Immunotherapy combined with chemotherapy as first-line treatment for HER2-negative advanced gastric or gastroesophageal junction cancer: systematic review and Bayesian network meta-analysis based on specific PD-L1 expression levels

Abstract

Background

Gastric/gastroesophageal junction cancer (G/GEJC) is a major global health concern. HER2-negative advanced cases, which are common, have a poor prognosis and high peritoneal metastasis rates. First-line chemotherapy has limitations, while immunotherapy combined with chemotherapy shows promise, the optimal regimen for patients with different PD-L1 expression levels remains unclear.

Methods

We conducted a systematic review and Bayesian network meta - analysis by searching PubMed, Embase, Cochrane Library, and Web of Science databases. Eight RCTs involving 7619 patients and 7 immunotherapy-chemotherapy regimens were included. Outcomes included overall survival (OS), progression-free survival (PFS), objective response rate (ORR), and grade ≥ 3 adverse events (AEs). This analysis evaluated the efficacy and safety of various first-line immunotherapy-combination chemotherapy regimens for patients with HER2-negative advanced G/GEJC. Additionally, we assessed treatment efficacy across different PD-L1 expression subgroups.This study was registered in the Prospective Register of Systematic Reviews (CRD420251028737).

Results

Conventional meta-analysis showed that compared with chemotherapy alone, immunotherapy - chemotherapy significantly improved OS (HR = 0.79, 95% CI: 0.74–0.84), PFS (HR = 0.72, 95% CI: 0.68–0.77), and ORR (RR = 1.61, 95% CI: 1.45–1.80) in HER2-negative advanced G/GEJC patients. However, it also led to a higher incidence of grade ≥ 3 adverse events (AEs, RR = 1.18, 95% CI: 1.13–1.23). Subgroup analysis revealed that in patients with PD - L1 expression ≥ 1%, ≥ 5%, and ≥ 10%, this combination therapy significantly improved OS and PFS, with greater benefits at higher combined positive scores (CPS). Network meta - analysis indicated that for PD-L1 unselected patients, cadonilimab combined with chemotherapy was superior in OS, PFS, and ORR. In PD - L1 ≥ 1% patients, pembrolizumab plus chemotherapy had the best OS; in PD-L1 ≥ 5% patients, cadonilimab plus chemotherapy was optimal for OS and nivolumab plus chemotherapy for PFS; in PD-L1 ≥ 10% patients, sugemalimab plus chemotherapy was most effective for both OS and PFS.

Conclusion

In conclusion, these findings support immunotherapy-combination chemotherapy as a superior first-line strategy for HER2-negative advanced G/GEJC. Different optimal regimens can be selected based on PD-L1 expression levels, providing evidence-based guidance for clinical decision-making.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-16084-1.

Introduction

Gastric/gastroesophageal junction cancer (G/GEJC) ranks as the fifth most prevalent malignancy globally in terms of both incidence and mortality [1]. In high-risk East Asian populations, particularly in China, over 60% of G/GEJC patients present with advanced stage disease at diagnosis, correlating with poor survival outcomes [2]. HER2, a classical therapeutic target in G/GEJC, is negative in around 80% of cases [3]. Notably, patients with HER2-negative G/GEJC have a higher incidence of peritoneal metastasis, ranging from 40% to 50% [4]. First-line treatment for advanced HER2-negative G/GEJC typically involves fluoropyrimidine-based and platinum-based doublet or triplet chemotherapy regimens. Nevertheless, these regimens can only achieve a median overall survival (OS) of less than 12 months and are accompanied by significant toxicity [5, 6].

Multiple pivotal studies in advanced G/GEJC have established that combining immune checkpoint inhibitors (ICIs) with chemotherapy prolongs OS by 2–4 months, solidifying this strategy as the new standard first-line treatment. Current guidelines recommend ICIs, such as programmed cell death 1 (PD-1) and programmed cell death ligand 1 (PD-L1) inhibitors, for managing patients with G/GEJC who exhibit overexpression of immune checkpoint ligand [7]. Research indicates that the combination of chemotherapeutic agents with ICIs may potentiate immunotherapy efficacy through dual mechanisms: enhancing host immune activation and inhibiting tumor immune evasion mechanisms, especially in tumors characterized by weak immunogenicity and poor chemosensitivity [8, 9]. The CheckMate-649 trial demonstrated that nivolumab in combination with chemotherapy improved median OS from 11.1 to 14.4 months (HR = 0.71; 95% CI: 0.58–0.86; P < 0.0001) and median progression-free survival (PFS) from 6.0 to 7.7 months (HR = 0.68; 95% CI: 0.56–0.81; P < 0.0001) in patients with a combined positive score (CPS) ≥ 5. Survival benefits of nivolumab-based therapy were observed to increase proportionally with higher CPS thresholds [10]. The KEYNOTE-059 and related studies showed pembrolizumab plus chemotherapy exhibited promising antitumor activity in PD-L1-positive advanced G/GEJC patients [11–13]. Moreover, the ORIENT-16 and COMPASSION-15 trials demonstrated significant clinical benefits of sintilimab plus chemotherapy and cadonilimab plus chemotherapy in patients with CPS ≥ 5 and CPS ≥ 10, respectively [14, 15].

Following the completion of large-scale clinical trials, multiple first-line ICI regimens have been developed for G/GEJC. However, the optimal regimen that balances efficacy and safety remains undefined. PD-L1 expression, a validated biomarker for predicting response to immunotherapy, enables the identification of patient subgroups most likely to derive maximal clinical benefit [16]. However, the optimal first-line ICI regimen for HER2-negative G/GEJC patients with different PD-L1 expression levels remains undetermined. Notably, the optimal first-line ICI regimen for HER2-negative G/GEJC patients stratified by PD-L1 expression levels has not been conclusively determined.

