Activity and Safety of Immune Checkpoint Inhibitors in Neuroendocrine Neoplasms: A Systematic Review and Meta-Analysis

Immune-checkpoint inhibitors (ICIs) have widened the therapeutic scenario of different cancer types. Phase I/II trials have been designed to evaluate the role of ICIs both as single agents and in combination in neuroendocrine neoplasms (NENs), but as yet no randomized controlled phase III trials have been carried out. A systematic review and meta-analysis of studies published could help to reduce the biases of single-phase II trials. Efficacy data were obtained on 636 patients. Pooled percentages of the overall response rate (ORR) and disease control rate (DCR) were 10% (95% CI: 6–15%, I2 = 67%, p < 0.1) and 42% (95% CI: 28–56%, I2 = 93%, p < 0.1), respectively. Median progression-free survival (mPFS) was 4.1 months (95% CI 2.6–5.4; I2 = 96%, p < 0.1) and median overall survival (mOS) was 11 months (95% CI 4.8–21.1; I2 = 98%, p < 0.1). Among the ICIs used as single agents, the anti-PD1 toripalimab achieved the highest ORR. Combination regimens were superior to monotherapy, e.g., the ICI combination nivolumab + ipilimumab, and the ICI + anti-angiogenetic combination atezolizumab + bevacizumab, both of which warrant further investigation. Promising efficacy and a good safety profile of ICIs represent a valid opportunity for expanding the therapeutic landscape of NENs. Predictive biomarkers are needed to identify the most suitable candidates for these regimens.

According to the latest WHO classification, gastroenteropancreatic G3 NENs are defined by a Ki67 proliferation index of >20%. In addition to proliferative activity, tumor differentiation appears to play a major role, further dividing the G3 NEN G3 group into G3 neuroendocrine tumors (NETs) and NECs, characterized by different prognosis and response to available therapies [40,41].
In the present systematic review and meta-analysis, we summarize what is currently known about the efficacy and safety of ICIs in NENs, and the possible implications for clinical practice.

Data Extraction
This systematic review and meta-analysis were conducted in accordance with the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [42,43]. A search of major databases, e.g., MEDLINE/PubMed, Cochrane, and Embase, was performed to identify studies on ICIs for the treatment of NENs published up to October 2020. The following keywords were used: "immune checkpoint inhibitors" or "ICIs" or "nivolumab" or "pembrolizumab" or "durvalumab" or "avelumab" or "atezolizumab" or "spartalizumab" or "toripalimab" or "tremelimumab" or "ipilimumab" or "anti PD1" or "anti PD-L1" or "anti-CTLA4" and "neuroendocrine tumors" or "neuroendocrine neoplasms" or "neuroendocrine carcinomas". An additional search for Meeting Abstracts published by the American Society of Clinical Oncology (ASCO), ASCO Gastrointestinal Symposium (ASCO-GI), European Society for Medical Oncology (ESMO), and European Neuro-Endocrine Tumors Society (ENETS) annual meetings was also performed. Finally, we manually checked the citations of the included publications. Original studies published in English were considered. Reviews, letters, and personal opinions were excluded. A flowchart of the selection process is shown in Figure 1.
Two authors independently conducted a preliminary screening of titles and abstracts, followed by a second-round screening when they read the full texts of the potentially relevant articles. In the event of disagreement, a third reviewer was consulted to facilitate the final decision. Data on the population treated, treatment efficacy, and toxicity parameters were extracted and pooled from the selected studies. The following data were also collected from each study: year of publication, name of first author, country of the study; study design (phase, randomization, independent review); baseline characteristics of the included patients (median age, histologic classification, grading, staging, primary tumor location); intervention including treatment regimens, dosage and number of administered cycles; complete response (CR), partial response (PR), stable disease (SD) and progressive disease (PD) absolute frequency or, when available, outcomes expressed as overall response rate (ORR) and disease-control rate (DCR); progression-free survival (PFS) and overall survival (OS); toxicities (grades and types of trAEs and irAEs). When response and toxicities were reported at the subgroup level, the values were combined to have endpoints at the single-cohort level. PFS and OS were maintained at the subgroup level because their nature (medians) does not allow for an overall merging value.

Quality Assessment
The risk of bias was assessed by the Risk Of Bias in Non-randomized Studies of Interventions (ROBINS-I) tool for non-randomized clinical trials [44]. In this tool, risk of bias is assessed within specified domains, including (1) bias due to confounding, (2) bias in the selection of participants for the study, (3) bias in the classification of interventions, (4) bias due to deviations from intended interventions (5) bias due to missing data, (6) bias in the measurement of outcomes, (7) bias in the selection of the reported results, and (8) overall bias. Given that assessments are inherently subjective and that there are no strict or objective criteria to judge bias within the ROBINS-I tool, disagreements were resolved by the intervention of a third investigator.

