Neoadjuvant immune checkpoint inhibitors for localized dMMR/MSI-H gastric cancer: a meta-analysis.

Introduction
Perioperative chemotherapy followed by surgery is the standard of care for resectable gastric and gastroesophageal junction (GEJ) cancers.1, 2, 3 The prevalence of deficient mismatch repair (dMMR) or microsatellite instability-high (MSI-H) in gastric GEJ tumors ranges from 3.9% to 12.4%, with higher rates observed in early-stage disease.4, 5, 6 Localized dMMR/MSI-H gastric and GEJ cancers respond poorly to fluoropyrimidine-based chemotherapy but are highly sensitive to immune checkpoint inhibitors (ICIs),7,8 prompting reconsideration of perioperative chemotherapy in this subgroup.
In the metastatic setting, ICIs have transformed the treatment paradigm for dMMR/MSI-H gastrointestinal malignancies.9, 10, 11, 12, 13, 14 In resectable tumors, neoadjuvant ICIs have demonstrated compelling efficacy, particularly in rectal and colorectal adenocarcinomas.15,16 In gastric and GEJ cancers, early-phase trials have produced promising findings. The phase II NEONIPIGA trial reported a pathologic complete response (pCR) rate of 59% (17/29) with nivolumab and ipilimumab.17 Similarly, the phase II INFINITY trial demonstrated a pCR of 60% (9/15) in surgical patients and a clinical complete response (cCR) rate of 76% (13/17) among nonoperative patients treated with tremelimumab and durvalumab. Notably, only 1 of 13 (7.7%) patients in the nonoperative group experienced local regrowth at a median follow-up of 11.5 months.18 Additionally, Cercek et al.19 reported a cCR of 50% (9/18) following neoadjuvant dostarlimab, with local recurrence in 2 of 9 patients (22.2%) at a median follow-up of 14.9 months. While these findings are encouraging, the limited sample sizes and early-phase design of these studies highlight the need for larger, more definitive investigations to clarify the outcomes of neoadjuvant ICIs in this select population.
Since the publication of these early-phase trials, neoadjuvant ICIs for localized dMMR/MSI-H gastric and GEJ cancers has gained increasing attention, particularly in the context of potential organ-sparing strategies.20, 21, 22 Several ongoing trials aim to expand upon the current body of evidence and further delineate oncologic outcomes in this population. In the interim, given the paucity of large-scale data, we conducted a systematic review and meta-analysis to synthesize existing evidence and evaluate the efficacy and safety of neoadjuvant ICIs in this clinical context.

Methods

Eligibility criteria
Studies were eligible for inclusion if they: (i) used prospective or retrospective design; (ii) enrolled patients with localized/potentially resectable dMMR/MSI-H gastric/GEJ cancer; and (iii) investigated neoadjuvant ICIs. At least one clinical outcome was required. Exclusion criteria included: metastatic-only studies; non-gastric/GEJ cohorts; sample size <5; and unclear regimens. Case series, review articles, editorials, and previously published meta-analyses were also excluded

Search strategy and data extraction
We systematically searched PubMed, EMBASE, Scopus, Web of Science, and Cochrane Reviews from inception to June 2025. The detailed search strategy is as described in the Supplementary Material, available at https://doi.org/10.1016/j.esmoop.2026.106066. The references from all included studies were also searched manually for any additional studies. Two authors (RA and WKS) independently extracted the data. This systematic review and meta-analysis was registered with the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD420251122340).

Endpoints
Efficacy outcomes included pCR, cCR, and major pathologic response (MPR). cCR was defined as no evidence of residual disease on tumor-specific imaging and on endoscopy with biopsy. Safety outcomes included grade 3-4 immune-related adverse events (irAEs). Nonoperative management (NOM) was assessed exploratorily.

Quality assessment
A risk of bias assessment was conducted using The Risk Of Bias In Non-randomized Studies – of Interventions, Version 2 (ROBINS-I V2) assessment tool, which is specifically designed for follow-up (cohort) studies. To apply this framework consistently across various study designs, including single-arm studies and specific intervention arms from randomized studies, the subgroup of interest (e.g. the ICIs arm) of each study was considered as a nonrandomized cohort. Seven domains were assessed and rated as having low, moderate, serious, or critical risk of bias. An overall judgment of risk of bias (low, moderate, serious, or critical) was then derived from the domain-level judgments using a predefined algorithm.23 The GRADE framework was not applied because the included studies were predominantly single-arm, early-phase (phase I-II) trials without comparator groups, for which standard GRADE certainty ratings are less applicable.

