Efficacy of PD-1/PD-L1 inhibitors in gastric or gastro-oesophageal junction cancer based on clinical characteristics: a meta-analysis

Gastric or gastroesophageal junction cancer (GC/GEJC) is the third common cause of cancer-related deaths globally and the fifth most prevalent, with 63% of these cases being locally advanced or metastatic, and more than one million cases were newly diagnosed in 2018 [1, 2]. Currently, patients with GC are mainly treated with surgery and chemotherapy [3‐7]. However, most patients with advanced GC/GEJC continue to progress after receiving treatment and have a poor prognosis [8]. The median overall survival (OS) of individuals with HER2-negative GC is 8–12 months [9], while the median OS of individuals with HER2-positive GC is as high as 13 months. These individuals may benefit from 1st-line targeted therapy [10], where treatment choices were further constrained [3, 11‐13]. Subsequently, immunotherapy has rapidly developed and gradually become a new frontier of advanced GC/GEJC therapy due to the genetic complexity and heterogeneity of cancer, wholly innovating the treatment outlook over the past decade, in which immune escape has played a significant role [14]. Immune checkpoint inhibitors (ICIs), also known as programmed death-1 (PD-1) or its ligand (PD-L1), as promising cancer therapeutic targets in the field of oncology, have caused great concern [15]. ICIs target critical regulators that help tumor cells escape immune attack, thereby enhancing the cytotoxic activity of anti-tumor T cells [16, 17]. To date, several extensive randomized controlled trials (RCTs) have confirmed that many tumor types can achieve disease control and improve OS from PD-1/PD-L1 inhibitors treatment [18‐21]. This rekindled our interest in targeting GC/GEJC patients with ICIs.

However, the response obtained from ICIs is observed in only a minority of GC/GEJC patients (approximately 10 to 20%) [22], and there may be potential for drug resistance and rapid progression of the disease. In addition, existing findings from different clinical trials in this area appear to be controversial, and there is no clear conclusion regarding the role of immunotherapy in individuals with GC/GEJC. Therefore, it is becoming even more urgent to look for dependable predictors to help clinicians screen individuals who are more likely to benefit from ICIs, both to restrict the number of patients exposed to latent autoimmunity by the effects of targeted axis drugs and to protect patients from ineffective treatment [23].

The expression of PD-L1 has been seen as a prognostic biomarker, which is expressed in 30 to 65% of invasive GCs, and is associated with some factors, such as EBV infection, tumor size, tumor depth of invasion, lymph nodes, and distant metastasis [24‐26]. Currently, PD-L1 positive expression as a biomarker for pembrolizumab 3rd-line treatment regimen in GC has been approved by the FDA [27], and nivolumab has been approved in many regions as a treatment regimen for unresectable late or recurrent GC, regardless of PD-L1 expression, which indicates that the potential and application range of immunotherapy are expanding. Unfortunately, a predictive biomarker for anti-PD-1/PD-L1 antibodies therapy, PD-L1 level, remains challenging in clinical trials. For instance, PD-L1 in GC/GEJC was defined according to the combined positive score (CPS), involving the expression of tumor cells, macrophages, and lymphocytes, and this differs from the PD-L1 definition in carcinoma of the lungs [28]; The cut-off value of PD-L1 positive expression has not been determined; PD-L1 expression is impacted by a number of factors, such as anti-tumor therapy, standardized detection methods, and the immune response of the host [29, 30]. Another controversial biomarker is tumor mutation burden (TMB). Studies have shown that different TMB levels in GC lead to different therapeutic effects and prognoses [31, 32]. Therefore, whether TMB has an ideal truncation value and its predictive effect on GC/GEJC requires further study.

