The Optimal Time of Applying Enteral Immunonutrition in Esophageal Cancer Patients Receiving Esophagectomy: A Network Meta-Analysis of Randomized Clinical Trials

Background: Enteral immunonutrition (EIN) has been extensively applied in cancer patients, however its role in esophageal cancer (EC) patients receiving esophagectomy remains unclear. We performed this network meta-analysis to investigate the impact of EIN on patients undergoing surgery for EC and further determine the optimal time of applying EIN. Methods: We searched PubMed, EMBASE, Cochrane library, and China National Knowledgement Infrastructure (CNKI) to identify eligible studies. Categorical data was expressed as the odds ratio with 95% condence interval (CI), and continuous data was expressed as mean difference (MD) with 95% CI. Pair-wise and network meta-analysis was performed to evaluate the impact of EIN on clinical outcomes using RevMan 5.3 and ADDIS V.1.16.8 softwares. The surface under the cumulative ranking curve (SUCRA) was calculated to rank all nutritional regimes. Results: Total 14 studies involving 1071 patients were included. Pair-wise meta-analysis indicated no difference between EIN regardless of the application time and standard EN (SEN), however subgroup analyses found that postoperative EIN was associated with decreased incidence of total infectious complications (OR=0.47; 95%CI=0.26 to 0.84; p=0.01) and pneumonia (OR=0.47; 95%CI=0.25 to 0.90; p=0.02) and shortened LOH (MD=-1.01; 95%CI=-1.44 to -0.57; p<0.001) compared to SEN, which were all supported by network meta-analyses. Ranking probability analysis further indicated that postoperative EIN has the highest probability of being the optimal option in terms of these three outcomes. Conclusions: Postoperative EIN should be preferentially utilized in EC patients undergoing esophagectomy because it has optimal potential of decreasing the risk of total infectious complications and pneumonia and shortening LOH.


Sources of identi cation
A systematic search was conducted by two independent reviewers in PubMed, EMBASE, Cochrane Library, and China National Knowledgement Infrastructure (CNKI) in order to identify potentially eligible studies from their inception util to December 30, 2020. Medical subject heading (MeSH) and text words were simultaneously used to develop the search strategy according to the speci ed criteria of each database. We summarized search strategies of all databases in Table S1. Additionally, we also manually checked the references of all included studies and two topic-related meta-analyses to identify any eligible studies which were missed at the electronical search stage. Moreover, we updated our search weekly, and the latest update was performed on January 23,2021. Any divergence about identi cation of sources was resolved based on the consensus principle.

Selection of studies
Two independent reviewers conducted the selection of studies according to the following developed criteria: (a) adult patients undergoing esophagectomy for EC; (b) patients were instructed to intake EIN or standard EN; (c) study reported at least one of the following clinical outcomes including total infectious complications, pneumonia, wound infection, sepsis, urinary tract infection, anastomotic leakage, and length of hospitalization (LOH); (d) only randomized controlled trial was eligible for our inclusion criteria; (e) language was limited to English and Chinese; and (f) study reported in Chinese must be published in core journal. We excluded a study when it covered at least one of the following criteria: (a) experimental and animal studies; (b) studies without insu cient information; and (c) duplicate study with poor quality or insu cient data. Any divergence about the selection of studies was resolved based on the consensus principle.

Information extraction
We designed information extraction sheet in advance, and two independent reviewers were assigned to extract the following information with our sheet: (a) characteristics of eligible study including name of the rst author, country, and year of publication; (b) characteristics of statistical design including sample size and outcomes; (c) characteristics of participants including age and gender; (d) details of nutritional regimes; and (e) information of risk of bias.
Any divergence about data extraction was resolved based on the consensus principle.
In this network meta-analysis, we only considered clinical outcomes because other outcomes such as biochemical parameters and immune parameters are the surrogate variable for developing clinical decision. Therefore, we de ned total infectious complications, anastomotic leakage, and LOH as the primary outcomes.
Remaining outcomes including pneumonia, wound infection, sepsis, and urinary tract infection were de ned as the secondary outcomes. If an outcome was reported as median and range or interquartile range, we estimated the mean and standard difference (SD) using the method proposed by Hozo and colleagues after extracting data (24).

