Value of Tumor Growth Rate (TGR) as an Early Biomarker Predictor of Patients’ Outcome in Neuroendocrine Tumors (NET)—The GREPONET Study

Gastroenteropancreatic neuroendocrine tumors are rare, and advanced disease cannot be cured. This article explores the role of tumor growth rate as a novel radiological biomarker in patients with advanced neuroendocrine tumors.

Abstract

Introduction.

Tumor growth rate (TGR; percent size change per month [%/m]) is postulated to be an early radiological biomarker to overcome limitations of RECIST. This study aimed to assess the impact of TGR in neuroendocrine tumors (NETs) and potential clinical and therapeutic applications.

Materials and Methods.

Patients (pts) with advanced grade (G) 1/2 NETs from the pancreas or small bowel initiating systemic treatment (ST) or watch and wait (WW) were eligible. Baseline and follow‐up scans were retrospectively reviewed to calculate TGR at pretreatment (TGR₀), first follow‐up (TGR_first), and 3(±1) months of study entry (TGR_3m).

Results.

Out of 905 pts screened, 222 were eligible. Best TGR_first (222 pts) cutoff was 0.8 (area under the curve, 0.74). When applied to TGR_3m (103 pts), pts with TGR_3m <0.8 (66.9%) versus TGR_3m ≥ 0.8 (33.1%) had longer median progression‐free survival (PFS; 26.3 m; 95% confidence interval [CI] 19.5–32.4 vs. 9.3 m; 95% CI, 6.1–22.9) and lower progression rate at 12 months (7.3% vs. 56.8%; p = .001). WW (vs. ST) and TGR_3m ≥ 0.8 (hazard ratio [HR], 3.75; 95% CI, 2.21–6.34; p < .001) were retained as factors associated with a shorter PFS in multivariable Cox regression. TGR_3m (HR, 3.62; 95% CI, 1.97–6.64; p < .001) was also an independent factor related to shorter PFS when analysis was limited to pts with stable disease (81 pts). Out of the 60 pts with TGR₀ data available, 60% of pts had TGR₀ < 4%/month. TGR₀ ≥ 4 %/month (HR, 2.22; 95% CI, 1.15–4.31; p = .018) was also an independent factor related to shorter PFS.

Conclusion.

TGR is an early radiological biomarker able to predict PFS and to identify patients with advanced NETs who may require closer radiological follow‐up.

Implications for Practice.

Tumor growth rate at 3 months (TGR_3m) is an early radiological biomarker able to predict progression‐free survival and to identify patients with advanced neuroendocrine tumors who may require closer radiological follow‐up. It is feasible to calculate TGR_3m in clinical practice and it could be a useful tool for guiding patient management. This biomarker could also be implemented in future clinical trials to assess response to therapy.

Introduction

Gastroenteropancreatic neuroendocrine tumor (GEP‐NET) is a relatively rare and heterogeneous group of neoplasms [1], [2]. Advanced disease cannot be cured and it is a lethal condition associated with impaired quality of life [3], [4]. Disease stage as well as tumor location and grade provide prognostic information and are major determinants in the therapeutic algorithm [5], [6]. Tumor grade according to the World Health Organization is determined from percentage of tumor cells with nuclear expression of Ki67, grade (G) 1 <2%, G2 3%–20%, and G3 >20%. A majority of GEP‐NETs demonstrate a relatively slow growth rate and fall into G1–G2 categories.

Slow tumor growth can confound the accurate assessment of treatment response using the current standard protocol in GEP‐NETs: response evaluation criteria in solid tumors (RECIST) [7]. A majority of patients with GEP‐NETs undergoing systemic therapy show stable disease, and even nonresponders (defined as patients not reaching partial response) may have a relatively long time until disease progression. Tumor growth rate (TGR) is an alternative imaging‐based calculation that provides quantitative information on the change in tumor size over time (% per month), based on data from two imaging scans [8]. Decrease in TGR after start of treatment has been shown to be independently associated with shorter progression‐free survival (PFS) using data from multiple phase‐1 studies in solid tumors [8], [9]. High TGR was a negative prognostic factor for overall survival (OS) in phase‐3 studies of sorafenib and everolimus in metastatic renal cell carcinoma [10].

