A systematic review of studies that evaluated the prognostic value of RDW in EC patients was performed. Four databases were electronically searched: Medline (host: OVID) from 1946 to April 2017; EMBASE (host: OVID) from 1974 to April 2017; and Web of Science and Cochrane Database of Systematic Reviews from 2005 to June 2017. The search terms used in our study were “RDW”, “red cell distribution width with esophageal neoplasm”, “esophagus neoplasm”, “esophagus neoplasms”, “cancer of esophagus”, “cancer of the esophagus”, “esophagus cancer”, “esophagus cancers”, “esophageal cancer”, “esophageal cancers”, “esophageal squamous cell cancer” (ESCC), and “esophageal adenocarcinoma”. Both free text and MeSH terms were used as keywords. The search strategy used for the PubMed database is shown in Table 1, and the presented search strategy was also used for the other electronic databases.

The search was performed by two investigators (Xu and Wang) who also evaluated the titles and abstracts of all candidate articles. Full-text was reviewed when the articles could not be categorized based on the title and abstract. Articles were included and excluded in accordance with the corresponding criteria defined in this study. Any disputes during the selection period were discussed with and resolved by a third investigator (Xiong). A flowchart demonstrating the details of the study selection according to the PRISMA guidelines is shown in Figure 1.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Methodological flow diagram of the meta-analysis.

Studies that met the following criteria were considered eligible: (1) A study of EC patients who underwent radical esophagectomy; (2) a study of patients with localized disease without distant metastasis; (3) a study of patients without preoperative neoadjuvant therapy; (4) a study of patients without previous antiinflammatory therapies and with available preoperative laboratory outcomes; (5) a study of the association between the preoperative RDW and OS/disease-free survival (DFS)/cancer-specific survival (CSS); and (6) a complete paper published in English. Studies that met the following criteria were excluded: (1) Letters, case reports, reviews or preclinical studies; (2) studies describing a repeated analysis or duplicate data; (3) studies lacking key information for further analysis; and (4) nonhuman studies.

We used predesigned extraction forms for data collection. The following information was extracted from each study: first author’s name, year of publication, country of the patients, research type, number of male/female patients included in the study, pathological types, RDW cut-off value, hazard ratio (HR) of elevated RDW for OS, CSS and DFS with 95% confidence interval (CI) and P-value. Assuming that most of the deaths were related to cancer, in the case of unavailability of OS information, data for CSS were extracted. HRs from multivariable analyses were extracted when available; otherwise, HRs from univariable analyses were extracted or estimated from Kaplan-Meier survival curves as described by Parmar and colleagues[27]. HRs for subgroups were compared defined by different markers.

HRs and the corresponding 95%CIs were directly obtained from each publication. If the values were not directly reported, the values were calculated according to the method described by Parmar and colleagues[27]. The meta-analysis was performed with STATA software version 12.0 (STATA Corporation, College Station, TX, United States) to combine the HRs with 95%CIs for categorical data and the weighted mean difference or standardized mean difference with 95%CIs for continuous data. All statistical tests were bilateral, and a P value < 0.05 was considered as statistical significance. If the data were not suitable for pooling, a systematic narrative synthesis of the information was performed, which was presented in the text to understand and to summarize the findings as well as characteristics of the included studies.

The heterogeneity of the pooled results was assessed through Cochran’s Q test and Higgins I-squared statistic. Significant heterogeneity was identified by P < 0.05 and/or I² > 50%, and the random-effects model (DerSimonian-Laird method) was used to combine the data. Otherwise, the fixed-effects model (Mantel-Haenszel method) was employed. To explore the potential source of heterogeneity among studies, subgroup analyses were performed according to various variables, such as the RDW cut-off value, the patient number in each study, and the study type and quality.

The quality of the included studies was assessed by using the Newcastle-Ottawa quality scale (NOS)[28]. Three aspects, namely, selection, comparability and outcomes, were assessed on this scale, which had a maximum score of 9. Studies with scores ≥ 7 were considered to be of high quality.

If significant heterogeneity was observed, a sensitivity analysis was performed after data extraction and subgroup analyses. This sensitivity analysis, which included the sequential omission of each study using the “metaninf” STATA command, aimed to validate the findings of this meta-analysis.

Begg’s funnel plot and Egger’s linear regression test were performed to evaluate publication bias, and a P value of < 0.05 was considered statistically significant.

Initially, 17 studies were selected from the electronic databases, and 10 studies remained after the removal of duplicates. After reading the titles and/or abstracts, four unrelated studies were excluded, and six full-text articles[21-26] were further assessed. None of these studies were excluded after thorough review of full-text. These six studies[21-26], which included 3826 patients, were included in this meta-analysis. The detailed search method and a flowchart representing the selection process are shown in Figure 1. These studies included five retrospective studies and one prospective study. The sample sizes varied from 144 to 2396, with a median value of 638. All six studies were conducted in Asian countries. The cut-off values for the RDW ranged from 12.2% to 15.3%. All six studies reported a correlation between RDW and OS/CSS, and two of the studies also investigated the association between RDW and DFS. With the exception of one study, the NOS scores of all the other studies were > 5. The general characteristics of the six included studies are summarized in Table 2.