Given that most randomized controlled trials (RCTs) focus on direct comparisons between ICIs and standard therapy rather than head-to-head evaluations of different ICIs, this study utilizes a Bayesian framework to perform indirect comparisons of efficacy and safety across various ICI regimens. Furthermore, we conducted a network meta-analysis (NMA) for patients with different PD-L1 expression levels, aiming to identify the optimal first-line treatment strategy for specific patient subgroups and provide evidence-based support for clinical decision-making.

Materials and methods

This NMA was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension statement for Network Meta-Analyses (PRISMA) guidelines (Supplementary Table 1) [17]. Due to the current lack of RCTs directly comparing different immunotherapy-combination chemotherapy regimens, we employed Bayesian methods for indirect comparisons to probabilistically predict the ranking of efficacy and safety among various immunotherapy-combination regimens [18]. This study has been registered in the International Prospective Register of Systematic Reviews (CRD420251028737).

Data sources and search strategy

We conducted computerized searches of PubMed, EMBASE, the Cochrane Library, and Web of Science databases to identify published RCTs on first-line immunotherapy combined with chemotherapy for HER2-negative advanced G/GEJC. The search keywords included: “advanced gastric cancer”, “Stomach Neoplasms”, “adenocarcinoma of the gastroesophageal junction”, “gastroesophageal-junction cancer”, “randomized clinical trial”, “immune checkpoint inhibitors”, “PD-L1 inhibitor”, " PD-1inhibitor”, “CTLA-4 Inhibitor”, “pembrolizumab”, “camrelizumab”, “nivolumab”, “ipilimumab”, “durvalumab”, “tremelimumab”, “toripalimab”, “tislelizumab” (Supplementary Table 2). The literature search covered all English publications from database establishment to April 1, 2025, utilizing a combined approach of free-text words and controlled vocabulary keywords for retrieval.

Selection criteria

Inclusion criteria

1. Patients with histologically or cytologically confirmed HER2-negative (or ERBB2-negative) advanced G/GEJC.

2. Multicenter, randomized, controlled phase III clinical trials evaluating ICIs as monotherapy or combination therapy for first-line treatment.

3. The included types of ICIs were PD-1, PD-L1, and CTLA-4, among others; the chemotherapy regimens primarily consisted of platinum-based (oxaliplatin, cisplatin) combination chemotherapy or in combination with fluorouracil agents (capecitabine, 5-FU, S-1).

4. RCTs comparing ICIs with chemotherapy alone or placebo plus chemotherapy in HER2-negative advanced G/GEJC.

5. Trials must report at least one of the following outcomes: ①OS: Time from randomization to death from any cause.②PFS: Time from randomization to disease progression or death from any cause.③ORR: The proportion of patients achieving an objective response.④Grade ≥ 3 adverse events (AEs): As defined by the National Cancer Institute Common Terminology Criteria for Adverse Events (NCI-CTCAE).

Exclusion criteria

1. Article types including reviews, case reports, conference abstracts, or animal studies.

2. Articles with incomplete data or unavailable outcome measures.

3. Studies with overlapping patient populations.

Prior to inclusion of RCTs, initial screening was performed based on titles and abstracts. All included RCTs underwent independent dual review by two investigators to ensure the incorporated data represented the most recently published evidence.

Data extraction and quality assessment

Two reviewers (Wen Yi, Kalima Muhetaer) independently performed literature screening and data extraction following PRISMA guidelines, with cross-verification of results. Discrepancies were resolved through consultation with a third reviewer. Extracted data included: clinical trial name, NCT number, journal, authors, country, publication year, tumor stage, histology, sample size, sex, age, PD-L1 expression status, and treatment regimens for experimental and control groups.The outcome measures included: HR with 95% CI for OS and PFS.Risk ratios (RR) with 95% CI for ORR and grade ≥ 3 AEs.The included RCTs were assessed for risk of bias using the Cochrane Risk of Bias Tool (RoB 2.0) [19]. The assessment workflow is as follows: First, two principal investigators conduct independent evaluations and discuss their findings. If a consensus cannot be reached, a third senior researcher is consulted for arbitration to make the final decision. This evaluation tool examines the following five domains: risk of bias arising from the randomization process, risk of bias owing to deviations from the intended interventions, risk of bias from missing outcome data, risk of bias in the measurement of the outcome, and risk of bias in the selection of the reported result. The risk-of-bias assessment for RCTs across different domains was categorized into three levels: low risk, high risk, and having “some concerns”.

Statistical analysis

The primary outcomes were OS and PFS, with the secondary outcomes comprised ORR and grade ≥ 3 AEs. HR with 95% CI were used as effect measures for OS and PFS, while OR with 95% CI served as effect measures for ORR and grade ≥ 3 AEs.