Emerging data suggest that TGR could be of value in patients with GEP‐NETs; the Controlled Study of Lanreotide Antiproliferative Response in Neuroendocrine Tumors (CLARINET) study evaluated lanreotide versus placebo in nonfunctioning small intestinal or pancreatic NETs with a Ki67 of <10 % [11]. After 12 weeks of treatment, almost all patients achieved stable disease on RECIST (lanreotide, 90/96 patients; placebo, 94/98 patients). A post hoc analysis reported an immediate reduction in TGR after 12 weeks of treatment in the lanreotide arm but not in those treated with placebo [12]. Mean (95% confidence interval [CI]) TGR at 12 weeks was 1.2% (–0.4 to 2.7) with lanreotide and 4.1% (2.6–5.6) with placebo. This corroborates findings from a post hoc analysis of a phase II study of lanreotide in G1 and G2 GEP‐NETs that displayed a reduction in TGR after start of therapy [13]. In their study, Ito et al. also noted a trend toward longer PFS in patients with a lower pretreatment TGR, which was determined from comparison of the baseline scan to a scan before study inclusion. Impact of pretreatment TGR is also supported by retrospective studies, both as a predictive factor of disease stability on lanreotide treatment [14] and as a prognostic factor that correlated to survival in patients with NETs who were either therapy naive or under somatostatin analogue (SSA) treatment [15].

Together, these findings suggest that TGR could be of value both as a predictive factor of tumor progression and as a prognostic factor in terms of survival in patients with NETs who are undergoing SSA treatment. We hypothesized that TGR could also be used to determine anti‐tumour activity in NETs and provide refined prognostic information in other treatment settings.

The aim of this study was to explore the role of TGR as a novel radiological biomarker in patients diagnosed with advanced NETs.

Materials and Methods

Study Design

This study was designed as a retrospective multicenter study. Ethical approval was obtained (as applicable) in each center. Clinicians screened consecutive patients and collected clinical and outcome data. In addition, imaging data for all examinations performed between baseline (defined as the time of treatment initiation) until progression or up to 2 years of follow‐up was collected locally by radiologists with an expertise in the field of NETs who were blinded from clinical data. Imaging data was also collected by a second reader blinded to measurements retrieved from the first reader (allowed to be a radiologist or a clinician, based on local availability). A training session was organized to ensure optimal choice of target lesion and measurement as well as high quality data collection.

Study Population

Eligible patients were those who met the following inclusion criteria: diagnosed with grade 1 or grade 2 NETs (biopsy or cytology confirmed) arising from the pancreas or the small bowel; diagnosed with advanced stages (not amenable for curative resection) at study entry; due to start any line of treatment with systemic therapies (including SSAs), chemotherapy, targeted agents, and peptide receptor radionuclide therapy [PRRT]) or watch and wait; available imaging data from baseline to progression or up to 2 years (in the absence of progression or death). Only patients with presence of at least one target lesion meeting the modified RECIST criteria so called RECIST_mTGR1.1 criteria for the purpose of this study (see below) and with contrast‐enhanced computerized tomography (CT) examinations with acquisition on a multidetector CT of at least 16 rows and/or magnetic resonance imaging (MRI) were eligible.

Patients with other concomitant malignancy (advanced stage with disease present at time of study entry) or treatment with local therapies (i.e., liver embolization) or other additional systemic treatments during follow‐up as part of this study were excluded.

