Prognostic significance of CXCR7 in cancer patients: a meta-analysis

This meta-analysis was performed according to the preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement [27]. For this study, we retrieved publications before 11 June 2018 from Embase, Web of Science, and PubMed using the following search terms: “CXCR7 or CXC chemokine receptor 7 or RDC1 or ACKR3 or GPRN1” AND “neoplasm or tumor or cancer or malignancy or carcinoma” AND “survival or prognosis or outcome or prognostic”. All relevant publications in reference lists were also searched manually to identify potentially qualified papers.

In the present meta-analysis, the eligible studies or cohorts must have meet the following criteria: (1) The expression level of CXCR7 was detected in human tissues or plasma samples; (2) Tumors were diagnosed accurately by histopathology; (3) The relationship between CXCR7 expression level and survival rates of patients was evaluated; (4) Studies provided sufficient information to calculate the hazard ratio (HR) for survival rates, with 95% confidence intervals (CI). Studies were excluded if they met any of the following criteria: (1) duplicate publications; (2) conference abstracts, case reports, reviews, patents, letters, non-English or unpublished articles; (3) studies merely concerned with cancer cell lines or animal models; (4) HRs and 95% CIs could not be extracted or calculated due to insufficient data.

Two researchers independently extracted all necessary data and reached an agreement on all contents. The third author made any decisions about confusing items. The extracted information in each study included: the first author’s name, the year of publication, region of the population enrolled, tumor type, sample size (high/low), follow-up time, type of sample detected, the endpoints, high or low expression accounting for poor prognosis, cut-off value, methods of obtaining HRs (directly or indirectly), survival analysis method and Newcastle–Ottawa Scale (NOS) score. Overall survival (OS), disease-free survival (DFS), progression-free survival (PFS) and recurrence-free survival (RFS) were considered to be endpoints. HRs that were directly obtained or calculated from Kaplan–Meier curves served as parameters to evaluate the relationship between CXCR7 expression level and prognosis of cancer patients [28].

The NOS score ranging from 0 to 9 was used to assess the quality of the cohort studies. This system included the following three categories to evaluate each study: selection of study groups, comparability of groups and ascertainment of outcomes. A NOS scores ≥ 7 indicated high quality, and a NOS scores < 7 indicated low quality.

Pooled HRs (high/low) and their associated 95% CIs which were calculated using Stata version 14.0 (Stata Corporation, College Station, TX, USA) demonstrated the relationship between CXCR7 expression level and prognosis of tumor patients. The heterogeneity among studies was estimated by the Cochrane’s Q test and the Higgins I² statistic (p < 0.10 or I² > 50% was considered significant) [29]. When heterogeneity was not significant (p > 0.10 and I² < 50%), the fixed-effect model was used for analysis. Otherwise, the random-effect model seemed to be more appropriate. Furthermore, meta-regression and subgroup analyses were conducted to identify potential sources of heterogeneity, and the included cohorts were divided into two subgroups based on similar characteristics. Sensitivity analysis was performed by removing each cohort sequentially to explore possible explanations for heterogeneity. Publication bias was evaluated using Begg’s test and Egger’s test [30]. Due to the publication bias in this meta-analysis, we also conducted trim and fill analysis.

As shown in Fig. 1, a total of 1318 articles were obtained from Embase, PubMed and Web of Science utilizing the search strategy described above. First, 373 duplicate reports were removed. Subsequently, 134 meeting abstracts, 16 patents, 7 non-English articles, 102 reviews, 439 studies focusing on non-CXCR7 topics and 51 studies about noncancer topics were excluded by skimming titles and abstracts. Furthermore, we ruled out 93 basic-research studies and 75 studies lacking relevant data through close reading. Ultimately, 28 studies were included and the relevant data were extracted. Detailed information of these qualified articles is presented in Table 1. The total number of studies was 28 comprising 33 cohorts and 5685 patients in this current meta-analysis. In summary, the sample size of all eligible studies was between 30 and 840 and the follow-up time ranged from 26 to 266 months. Among all the cohorts, China (n = 13) was the most common region of studies, followed by Italy (n = 6), USA (n = 4), Japan (n = 3) Germany (n = 3), UK (n = 2), Singapore (n = 1), and Netherlands (n = 1). In terms of disease outcomes, 24 cohorts reported OS, 4 cohorts reported DFS, 8 cohorts reported RFS and 3 cohorts reported PFS. To assess the expression level of CXCR7, most studies used immunochemistry (IHC). Other methods such as RT-PCR, mRNA microarray, and cDNA-array were also applied. The types of cancer evaluated in the current meta-analysis were glioma (n = 3) [31‐33], thyroid carcinoma (n = 1) [34], esophageal cancer (n = 3) [35‐37], breast cancer (n = 4) [38‐41], hepatocellular carcinoma (n = 2) [42, 43], lung cancer (n = 2) [44, 45], pancreatic cancer (n = 2) [23, 46], renal cancer (n = 2) [10, 47], oral carcinoma (n = 1) [48], chondrosarcoma (n = 1) [49], gastric cancer (n = 1) [50], colorectal carcinoma (n = 3) [24, 51, 52], gallbladder cancer (n = 1) [53], extramammary Paget disease (n = 1) [26] and cervical cancer (n = 1) [54].