Bayesian network meta-analysis was carried out in R (version 4.2.0) using the “rjags” and “gemtc” packages. A fixed-effect framework was adopted, with three parallel Markov chains run sequentially: each chain included 20,000 burn-in iterations followed by 100,000 sampling iterations (thin = 1). Publication bias was assessed via funnel plots created in STATA 18.0. To confirm the soundness of the indirect comparisons, conventional frequentist pairwise meta-analyses of direct head-to-head trials were performed and juxtaposed with the corresponding indirect estimates from the network meta-analysis (Supplementary Table 3). A separate frequentist pairwise meta-analysis was conducted in RevMan 5.4 to re-evaluate first-line immunotherapy plus chemotherapy versus chemotherapy alone in HER2-negative advanced G/GEJC patients, both irrespective of PD-L1 status and within predefined PD-L1-positive subgroups. To test the consistency assumption of the network structure, this study further compared the DIC values between the consistency model and the inconsistency model. If the differences in DIC values between the consistency and inconsistency models were all below 5, it indicated no significant inconsistency in the network, suggesting good exchangeability between indirect and direct evidence. For head-to-head meta-analyses, study heterogeneity was quantified using Cochran’s Q statistic and the I² metric. Fixed-effect modeling was applied when P > 0.10 and I² <50%; otherwise (P ≤ 0.10 or I² ≥50%), a random-effects model was used, followed by leave-one-out sensitivity analyses to identify potential sources of heterogeneity. Subgroup analyses stratified by PD-L1 expression thresholds were prespecified for all primary endpoints. All statistical tests were two-sided, with statistical significance defined as α = 0.05.

Results

Systematic review and characteristics of the included studies

The initial search yielded 1123 potentially relevant records. Following deduplication and abstract screening to exclude irrelevant studies, 282 articles were selected for full-text review, with 8 studies ultimately included in the meta-analysis (Fig. 1). These eight studies enrolled 7619 patients with HER2-negative advanced G/GEJC, with 3829 allocated to the experimental group receiving first-line combined immunotherapy and chemotherapy and 3790 allocated to the control group receiving either chemotherapy alone or placebo plus chemotherapy.

Fig. 1.

Study selection

In the CheckMate-649 trial, over 90% of patients with a tumor proportion score (TPS) ≥ 1% (230 of 253 cases) exhibited CPS ≥ 5. Therefore, the TPS ≥ 1% cohort was analyzed under the assumption of biological equivalence to CPS ≥ 5 population. The KEYNOTE-062 trial excluded patients with CPS < 1, thus precluding the inclusion of PD-L1-negative subgroups in the analyses of CPS < 5 and CPS < 10 cohorts. The RATIONALE-305 trial reported a CPS ≥ 5 prevalence of 60%, comparable to CheckMate-649 (61%) and ORIENT-16 (61%), justifying the use of tumor-associated positivity (TAP) as a surrogate for CPS. Detailed characteristics of the included studies are presented in Tables 1 and 2. Methodological quality assessment of all 8 studies was performed using the Cochrane Risk of Bias 2.0 (ROB 2.0) tool, with results shown in Supplementary Fig. 1.

Table 1.

Baseline characteristics of studies included in the network meta-analysis

Study	Source	Registered ID	Sample Size	Stage
(Phase, Design)	(y)	(Randomization)	(Median	(Male/	Histology	Ethnicity (%)	Intervention Arm(s)	Control Arm
			Age/y)	Female)
CheckMate 649 [10]		NCT02872116	789/792	IV	G/GEJC	Asian (23)	Nivolumab (360 mg Q3W or 240 mg Q2W)+
(open-label, III)	2021	(1:1)	(62/61)	1100/481		United States and Canada(17)	[XELOX]capecitabine 1000 mg/m2,bid, days1-14 + oxaliplatin130mg/m2,day 1,Q3W
						Rest of world (61)	[FOLFOX ]leucovorin 400 mg/m2, day 1;fluorouracil 400 mg/m2, day 1 and 1200 mg/m2, days 1–2;oxaliplatin 85 mg/m2 day 1 Q2W
ATTRACTION-4 [20]		NCT02746796	362/326	IV	G/GEJC	Japan(55)	Nivolumab 360 mg Q3W +	placebo+
(open-label, III)	2024	(1:1)	(64/65)	523/201		South Korea(41)	[SOX]oxaliplatin 130 mg/m² on day 1 +  oral S-1 40 mg/m²
						Taiwan(4)	[CAPOX]oral capecitabine 1000 mg/m² bid on days 1–14
KEYNOTE-062 [13]		NCT03745170	257/259	IV	G/GEJC	Europe/North America/ Australia(58)	pembrolizumab 200 mg Q3W+	placebo+
(open-label, III)	2020	(1:1)	(62/62.5)	374/142		Asia(25)	[FP]cisplatin 80 mg/m2/d on day 1 plus fluorouracil 800 mg/m2/d on days 1 to 5
						Rest of world(17)	[XP ]capecitabine 1000 mg/m² bid on days 1–14
ORIENT-16 [14]		NCT03777657	327/323	IV	G/GEJC	Asia	sintilimab 3 mg/kg for body weight < 60 kg+	placebo+
(open-label, III)	2023	(1:1)	(60/60)	483/176			[XELOX]capecitabine 1000 mg/m2,bid, days1-14 + oxaliplatin130mg/m2,day 1,Q3W
KEYNOTE-859 [21]		NCT03802591	790/789	IV	G/GEJC	multiple	pembrolizumab 200 mg Q3W +	placebo+
(open-label, III)	2023	(1:1)	(61/62)	1071/508			[FP]cisplatin 80 mg/m2/d on day 1 plus fluorouracil 800 mg/m2/d on days 1 to 5
							[XELOX]capecitabine 1000 mg/m2,bid, days1-14 + oxaliplatin130mg/m2,day 1,Q3W
Rationale 305 [22]		NCT03777657	501/496	IV	G/GEJC	Asia(73)	Tislelizumab 200 mg Q3W +	placebo+
(open-label, III)	2024	(1:1)	(61/62)	NA		Europe/ North America(26)	[XELOX]capecitabine 1000 mg/m2,bid, days1-14 + oxaliplatin130mg/m2,day 1,Q3W
							[FP]cisplatin 80 mg/m2/d on day 1 plus fluorouracil 800 mg/m2/d on days 1 to 5
COMPASSION-15 [15]		NCT03745170	305/305	IV	G/GEJC	Asia	Cado 10 mg/kg Q3W+	placebo+
(open-label, III)	2025	(1:1)	(63/64)	474/143			[XELOX]capecitabine 1000 mg/m2,bid, days1-14 + oxaliplatin130mg/m2,day 1,Q3W
GEMSTONE-303 [23]		NCT03802591	241/238	IV	G/GEJC	Asia	Sugemalimab 1200 mg Q3W+	placebo+
(open-label, III)	2025	(1:1)	(63/63)	350/129			[CAPOX]oral capecitabine 1000 mg/m² bid on days 1–14 and in travenous oxaliplatin 130 mg/m2 on day 1

Table 2.