Target Lesions: RECIST_mTGR1.1 Criteria

Specific definition of target lesions applied, so called RECIST_mTGR1.1 for the purpose of this study. Such criteria followed guidelines from RECIST 1.1 for definition and measurement of target lesions but allowed the number of lesions to be a maximum of 10 target lesions (5 per organ) to maximize information retrieved from each imaging performed. In addition, for a lesion to be considered target lesion on baseline examination, it should have not received local treatment (including, but not limited to, percutaneous ablation, transarterial chemoembolization, selective internal radiation therapy, and stereotactic radiotherapy) within 6 months of study entry. Finally, patients in whom necrosis in one of the target lesions was seen were excluded to avoid confounding factors (i.e., increase in size owing to necrosis rather than to progression). For each patient/target lesion, same imaging technique was selected for each time point (CT and/or MRI).

Calculation of TGR

TGR was centrally calculated for all patients and expressed as the percentage change in tumor size over one month (%/m) using a previously published formula [8], [10]:

$$\begin{array}{c}\mathrm{TGR}=100\times \left(\exp \left[\mathrm{TG}\right]–1\right)\\ \mathrm{TG}=\left(3\times \log \left[\mathrm{D}2/\mathrm{D}1\right]\right)/\text{time}\left(\text{months}\right)\end{array}$$

where TG = tumour growth, D1 = tumor size at date 1, D2 = tumor size at date 2, and time (months) = (date 2 – date 1 + 1)/30.44. Tumor size was determined using the sum of the longest diameters of target lesions only.

TGR was calculated for every radiological examination available for every patient. Special interest was focused on the following three time‐points: (a) TGR_3m: TGR at 3 months (+/−1) of starting treatment/follow‐up (for patients on “watch and wait”) represented the comparison between baseline and examination at 3 months; (b) TGRfirst: TGR at first follow‐up imaging compared baseline and first radiological assessment after starting treatment/follow‐up (for patients on “watch and wait”) whenever this was performed; (c) TGR₀: Pretherapeutic TGR compared baseline and any previous imaging performed over the 12 months before the baseline examination (prebaseline).

Objectives

The primary objective was to evaluate whether TGR_3m after starting treatment and/or follow‐up (for patients on watch and wait) was a factor predictive of progression‐free survival (PFS).

Secondary objectives included the following: (a) to identify the optimum cutoff value (%/m) for both TGR₀ and TGR_3m to predict 1‐year progression‐free survival; (b) to identify the impact of TGR_3m on OS; (c) to assess whether the role of TGR was affected by using different readers (between oncology/radiology and between radiologists); (d) to analyze the impact of TGR_3m in patients who achieved stable disease as best response; (e) to evaluate if TGR₀ was predictive of PFS; (f) to analyze TGR₀ and TGR_3m between subgroup populations (according to primary tumor site, type of treatments administered and pathological factors (grade and Ki67); (g) to analyze changes on TGR during treatment and/or follow‐up; (h) comparison between TGR and RECIST.

Statistical Analysis

Demographic and clinical characteristics were analyzed by producing tables of frequency for categorical variables and by calculation of the median and 95% confidence interval (CI) for continuous variables. Statistical t test (including paired t test), chi‐squared test, and the Mann‐Whitney test (in case of non‐normal distribution as per Shapiro‐Wilk test) were applied as appropriate for comparison of characteristics (including TGR) between subgroups/time‐points.

Receiver operating characteristic (ROC) curves and logistic regression were employed for identification of the most informative TGR_first and TGR_3m cutoffs for prediction of 12 month progression rate. For TGR₀, a previously identified cutoff of 4%/m was employed for survival analysis [11]. Comparison of this cutoff with other alternative cutoffs identified by median and ROC analyses was performed. PFS and OS were defined as the time from starting the new treatment or watch and wait period to the time of progression (defined as the time to progression as per radiological/clinical criteria based in current clinical practice [RECIST v1.1]) and the date of death/last follow‐up, respectively. Best radiological response was defined as per RECIST1.1. Median PFS and OS were estimated by the Kaplan‐Meier methods; such medians were compared (subgroup analysis) using log‐rank test. Survival analysis was performed by univariate Cox regression; mutivariable analysis was undertaken including prognostic factors identified to be significant in the univariate analysis (p < .05). The landmark method was employed for survival analyses [16], [17] to avoid the potential bias introduced by the fact that “responders must live long enough for a response to be observed and for TGR to be measured.” Current guidelines [18], [19], advice to perform a radiological reassessment after 3 months of start of a new treatment line; on this basis, 84 days (3 months) was used as landmark point for survival analysis and patients dying or progressing before that time were excluded from survival analysis.