It makes sense to look for other practical and economic biomarkers to predict the potency of PD-1/PD-L1 inhibitors in GC/GEJC. In previously published meta-analyses, the potency of immunotherapy against PD-1/PD-L1 was found to be superior to that of chemotherapy in the treatment of advanced gastric esophageal cancer [33‐35]. These conclusions were drawn after the inclusion of patients with esophageal cancer (EC) who received ICIs or chemotherapy; thus, immunotherapy has a generally good prognosis. However, accurate predictions of the GC/GEJC should not be obtained from such comparisons. Several studies have shown that immunotherapy has a remarkable effect on EC [36, 37], which may lead to false-positive results when included in analyses based on GC. Therefore, we performed a meta-analysis that only included individuals with GC/GEJC accepting PD-1/PD-L1 inhibitors, and all those individuals were compared. The CheckMate 649 trial [38] that involved patients with EC was not included. We could scientifically determine which factors among the different clinical characteristics make GC/GEJC individuals more likely to benefit from anti-PD-1/anti-PD-L1 antibodies and guide the choice of treatment. We present these results based on the Preferred Reporting Items for Systemic Reviews and Meta-Analyses (PRISMA) guidelines [39].

At present, a large number of retrospective immunotherapy studies have been carried out or are underway to identify new biomarkers. Fehrenbacher et al. showed that atezolizumab was beneficial for survival in patients with high expression of tumor T effector factor and the gene marker IFN-γ [44]. Another study in the US found that patients with a higher proportion of central memory T cells to effector T cells had a longer PFS [45]. Preconditioning lactate dehydrogenase (LDH), C-reactive protein (CRP), neutrophil-to-lymphocyte ratio (NLR), lung immunoprognostic index (according to derived NLR and LDH levels), and the intestinal microbiome may also be potential signatures [46‐48]; however, there remains no consensus standard for defining the relevant threshold for some of them. In addition, the potency predictive value of these biomarkers is further proven to be necessary through prospective trials, as they are only found in exploratory or retrospective analyses of small samples. As up-to-date clinical outcomes accumulate, we sought to determine whether additional clinical factors are available and cost-effective to predict the efficacy of immunotherapy in GC/GEJC.

To the best of our knowledge, our meta-analysis is the most accurate and adequate research, and it includes the most recent RCTs, to provide guidance on better identifying patients with GC/GEJC probably benefiting from PD-1/PD-L1 inhibitors.

Previous studies have indicated that ICI-based therapy can prolong the survival of individuals with melanoma or non-small cell lung cancer in any age group [49]. A study also showed that ICIs may be more effective in the elderly, and different mechanisms of action of ICIs in different age groups have been reported [50, 51]. In our meta-analysis, we used 65 years old as the most commonly used age cut-off point to distinguish between older and younger patients and demonstrated that PD-1/PD-L1 inhibitors substantially prolonged survival in patients aged < 65 years compared to chemotherapy. However, PD-1/PD-L1 inhibitors did not substantially improve survival in elderly individuals aged ≥ 65 years. This may be due to the fact that older individuals are closely related the falling of immune system function known as immune-senescence [52], making them unable to regain antitumor activity, and may suffer more frequent or more serious immunotherapy toxicities than younger patients [53, 54]. Therefore, it is necessary to cautiously use anti-PD-1/PD-L1 antibodies, and our current analytical data do not support the large-scale clinical use of ICIs in GC/GEJC patients aged ≥ 65 years.

It is well known that sex differences influence innate and adaptive immune responses [55, 56], possibly reflecting the different potencies of anti-PD-1/PD-L1 antibody therapy in male and female GC/GEJC individuals. Our findings indicate that there may be a correlation between ICIs and sex, with males being more effective. This differential response to immunotherapy has also been observed in male patients with lung cancer [57]. This may be due to the fact that female cancer patients have a stronger immune escape mechanism than male, and therefore more probably develop resistance to immunotherapy [56‐58]. Nevertheless, further sub-population analyses showed that a survival benefit was observed with immunotherapy only in the treatment of 1st-line combination, whereas no survival benefits were observed in ≥ 2nd-line or monotherapy. Thus, early combination therapy based on ICIs is recommended for male patients with GC/GEJC. In summary, sex appears to be an appropriate clinical predictor of ICIs. It is worth noting that sex, as a biological variable, should be emphasized in subsequent clinical trials with respect to GC/GEJC of checkpoint inhibitors, and the relationship between sex and efficacy of immunotherapy should no longer be ignored by the scientific community.