Assessment of risk of bias
The risk of bias of individual study was assessed by two independent reviewers with the Cochrane Risk of Bias assessment tool(25) from the following six domains: random sequence generation; allocation concealment; blinding of participants and personnel; blinding of outcome assessment; incomplete outcome data; selective reporting; and other bias. A study was labeled with low, unclear, or high risk of bias according to the matching degree between actual information and assessment criteria. Any divergence about the assessment of risk of bias was resolved based on the consensus principle.

Statistical analysis
For traditional pair-wise meta-analysis, we used Review Manager 5.3 (Cochrane Collaboration, Copenhagen, Denmark) to conduct all statistical analyses (26). In our study, only LOH was continuous data, and it therefore was expressed as the mean difference (MD) with 95% con dence interval (CI). Remaining outcomes were categorical data, and all were expressed as odds ratio (OR) with 95% CI. We rstly qualitatively evaluated the heterogeneity across studies with Cochrane Q test (27), and then quantitatively estimated the level of heterogeneity with I2 statistic (28). We adopted random-effects model to perform meta-analysis because variations across studies in the real world can not be ignored. We designed subgroup analysis basing on the time of applying EIN in order to speci cally investigate the pure effectiveness of each EIN regime compared to standard EN. Moreover, we draw funnel plots of primary outcomes to qualitatively inspect the possibility of existence of publication bias when accumulated number of eligible studies was more than 10 (29).
In order to determine the optimal time of applying EIN, we further conducted a Bayesian network analysis with the Aggregate Data Drug Information System (ADDIS V.1.16.8, Drugis, Groningen, NL), which was developed based on Markov Chain Monte Carlo (MCMC) method (30,31). The following parameters were set for achieving good convergence: 4 chains for simulation, 50,000 simulation iterations, 2.5 variance scaling factor, 10 thinning interval, and 20000 tuning iterations. With ADDIS software, Brooks-Gelman-Rubin method was automatically performed to evaluate the convergence degree through calculating the potential scale reduction factor (PSRF) (32), and a PSRF indicated a better convergence when it was greatly closed to 1 (32). We rstly estimated the effect size based on consistency and inconsistency assumptions respectively when two types of comparisons were simultaneously available (31), and then we used node-splitting method to test whether effect sizes from two assumptions were inconsistent (33). The assumption of consistency between direct and indirect comparisons was right when a P of more than 0.05 was generated (33). For network meta-analysis, LOH was expressed as MD with 95% credible interval (CrI), and categorical data was expressed as OR with 95% CrI. Finally, we estimated the surface under the cumulative ranking curve values in order to determine the possibility of considering a regime as the best option. Based on the results calculated from ADDIS software, a regime was considered to be worse if the corresponding possibility of ranking was greatly closed to 100% (34). We utilized league gures to document the results of network meta-analysis. Moreover, we also used STATA software 14.0 (Stata Corp LP, College Station, Texas, USA) to create evidence structure of primary outcomes (35).

Identi cation and selection of study
We identi ed 714 records after initially searching PubMed, EMBASE, Cochrane library, and CNKI databases.
After removing 71 duplicate records, we continue to remove 612 ineligible records through checking the title and abstract. Then, 15 studies were excluded after checking eligibility based on full-text due to following reasons: lack of outcome (n = 1), ineligible nutritional regimes (n = 6), Chinese study with poor quality (n = 1), ineligible study design (n = 1), and abstract (n = 1). Moreover, we also identi ed additional two studies from meta-analysis published in China. Finally, 14 studies(13, 14, 36-47) met our inclusion criteria. The process of identi cation and selection of studies was delineated in Fig. 1.

