A systematic review was performed to assess the effectiveness and safety of first-line treatments for esophageal cancer. This review entailed a computer-based search of randomized controlled trials (RCTs) within the Cochrane, PubMed, Embase, and Web of Science databases, culminating on November 17, 2023. The search methodology incorporated a blend of controlled vocabulary terms and free-text keywords including “esophageal cancer,” “newly diagnosed,” “untreated,” “first-line,” “front-line,” and “initial.” The specifics of the search strategies are delineated in Suplementary Table S1 .

Inclusion Criteria: This review rigorously selected studies involving adult patients who have been diagnosed with esophageal cancer. The interventions scrutinized included Cisplatin and Docetaxel (CD), Cisplatin, Epirubicin, and Capecitabine (CET-CF), Nedaplatin and Capecitabine (N-CF), Nedaplatin and Irinotecan (NI), Nimotuzumab and Capecitabine (Nim-CF), Paclitaxel and Capecitabine (P-CF), and S-1 and Capecitabine (Ser-CF), compared against a control group receiving Capecitabine (CF) alone. The primary objectives of this review were to assess overall survival (OS), progression-free survival (PFS), the objective response rate (ORR), and the disease control rate (DCR). Secondary outcomes were focused on the evaluation of adverse events. Only studies designed as randomized controlled trials (RCTs) met the criteria for inclusion.

Exclusion Criteria: The review excluded studies if they were redundant publications, animal research, case reports, conference abstracts, or reviews. Furthermore, studies were excluded if they included patients with concurrent organic diseases.

The selection of studies was meticulously undertaken by two authors, who independently applied the pre-established inclusion and exclusion criteria. Any discrepancies that arose during this process were amicably resolved through mutual discussion between the authors. If a consensus could not be reached, the matter was escalated to a third, neutral party for arbitration. The data harvested from the studies that met these criteria encompassed critical variables, including but not limited to the primary author’s name, the year of publication, the country of origin, the sample size, the distribution of participants by gender, their ages, the ECOG (Eastern Cooperative Oncology Group) performance status, the nature of the interventions administered, and the metrics used to gauge the outcomes.

The evaluation of bias risk was conducted in accordance with the most current guidelines stipulated by the Cochrane Risk of Bias Assessment Tool 2.0 (ROB 2.0) (9). This comprehensive tool outlines five critical domains to scrutinize: bias emerging from the randomization process, bias resulting from deviations from the intended interventions, bias associated with missing outcome data, bias in the measurement of outcomes, and bias in the selection of reported results. The assessment of quality categorized the studies into three distinct levels: “low risk of bias,” “some concerns,” or “high risk of bias.” To ensure reliability, the findings were independently verified by two reviewers. Any inconsistencies identified during this process were meticulously addressed, either by reaching a consensus through discussion or, if necessary, by seeking the judgment of a third party.

A Bayesian network meta-analysis was performed utilizing R software version 4.2.3 (R Foundation for Statistical Computing) (10, 11), adopting a priori vague random-effects models. The analysis employed Markov chain Monte Carlo techniques, as described by (12), to derive optimal combined estimates and associated probabilities for each treatment option. For representing survival data, hazard ratios (HRs) and their 95% credible intervals (CIs) were used, while binary outcomes were depicted through odds ratios (ORs) and their 95% CIs.

The Surface Under the Cumulative Ranking curve (SUCRA) was calculated to gauge the likelihood of each intervention being the most effective. Network and funnel plots were generated using STATA version 15.0, utilizing a direct macro command for their creation. In the network plot, drugs are represented by circles, with the size of each circle being proportional to the number of patients included in the respective trials. The links between circles (edges) denote the direct comparisons available between drugs. Furthermore, cumulative probability distributions were visualized using the ggplot2 package in R.

An initial search of the databases retrieved a total of 1,737 articles. After the elimination of duplicates, this number was reduced to 407. A detailed screening of titles and abstracts led to the exclusion of 1,263 articles. Further scrutiny through full-text review resulted in the removal of 69 more articles. Ultimately, eight articles (13–20) were selected for in-depth analysis, as depicted in Figure 1 .