Since most RCTs related to PD-1/PD-L1 inhibitors in GC/GEJC exclude individuals with PS ≥ 2, the function of ICIs in these individuals was not analyzed. When we focused on PS 0 and 1, our findings showed that patients with PS 1 seemed to have a longer survival benefit for the treatment of anti-PD-1/anti-PD-L1, with an HR, of 0.93 for patients with PS 0 and 0.89 for patients with PS 1. The preclinical theoretical basis of our findings remains unknown. Results from a previous phase II clinical subgroup analysis of pembrolizumab emphasized that a better PS was related to a higher response rate and longer OS [59]. Mishima et al. also indicated that patients with advanced GC with a PS of 0 had better survival than those with a PS of 1, suggesting that ECOG PS could be used as a predictor of immunotherapy potency [60], which is inconsistent with our results and involves relatively small sample sizes. This may be due to the lack of relevant studies, data on GC/GEJC immunotherapy, and insufficient evidence, which is worthy of close attention in subsequent studies and requires further clarification. Furthermore, patients with PS 1 should be given preference for 1st-line or combination therapy based on ICI as alternative regimens. Therefore, ECOG PS 0 or PS 1 may not be independent predictors of anti-PD-1/PD-L1 antibodies potency.

The Lauren classification distinguishes between two main subtypes of GC: intestinal and diffuse [61]. Diffuse-type GC accounts for approximately 30% of GC cases and is more common in younger individuals [62]. However, there is no systematic review of whether histological type could be meaningful predictors of anti-PD-1/anti-PD-L1 antibodies. We did not find an improvement in survival after ICI therapy in patients with intestinal or diffuse types. In addition, our results showed that individuals did not benefit from anti-PD-1/PD-L1 antibody therapy regardless of previous gastrectomy status or primary tumor sites. Nevertheless, Dong et al. found that individuals who had previously undergone gastrectomy had a relatively better prognosis than those who had not. They thought gastrectomy leads to an imbalance in Th17/Treg and increased PD-1/PD-L1 expression. Blocking the PD-1 and PD-L1 pathways can alleviate Th17/Treg cell imbalance resulting from gastric surgery [63]. Therefore, the following researches may pay more attention to whether gastrectomy affects the efficacy of immunotherapy. For GC patients treated with anti-PD-1/PD-L1 antibodies, early combination treatment based on ICI is recommended. Furthermore, our analysis suggested that GEJC patients obtained OS improvement in ≥ 2nd-line therapy based on ICI, which further illustrated the significance of follow-up treatment.

The expression of PD-L1 has been investigated and demonstrated to be a moderately effective predictor of ICIs potency in GC patients, but its prognostic significance is not clear. A review of 15 published studies showed that in 11 studies, the expression of PD-L1 on the combined positive score (CPS) was a negative prognostic factor for OS, and in three studies, it was a positive prognostic factor, one of which reported no prognostic significance [64]. In our analysis, the potency of the anti-PD-1/anti-PD-L1 antibody on OS in the subgroup with a higher CPS cut-off was better than that in the subgroup with a lower CPS cut-off. Further, the benefits of ICIs based 1st-line treatment and monotherapy on OS were observed in subpopulations with a CPS ≥ 10. In addition, our analysis results revealed that in terms of the tumor proportion score (TPS), that is, the percentage of positive tumor cells only, negative PD-L1 expression (TPS < 1%), or positive PD-L1 expression (TPS ≥ 1%) showed no improvement in OS after application of anti-PD-1/anti-PD-L1 antibodies. This may be due to the fact that a small number of patients were analyzed, or to inherent limitations of the trials themselves, such as the confounding factors and systematic bias risk, as well as the performance differences between different PD-L1 measurement methods [65], all caused the result of which show that PD-L1 TPS does not appear to be a suitable biomarker for predicting ICIs. Kulangara et al. assessed the capacity of PD-L1 CPS and TPS as biomarkers to predict pembrolizumab potency in the KEYNOTE-059 trial, estimating pembrolizumab monotherapy and combination therapy in patients with recurrent or metastatic GC/GEJC [66]. The study revealed that the objective response to pembrolizumab was substantially correlated with the CPS, according to the PD-L1 CPS truncation value of 1, but not with the TPS. Although the clinical significance of CPS and the best cut-off point for ICI treatment have not been determined due to the rapid clinical development of ICIs and based on limited data, these observations support ICIs as another treatment option for GC/GEJC in PD-L1 CPS ≥ 1 individuals [67] according to the immunohistochemistry of PD-L1 [68]. Compared with TPS, PD-L1 CPS is deemed as a highly reliable and reproducible scoring method to predict the response to PD-1/PD-L1 inhibitors in GC/GEJC individuals [66, 69]. Currently, CPS combines PD-L1 expression in tumor and immune cells into a single score, and pathologists are able to evaluate them directly without having to score them [70]. This allows CPS to play a role in GC, where the expression of PD-L1 is driven by immune cells, tumor cells, or both. In addition, CPS has nothing to do with organizational structure [71], and it is hoped that CPS will play a role in cytology or digestion of tumor samples. Therefore, we expect that CPS could be an effective diagnostic tool in the field of GC/GEJC immunotherapy.