Statistical analysis
This systematic review and meta-analysis was carried out in accordance with the Cochrane Collaboration and the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement guidelines.24 Proportions with 95% confidence intervals (CIs) were used to pool event rates for occurrence of cCR, pCR, MPR, and grade 3-4 irAEs. Cochran Q test and I2 statistics were used to assess for heterogeneity; P values inferior to 0.10 and I2 > 25% were considered significant for heterogeneity. We used a DerSimonian and Laird random-effects model. Subgroup analyses were conducted based on neoadjuvant therapy duration and strategy to explore sources of heterogeneity and assess consistency across study designs. Sensitivity analyses were carried out using a leave-one-out approach to evaluate the robustness of the overall estimates by sequentially removing individual studies; a Baujat plot was generated to identify studies contributing most to heterogeneity; and Egger’s test was applied to assess the risk of publication bias.25 RStudio 2025.05.1 + 513 was used for statistical analysis.

Results

Study selection
As illustrated in Supplementary Figure S1, available at https://doi.org/10.1016/j.esmoop.2026.106066, the initial search yielded 4031 results. After removal of duplicate records and ineligible studies, 115 remained and were fully reviewed based on inclusion criteria. Of these, 20 studies met the predefined inclusion criteria and were included in the quantitative synthesis. Of the 20 included studies, 19 were identified through database screening and 1 through snowballing of references/recently published work. The KEYNOTE-585 trial was not included because it did not report absolute pCR rates specifically for the MSI-H pembrolizumab group, but instead presented the difference in pCR between the pembrolizumab and the placebo arm.

Study characteristics
The 20 studies comprised 396 patients with localized dMMR/MSI-H gastric/GEJ cancer from 12 clinical trials (3 phase III,26, 27, 28 8 phase II17, 18, 19,29, 30, 31, 32, 33, and 1 phase I34) and 8 retrospective studies (5 retrospective analysis,35, 36, 37, 38, 39 2 retrospective cohorts40, and 1 retrospective propensity score-matched analysis41). Eight studies evaluated neoadjuvant ICIs alone {five with single agent19,30,34, 35, 36 and three with dual blockade using anti-programmed cell death protein 1/programmed death-ligand 1 [anti-PD(L)1] plus anti-cytotoxic T-lymphocyte antigen 4 (anti-CTLA4)}.17,18,31 Ten studies investigated neoadjuvant immunochemotherapy22,26,27,29,32,33,37, 38, 39, 40 and two assessed immunochemotherapy combined with targeted therapy.28,41 Five studies entertained NOM.18,19,22,38,39 Study characteristics are reported in Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2026.106066.

Overall pooled analyses
Among 19 studies (n = 313), the pooled pCR was 41.9% (95% CI 33.2% to 51.1%, I2 = 42.9) (Figure 1A). The pooled MPR was 64.2% (95% CI 49.1% to 76.9%, I2 = 41.3) (Figure 1B). The pooled cCR was 63.8% (95% CI 45.6% to 78.8%, I2 = 53.8) (Figure 1C). Grade 3-4 irAEs occurred in 6.7% (95% CI 1.7% to 13.8%, I2 = 58.1) and are displayed by regimen in Figure 2. Among 63 patients from four studies,18,36,37,38 pCR did not differ significantly between cT1-T3 tumors (50.2%, 95% CI 27.1% to 73.2%) and cT4 tumors (33.3%, 95% CI 17.6% to 53.9%, P = 0.29) (Supplementary Figure S2, available at https://doi.org/10.1016/j.esmoop.2026.106066). No irAEs were reported that precluded surgery, and no grade 5 events occurred. One patient died from early postoperative complications22 and two from late postoperative complications.18

Subgroup analyses by neoadjuvant therapy strategy
For pCR, the pooled rate was 32.9% (95% CI 16.9% to 54.3%, I2 = 71.9) among patients receiving single-agent ICI, 52.8% (95% CI 39.5% to 65.8%, I2 = 44.0) with dual ICI therapy, 42.1% (95% CI 28.5% to 56.9%, I2 = 37.9%) with single-agent ICI with chemotherapy, and 46.7% (95% CI 34.5% to 59.2%, I2 = 0.0) with single-agent ICI plus chemotherapy and targeted therapy. Between-subgroup differences for pCR across all studies were not statistically significant (P = 0.43) (Figure 3A).
For MPR, pooled rates were 69.8% (95% CI 56.3% to 86.1%, I2 = 39.5) for dual ICI therapy and 61.3% (95% CI 33.6% to 83.2%, I2 = 57.0) for single-agent ICI plus chemotherapy. No statistically significant differences were observed between subgroups (P = 0.97) (Figure 3B).

