanicus Hansa-Wilkinson, MSc, of Watermeadow Medical, an Ashfield company, for providing medical writing and editorial support, which was funded by Ipsen in accordance with Good Publication Practice guidelines.
CLARINET Study Group
Austria M. Raderer; Belgium I. Borbath, D. Ysebaert; Czech Republic E. Sedláčková, P. Vítek; Denmark H. Grønbæk; France A. Adenis, L. Buscail, G. Cadiot, S. Dominguez, M. Ducreux, C. Lombard-Bohas, E. Mitry, P. Ruszniewski, J.F. Seitz; Germany N. Begum, I. Harsch, M. Pavel, C. Schöfl, M. Weber, B. Wiedenmann; India M. Mallath, P. Patil, K. Sambasivaiah, R. Saxena; Italy E. Bajetta, A. Buonadonna, R. Buzzoni, R. Cannizzaro, A. Colao, C. De Angelis, P. Tomassetti; Poland J. Ćwikła, B. Kos-Kudła; Slovakia T. Salek; Spain J. Capdevila, G. Soler, J.M. Tabernero; Sweden H. Ahlman, M. Kjellman; UK G. Aithal, A. Anthoney, M. Caplin, A. Grossman, J. Newell-Price, J. Ramage, N. Reed, A. Rees, W. Steward, L. Wall; USA M. Choti, A.T. Phan, E.M. Wolin.