Two‐sided significance tests with p < .05 were considered significant. Stata version 12.0 software was employed for the statistical analysis [20].

Results

Patients were identified across seven international centers. Out of 905 patients screened, 222 were eligible (supplemental online Appendix 1). A total of 948 scans (76.6% CT, 23.4% MRI) and 790 individual target lesions were analyzed. First reader was a radiologist in all centers, second reader was a radiologist in 3 centers and a clinician in 5 centers.

Patients’ Characteristics

Baseline characteristics for these patients are summarized in Table 1. Median follow‐up and estimated OS were 40.5 months (95% CI, 36.9–43.6) and 99.4 months (95% CI, 77.7–134.2), respectively. Estimated median PFS was 26.9 months (95% CI, 22.2–31.9); 21.6% of the patients had progressed at 12 months of study entry. PRRT (partial response rate 35.7%) and chemotherapy (complete response rate 2.5%, partial response rate 27.5%) achieved the highest rate of objective responses when compared with other treatments (p < .001; full data not shown).

Table 1. Patients’ baseline characteristics.

^aOne patient was treated with IFN in this group of patients.

Abbreviations: IFN, interferon; NET, neuroendocrine tumor; PRRT, peptide receptor radionuclide therapy; SSA, somatostatin analogue.

Identification of a TGR Cutoff Able To Predict Rate of Progression at 12 Months

To increase the power of the analysis, data of TGR_first (222 observations) was used for identification of the best TGR cutoff. Median time between baseline scan and first assessment was 3.8 months (range, 0.2–26.51; 95% CI, 3.6–4.2). Median TGR_first was 0%/m (95% CI, −0.01 to 0). Most informative TGR_first cutoff for prediction of progression at 12 months was 0.8%/m (area under the curve [AUC], 0.74; 95% CI, 0.65–0.84; Fig. 1; sensitivity 72.9% and specificity 78.7%; correctly classified patients 77.5%). Logistic regression confirmed the increased risk of progression at 12 months for patients with TGR_first ≥0.8 (odds ratio [OR] 8.99; 95% CI, 4.37–18.48; p < .001).

Figure 1.

ROC curve analysis of TGR_first for prediction of progression‐free patients at 12 months of starting treatment and/or follow‐up.

Abbreviations: ROC, receiver operating characteristic; TGR_first, tumor growth rate at first follow‐up.

A sensitivity analysis was performed to confirm the robustness of the identified cutoff. For this purposes, ROC analysis was repeated using observations for TGR_3m (103 patients) and TGR_4m (calculated between baseline and imaging after 4 month [±2]; 178 patients). TGR_3m reached the highest AUC (0.80; 95% CI, 0.68–0.92) and was confirmed to be the most informative time point. Most suitable cutoff did not vary when such analysis was performed (supplemental online Appendix 2).

Impact of TGR_3m on Patients’ Outcome

When the previously identified TGR cutoff was applied to TGR_3m (103 patients with data available), patient population was divided in two groups (TGR_3m <0.8 [66.9%] and TGR_3m ≥0.8 [33.1%]) with different outcomes. Patients with TGR_3m ≥ 0.8 (vs. <0.8) had shorter PFS (hazard ratio [HR] 2.13; 95% CI, 1.34–3.37; p = .001; Fig. 2A) and higher risk of progression at 12 months (OR, 14.99; 95% CI, 5.09–44.19; p < .001). Table 2 summarizes predicted progression‐free rate at different time‐points for both groups. At 12 months, 92.74% and 43.18% of patients in the TGR_3m <0.8 and TGR_3m ≥0.8 groups are expected to be free of progression (equivalent to 12 month progression rate of 7.26% and 56.82%, respectively).