Subgroup analyses by duration of neoadjuvant ICI exposure
To examine whether pCR varied by length of neoadjuvant ICI, studies were stratified into ≥3 months and <3 months of neoadjuvant ICI exposure. The pooled pCR for ≥3 months was 50.5% (95% CI 42.3% to 58.7% I2 = 30.1%). For <3 months, the pooled pCR was 28.4% (95% CI 18.9% to 40.2%, I2 = 16.5%). The test for subgroup differences was significant (χ2 = 8.8, df = 1, P = 0.003), indicating higher pCR with longer ICI duration (Figure 4).

Nonoperative management
NOM was explored in five studies,18,19,22,38,39 encompassing 44 patients (11.1% of the pooled cohort). Two were prospective clinical trials that implemented NOM in patients achieving cCR following single-agent or dual ICI blockade.18,19 The remaining three studies were retrospective cohorts or analyses that adopted NOM case-by-case driven by patient refusal, medical comorbidities, or multidisciplinary consensus.22,38,39 Across the five studies, median follow-up durations ranged from 11.5 to 28 months. Pooled cCR for patients who underwent NOM was 86% (31/36).18,19,22 Notably, in the study by Cercek et al.,19 one patient developed early local tumor regrowth that was successfully treated with salvage surgery, while in the study by Sahvan et al.,22 one patient experienced local recurrence 12 months after achieving cCR and likewise underwent salvage surgery. These findings are summarized in Supplementary Table S2, available at https://doi.org/10.1016/j.esmoop.2026.106066. In the INFINITY trial, 1-year overall survival (OS) and progression-free survival (PFS) among NOM patients were 100% and 93.8%, respectively. In the study by Sahvan et al.,22 2-year OS was 92.1%, event-free survival (EFS) 87.3%, and metastasis-free survival (MFS) 88.2% at a median follow-up of 19.3 months. This is illustrated in Supplementary Table S3, available at https://doi.org/10.1016/j.esmoop.2026.106066.

Quality assessment
Individual appraisal of randomized and nonrandomized studies is reported in Supplementary Table S4, available at https://doi.org/10.1016/j.esmoop.2026.106066. Two randomized studies were rated as low risk of bias and one as moderate. For nonrandomized studies, 13 were rated as critical risk of bias mostly driven by their single-arm design and consequent lack of comparator arm which resulted in critical risk of bias in the first domain of bias due to confounding. Risk of bias assessment was not carried out in three studies because they were conference abstracts. Leave-one-out analyses for pCR revealed the IMHOTEP and NICE trials as principal sources of heterogeneity (Supplementary Figure S3, available at https://doi.org/10.1016/j.esmoop.2026.106066). Excluding IMHOTEP reduced I2 from 42.9% to 27.3% and excluding NICE lowered it to 34.6%. As shown in Supplementary Figure S4, available at https://doi.org/10.1016/j.esmoop.2026.106066, heterogeneity and influence analyses using the Baujat plot for pCR revealed that IMHOTEP 2024 and NICE 2024 contributed disproportionately to overall heterogeneity and effect size estimates. The funnel plot for pCR showed moderate asymmetry, with IMHOTEP 2025 and NICE 2024 falling outside the pseudo-95% confidence limits, suggesting possible small-study effects (Supplementary Figure S5, available at https://doi.org/10.1016/j.esmoop.2026.106066). Egger’s test, however, did not indicate significant funnel plot asymmetry (t = −1.03, P = 0.3165), thus provided no statistical evidence of publication bias (Supplementary Figure S6, available at https://doi.org/10.1016/j.esmoop.2026.106066).