Characteristics of included randomized controlled trials

Study		PD-L1 ≥ 1 Patients (%)		PD-L1 ≥ 5 Patients (%)		PD-L1 ≥ 10 Patients (%)
	PD-L1 Detection	Intervention(s), n(%)	Control, n (%)	Intervention(s), n(%)	Control, n (%)	Intervention(s), n(%)	Control, n (%)	Reported Outcomes
CheckMate 649	CPS	641(81)	655(82)	473 (60)	482 (61)	/	/	OS, PFS, ORR, grade ≥ 3 AEs
ATTRACTION-4	TPS	219 (61.0)	260 (72.6)	/	/	/	/	OS, PFS, ORR, grade ≥ 3 AEs
KEYNOTE-062	CPS	257(100)	250(100)	/	/	99 (38.5)	90 (36.0)	OS, PFS, ORR, grade ≥ 3 AEs
ORIENT-16	CPS	/	/	197(60)	200(61.9)	/	/	OS, PFS, ORR, grade ≥ 3 AEs
KEYNOTE-859	CPS	1235		/	/	551		OS, PFS, ORR, grade ≥ 3 AEs
Rationale 305	TAP/CPS	/	/	274 (55)	272 (55)	/	/	OS, PFS, grade ≥ 3 AEs
COMPASSION-15	CPS	420		256		153		OS, PFS, ORR, grade ≥ 3 AEs
GEMSTONE-303	CPS	/	/	241(100)	238(100)	130 (53.9)	128 (53.8)	OS, PFS, ORR, grade ≥ 3 AEs

Pairwise meta-analysis

Comparisons of OS

Among the six included studies, inter-study heterogeneity was not statistically significant (P = 0.40, I² = 2%), and a fixed-effect model was used for meta-analysis (Supplementary Fig. 2). The results demonstrated that, compared to chemotherapy alone, first-line immunotherapy combined with chemotherapy significantly prolonged OS in unselected PD-L1 expression patients (HR = 0.79, 95% CI: 0.74–0.84). Compared to chemotherapy, subgroup analyses showed OS benefits with immunotherapy-chemotherapy in HER2-negative advanced G/GEJC patients across increasing PD-L1 thresholds: PD-L1 expression ≥ 1% (HR = 0.78, 95% CI: 0.72–0.86), PD-L1 expression ≥ 5% (HR = 0.71, 95% CI: 0.64–0.78), and PD-L1 expression ≥ 10% (HR = 0.68, 95% CI: 0.59–0.79), with higher CPS correlating to more significant survival benefits (Fig. 2 and Supplementary Fig. 3).

Fig. 2.

Pooled forest plots. A The pooled forest plots for OS and PFS. B The pooled forest plots for ORR and grade ≥3 AEs

Comparisons of PFS

Among the six included studies, significant statistical heterogeneity was observed (P = 0.01, I² = 66%), and a random-effects model was used for meta-analysis (Supplementary Fig. 4). The results demonstrated that, compared to chemotherapy alone, first-line immunotherapy combined with chemotherapy significantly prolonged PFS in patients unselected for PD-L1 expression (HR = 0.72, 95% CI: 0.68–0.77), PD-L1 expression ≥ 1% (HR = 0.75, 95% CI: 0.68–0.83), PD-L1 expression ≥ 5% (HR = 0.66, 95% CI: 0.60–0.74), and PD-L1 expression ≥ 10% (HR = 0.73, 95% CI: 0.58–0.92) compared to chemotherapy alone (Fig. 2 and Supplementary Fig. 5).

Comparisons of ORR

The meta-analysis incorporated seven studies, with no significant statistical heterogeneity observed among them (P = 0.69, I² = 0%), and a fixed-effects model was applied (Supplementary Fig. 6). The results indicated that the immunotherapy combined with chemotherapy group achieved a 1.23-fold higher ORR compared to chemotherapy alone in patients with HER2-negative advanced G/GEJC, with a statistically significant difference between the two groups (RR = 1.61, 95% CI: 1.45–1.80).

Safety and toxicity

The incidence of grade ≥ 3 AEs was used to evaluate the safety and toxicity of first-line ICIs. All eight included studies on first-line therapy reported the incidence of grade ≥ 3 AEs, with substantial statistical heterogeneity among studies (P = 0.06, I² = 49%), and a fixed-effects model was applied for meta-analysis (Fig. 2 and Supplementary Fig. 7). The results revealed that HER2-negative advanced G/GEJC patients treated with ICIs had a higher incidence of grade ≥ 3 AEs compared with the chemotherapy group (RR = 1.18, 95% CI: 1.13–1.23).