Figure 2.

Kaplan‐Meier curved for PFS. (A): Kaplan‐Meier PFS for TGR_3m: TGR_3m separated patients into two groups with different progression‐free survival (PFS). Adjusted (multivariable analysis) Cox p is provided. (B): Kaplan‐Meier PFS for TGR_3m for patients with stable disease as best response: TGR_3m separated patients into two groups with different PFS. Adjusted (multivariable analysis) Cox p is provided. (C): Kaplan‐Meier PFS for TGR₀: TGR₀ separated patients into two groups with different PFS. Adjusted (multivariable analysis) Cox p is provided.

Abbreviations: CI, confidence interval; HR, hazard ratio; TGR, tumor growth rate; TGR₀, TGR at pretreatment; TGR_3m, TGR at 3 months.

Table 2. Estimated risk of being free of progression during follow‐up.

Abbreviations: CI, confidence interval; pts, patients; TGR_3m, tumor growth rate at 3 months.

Univariate and multivariable analysis was performed looking for prognostic factors for PFS. In addition to TGR_3m, univariate Cox regression identified other prognostic factors for PFS (Table 1, 3, column A) such as site of primary tumor and type of treatment. Watch and wait treatment (vs. other) and TGR_3m ≥0.8 (vs. TGR_3m <0.8; HR, 3.75; 95% CI, 2.21–6.34; p < .001) were retained as factors associated with a shorter PFS in the multivariable analysis (Table 1, 3, column B; Fig. 2A). Findings were confirmed by landmark analysis (3 patients excluded): TGR_3m HR 3.45 (95% CI, 2.00–5.95); p < .001. Figure 3 shows a swimmer plot with TGR_3m, PFS and response data for individual patients.

Table 3. Univariate and multivariable survival analysis Cox regression analysis (progression‐free survival).

Abbreviations: CI, confidence interval; HR, hazard ratio; IFN, interferon; PRRT, peptide receptor radionuclide therapy; Ref, reference category; SSA, somatostatin analogue; TGR_3m, tumor growth rate at 3 months.

Figure 3.

Swimmer plot shows individual results for TGR_3m, PFS, and achieved best radiological response for all those patients with TGR_3m data available. Patients with low TGR (blue) and high TGR (red) are compared. (A): Shows TGR_3m as a continuum, showing how majority of patients achieved a TGR_3m of 0 or negative value after starting treatment. Interestingly, TGR_3m corresponded with prolonged PFS (B); triangles highlight patients free of progression at data analysis, dashed lines represent the estimated median PFS for each group of patients (longer PFS for patients with low TGR_3m). (C): Shows how higher rate of PR, CR, and SD were achieved in the group of patients with low TGR (blue), whereas patients with PD are concentrated in the group with higher TGR_3m (red).

Abbreviations: CR, complete response; PD, progressive disease; PFS, progression‐free survival; PR, partial response; SD, stable disease; TGR, tumor growth rate; TGR_3m, TGR at 3 months.

TGR_3m did not impact on OS (HR, 0.81; 95% CI, 0.32–2.04; p = .652); thus, no further analysis with OS was performed.

Role of TGR_3m in Patients with Stable Disease

When the analysis was limited to patients with TGR_3m data available and stable disease as best achieved response (81 patients), median TGR_3m was 0%/m (95% CI, −28.7 to 71.5). Once again, TGR_3m cutoff of 0.8%/m stratified stable disease patients in two groups with different PFS (Fig. 2B): TGR_3m <0.8 (67.9%); medianTGR_3m − 0.8 (95% CI, −1.7 to 0; median PFS, 27.4 months; 95% CI, 19.7–39.1) and TGR_3m ≥0.8 (32.1%); median TGR_3m 3.8 (95% CI, 2.9–6.2; median PFS, 9.8 months; 95% CI, 6.4–23.6). Survival analysis showed that TGR_3m and treatment were significant in univariate Cox regression (Table 2, 3; column A) and included in multivariable Cox regression (Table 2, 3; column B). TGR_3m ≥0.8 was an independent factor related to shorter (PFS, 3.62; 95% CI, 1.97–6.64; p < .001) in the multivariable analysis (Fig. 2B).