The HRs and 95%CIs from the six studies involving 3826 patients were extracted and then pooled. The pooled results showed that the RDW was not associated with OS or CSS (HR = 1.27, 95%CI: 0.97-1.57, P = 0.000; Figure 2), with significant heterogeneity among the six studies (I² = 53.6%, P = 0.056; Figure 2); thus, the random-effects model was adopted for further analyses. The correlation between the RDW and DFS in EC patients were further investigated based on the pooled HRs and 95% CIs from two studies comprising 647 patients. As a result, RDW was not associated with DFS (HR = 1.42, 95%CI: 0.96-1.88, P = 0.000; Supplementary Figure 1), and no heterogeneity was observed (I² = 0.0%, P = 0.423; Supplementary Figure 1). In consideration of the significant heterogeneity of the pooled results regarding the effect of the RDW on OS/CSS, subgroup analyses, sensitivity analyses and Begg’s funnel plot and Egger’s linear regression analyses were conducted to further identify the heterogeneity source.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Forest plots of studies evaluating HR with 95%CI of red cell distribution width for overall survival in esophageal cancer patients. CI: Confidence interval; HR: Hazard ratio.

Sensitivity analyses was carried out by sequentially omitting each study to investigate its influence on results, indicating that the study conducted by Hu et al[26] was the primary source of heterogeneity (Figure 3). After exclusion of this study, the heterogeneity was effectively reduced or eliminated (I² = 0.0%, P = 0.926; Supplementary Figure 2). Surprisingly, the corresponding pooled HR varied with the inclusion (HR = 1.27, 95%CI: 0.97-1.57, P = 0.000; Figure 2) and omission of this study (HR = 1.42, 95%CI: 1.16-1.69, P = 0.000; Supplementary Figure 2). After reviewing the six studies included in our meta-analysis, we found that the study conducted by Hu et al[26] was the only prospective study, whereas the other five studies were retrospective ones. The sensitivity analyses indicated that the study type might be a source of heterogeneity. Therefore, we performed a subgroup analysis based on the study type (Supplementary Figure 3) and found that a high RDW was significantly associated with poor OS/CSS in patients with EC in the subgroup of retrospective studies.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Effect of individual studies on the pooled HR for red cell distribution width and overall survival of esophageal cancer patients. HR: Hazard ratio; RDW: Red cell distribution width.

To further explore other sources of heterogeneity, we performed a subgroup analysis based on the RDW cut-off values (≤ 13% or > 13%). For RDW cut-off value > 13%, heterogeneity was effectively reduced or eliminated after exclusion of the study in which the cut-off value was ≤ 13%, (I² = 0.0%, P = 0.850; Figure 4), and the corresponding pooled HR was increased (HR = 1.45, 95%CI: 1.13-1.76, P = 0.000; Figure 4). This finding indicated that the RDW cut-off value might be another source of heterogeneity. Thus, at an RDW cut-off value > 13%, a high RDW is significantly associated with poor OS/CSS in patients with EC.

Open in New Tab   Full Size Figure   Download Figure

Figure 4 Forest plots of RDW > 13% vs RDW ≤ 13% evaluating HR with 95%CI of red cell distribution width for overall survival in esophageal cancer patients. CI: Confidence interval; HR: Hazard ratio; RDW: Red cell distribution width. 0: RDW > 13%; 1: RDW ≤ 13%.

Furthermore, when the subgroups were stratified by patient number (≤ 400 or > 400), the heterogeneity was effectively reduced or eliminated after excluding the studies with > 400 patients (I² = 0.0%, P = 0.850; Supplementary Figure 4), and the corresponding pooled HR was increased when the patient number was ≤ 400 (HR = 1.45, 95%CI: 1.13-1.76, P = 0.000; Supplementary Figure 4). Hence, the subgroup analyses indicated that the patient number might also be a source of heterogeneity. For studies with patient number ≤ 400, a high RDW was significantly associated with poor OS/CSS in patients with EC.

Finally, when the subgroups were stratified by NOS scores (≤ 6 or > 6), heterogeneity was effectively reduced or eliminated after omitting the studies with NOS scores > 6 (I² = 0.0%, P = 0.876; Supplementary Figure 5), and the corresponding pooled HR was increased (HR = 1.42, 95%CI: 1.09-1.74, P = 0.000; Supplementary Figure 5). Thus, the study quality might be another source of heterogeneity.

Begg’s funnel plot and Egger’s linear regression analyses were performed to estimate the potential publication bias in the present meta-analysis. The P values regarding OS/CSS were 0.133 (Begg’s test; Supplementary Figure 6) and 0.005 (Egger’s test; Supplementary Figure 7). Due to the small sample sizes of the included studies, there was significant publication bias in our study, which was also demonstrated by the funnel plot (Funnel plot; Supplementary Figure 8). Therefore, publication bias might be another source of heterogeneity.