Network meta-analyses

Comparisons of OS

The NMA demonstrated that patients receiving immunotherapy combined with chemotherapy had greater OS benefits compared to those treated with chemotherapy alone. Specifically, in populations unselected for PD-L1 expression, cadonilimab combined with chemotherapy (HR = 0.66, 95% CI: 0.54–0.81) and sintilimab combined with chemotherapy (HR = 0.77, 95% CI: 0.63–0.94) significantly prolonged OS relative to chemotherapy alone (Figs. 3 and 5A).

Fig. 3.

Network diagrams of comparisons on different outcomes of treatments in specific groups. Color yellow indicates OS, purple represents PFS, green denotes ORR, and blue corresponds to AE ≥ 3

Fig. 5.

Network meta-analysis pooled outcomes. A Upper triangular section: 95% CrI for PFS; lower triangular section: 95% CrI for OS. B Upper triangular section: 95% CrI for PFS for patients with PD-L1 expression ≥ 1.lower triangular section: 95% CrI for OS for patients with PD-L1 expression ≥ 1. C Upper triangular section: 95% CrI for PFS for patients with PD-L1 expression ≥ 5.lower triangular section: 95% CrI for OS for patients with PD-L1 expression ≥ 5. D Upper triangular section: 95% CrI for PFS for patients with PD-L1 expression ≥ 10.lower triangular section: 95% CrI for OS for patients with PD-L1 expression ≥ 10. E Upper triangular section: 95% CrI for ORR.Lower triangular section: 95% CrI for grade ≥ 3 AEs.The data within each cell represent the hazard ratios or 95% CrI for the comparison between the treatment defined by the row and by the column. A hazard ratio of less than 1 and an odds ratio of greater than 1 indicates a favourable effect for the row-defining treatment. Significant results are highlighted in bold

Comparisons of PFS

The NMA showed that patients receiving immunotherapy in combination with chemotherapy had greater PFS benefits compared to those treated with chemotherapy alone. Cadonilimab combined with chemotherapy was identified as the most effective regimen (HR = 0.53, 95% CI: 0.44–0.64), a finding consistent with the observed OS outcomes (Figs. 3 and 5A).

Comparisons of ORR

The NMA demonstrated that patients receiving immunotherapy combined with chemotherapy achieved significantly higher ORR compared to those treated with chemotherapy alone. Cadonilimab combined with combination (RR = 0.75; 95% CI: 0.65–0.90) exhibited superior ORR benefits relative to chemotherapy monotherapy. Both nivolumab combined with chemotherapy (RR = 1.24; 95% CI: 1.14–1.35) and pembrolizumab combined with regimens (RR = 1.24; 95% CI: 1.13–1.36) showed comparable efficacy advantages over conventional chemotherapy, with all comparisons reaching statistical significance (Figs. 3 and 5E).

Safety and toxicity

Safety and toxicity assessments were based on grade ≥ 3 AEs. The NMA incorporated eight studies evaluating seven first-line immunotherapy combined with chemotherapy combination regimens. Results indicated that all chemotherapy-alone regimens had a significantly lower incidence of grade ≥ 3 AEs compared with immunotherapy combination groups. Among these, cadonilimab plus chemotherapy exhibited the least favorable safety profile versus chemotherapy alone (RR = 0.93; 95% CI: 0.71–1.21), though this difference lacked statistical significance (Figs. 3 and 5E).

On the basis of PD-L1 expression level

For patients with any PD-L1 status and those with ≥ 5% expression, cadonilimab combined with chemotherapy was the optimal treatment compared to chemotherapy alone (HR = 0.66, 95% CI: 0.54–0.81 for any status; HR = 0.58, 95% CI: 0.41–0.82 for ≥ 5% expression. In patients with PD-L1 expression ≥ 1% and ≥ 5%, nivolumab combined with chemotherapy showed significant OS benefits compared with chemotherapy alone (for PD-L1 expression ≥ 1%, HR = 0.81, 95% CI: 0.69–0.96; for PD-L1 expression ≥ 5%, HR = 0.71, 95% CI: 0.59–0.86), with greater clinical benefits observed at higher CPS (Figs. 4 and 5B, C and D). The pembrolizumab combined with chemotherapy demonstrated significant OS prolongation in patients with PD-L1 expression ≥ 1% (HR = 0.77, 95% CI: 0.69–0.86) and PD-L1 expression ≥ 10% (HR = 0.7, 95% CI: 0.59–0.83), with greater OS benefits observed in populations exhibiting higher PD-L1 CPS, indicating that immunotherapy combined with chemotherapy becomes the preferred treatment strategy as PD-L1 CPS increases. In PD-L1 expression ≥ 1% and ≥ 5% populations, first-line nivolumab combined with chemotherapy showed significant PFS prolongation versus chemotherapy alone (for PD-L1 expression ≥ 1%, HR = 0.72, 95% CI: 0.64–0.8; for PD-L1 expression ≥ 5%, HR = 0.68, 95% CI: 0.57–0.82), with enhanced PFS benefits at higher PD-L1 CPS levels. Compared to chemotherapy alone, the pembrolizumab combined with chemotherapy did not achieve significant PFS benefits in PD-L1 expression ≥ 1% (HR = 0.84, 95% CI: 0.7–1.01) or PD-L1 expression ≥ 10% (HR = 1.10, 95% CI: 0.75–1.61).

Fig. 4.