Impact of Readers on the Performance of TGR_3m as a Prognostic Factor

Even though there were discrepancies between readers at the time of selection and measurement of target lesions, thus impacting on TGR absolute results for individual patients, agreement between readers when TGR_3m was dichotomized using the previously defined cutoff was good (81.6%) and did not differ significantly according to the type of second reader (agreement between first and second reader was 81.7% and 81.4% if second reader was a radiologist or a clinician, respectively).

More importantly, TGR_3m remained as an independent factor for PFS when multivariable analysis was repeated using TGR_3m measured by second reader (HR, 4.22; 95% CI, −7.47 to 2.38; p < .001); being the type of second reader not a significant factor (p = 0.4).

TGR_3m Versus RECIST_3m

Percentage change on tumor size at 3 months assessed by RECIST 1.1 (RECIST_3m) and paired TGR_3m could be calculated for 101 patients. The accuracy for prediction of 12‐month progression was compared between TGR_3m and RECIST_3m. Both tools seemed to be as informative when analyzed as a continuous variable [21]. The standard 20% cutoff for RECIST at 3 months did not provide any useful information (supplemental online Appendix 3), whereas TGR_3m was useful for prediction of patient's outcome. We were able to identify two alternative cutoffs for definition of progression by RECIST_3m that could be useful for identification of patients with risk of early progression: 10% (based on previous experience in NETs [22]) and 3.6% (based on ROC curve analysis; supplemental online Appendix 3, 4).

The main advantage in favor of TGR_3m vs. RECIST_3m was the fact that TGR_3m had less variability (standard deviation (SD), 9.22 vs. 17.06, respectively).

Pretherapeutic TGR

Out of the 60 patients with pretherapeutic TGR (TGR₀) data available, median TGR₀ was 2.1%/m (95% CI, 0.9–4.9). There was a median of 3.9 months (range, 0.6–8.6; 95% CI, 3.2–4.4) between prebaseline and baseline imaging. A total of 36 patients (60.0%) had TGR₀ < 4%/m, and 24 patients (40.0%) had TGR₀ ≥ 4%/m.

TGR₀ correlated with PFS. Primary tumor, type of treatment, and TGR₀ were significant in the univariate survival analysis for PFS and were included in multivariable analysis. Multivariable analyses showed that small bowel primary tumor (vs. pancreas; HR, 0.41; 95% CI, 0.18–0.94; p = .036) and TGR₀ ≥ 4%/m (HR, 2.22; 95% CI, 1.15–4.31; p = .018; Fig. 2C), were independent factors related to longer and shorter PFS, respectively (supplemental online Appendix 5). Findings were confirmed by landmark analysis (two patients excluded; TGR₀ HR, 2.15; 95% CI, 1.08–4.28; p = .029).

Alternative Cutoff for TGR₀

In addition to the 4%/m cutoff previously defined, alternative cutoffs for TGR₀ were identified within the 60 patients with TGR₀ data available. Such alternatives derived from the median TGR₀ (2%/m cutoff derived from median TGR₀ of 2.1%/m) and the most informative cutoff (3%/m) identified with ROC analysis. Comparison of the performance of such cutoffs is summarized in supplemental online Appendix 6; previously defined cutoff of 4 was superior, especially in terms of specificity (67.4%), with no significant detriment in estimation of PFS.

TGR Within Subgroups

A subgroup analysis did not reveal any difference in TGR₀ across different primary sites, tumor grade, or treatment categories. On the contrary, TGR_3m was higher on patients with small bowel primary (p = 0.053), and on treatment with SSA/interferon (IFN; p = .363). No differences by grade or Ki67 were identified for TGR_3m. See supplemental online Appendix 7 for full detail.