Statistical Analysis
The main endpoints of ORR, DCR, any trAE, any irAE, trAEs of grade ≥ 3, and irAEs of grade ≥ 3 were pooled together using the Meta package in R software (version 3.6.1). Relative risks and 95%CIs describing the ORR of PD-L1 biomarkers were synthesized calculated using the Meta package in R. In addition, the median PFS and OS values, accompanied by their 95%CIs, were extracted and pooled estimates obtained according to the method used by McGrath et al., [45]. This meta-analysis was performed in R software using the metamedian package. Heterogeneity between study outcomes was evaluated using the I 2 index and Cochrane's Q-test (reported with p-value). Values of I 2 > 50% and p < 0.1 indicated a substantial heterogeneity between studies. The pooled estimates with their 95% CI were determined using the fixed-effects model (p > 0.1) or the random-effects model (p < 0.1). While the Meta R package automatically performs the I 2 and Q-test calculus, ad hoc code was written to evaluate the I 2 and Q-test on PFS and OS. Subgroup analyses of primary efficacy endpoints were included to investigate the possible sources of heterogeneity and to identify differences in subsets of patients. A funnel plot of the main endpoint ORR was generated to assess potential publication bias, and its asymmetry was evaluated via linear regression test using the Meta package in R software (3.6.1).

Patient Characteristics at Baseline
Six hundred and thirty-six patients were treated with ICIs either as monotherapy or in combination. The median age at enrolment ranged across studies from 41 to 67 years. Eastern Cooperative Oncology Group performance status (ECOG PS) at screening was 0-1 in 10/14 studies and 0-2 in the remaining 4. The most frequent site of origin of NENs was the pancreas (219 patients, 34.2%) followed by the gastrointestinal tract (201 patients, 31.4%) and lung (100 patients, 15.6%). The remaining 72 (11.3%) patients had NENs originating from other sites or of unspecified/unknown origin. The majority of patients (418, 65.3%) had grade (G) 1 or G2 NENs. Eighty-six (13.4%) patients had G3 neuroendocrine tumors (NETs) and 114 patients (17.8%) had neuroendocrine carcinomas (NECs); in 13 (2%) cases, the distinction between NET G3 and NEC was not specified. Grading was unknown in 9 (1.5%) cases.
The patients included in the studies were all pre-treated, but data on previously administered therapies were available in only 3 studies for a total of 193 patients [26,38,39]. Somatostatin analogs (SSAs) were the most widely used drugs (90 patients), followed by chemotherapy (110 patients), everolimus (76 patients), and TKIs (26 patients). Functional status was not mentioned in just under half of the studies (6/14). In the 8 articles in which this information was given, all the patients had non-functioning tumors, except the study by Strosberg et al. that included 8 functioning tumors [27].
In 7 studies, 370 patients were evaluated for PD-L1 expression [26,27,30,31,33,36,39]. However, a PD-L1 positivity of ≥1% was an inclusion criterion in only one study [26]. ORR was the main primary endpoint, while secondary endpoints were DCR, OS, PFS, and safety. Figure 2 shows the forest plots of ORR (A) and DCR (B), reported as percentages, along with their 95% confidence interval (95%CI). The efficacy analysis of the ORR was carried out on 636 patients, while only 596 patients were included in the DCR analysis. In one study, information on DCR or stable disease was not available [32]. The best response obtained was a partial response (PR) in 39 patients and stable disease (SD) in 210 patients. Three complete responses (CR) were registered. The pooled proportions of ORR and DCR were 0.10 (95% CI: 0.06-0.15, I 2 = 67%, p < 0.1) and 0.42 (95% CI: 0.28-0.56, I 2 = 93%, p < 0.1), respectively. The linear regression test (p = 0.212) indicated that no publication bias existed in this meta-analysis for ORR ( Figure S1).

PD-L1 Biomarker for ORR
Four studies with PD-L1 expression data were included in an additional analysis to analyze the prognostic role of PD-L1 for ORR. Figure 3 shows the forest plots of ORR comparisons based on PD-L1 expression. PD-L1-positive patients had a higher ORR than those with PD-L1 negative tumors (RR = 1.22, 95% CI: 1.22-8.89, I 2 = 2%, p = 0.36).