Discussion
In this systematic review and meta-analysis of 20 studies involving 396 patients with localized dMMR/MSI-H GEJ cancer, we evaluated the efficacy and safety of neoadjuvant ICIs. The main findings include high pCR, cCR, and MPR rates of 41.9%, 63.8%, and 64.2%, respectively, and a low incidence of grade 3-4 irAEs, reported in only 6.7% of patients. Longer neoadjuvant ICI exposure (≥3 months) yielded a higher pCR than <3 months [50.2% (95% CI 42.3% to 58.7%) versus 28.4% (18.9% to 40.2)]. Additionally, our qualitative analysis identified a subset of 44 patients managed with NOM, of whom only 2 experienced local tumor regrowth. These findings underscore the strong efficacy and favorable safety profile of neoadjuvant ICIs in this population, as well as the potential feasibility of NOM as a management strategy in selected patients.
dMMR/MSI-H status carries distinct therapeutic and prognostic implications.7 Tumors with this molecular subtype exhibit reduced benefit from fluoropyrimidine-based chemotherapy due to inherent chemoresistance42 but demonstrate remarkable responses to ICIs, attributed to their high tumor mutational burden and increased neoantigen load.43 Notably, MSI-H has also emerged as a favorable prognostic factor in resectable gastric cancer, suggesting a rationale for minimizing or omitting chemotherapy in this subset.7 While ICIs are already integrated into guidelines for dMMR/MSI-H GI malignancies,1,2 the NOM strategy in those who achieve cCR remains underexplored.
The promising pooled rates of complete pCR, cCR, and MPR in our systematic review and meta-analysis align with pivotal studies in the field. The NEONIPIGA trial reported a pCR rate of 58.6%, while the INFINITY trial demonstrated pCR and cCR rates of 60.0% and 76.5%, respectively. The Cercek et al.19 study reported a cCR of 45.0% among patients with gastroesophageal cancer. These findings are further supported by an individual patient data pooled analysis of seven studies, which highlighted higher rates of cPR and MPR with ICI compared with chemotherapy.44 The IMHOTEP study, however, appears to have lower pCR rates, likely due to the abbreviated single-agent ICIs duration (one to two cycles of pembrolizumab).30 Across studies included in our analysis, neoadjuvant ICI exposure ranged from 1 to 6 months, with subgroup analysis showing higher pCR with ≥3 months versus <3 months of ICI. This aligns with data from colorectal cancer. A recent systematic review demonstrated a positive association between longer treatment duration and higher rates of complete response in patients with localized dMMR colorectal cancer.45
Subgroup analyses stratified by ICI strategy did not show differences in pCR rates. While NEONIPIGA and INFINITY employed dual blockade and reported high pCR rates (58.6% and 60.0%, respectively), our pooled estimate for dual regimens was lower at 52.8%, driven by Kang et al.31 who enrolled a small cohort with predominantly cT4 disease and reported only 22.2% pCR. Retrospective cohorts such as Sahvan et al.22 and Shannon et al.39 also demonstrated encouraging pooled pCR rates around 60.0%, despite using variable regimens with or without chemotherapy. Among studies using single-agent ICI plus chemotherapy, pCR rates ranged broadly from 0% in the MONEO study (n = 6) to 78.5% in the NICE trial.32,33 Since we lack individual-level patient data, however, one must note the lack of observed difference between dual- and single-agent strategies may be confounded by variations in baseline clinical characteristics, such as TNM (tumor–node–metastasis) stage.46 While data suggest chemotherapy may potentiate immunogenicity in certain subgroups by augmenting the tumor-to-T-cell ratio, the optimal ICI regimen and role of combination therapy remain to be defined in prospective cohorts.
It is important to note that surrogate endpoints such as pCR and cCR must be interpreted with caution. While these markers are well-validated predictors of long-term outcomes in rectal cancer,47,48 their applicability to upper GI malignancies is uncertain.49 Importantly, pCR and cCR can be markedly discordant, as evidenced by a large cohort study in which only 31.0% (67/218) of patients with cCR also achieved pCR.50 Data from the studies included in this review further illustrate a bidirectional discordance specific to the post-immunotherapy setting, reflecting the limitations of current diagnostic modalities. First, clinical assessment frequently underestimates the depth of response, as immune-mediated inflammation or healing ulcers can mimic residual tumor on standard imaging and endoscopy. For instance, Wang et al.41 reported a cCR rate of only 3.6% despite a pCR rate of 47.7%, suggesting that radiological evaluation missed the vast majority of complete responders. Similarly, Shannon et al.39 observed that all resected specimens achieving pCR still exhibited gross ulceration on endoscopy, complicating visual assessment. Conversely, clinical assessment may overestimate response, posing a risk for NOM. Even with strict definitions requiring negative biopsy, false negatives occur. Zhang et al.38 described a patient achieving radiological complete response who harbored residual ypT2N0 disease, and the INFINITY trial noted that standard tumor biopsies had reduced sensitivity for detecting residual disease.