Characteristics of eligible studies
Most of the 14 eligible studies were conducted in Japan (13,14,36,39,41,42,44) and China (38,(45)(46)(47). The sample size of individual study was between 29 and 191, with a total participants of 1071. All studies were two-arm design except for one study (44), which was designed as three-arm. Details of characteristics of included studies were summarized in Table 1.
Inconsistency test based on node-splitting analysis was not performed. Ranking probability suggested that postoperative EIN has the highest probability of being optimal option (0.70), followed by perioperative EIN (0.39), SEN (0.52) and preoperative EIN (0.76) ( Figure S14 in the appendix).

Publication bias
Among three primary outcomes, the accumulated number of two outcomes including total infectious complications ( Figure S15 in the appendix) and anastomotic leakage ( Figure S16) was more than 10, and symmetric funnel plots indicated absence of publication bias.

Discussion
As one of the most common gastrointestinal malignant tumors(1), EC has been estimated to have 0.60 million new cases and 0.54 million cancer-related deaths in 2020(3). To date, esophagectomy remains the preferred therapeutic option for the treatment of EC patients. However, patients will experience a series of serious complications after surgery such as infection and anastomotic leakage due to immunosuppression and in ammatory response (8,10). Therefore, it is critically important to supply immune-modulating substances such as arginine and omega-3-fatty acids to SEN, which was de ned as EIN (7,17). Although the effectiveness of EIN in patients receiving gastrointestinal surgery has been established (8)(9)(10), the role of EIN in EC patients undergoing esophagectomy remains controversial. We therefore performed this network metaanalysis to determine the effectiveness of EIN for improving clinical outcomes among patients undergoing surgery for EC. Our ndings suggested that postoperative EIN decrease the incidence of total infectious complications and pneumonia and shorten the LOH after surgery compared to SEN. Meanwhile, ranking probability suggested that postoperative EIN has the highest probability of being optimal nutritional prescription.
To date, two meta-analyses (17,18) investigating the effectiveness of EIN for treating EC patients receiving esophagectomy have been published. Li and colleagues performed a meta-analysis of six articles to compare the effectiveness between EIN and SEN in patients receiving oesophagectomy, and found that impact of EIN on immunological status, biological status or clinical outcomes remains unclear (17). However, this nding must be cautiously interpreted because this meta-analysis missed four eligible studies (37,39,43,44). In the same year, Wang and colleagues also reported a meta-analysis of investigating the comparative effectiveness between perioperative EIN and SEN, suggesting no statistical signi cance (18). It must be noted that this meta-analysis correctly included a study which focused on preoperative EIN into analysis (14) and also missed an eligible study (44). Compared to previous meta-analyses, the present network meta-analysis has several strengths: (a) we incorporated Chinese studies with high quality to increase the statistical power; (b) we designed subgroup analysis for speci cally investigating the comparative effectiveness of various EIN regimes, which were categorized according to the initiation time and duration, and SEN(10); and (c) we used network meta-analysis method to determined the ranking of various EIN regimes. Based on these strengths stated above, the present network meta-analysis generated more reliable, accurate and detailed results, which can provide golden references for clinical decision.
We must acknowledge several limitations in our network meta-analysis. First, regardless of the fact that more eligible studies were included in this network meta-analysis, the sample size of individual study was small, which may have negative impact on the robustness of pooled results. Second, most eligible studies investigated the comparative effectiveness between postoperative or perioperative EIN and SEN, however only one study compared preoperative EIN with SEN, and thus more future studies investigating the direct comparison between preoperative EIN and SEN must be performed in order to further establish our ndings. Third, formulas of EIN were different across eligible studies, however we did not design subgroup or sensitivity analysis to determine the comparative effectiveness of various formulas(11) due to insu cient number of eligible studies. Forth, duration of each individual regime such as perioperative EIN was also variation across studies, insu cient number of eligible studies failed to support further designing subgroup analysis.

Conclusions
At the basis of available best evidence, we concluded that postoperative EIN is the optimal nutritional regime for the treatment of patients undergoing surgery for EC because this regime was associated with decreased incidence of total infectious complications and pneumonia and shortened LOH.

Declarations
Ethics approval and consent to participate: Not applicable.
Consent for publication: Not applicable.
Availability of data and materials: The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.
Competing interests: No con ict of interest has been declared by the authors.