2.1 Study registration
The meta-analysis complies with the PRISMA statement[6].
2.2 Criteria for considering studies
2.2.1 Study types
This meta-analysis only included fully published RCTs. Other types of studies (such as animal studies, case reports, retrospective studies, comments) were excluded.
2.2.2 Types of patients
Cancer patients aged ≥18 years old, with histological confirmed of EC refractory or intolerant to previous chemotherapy.
2.2.3 Types of interventions
Treatment intervention: PD-1 inhibitor was given for the management of EC.
Control intervention: chemotherapy.
2.2.4 Outcomes of interest
The primary outcomes included overall survival(OS),progression-free survival(PFS) ,objective response rate and completed response rate. The secondary outcomes included grade 3 or more adverse events rate.
2.3 Searching methods for Eligibility studies
The following databases were included for searching: PubMed (since its inception), Cochrane library (since its inception), EMBASE (since its inception), and ClinicalTrials.gov (since its inception). The lasted search was December 1st,2020. The language was restricted to English only. The main search terms included “PD-1”,” Immune checkpoint inhibitors”,and “esphagus cancer” . Furthermore, references of all relevant identified studies were examined for yielding more RCTs.
2.4 Selection of study
The studies from initial search were import into the Note express software. By using this software, duplicated references were excluded. The remaining studies were required a title and abstracts screen by two independent researches (G.L and Z.Z) in accordance to the inclusion criteria. The potential meet inclusion criteria studies retrieved as a full text form will be screen and evaluated by two independent reviewers for the final inclusion. Any disagreement was discussed and resolved by the author group. References of the studies identified for including analysis was evaluated for the potential missing studies.
2.5 Data extraction
Data from the inclusion studies were extracted using a pre-piloted and standardized form. The extracted information included the following: first author of the study , country location, years of publication, samples sizes of each interventions, details of intervention methods, details of patients in each intervention (including ages, sex, body mass index, cancer types, dose and duration of radiation), and outcomes of interest Adverse events were defined as any untoward medical occurrence in a patient or clinical investigation subject administered a pharmaceutical product and which does not necessarily have a causal relationship with this treatment. Data extraction was carried out by two independent reviewers. Any disputes were resolved by discussion by the reviewer group.
2.6 Assessment risk of bias for the inclusion RCTs
Risk of bias were appraised using the Cochrane tools[7] by pair reviewers.
2.7 Measurement of treatment effect
For dichotomous parameters (e.g. response rate and adverse events), the events in each arm was recorded. The analysis was using odds ratio (OR) and its 95% confidence interval (CI). Hazard ratio(HR) and its 95% CI were applied for measuring survival outcomes.
2.8 Publication bias assessment
Publication bias was evaluated using funnel plot of the outcomes of interest. In addition, the Egger’s test and Begg’s test [8]were also used for detection of publication bias.
2.9 Data synthesis
STATA 12.0 and RevMan 5.3 software will be utilized for the data analysis. As we list above, Chi2 test will be adopt for measuring the heterogeneity. The I2 values [7] of 25%, 50%, and 75% were considered low, moderate, and high, respectively. Continuous variables were measured using weighted mean difference (MD) with a 95% confidence interval (CI). Dichotomous variables, like the incidence of complications, were analyzed using odds ratios (ORs) with a 95% CI. If I2<25%, a fixed-effects models will be adopted for evaluation. Otherwise, a random-effects model was utilized. Subgroup analysis was performed to validate the reliable of the results and to find out the potential source of heterogeneity. Sensitivity analysis was carried out without trials of great risk of bias.