The analysis encompassed eight articles, comprising a total of 3051 untreated esophageal cancer patients. Specific characteristics of the included studies can be found in Table 1 . All studies included in this analysis clearly described their blinding methods. High risk primarily stemmed from deviations in the expected intervention measures, and the risk of bias assessment for the included studies can be found in Figures 2 and 3 .

Seven studies (13, 14, 16–20) evaluated progression-free survival (PFS), as depicted in Figure 5 . The network diagram ( Figure 5A ) illustrates direct comparisons among various treatment combinations: CF with NI, N-CF, P-CF, Ser-CF, CD, and CET-CF. When compared to CF, the hazard ratios (HRs) were as follows: for CD, HR = 1.35 with a 95% confidence interval (CI) of (0.12, 15.18); for CET-CF, HR = 1.08, 95% CI (0.13, 8.8); for N-CF, HR = 0.79, 95% CI (0.37, 1.7); for NI, HR = 1.2, 95% CI (0.64, 2.3); for P-CF, HR = 0.64, 95% CI (0.35, 1.2); and for Ser-CF, HR = 0.60, 95% CI (0.25, 1.5). These treatments showed marginal differences in enhancing PFS ( Table 2B ). Notably, Ser-CF achieved the highest area under the cumulative ranking curve at 73.5%, followed closely by P-CF at 73.3%, N-CF at 57.4%, and NI at 27.0% ( Figure 5B ; Table 3 ).

In this analysis, seven studies (13, 14, 16–20) were examined to assess the objective response rate (ORR), as depicted in Figure 6 . The network diagram ( Figure 6A ) delineated direct comparisons among various treatments: Cetuximab and Fluoropyrimidine (CF) versus Nitroglycerin Patch (NI), Non-Cetuximab and Fluoropyrimidine (N-CF), Preoperative Cetuximab and Fluoropyrimidine (P-CF), Serotonin and Cetuximab and Fluoropyrimidine (Ser-CF), Cetuximab and Docetaxel (CD), and Cetuximab monotherapy (CET-CF).

The analysis of odds ratios (ORs) revealed varying effectiveness of treatments compared to CF. Specifically, CD exhibited an OR of 0.74 [95% CI (0.20, 2.6)], indicating less effectiveness, while CET-CF showed an OR of 1.2 [95% CI (0.42, 3.7)], suggesting a marginal improvement. Notably, N-CF [OR=2.8, 95% CI (2.2, 3.8)], NI [OR=1.2, 95% CI (0.92, 1.6)], P-CF [OR=2.0, 95% CI (1.5, 2.6)], and Ser-CF [OR=1.9, 95% CI (1.3, 2.7)] demonstrated more promising results, with N-CF, P-CF, and Ser-CF notably enhancing ORR, as highlighted in Figure 6C .

However, treatments such as CET-CF, NI, and CD did not significantly improve ORR, as detailed in Table 2C . Among the analyzed treatments, N-CF emerged as the most favorable in enhancing ORR. In terms of the cumulative ranking curve, CD achieved the highest area under the curve (AUC) of 84%, followed by CF (81.3%), NI (62.9%), and N-CF (2.83%), as shown in Figure 6B and Table 3 . This comprehensive analysis underscores the varied efficacy of the treatments, with N-CF standing out for its potential in improving ORR.

Seven studies (13, 15–20) reported neutropenia as an adverse event, as illustrated in Figure 8 . The network diagram ( Figure 8A ) displayed direct comparisons between CF and NI, CF and N-CF, CF and Nim-CF, CF and P-CF, CF and Ser-CF, and CF and CET-CF. Compared to CF, CET-CF [OR=5.6, 95% CI (1.9, 18.0)], N-CF [OR=1.3, 95% CI (0.92, 1.7)], NI [OR=0.014, 95% CI (0.002, 0.048)], Nim-CF [OR=1.6, 95% CI (0.63, 3.9)], P-CF [OR=1.3, 95% CI (1.0, 1.8)], and Ser-CF [OR=1.1, 95% CI (0.76, 1.6)] were analyzed. In comparison to CF, P-CF and CET-CF increased the probability of neutropenia, with CET-CF having the highest likelihood. However, Ser-CF, Nim-CF, NI, and N-CF had a lower probability of causing neutropenia ( Figure 8C ). The area under the cumulative ranking curve indicated that NI had the highest impact (100%), followed by CF (73.6%), Ser-CF (60.2%), and CET-CF (0.9%) ( Figure 8B ; Table 3 ).