Changes on TGR over Time

Comparisons between mean TGR₀ (60 observations; mean, 4.89%/m) versus mean TGR_3m (103 observations; mean, −0.44%/m) showed a significant reduction on TGR after starting treatment and/or follow‐up (p < .001). In contrast, when intrapatient comparison was performed (paired t test), such differences were not statistically significant (TGR₀ vs. TGR_3m, 30 observations; mean, 7.51 vs. 2.47, p = .283). Further comparisons and detail are provided in supplemental online Appendix 8.

Discussion

Challenges for radiological assessment of response in NETs are well known to the NET community, especially in well‐differentiated tumors [21]. We hypothesized that TGR could be a helpful predictor of patients’ outcome, with future utility for treatment decision making and better understanding of biological behavior. Our results confirm this hypothesis and show that it is feasible to calculate TGR_3m in the real clinical practice using standard‐of‐care imaging and that its value is not affected by the expertise of the reader (radiologist vs. clinicians). We showed that TGR can be used for risk stratification of patients at an early stage (3 months) after starting systemic treatment. Based on our data, TGR_3m is able to classify patients into two groups, and our cutoff for such classification has been shown to be robust based on repeated analyses reaching the same findings. Even though previous research in cancer has explored the role of TGR, the practicality of such findings was limited by the absence of a cutoff able to be implemented in clinical practice. This is the first time that a TGR_3m cutoff is confirmed to impact patients’ outcome.

One of the utilities of implementing TGR_3m in clinical practice would be the fact that TGR_3m is able to identify patients at high risk of tumor progression early on after treatment initiation. Understanding that, clinicians will still wait to have radiological confirmation of progression before implementing a change on the treatment strategy. Thus, TGR_3m will not affect the treatment administered to patients. We do, however, believe that TGR_3m could impact on the patients surveillance plan. By these means, patients with high TGR_3m should be followed up more regularly than patients with low TGR (i.e., 3‐monthly) because of the significantly shorter PFS identified in this patient population. In contrast, patients with low TGR_3m could have less regular imaging (i.e., 6‐monthly) in an attempt of avoiding unnecessary radiation in a population of patients unlikely to progress in the short term. External validation of our findings would be required before implementing this into clinical practice.

Moreover, the role of TGR_3m as an early radiological biomarker was confirmed in the subgroup of patients who achieved stable disease as best response. This analysis is of special interest because of many reasons. First, it shows the utility of a biomarker in one of the most likely scenarios in NETs, where partial responses are rarely seen. Second, the fact that results are replicated when patients with partial responses (mostly treated with PRRT in this series) are excluded from the analysis confirms that TGR is measuring changes in tumor size that RECIST is not able to identify, thus providing a more granulated and accurate assessment. In addition, the fact that TRG_3m was superior to the currently used RECIST_3m definition of progression (+20% change) confirms that TGR_3m is more accurate and reliable and shows, as hypothesized in other scenarios, that alternative RECIST cutoffs for definition of progression and response may need to be explored in NETs [22].

In addition to TGR_3m, this study also explored the impact of TGR₀ and confirmed previous findings from the CLARINET study, showing that TGR₀ is able to predict response to treatment in terms of PFS in patients with NET G1/G2, regardless of systemic treatment [12]. Based on our findings, we would recommend using the TGR₀ cutoff of 4%/m (rather than other alternatives) for future studies because of its better performance when compared with others. Whether TGR₀ could help tailoring options of treatment for our patients remains a possibility but is still not confirmed and would require further analysis with bigger sample of patients with paired TGR₀ and TGR_3m data.