The network meta-analysis diagram comparing the efficacy of first-line immune checkpoint inhibitors (ICIs) combined with chemotherapy in HER2-negative advanced G/GEJC across varying PD-L1 expression thresholds. A Patients with PD-L1 expression ≥ 1% treated with ICIs-based regimens. B Patients with PD-L1 expression ≥ 5% treated with ICIs-based regimens. C Patients with PD-L1 expression ≥ 10% treated with ICIs-based regimens

Rank

Ranking analysis was performed based on Bayesian ranking spectrum (Figs. 6 and 7,Supplementary Tables 4,5). In HER2-negative advanced G/GEJC patients unselected for PD-L1, cadonilimab combined with chemotherapy as first-line therapy versus chemotherapy alone was most likely to achieve optimal OS (69.24%) and PFS (90.26%) benefits. In HER2-negative advanced G/GEJC patients with PD-L1 expression ≥ 1%, pembrolizumab combined with chemotherapy versus chemotherapy alone showed the highest probability of optimal OS (69.70%), while nivolumab-chemotherapy combination was most likely to obtain PFS (92.02%) benefits. In the population with PD-L1 expression ≥ 0%, the combination of sugemalimab and chemotherapy, compared with chemotherapy alone, demonstrated the best efficacy, ranking first in both OS (70.53%) and PFS (99.52%). Cadonilimab combined with chemotherapy as first-line therapy, compared with chemotherapy alone, was most likely to achieve optimal ORR (35.45%), with a ranking comparable to that of the sugemalimab-chemotherapy combination (36.61%).

Fig. 6.

Bayesian ranking profiles illustrating efficacy and safety outcomes of first-line immunotherapy regimens in PD-L1-unselected patients with HER2-negative advanced G/GEJC. A OS ranking; B PFS ranking; C ORR ranking; D Grade ≥ 3 AEs ranking

Fig. 7.

Bayesian ranking profiles comparing efficacy and safety outcomes of first-line immunotherapy regimens in HER2-negative advanced G/GEJC patients stratified by PD-L1 expression levels. A OS ranking for PD-L1 expression ≥ 1%; B OS ranking for PD-L1 expression ≥ 5%; C OS ranking for PD-L1 expression ≥ 10%; D PFS ranking for PD-L1 expression ≥ 1%; E PFS ranking for PD-L1 expression ≥ 5%; F PFS ranking for PD-L1 expression ≥ 10%

Heterogeneity and inconsistency

Pairwise meta-analysis using frequentist methods yielded results consistent with the corresponding Bayesian framework synthesis (Supplementary Table 3). To test the consistency assumption of the network structure, this study further compared the Deviance Information Criterion (DIC) values between the consistency model and the inconsistency model. As shown in Table S6, the differences in DIC values between the consistency and inconsistency models were all below 5 for all outcome measures (e.g., OS: 10.80 vs. 11.52; PFS: 11.50 vs. 11.78), indicating no significant inconsistency in the network and good exchangeability between indirect and direct evidence. Furthermore, all I² values for the outcomes were below 50%, supporting low heterogeneity among the included studies and demonstrating favorable homogeneity and stability of the network structure.Sensitivity analyses were conducted for outcomes with I² >50% to investigate heterogeneity sources. Funnel plot analysis using OS as the outcome measure revealed symmetrical distribution of study points without scattered outliers, indicating low likelihood of publication bias in this study(Supplementary Fig. 8).

Discussion

G/GEJC ranks among the top five malignancies globally in terms of both incidence and mortality, with the treatment challenges being particularly pronounced in patients with advanced HER2-negative disease. Although the combination strategy of ICIs with chemotherapy has gradually emerged as a new first-line treatment option in recent years, the optimal regimen for patients with different PD-L1 expression levels remains unclear. This study conducted a systematic review and Bayesian network meta-analysis, incorporating 8 RCTs involving a total of 7,619 patients with advanced HER2-negative G/GEJC. It compared the efficacy and safety of seven different immunotherapy-chemotherapy combination regimens across various PD-L1 expression strata (≥ 1%, ≥ 5%, ≥ 10%), providing critical evidence to inform clinical practice and guideline updates.

The results of this traditional meta-analysis show that, compared to chemotherapy alone, first-line immunotherapy combined with chemotherapy significantly improves OS (HR = 0.79, 95% CI: 0.74–0.84), PFS (HR = 0.72, 95% CI: 0.68–0.77), and ORR (RR = 1.61, 95% CI: 1.45–1.80) in patients with HER2-negative advanced G/GEJC. However, it should be noted that the immunotherapy combination regimen also significantly increased the incidence of grade ≥ 3 AEs (RR = 1.18, 95% CI: 1.13–1.23). A meta-analysis by M.S. Beshr et al. in 2024 [24], which included 10 RCTs and 6,782 patients, on PD-1/PD-L1 inhibitors combined with chemotherapy or monotherapy versus chemotherapy alone for advanced unresectable HER2-negative gastric, gastroesophageal junction, and esophageal adenocarcinomas, showed that immunochemotherapy significantly improved OS and PFS, Our findings are consistent with this observation. Similarly, a study by Wenwei Zhang et al. in 2025 [25] on immunotherapy combined with chemotherapy as a first-line treatment for advanced gastric cancer indicated that all ICI-chemotherapy combinations improved OS and PFS compared to chemotherapy alone. Specifically, nivolumab combined with chemotherapy was associated with an increase in grade ≥ 3 AEs, and cadonilimab combined with chemotherapy had a toxicity profile that more frequently led to treatment discontinuation. Another meta-analysis by Lin et al. [26] showed that, compared to chemotherapy alone, chemotherapy combined with ICIs provided better PFS (HR = 0.71; 95% CI: 0.65–0.79), OS (HR = 0.79; 95% CI: 0.75–0.83), and higher ORR (OR = 1.57; 95% CI: 1.43–1.72) in patients with advanced G/GEJC. However, the combination regimen also increased the risk of grade ≥ 3 AEs (RR = 1.15; 95% CI: 1.09–1.22) and serious adverse events (RR = 1.44; 95% CI: 1.21–1.70),This finding is consistent with our results.