Risk of Bias
All the included non-randomized clinical trials were considered to be at low risk of bias, except the study by Halperin et al., which was deemed to have a moderate risk of bias [32]. More details on the risk of bias assessment, with reasons supporting each study assessment, can be found in Supplementary Table S1.

Discussion
In recent years, research into immuno-oncology has led to important breakthroughs in the treatment of solid tumors [21]. Although some immunological therapies, such as interferon, have already been used successfully in patients with NETs, the role of ICIs for NENs has yet to be clarified [5]. Several phase Ib/II trials recently explored single-agent and combination therapy with ICIs in this disease setting. However, the "one-fits-all" approach, albeit useful to reach the planned study accrual due to the rarity of the disease, has not provided conclusive results [46]. Our meta-analysis confirmed the heterogeneity of results across studies as a consequence of the clinical characteristics of enrolled patients (i.e., primary site, grade, percentage of metastatic cases) and the type of samples analyzed. It also enabled us to identify future directions for research into the use of ICIs in NENs.
In our meta-analysis, anti-PD1 compounds such as toripalimab produced an ORR of 23%, while the combination of anti-PD1 nivolumab with anti-CTLA4 ipilimumab reached 25% ORR in a mixed population in 2 studies [29,33,38]. Conversely, anti-PDL1 therapies generally failed in their main objective [29,37,39], which might be attributable to the heterogeneity of PD-ligand expression, PD-L1 being prevalent in lung NETs and PD-L2 in pNETs, suggesting a different clinical resistance to PD-1/PD-L1 checkpoint inhibitors [47]. The ORR was higher in PDL1 positive patients than in PDL1-negative cases.
Neo-angiogenesis appears to play a key role in NEN development and progression. NETs are, in fact, characterized by a high vascular supply and overexpression of VEGF-A. The latter induces immunosuppressive cells, e.g., tumor-associated macrophages, regulatory T cells, and myeloid-derived suppressor cells, leading to an immunosuppressive phenotype that inhibits the maturation of dendritic cells and the activation and proliferation of T cells [48]. Thus, treatment with an anti-VEGF antibody could reverse these immunosuppressive effects and promote T cell activation and dendritic cell maturation. In fact, in the study by Halperin et al., the addition of bevacizumab to atezolizumab led to an ORR of 15-20% and a PFS of 14.9-19.6 months in extra-pancreatic and pancreatic NETs, suggesting the synergistic activity of bevacizumab in converting an immune 'cold' tumor into a 'hot' one [32].
Another important issue is the short-term PFS obtained from the use of ICIs (except for Halperin's study), similar to that observed in the placebo arms of studies on everolimus and sunitinib [12,15]. A possible explanation for this could be that ICIs are more beneficial in the early phase of the natural history of NENs, as observed in Halperin's study. This benefit may be lost in heavily pre-treated patients due to modifications occurring in both the tumor and the microenvironment. In addition to the need for larger studies, a potential future strategy could involve the use of anti-PD1 therapy, perhaps in combination with antiangiogenic therapy, in patients selected based on grading and site of origin.
The safety profile of ICIs is also an important aspect to be considered. Our analysis revealed a range of side effects that are consistent with those previously reported in other solid tumors.
Our study has a number of limitations. First, all of the studies selected were nonrandomized phase I/II trials. Furthermore, to avoid publication bias, we were obliged to include studies characterized by high population heterogeneity because of the poor accrual potential for a rare disease such as neuroendocrine neoplasia. Thus, although the meta-analysis did not yield conclusive results on the use of ICIs in NENs, it nevertheless shed light on interesting therapeutic possibilities such as the use of ICI combinations with anti-VEGF or anti-CTLA-4 therapies. Of note, even if we had conducted a comprehensive literature search with a sensitive search algorithm and an extensive manual search of refer-ence lists and conference proceedings, we would not have been able to find unpublished data on every specific site of origin of the disease. We are fully aware that a substantial amount of information is not available to the public and, as such, cannot completely rule out publication bias.

Conclusions
Overall, our meta-analysis of data on ICIs in NENs revealed a high heterogeneity of treatment response. Although our results are not conclusive, they suggest that ICIs are active in NENs. Some information obtained on the different activity of anti-PD1 and anti-PDL1, and on the association with other compounds (i.e., anti-CTLA4 or anti-VEGF) could be useful for the design and development of future clinical trials.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/ph14050476/s1, Figure S1: Funnel plot of overall response rate (ORR) for publication bias, Table S1. Risk of bias for non-randomized studies. Data Availability: The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.