Our study has many strengths. Our population of patients was identified through centers with an expertise in the field, and baseline characteristics matched with previous series, thus confirming that our population is representative of patients with NETs across the world [1]. The expertise of radiologist involved in this study and the attention to detail at time of patient selection are also reinforcing the robustness of our study and our findings. Unfortunately, and even though this study confirms the utility of TGR, some questions remain unanswered and limitations to this study also apply. Main limitation was the limited sample size for subgroup analysis (i.e., analysis of TGR changes over time and differences between treatment groups). We still need to validate this findings in other independent series, and the impact of primary tumor and treatment on TGR₀ and TGR_3m would need to be explored to identify whether tailoring treatment before treatment initiation based on TGR₀ is feasible or whether we need to resign to modify the follow‐up strategy according to TGR_3m. Similarly, the study was not designed or powered to identify variations in TGR_3m by type of treatment; specific series built for such purposes would be able to address such issues. In fact, the currently recommended TGR_3m cutoff was determined based on a cohort of patients receiving a variety of different treatments; whether this is the most appropriate cutoff regardless of treatment administered (i.e., chemotherapy vs. SSA) will need validation. In addition, differences on TGR_3m between primary tumor (higher on patients with small bowel primary vs. pancreas) and treatment groups (higher on patients receiving SSA/IFN vs. watch and wait [WW]) could be reflection of a selection bias (more aggressive/cytotoxic treatment selected for pancreatic patients [vs. small bowel] and SSA/IFN selected for patients with a more aggressive tumors [vs. WW]). It is worth mentioning that TGR_3m did not impact on OS, probably because of the impact of previous and consecutive treatments administered.

Conclusion

TGR_3m is an early radiological biomarker able to predict PFS and to identify patients with advanced NETs who may require closer radiological follow‐up. It is feasible to calculate TGR_3m in clinical practice, and it could be a useful tool for guiding patient management. This biomarker could also be implemented in future clinical trials to assess response to therapy.

See http://www.TheOncologist.com for supplemental material available online.

Author Contributions

Conception/design: Angela Lamarca, Joakim Crona, Maxime Ronot, Marta Opalinska, Carlos Lopez Lopez, Daniela Pezzutti, Louis de Mestier, Frederico Costa, Marianne Pavel, Clarisse Dromain

Provision of study material or patients: Angela Lamarca, Joakim Crona, Maxime Ronot, Carlos Lopez Lopez, Daniela Pezzutti, Pavan Najran, Luciana Carvhalo, Regis Otaviano Franca Bezerra, Philip Borg, Naik Vietti Violi, Hector Vidal Trueba, Louis de Mestier, Niklaus Scaefer, Anders Sundin, Frederico Costa, Clarisse Dromain

Collection and/or assembly of data: Angela Lamarca, Joakim Crona, Maxime Ronot, Marta Opalinska, Carlos Lopez Lopez, Daniela Pezzutti, Pavan Najran, Luciana Carvhalo, Regis Otaviano Franca Bezerra, Philip Borg, Naik Vietti Violi, Hector Vidal Trueba, Louis de Mestier, Niklaus Scaefer, Anders Sundin, Frederico Costa, Marianne Pavel, Clarisse Dromain

Data analysis and interpretation: Angela Lamarca, Marianne Pavel, Clarisse Dromain

Manuscript writing: Angela Lamarca, Joakim Crona

Final approval of manuscript: Angela Lamarca, Joakim Crona, Maxime Ronot, Marta Opalinska, Carlos Lopez Lopez, Daniela Pezzutti, Pavan Najran, Luciana Carvhalo, Regis Otaviano Franca Bezerra, Philip Borg, Naik Vietti Violi, Hector Vidal Trueba, Louis de Mestier, Niklaus Scaefer, Anders Sundin, Frederico Costa, Marianne Pavel, Clarisse Dromain

Disclosures

Angela Lamarca: Ipsen, Pfizer (H), Ipsen (RF); Joakim Crona: Novartis (H); Marianne Pavel: Ipsen, Novartis, Pfizer, Lexicon (C/A, SAB), Ipsen, Novartis (former Institution, Charité) (RF). The other authors indicated no financial relationships.

(C/A) Consulting/advisory relationship; (RF) Research funding; (E) Employment; (ET) Expert testimony; (H) Honoraria received; (OI) Ownership interests; (IP) Intellectual property rights/inventor/patent holder; (SAB) Scientific advisory board