Additionally, in these seven articles (13, 15–20), nausea as an adverse event was mentioned, as shown in Figure 9 . The network diagram ( Figure 9A ) did not form a closed-loop structure between the various interventions. Compared to CF, CET-CF [OR=2.6, 95% CI (0.92, 7.6)], N-CF [OR=1.3, 95% CI (1.0, 1.7)], NI [OR=0.066, 95% CI (0.043, 0.099)], Nim-CF [OR=0.64, 95% CI (0.28, 1.4)], P-CF [OR=1.2, 95% CI (0.93, 1.6)], and Ser-CF [OR=1.0, 95% CI (0.71, 1.5)] were evaluated. Compared to CF, CET-CF and N-CF had an increased probability of causing nausea, with CET-CF having the highest likelihood. Conversely, NI, Nim-CF, P-CF, and Ser-CF had a lower probability of causing nausea, with NI having the lowest probability ( Figure 9C ). The area under the cumulative ranking curve indicated that NI had the highest impact (100%), followed by Nim-CF (76.3%), CF (59.6%), and CET-CF (5%) ( Figure 9B ; Table 3 ).

Furthermore, skin disorders were mentioned in six articles (13, 15–20), as shown in Figure 10 . The network diagram ( Figure 10A ) did not form a closed-loop structure between CF and the various interventions. Compared to CF, CET-CF [OR=5.58e+22, 95% CI (1.04e+03, 4.20e+71)], N-CF [OR=3.99, 95% CI (1.87, 9.55)], NI [OR=11.7, 95% CI (5.85, 27.2)], Nim-CF [OR=3.77, 95% CI (1.02, 18.4)], P-CF [OR=1.91, 95% CI (1.07, 3.51)], and Ser-CF [OR=2.30, 95% CI (0.921, 6.96)] were analyzed. It was observed that all interventions had a higher probability of causing skin adverse events compared to CF, with CET-CF having the highest probability and Ser-CF having the lowest probability. The area under the cumulative ranking curve indicated that CF had the highest impact (98.7%), followed by P-CF (73.2%), Ser-CF (65.0%), and CET-CF (0.00%) ( Figure 10B ; Table 3 ).

Six articles (15–20) reported instances of anorexic adverse reactions, as depicted in Figure 11 . The network diagram ( Figure 11A ) indicates that the various treatment regimens did not form a closed-loop structure. Direct comparisons exist between CF and NI, CF and N-CF, CF and Nim-CF, CF and P-CF, and CF and Ser-CF. Compared to CF, the odds ratios (OR) and 95% confidence intervals (CI) for anorexic reactions are as follows: N-CF [OR=2.6, 95%CI (0.14, 59.0)], NI [OR=1.0, 95%CI (0.057, 23.0)], Nim-CF [OR=0.68, 95%CI (0.011, 40.0)], P-CF [OR=1.9, 95%CI (0.11, 33.0)], and Ser-CF [OR=1.1, 95%CI (0.020, 64.0)]. From this, it can be inferred that the likelihood of experiencing anorexia is higher with N-CF and lower with Nim-CF ( Figure 11C ). The cumulative ranking probability curves indicate that Nim-CF had the highest area under the curve (62.1%), followed by CF (57.6%), NI (54.4%), and N-CF (33.3%) being the lowest ( Figure 11B ; Table 3 ). Based on these four adverse reactions, it can be concluded that CET-CF has a higher probability of presenting the aforementioned adverse events, while NI has a lower probability.

In summary, based on the four types of adverse reactions mentioned above, CET-CF had a higher probability of causing these adverse events, while NI had a lower probability.

We used funnel plots to assess publication bias for ORR, DCR, and the four discussed adverse reactions in this article. The results indicated a significant likelihood of publication bias for ORR, DCR, and the four adverse reactions ( Suplementary Table S2 , Suplementary Figures S1 - S6 ).