Esophageal cancer (EC) was the eighth most common cancer globally, with about half of all cases occurring in China. Prominent symptoms usually do not appear until the cancer has infiltrated over 60% of the circumference of the esophageal tube, by which time the tumor is already in an advanced stage and the prognosis generally tends to be fairly poor. Therefore, finding a simple and effective prognostic indicator is particularly urgent for individualized treatment of EC patients. Recently, red blood cell distribution width (RDW), as an important complete blood count parameter which has a close correlation with cancer-related inflammation, has been investigated as an important prognostic factor for EC patients in more and more studies, but the conclusions of these studies have not been consistent. Therefore, we conducted this meta-analysis to explore and verify the real role of RDW in the prognosis of patients with EC.

We systematically reviewed the existing studies regarding the role of RDW in the prognosis of EC patients and performed a meta-analysis with the extracted data to clarify the real impact of RDW on the outcomes of the EC patients. Identifying the real role of RDW in the prognosis of patients with EC and the defects existing in the previous and current studies can guide the future researchers to conduct more well-designed related studies on this topic and the upstream or downstream research related to RDW.

The main objectives of this article were to perform a meta-analysis of the data provided in these studies with inconsistent conclusions about the prognostic effect of RDW on EC patients, and to verify the real impact of RDW on the prognosis of EC patients by increasing the sample size. In the end, we could determine whether we need to conduct further studies on this topic according to the conclusion of this systematic review and meta-analysis.

First, we searched four related electronic databases (PubMed, EMBASE, Web of Science and Cochrane Library) using the identified MESH terms, and finally identified six studies which met the standards based on the inclusion and exclusion criteria of the selected literature, then we assessed the quality of the included studies according to Newcastle-Ottawa quality scale. Second, we used the electronic EXCEL table to collect the data from the included studies that we needed and utilized statistical software (STATA version 12.0) to conduct statistical analysis of the related data. Third, we performed the sensitivity analysis, subgroup analysis, Begg’s funnel plot and Egger’s linear regression test to explore the potential source of heterogeneity among studies, to find the influencing factors that affect the role of RDW in the prognosis of EC patients and point out the directions for further related research in the future.

Different from the traditional review, we used meta-analysis methods to synthesize data and perform statistical analysis to the relevant literature and quantify the effect of RDW on the prognosis of EC patients. Moreover, in addition to sensitivity analysis and subgroup analyses to find sources of heterogeneity, we also used the Begg’s funnel plot and Egger’s linear regression test to quantify publication bias rather than just using the traditional funnel plot for qualitative analysis. These were the characteristics and indicate the novelty of the research methods used in our study.

This systematic review and meta-analysis indicated that elevated RDW was not an independent risk factor for the worse outcome of EC patients overall, whether it’s for overall survival/cancer-specific survival [hazard ratio (HR) = 1.27, 95% confidence interval (CI): 0.97-1.57, P = 0.000] or disease-free survival (HR = 1.42, 95% CI: 0.96-1.88, P = 0.000). The prognostic value of RDW in patients with EC is only reflected in the retrospective study (HR = 1.42, 95%CI: 1.16-1.69, P = 0.000) of small samples (sample size ≤ 400, HR = 1.45, 95% CI = 1.13-1.76, P = 0.000) currently, and there is a need to choose the appropriate RDW cutoff value (RDW > 13%, HR = 1.45, 95%CI: 1.13-1.76, P = 0.000) as a prerequisite. Therefore, the actual effect of RDW on the prognosis of EC patients needs further prospective multicenter large-sample studies to be validated in the future.

Different from the traditional viewpoints, our systematic and meta-analysis demonstrated that RDW had no correlation with the prognosis of EC patients, no matter favorable or unfavorable. Therefore, such traditional theories and assumptions, that cancer-related inflammation leads to an increased RDW in the blood, and elevated RDW in turn suggests the occurrence of cancer, were challenged and questioned by the results of our meta-analysis. At the same time, it also suggests that we could perform the meta-analysis to statistically analyze the inconsistent result data of different types of small-sample studies and achieve a conclusion that is completely different from our previous understanding. This leads to the emergence of new theories and assumptions and provides direction for our future research design and potential mechanism research. Our systematic reviews and meta-analysis suggest that we should be more cautious and rational to see the impact of increased RDW on the prognosis of EC patients in our future clinical work.

From our study, we could learn that we can’t blindly believe in traditional ideas that already exist. When the opinions of previous studies are inconsistent and chaotic, we should use statistical methods to perform statistical clustering analysis on various data, and draw a scientific conclusion to guide our clinical work and indicate the future research direction. Moreover, through the systematic analysis of the previous research, we should carry out more multicenter, large-sample prospective studies in the future to overcome the defects of the current research in the study design to further verify the role of RDW in the prognosis of EC patients. In addition, we also need to conduct further basic experiments based on the results of such above-mentioned optimized research to uncover its underlying mechanisms.