We identified eligible studies comparing the neoadjuvant immunotherapy with routine neoadjuvant therapy in the treatment of locally advanced EC in the MEDLINE (via PubMed), Embase (via OVID), ISI Web of Science database and Cochrane Library, between January, 2000 and February, 2023. The language was limited to English. The following search terms or keywords were used: esophageal cancer (MeSH) OR esophageal squamous cell carcinoma OR esophageal adenocarcinoma AND neoadjuvant OR preoperative AND programmed cell death 1 (PD-1) OR programmed cell death ligand 1 (PD-L1) OR immunotherapy ( Supplement Table S2 ). The last search was conducted on February 6, 2023. All titles and abstracts were screened and reviewed carefully.

Two authors (H.D.X. and K.L.) independently retrieved the available literature to identify the eligible studies. The studies were chosen on the basis of the following criteria: (a) studies only including patients with esophagus cancer or esophagogastric cancer; (b) the primary efficacy outcomes were pathological complete response rate; complete (R0) tumor resection rate; adverse events of neoadjuvant treat; (c) Randomized Controlled Trials (RCTs) or Retrospective experiments comparing neoadjuvant ICIs combined with nCT or nCRT for treating EC and (d) The experimental design met the requirements and included patients with ESCC and EAC. Exclusion criteria were as the following criteria: (a) studies reporting incomplete or inconsistent outcomes; and (b) duplicate studies, studies reporting animal experiments, case reports, cohort studies, and review articles.

Data extraction was respectively and carefully performed by two reviewers (H.D.X. and K.L.). The following information was collected: first author, year of publication, region, characteristics of the study population (number, sex and age), TNM stage, treatment therapy, adverse events of neoadjuvant therapies, postoperative complications, and pathological response. If the HR and its 95% CI were not directly provided in the original articles, the extracted survival information and the published risk table were used to reconstruct the survival curve for each included study using the method of David (20). The extraction of information was repeated if there were apparent discrepancies. Reviewers would contact the corresponding authors of the studies to access relevant data to analysis, when no sufficient data in publications were extracted. The methodological quality was assessed by reviewers (H.D.X. and K.L.) using the Newcastle-Ottawa Scale (NOS). Moderate quality was defined as 4-6 scores, and 7-9 scores was high quality. An additional adjudicator (L.W.) would be invited into the discussion to resolve the discrepancies between the reviewers. To ensure that patients were not counted several times, we selected data with the largest number of participants if a medical database was used by multiple studies in adjacent time periods and the number of patients were similar.

The neoadjuvant immunotherapy comprised neoadjuvant immune checkpoint inhibitor in combination with chemotherapy (nICT) and neoadjuvant immune checkpoint inhibitor in combination with chemoradiotherapy (nICRT). The routine neoadjuvant therapy included neoadjuvant chemotherapy (nCT) and neoadjuvant chemoradiotherapy (nCRT).

The pathological TNM stage was staged according to the 8th edition American Joint Committee on Cancer/Union for International Cancer Control staging system (21). We used Response Evaluation Criteria In Solid Tumours guideline version 1.13 system to classify regressive changes after neoadjuvant treatment based on histopathological results to reveal prognostic information (22). The treatment related adverse events (TRAEs) were assessed by Common Terminology Criteria for Adverse Events, version 4.0 (23).

Pathologic complete response (pCR) was defined as no evidence of residual tumor cells of the complete resected tumor specimen of neoadjuvant therapy and resection. The major pathological response (MPR) was defined as less than 10% of residual tumor cells. In the present study, the pCR and MPR rates were considered to be the primary outcomes. R0 resection was defined as a microscopically margin-negative resection without microscopic tumor on the primary tumor bed. The R0 surgical resection rate was set as the secondary outcome for comparing neoadjuvant immunotherapy plus chemotherapy with chemotherapy alone for patients.

The primary outcome of interest was pathologic response (pCR and MPR). The second outcome of interest was R0 resection rate. Odds ratio (OR) and associated 95% CI were used as the effect indicators comparing the safety and efficiency of the neoadjuvant immunotherapy with the routine neoadjuvant therapy. To minimize the influence of recall and selection bias that occur in retrospective studies, we performed stratified analyses to assess the association in all cohort studies. The heterogeneity between studies was evaluated with Q and I² statistics (24). The results were calculated using a random-effect model (Mantel-Haenszel method) when statistically heterogeneity (I² >50%) between studies were found. If low heterogeneity (I² ≤50%) was between studies fixed-effect model (Inverse Variance) was performed.

Sensitivity analysis, subgroup analysis and meta-regression were all performed to explore the sources of heterogeneity. The potential publication bias was further validated by the Egger’s and Begg’s test (25). All statistical analyses were two sides; and P value less than 0.05 was considered statistically significant. Statistical analysis was performed using the STATA version 15.0 (Stata Corp LP, College Station, Texas, USA).