In total, 24 cohorts from 20 studies assessed the relationship between CXCR7 expression level and OS of cancer patients in 3182 participants. The random-effect model was used because of significant heterogeneity (I² = 41.0%; p = 0.02). The pooled HR for OS of patients with high CXCR7 level compared with low expression was 1.72 (95% CI 1.49–1.99, p < 0.001), indicating that high CXCR7 level was markedly related to reduced OS of cancer patients (Fig. 2). Moreover, to explore the sources of heterogeneity, subgroup analyses according to the type of cancer (digestive system or nondigestive system carcinoma), sample size (fewer than 100 or more than 100), follow-up time (fewer than 100 or more than 100 months), region (Asia or elsewhere), methods of obtaining HRs (directly or indirectly) and paper quality (NOS scores ≥ 7 or < 7) were conducted (Fig. 3a–f). All results of subgroup analyses demonstrated significant relationship between CXCR7 overexpression and poor OS of tumor patients. We also applied meta-regression by the covariates including all mentioned factors to explore the sources of heterogeneity. However, all the above factors did not account for the sources of heterogeneity since the p values of all above factors were not less than 0.05 (Table 2). Additionally, we conducted Cox multivariate analysis from 8 studies including 9 cohorts to obtain HR. The result indicated that elevated CXCR7 expression could independently predict OS for the prognosis of tumor patients (HR 1.49, 95% CI 1.24–1.79, p < 0.001).

Subsequently, we explored the relationship between CXCR7 expression level and OS in certain types of tumor. Significant associations were detected between higher CXCR7 expression level and worse OS of glioma (HR 1.77; 95% CI 1.43–2.19, p < 0.001) (Fig. 4a), breast cancer (HR 1.45; 95% CI 1.28–1.63, p < 0.001) (Fig. 4b), esophageal cancer (HR 2.72; 95% CI 1.11–6.66, p = 0.029) (Fig. 4c), and pancreatic cancer (HR 1.46; 95% CI 1.12–1.90, p = 0.006) (Fig. 4d). However, for lung cancer (HR 2.40; 95% CI 0.34–17.07, p = 0.383) (Fig. 4e) and hepatocellular cancer (HR 1.37; 95% CI 0.84–2.24, p = 0.209) (Fig. 4f), no significant relationship was found between CXCR7 expression level and OS of tumor patients.

As shown in Fig. 5a, three cohorts were included in the meta-analysis of PFS. The results illuminated an obvious association between higher CXCR7 expression level and shorter PFS (HR 2.83, 95% CI 1.66–4.85, p < 0.001). Similarly, the analysis of RFS, which covered eight cohorts, showed that the tumor patients with elevated CXCR7 expression level had significantly worse RFS compared to those with lower CXCR7 (HR 1.58, 95% CI 1.34–1.88, p < 0.001) (Fig. 5b). Furthermore, an obvious association between higher CXCR7 expression level and shorter DFS was revealed by assessing four cohorts (HR 5.58; 95% CI 3.16–9.85, p < 0.001) (Fig. 5c). Considering the limited quantity of included cohorts, we did not conduct a subgroup analysis.

We conducted sensitivity analysis to determine the influences of individual studies and the stability of our results by removing single study sequentially. For OS, the sensitivity analysis showed that the three cohorts from Wu et al., Li et al. and Werner et al. influenced the results greatly, which showed that these cohorts might account for heterogeneity. All the values of HR in the list were greater than 1, indicating that our results were stable and reliable (Fig. 6a). For RFS (Fig. 6b), the sensitivity analysis revealed that the cohort from Ribas et al. [40] affected the results greatly. In addition, all included studies had a great influence on DFS (Fig. 6c) and PFS (Fig. 6d), which indicated that the results for DFS and PFS were relatively unstable.

We conducted Begg’s funnel plot and Egger’s test to evaluate the publication bias. The funnel plot was obviously asymmetric. Moreover, significant publication bias (p = 0.003, p = 0.035 respectively) was identified by Egger’s linear regression test (Fig. 7a) and Begg’s test (Fig. 7b). Furthermore, we conducted the trim and fill analysis (Fig. 7c). Seven studies that focused on the value of CXCR7 expression level for predicting the OS of tumor patients remained unpublished. It is worth noting that the filled results for OS (HR = 1.54, 95% CI 1.31–1.81, p < 0.001) were in accordance with our original results.