Indeed, the combination of chemotherapy and ICIs has now emerged as a highly effective immunotherapy strategy and has been incorporated into standard treatment guidelines for various malignancies [27]. This efficacy may be attributed to the following reasons: traditional chemotherapy not only kills tumor cells through direct cytotoxic effects but also remodels the tumor immune microenvironment through multiple pathways, creating favorable conditions for ICIs to exert their effects. After chemotherapy-induced tumor cell apoptosis, the body releases a large number of tumor-associated antigens [28], which activate the antigen-presenting function of dendritic cells [29], thereby promoting the activation and proliferation of effector T cells. Chemotherapy can selectively eliminate regulatory T cells and myeloid-derived suppressor cells, reduce the secretion of immunosuppressive factors (such as IL-10 and TGF-β), and alleviate the immunosuppressive state of the tumor microenvironment [30]. Third, chemotherapy drugs represented by platinum agents and fluorouracil can upregulate the expression of PD-L1 on the surface of tumor cells by activating the IFN-γsignaling pathway, providing more targets for ICIs [31, 32]. And ICIs counteract the immune escape mechanisms of tumor cells by blocking the PD-1/PD-L1 and CTLA-4 signaling pathways, restoring the ability of effector T cells to kill tumor cells [33, 34]. The two modalities synergize to achieve antitumor effects, which is particularly critical for G/GEJC with weak immunogenicity and poor chemotherapy sensitivity [35]. Meanwhile, the occurrence of adverse events with combination regimens highlights the necessity for personalized management strategies. In clinical practice, patients’ individual circumstances should be fully considered to balance the benefits and risks of treatment and develop rational therapeutic plans.

The NMA in this study further refined the optimal treatment choices for patients with different PD-L1 expression levels. The results indicated that in HER2-negative advanced G/GEJC patients with PD-L1-unselected, the PD-1/CTLA-4 bispecific antibody cadonilimab combined with chemotherapy demonstrated the best performance in terms of OS, PFS, and ORR: OS (HR = 0.66, 95% CrI: 0.54–0.81), PFS (HR = 0.53, 95% CrI: 0.44–0.64), ORR (RR = 0.75, 95% CrI: 0.65–0.90). Moreover, its safety profile showed no statistically significant difference compared to chemotherapy alone (RR = 0.93, 95% CrI: 0.71–1.21). As the first PD-1/CTLA-4 bispecific antibody approved in China for the treatment of advanced gastric cancer, the efficacy advantage of cadonilimab stems from its unique dual-target mechanism of action. Blocking the PD-1 pathway can alleviate T-cell exhaustion and restore the anti-tumor function of effector T-cells; while inhibiting the CTLA-4 pathway can suppress the activation and proliferation of regulatory T-cells, increase the number of tumor-infiltrating lymphocytes, and further enhance the immune response [36]. Compared with single PD-1 inhibitors, dual-target blockade can more comprehensively activate anti-tumor immunity, particularly in unselected populations with high heterogeneity in PD-L1 expression, covering more potential beneficiaries and reducing immune non-response due to low PD-L1 expression. Furthermore, the Fc segment of cadonilimab has been engineered to reduce antibody-dependent cell-mediated cytotoxicity, minimizing damage to effector T-cells [37]. This contributes to enhanced efficacy while maintaining a favorable safety profile, which is consistent with the findings in this study that showed no significant increase in the incidence of grade ≥ 3 AEs. In the CSCO Gastric Cancer Diagnosis and Treatment Guidelines (2025), the combination of XELOX regimen and cadonilimab has been added as a Category I recommendation, regardless of PD-L1 expression status.

This study also confirmed that pembrolizumab combined with chemotherapy significantly improved OS in patients with PD-L1 expression ≥ 1% and ≥ 10%, establishing it as the optimal treatment regimen. Moreover, the OS benefit exhibited an incremental trend with increasing PD-L1 CPS. These results indicate that immunotherapy combined with chemotherapy should be prioritized as PD-L1 CPS levels rise. This conclusion aligns with the findings of Ji et al.‘s meta-analysis, which underscored the clinical value of the pembrolizumab-based regimen, particularly in patients with PD-L1-positive tumors or those exhibiting high microsatellite instability [38]. Furthermore, this study observed that nivolumab combined with chemotherapy significantly prolonged PFS in patients with PD-L1 expression ≥ 1% and ≥ 5%, and this PFS benefit was further enhanced with increasing PD-L1 CPS. The ORIENT-16 trial [14] and the KEYNOTE-859 trial [21] reported significant improvements in median OS at a PD-L1 cutoff ≥ 10%, with HRs of 0.64 and 0.56, respectively (both P < 0.00001). The CheckMate-649 trial [10] also reported similar survival benefits in the population with PD-L1 expression ≥ 5%. Additionally, a meta-analysis investigating PD-1/PD-L1 inhibitors combined with chemotherapy as first-line treatment for unresectable, locally advanced, HER2-negative G/GEJC demonstrated that, compared to chemotherapy alone, the combination significantly improved OS, especially in patients with tumor cell PD-L1 expression ≥ 1% (HR = 0.62, 95% CI: 0.48–0.81) and those with PD-L1 CPS ≥ 10 (HR = 0.66, 95% CI: 0.57–0.77) [39],This finding further corroborates our hypothesis.