Our findings also reaffirm the favorable safety profile of neoadjuvant ICIs. The incidence of grade 3-4 irAEs was low at 6.7%, with no reports of grade 5 events. Although concerns have been raised about treatment-related delays or contraindications to surgery after neoadjuvant ICIs, this was not substantiated in our dataset. Cercek et al.19 explicitly reported that surgical eligibility was not compromised in any patient, and similar observations were echoed in other trials.17, 18, 19,26,27 One patient, however, died from early22 and two from late postoperative complications.18
An emerging paradigm in localized dMMR/MSI-H tumors is NOM.21 Initially pioneered in rectal cancer,15,51 NOM seeks to avoid the morbidity and mortality associated with surgery52 in complete responders. Tts application to gastric and GEJ malignancies, however, is challenged by the difficulty of accurately assessing cCR. In rectal cancer, cCR can be reliably evaluated using digital rectal examination, rectoscopy, and rectal cancer protocol pelvic magnetic resonance imaging (MRI),53 whereas in upper GI cancer the definition of cCR remains unclear.54 The current assessment typically depends on endoscopic biopsies and cross-sectional imaging [computed tomography (CT), positron emission tomography (PET)], both with limited accuracy in the setting of post-immunotherapy inflammatory changes. In the INFINITY trial, surveillance for NOM included multimodal follow-up every 12 weeks for 2 years with CT scans, endoscopic ultrasound (EUS) with bite-on-bite biopsies, nodal fine needle aspiration (FNA), and serial liquid biopsy assays. The study by Sahvan et al.22 adopted a similar approach using CT or PET/CT, endoscopy/EUS, and liquid biopsies when available, with one recurrence successfully salvaged endoscopically. Cercek at al.19 implemented surveillance every 4 months, utilizing PET imaging (PERCIST) and endoscopy with biopsy. Our qualitative synthesis identified 44 patients managed nonoperatively, with only two documented local recurrences.19,22 This underscores the potential feasibility of NOM in selected patients, particularly given the significant postoperative risks in upper GI oncologic surgery.
Beyond short-term response rates, durability of benefit is a key consideration. Early survival data are encouraging. In the INFINITY trial, 1-year OS and PFS among NOM patients were 100% and 93.8%, respectively, while 2-year OS and PFS in surgical patients were 73.3% and 66.7%, respectively. NEONIPIGA reported 97% OS at a median follow-up of 14.9 months, and Cercek et al.19 observed a 2-year recurrence-free survival of 85% in 67 patients with non-rectal MSI-H tumors. Sahvan et al.22 further reported 2-year OS of 92.1%, EFS of 87.3%, and MFS of 88.2% at a median follow-up of 19.3 months. These preliminary findings (Supplementary Table S3, available at https://doi.org/10.1016/j.esmoop.2026.106066) suggest that neoadjuvant ICIs may provide sustained clinical benefit, though confirmation in larger cohorts with longer follow-up is needed.
There are several limitations related to this meta-analysis and individual studies. First, our analysis was carried out with study-level data rather than individual patient data, which would limit the power of our analysis. Second, study heterogeneity, nonuniform outcome definitions, and small sample sizes constrain interpretability. Third, while our analysis indicates a benefit to longer treatment duration, these univariate findings must be interpreted with caution as the sample size precluded multivariate meta-regression to adjust for confounders like dual-agent immunotherapy or concurrent chemotherapy. Moreover, as previously mentioned, the endpoints used (pCR, cCR, and MPR) do not correlate strongly with durability of response, and there was a scarcity of data on long-term outcomes, precluding their inclusion in pooled analyses. Thus, the findings should be interpreted with caution. We employed rigorous methodology, however, including subgroup and sensitivity analyses, to mitigate bias and enhance validity. While the data are derived mostly from small phase II trials and retrospective cohorts, pooling enabled the most comprehensive synthesis to date in this rapidly evolving field.
In conclusion, neoadjuvant ICIs for localized dMMR/MSI-H GEJ tumors yield high response rates and minimal toxicity, suggesting a transformative potential in this unique patient population. In addition, our findings support the feasibility of surgery-sparing approaches for selected patients and highlight the need for standardization of response criteria and prospective validation in larger, controlled trials. As the field evolves, establishing optimal ICI regimens, refining selection for NOM, and defining durable endpoints will be crucial to guide clinical decision-making and inform future guidelines.

Funding
None declared.

Disclosure
The authors have declared no conflicts of interest.