After reviewing 557 publications found using the predefined search terms. All investigators finally agreed to include eight eligible studies (26–33) with 652 patients in our meta-analysis ( Table 1 ). The PRISMA flow chart of this meta-analysis was shown in Figure 1 . Among them, five studies were conducted on esophageal squamous cell carcinoma (ESCC) (26–28, 31, 32), and the other three addressed esophageal adenocarcinoma cancer (EAC) (29, 30, 33). About the neoadjuvant strategies, there were four studies that studied nICT vs nCT (26, 28, 31, 33), two studies that studied nICT vs nCRT (27, 32), two studies that studied nICRT vs nCRT (29, 30). The sample size was ranged from 47 to 168. The Newcastle-Ottawa scores are presented in the Supplement Table S2 .

Eight studies (26–33) were included in the pCR meta-analysis. Due to the heterogeneity between studies (I² = 32.8%, P=0.166), the data from the subgroups within a single study was pooled using a fixed-effect model. The estimated pCR rate was higher in the neoadjuvant immunotherapy group, including nICT and nICRT (OR =1.86; 95% CI, 1.25–2.79; Figure 2 ). As to the difference of the histologic subtypes, the studies were divided into two subgroups (the EAC group and the ESCC group). However, the different results were found in the ESCC and EAC subgroups, the estimated OR was 2.35 (95%CI, 1.20–4.64) in the EAC subgroup, and 1.65 (95% CI, 1.00–2.72) in the ESCC subgroup. The heterogeneity of two subgroups were (I² = 45.3%, P=0.161) and (I² = 30.9%, P=0.215), respectively. Interestingly, we found the common result (OR=1.93, 95% CI, 1.08–3.46; I² = 57.5%, P=0.094) (see Supplementary Material 3: Figure S1 ), when we deleted all studies included nCRT.

Six studies (27–31, 33) reported on the MPR. When pooling the studies, the pooled MPR was higher in the neoadjuvant immunotherapy group (OR =2.66; 95% CI, 1.69–4.19; Figure 3 ). Common results were showed in the subgroups, EAC and ESCC. The result was showed in Figure 3 .

No difference of R0 resection was founded between two groups (OR=1.79, 95% CI, 0.84–3.84; Figure 4 ), with moderate heterogeneity (I² = 39.9%, P=0.156).

Incidence of the overall grade ≥3 TRAEs was significantly higher in patients receiving neoadjuvant immunotherapy compared to patients receiving routine neoadjuvant therapy (neoadjuvant chemotherapy/chemoradiotherapy). Further analyses of individual grade ≥3 TRAEs showed that the neoadjuvant immunotherapy was associated with more pneumonitis/pneumonia (OR=3.46, 95% CI, 1.31–9.16; I² = 67.3%, P=0.005; Figure 5A ) and thyroid dysfunction (OR=4.69, 95% CI, 1.53–14.36; I² = 56.5%, P=0.032; Figure 5B ). Other individual grade ≥3 TRAEs including blood system, gastrointestinal system, and hypokalemia were comparable between the neoadjuvant immunotherapy and the routine neoadjuvant therapy (see Supplementary Material 3: Figure S2 ).

One death was reported in the patients received nICRT, and the death was due to pneumonitis (30).

Surgical resection rate (OR=0.74, 95% CI, 0.42–1.29; I² = 0.0%, P=0.478) and surgical delay rate (OR=1.24, 95% CI, 0.79–1.90; I² = 22.8%, P=0.255) were comparable between the neoadjuvant immunotherapy and the routine neoadjuvant therapy (see Supplementary Material 3: Figure S3 ). No surgical mortality was reported.

We conducted sensitivity analyses to ensure that the combined outcomes were not severely altered by the specific trials, and the overall estimates remained consistent across these analyses.

Egger’s test and Begg’s test were used to evaluate publication bias. Two regression intercept tests showed that the publication bias was not statistically significant ( Supplementary Material 3: Table S3 ).

There were several limitations in our study. Firstly, all included studies were descriptive study and the results have not been evaluated in large-scale controlled trials. Therefore, these findings required further validation by large RCTs. Only the RCTs were the golden standard of comparing the neoadjuvant immunotherapy and the routine neoadjuvant therapy. Secondly, researches for neoadjuvant immunotherapy in EAC remains fairly limited. The few researches were all performed in North America and Europe (29, 30). The diversity between ESCC and EAC might may lead to different responses to the neoadjuvant immunotherapy. Therefore, more clinical trials of neoadjuvant immunotherapy in EAC are needed, especially in East Asia. The main outcome measures are pCR and MPR, both would be typically increased by radiotherapy. A clear comparison between nICT vs nCT and nICRT vs nCRT is not achievable for the smaller sample size of the included studies. Thirdly, all eligible studies concentrated the pathological response rates, but no survival data was reported. The association between pathological response and survival in esophageal cancer deserves further investigation (51). Only the overall survival data was the gold standard to compare the neoadjuvant immunotherapy with routine neoadjuvant therapy. Another main limitation is the heterogeneity of the included studies, which is reflected in the different ICIs.