Beyond PD-L1 expression, additional biomarkers—including tumor mutational burden (TMB), Epstein-Barr virus-encoded RNA (EBER) status, microsatellite instability (MSI) status, and Claudin 18.2 (CLDN18.2)—may also possess predictive value for immunotherapy response. TMB acts as a predictive biomarker of ICI efficacy across diverse cancer types, with elevated TMB associated with a higher likelihood of clinical benefit [40]. EBER-positive tumors exhibit dense immune cell infiltration, rendering EBER-positive gastric cancer more responsive to ICI combined with chemotherapy [41]. As a transmembrane protein specifically expressed in normal gastric mucosal epithelial cells and abnormally upregulated in a subset of G/GEJC, CLDN18.2 not only serves as a target for monoclonal antibodies (e.g., zolbetuximab) but also correlates with immunotherapy response [42]. Studies have shown that CLDN18.2-positive tumors often have a more inflamed tumor microenvironment, with increased TIL infiltration and PD-L1 expression [43, 44], which may enhance the synergistic effect of ICIs combined with chemotherapy, though further validation in larger cohorts is needed.

Compared to other meta-analyses [45–48], this study excluded trials involving HER2-positive patients in its inclusion criteria to reduce heterogeneity and inconsistency, with all group comparisons based on identical PD-L1 CPS thresholds. Indeed, for HER2-positive patients, an updated result from the phase III KEYNOTE-811 trial presented at the 2024 European Society for Medical Oncology (ESMO) Annual Congress demonstrated that, compared to the trastuzumab plus chemotherapy group, the pembrolizumab combined with trastuzumab and chemotherapy group significantly prolonged median OS. In patients with a PD-L1 CPS ≥ 1, the pembrolizumab combination group showed an extended median OS (20.1 months vs. 15.7 months, HR = 0.79, 95% CI: 0.66–0.95), and the ORR in the overall population increased by 12.5% [49]. This indicates that the combination of ICIs remains effective for HER2-positive patients. Second, this study utilized a Bayesian framework to determine treatment rankings within each subgroup. This statistical approach employs Bayesian ranking by calculating the Surface Under the Cumulative Ranking Curve values for each regimen, overcoming the limitation of traditional meta-analysis that only allows pairwise comparisons. It is particularly suitable for network meta-analyses involving indirect comparisons, and the ranking results are consistent with the mechanisms of action of the drugs and the quality of the included studies. Additionally, this study stratified patients based on PD-L1 expression levels and identified the optimal immunotherapy regimens, providing evidence-based medical support for subsequent guideline updates.

Although this study derived significant conclusions, the following limitations warrant attention. First, the included studies were multicenter RCTs involving patients of diverse ethnicities, but we observed a predominance of Asian populations in the analyzed cohorts, indicating a potential population selection bias that may limit the generalizability of our findings to Western populations. Given differences in gastric cancer pathological characteristics and treatment responses between Eastern and Western populations, our findings may hold greater clinical relevance for Asian populations. Future RCTs incorporating larger Western or other minority populations would enhance the comprehensiveness and generalizability of the results. Second, significant treatment-related biases were observed across studies, including heterogeneity in chemotherapy regimens, treatment cycles, duration, and maintenance periods, which may introduce outcome variability and affect the comparability of results. Additionally, the current analysis was limited by insufficient data on PD-L1 low-expression populations, which precluded meaningful subgroup analysis. Future studies should prioritize including more PD-L1 low-expression patients to explore potential benefits of immunotherapy in this subgroup. Third, the included RCTs utilized both TPS and CPS to assess PD-L1 expression. As CPS more accurately characterizes the tumor microenvironment and PD-L1 expression status, CPS is the preferred evaluation method. The final results of the RATIONALE 305 trial provided outcomes based on both CPS and TAP scores. Notably, the OS for CPS ≥ 5 versus < 5 (0.72 vs. 0.91) and TAP score ≥ 5% versus < 5% (0.73 vs. 0.89) showed remarkable similarity, further supporting the interchangeability between CPS ≥ 5 and TAP ≥ 5% as implemented in our meta-analysis.

Conclusion

In conclusion, this study integrated data from the most comprehensive RCTs and validated the clinical benefits of immunotherapy combined with chemotherapy for HER2-negative advanced G/GEJC through network meta-analysis. By stratifying patients based on PD-L1 expression levels, we provide evidence-based guidance for clinicians in treatment decision-making. When balancing efficacy and safety, cadonilimab plus chemotherapy represents a favorable first-line therapeutic option for PD-L1-unselected G/GEJC patients. Furthermore, our results demonstrate distinct survival outcomes associated with specific immunotherapy regimens in HER2-negative advanced G/GEJC patients stratified by PD-L1 expression thresholds (≥ 1%, ≥ 5%, or ≥ 10%). These findings may supplement the National Comprehensive Cancer Network guidelines regarding optimal treatment strategies for HER2-negative advanced G/GEJC patients with varying PD-L1 expression levels.

Abbreviations

ICIs

Immune checkpoint inhibitors

OS

Overall survival

RCTs

Randomized controlled trials

PD−1

Programmed cell death 1

PFS

Progression-free survival

ORR

Objective response rate

G/GEJC

Gastric/gastroesophageal junction cancer

NMA

Network meta-analysis

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

HRs

Hazard ratios

CIs

Confidence intervals

Authors’ contributions

W.Y.: conceptualization, data curation, formal analysis, investigation, methodology, writing—original draft; K.M.: resources, validation; L.C.: formal analysisX.Ma.: investigation, visualization; Y.W.: software, validation; L.C.: data curation, project administration; Y.G. , C.Q.: funding acquisition, resources; M.H.: investigation, data curation; L.Z.: supervision, review & editing. All authors have read and agreed to the published version of the manuscript.

Funding

There is no funding source with authors to declare.

Data availability

All date were obtained from publicly available databases of relevant clinical trials.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.