In our meta-analysis, line of treatment also predicted the clinical results of ICIs. We found that when the number of previous system schemes was 0, the survival benefits of immunotherapy were achieved; however, in this study, two related studies in GC and GEJC on the treatment of ≥ 2nd-line, involving JAVELIN Gastric 300 [40] and KEYNOTE-061 [41], showed that OS did not improve with the patient’s HR, of 0.96. This may be due to latent immunosuppression and severe adverse reactions caused by conventional chemotherapy or previously applied targeted therapy, which is likely to have a negative effect on subsequent immunotherapy [72]. For GC/GEJC patients, OS may be affected by the previous treatment route rather than the efficacy of the experimental drugs themselves. Therefore, generally speaking, the treatment of anti-PD-1/anti-PD-L1 is not satisfactory in the ≥ 2nd-line application of GC and GEJC.

While our paper has produced promising insights, we must admit that the limitations of the present study are evident in many aspects. First, the limitation of this paper is that the number of trials that meet the analysis conditions is limited. However, the sample size was > 2000, so it was comparable in all trials. And the sample size of the JAVELIN 300 was the smallest (n = 371), indicating that there was no obvious publication bias or asymmetry. Second, we collected the trial outcomes from the published papers but could not obtain the raw data of each patient, possibly causing a bias in data analysis. Under the premise of more relevant research, extensive clinical studies and synthesis analyses of larger samples can be developed to further confirm the potency of ICIs therapy in GC/GEJC. Third, this meta-analysis was based on usable survival HR, assuming the proportion of hazards. In particular, in the KEYNOTE-061 and 062 trials, there was an obvious crossover between the two survival curves after 6–8 months, which makes the HR, measurement unreliable, and thus the combined HR, of the meta-analysis unreliable [73, 74]. The intersection of the survival curves occurred in immunotherapy trials, indicating that there are two different patient subgroups, and the response to ICI is the opposite [75]. Fourth, the limitations of further research were probably attributable to the lack of subgroup data on a few PD-1/PD-L1 inhibitors. Differences in efficacy among subgroups such as microsatellite instability (MSI) status, smoking status, and alcohol use may be the next direction for continued research. Data from the implementation of future trials may reinforce these observations. Fifth, the weakness and complexity of the results of this study lies in the probable differences in pharmacology between anti-PD-1 and anti-PD-L1 antibodies, and the different antibodies for evaluating PD-L1 status used in the RCTs. Finally, in view of the subgroup analysis, the findings should be interpreted with caution because of the small number of individuals analyzed. Hence, further validation studies are essential in GC/GEJC patients treated with PD-1/PD-L1 inhibitors.

In summary, PD-1/PD-L1 inhibitors improve OS but not PFS compared with chemotherapy in GC/GEJC. Age and sex could be used to predict the treatment potency of PD-1/PD-L1 inhibitors in GC/GEJC. With more phase III trials published in the future, the subgroup analyses may provide more accurate and comprehensive subgroup information to help us select the appropriate treatment population.