Literature search

Studies were identified via an electronic search of PubMed, EMBASE, and Chinese biomedicine databases using the following keywords: gastric cancer, gastric carcinoma, BIRC5, baculoviral inhibitor of apoptosis repeat-containing 5, and survivin. The search ended on December 21, 2012. The references for articles and reviews were manually searched for additional studies. We also hand-searched the journals that published articles most relevant to this review.

Inclusion and exclusion criteria

The systematic review generated complete databases from published studies dealing with the prognostic value of survivin in patients with gastric cancer. We placed no restrictions on language of publication. To be eligible for inclusion, studies had to meet the following criteria: (1) they measured survivin expression in gastric cancer with methods such as immunohistochemistry (IHC), reverse transcription-polymerase chain reaction (RT-PCR), or fluorescence in situ hybridization (FISH); (2) they compared overall survival between different expressions of survivin in gastric cancer; (3) they contained a hazard ratio (HR) and 95% confidence interval (CI) for overall survival according to survivin status which either were reported or could be computed from the data presented; (4) the prognostic effect was valued by mortality of the patients; (5) when the same author or group reported results obtained from the same patient population in more than one article, the most recent report or the most informative report was included; and (6) the study quality was evaluated >5 stars according to Newcastle-Ottawa quality assessment scale [16].

The exclusion criteria were as follows: (1) letters, reviews, case reports, conference abstracts, editorials, and expert opinion were excluded; (2) articles in which no information on overall survival was given or HR about overall survival from the given information could not be computed were excluded; and (3) articles in which prognostic effect was valued by recurrence of gastric cancer were excluded.

Data extraction

Two investigators (Liu J. L. and Gao W.) reviewed all of the research that met the inclusion and exclusion criteria. Data were extracted independently by two investigators (Liu J. L. and Gao W.) using a data extraction sheet. Data extracted concerned first author’s name, year of publication, source of patients, language of publication, number of patients, type of sample, assay method, location of expression, and survival data. In addition, controversial problems were resolved in a meeting called by Kang Q. M.

Assessment of study quality

Study quality was assessed independently by two investigators (Liu J. L. and Gao W.) by means of reading and evaluating according to the Newcastle-Ottawa quality assessment scale. Briefly, the overall star system assesses three main categories on the following: (1) selection of cohort, (2) comparability of cohort, and (3) ascertainment of outcome. A study can be awarded a maximum of one star for each numbered item within the selection and outcome categories. A maximum of two stars can be given for comparability. The total number of stars was accumulated last, with more stars reflecting higher methodological quality. A study can be awarded a maximum of nine stars.

Statistical analysis

The primary outcome revealed the overall survival in all populations; the outcome was then stratified by ethnicity of patients, assay method, language of publication, type of sample, HR estimate, and location of expression. HR and 95% CI were used to estimate the impact of survivin expression on overall survival. A combined HR>1 implied a worse survival for the group with survivin expression. This pejorative impact of survivin on survival was considered statistically significant if 95% CI for the combined HR did not overlap 1. If a direct report of HR and 95% CI was not available, estimated value was derived indirectly from Kaplan-Meier curves using the methods described by Tierney et al. [17]. Kaplan-Meier curves were read by Engauge Digitizer version 4.1 (http://digitizer.sourceforge.net/), then the survival data read from Kaplan-Meier curves were entered in the spreadsheet appended to Tierney’s paper [17]. This work was performed by two independent persons to reduce inaccuracy in the extracted survival rates. To assess heterogeneity among the studies, we used the Cochran Q and I² statistics. For the Q statistic, a P value <0.10 was considered statistically significant for heterogeneity [18]. Then the random effects model was calculated according to the DerSimonian-Laird method [19]. Otherwise, the fixed-effects model (Mantel-Haenszel method) was used. For I², a value >50% was considered a measure of severe heterogeneity [20], so the conclusion should be made with discretion, or a combination of HRs was given up. All statistical analyses were performed by Review Manager 5.0 (http://www.cochrane.org). A significant two-way P value for comparison was defined as P <0.05.

Literature Selection

A total of 730 potentially relevant citations were retrieved after the initial search of databases. Additionally, one study was found by hand-search of the journals that published the articles most relevant to this review [21]. The title and abstract of relevant articles were read by the two authors (Liu J. L. and Gao W.) independently. Six hundred twenty-three citations were excluded from analysis after the first screening based on abstracts or titles, leaving 108 available for further full-text review. After carefully reading the full-text articles, 88 studies were excluded; 63 studies were excluded because they were reviews or studies about the correlation with clinicopathological variables, not survival. Another 25 studies were excluded due to insufficient survival data. Last, three studies were excluded due to unexplainable data (data were contradictive in the studies) [22–24]. As a result, 17 eligible studies [21,25–40] were included in the qualitative synthesis, and a final meta-analysis of 1365 patients from 16 evaluable studies was performed (Figure 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flow diagram of study selection.

https://doi.org/10.1371/journal.pone.0071930.g001

Characteristics of the included studies

The basic feature descriptions of the 17 studies are summarized in Table 1. Briefly, study sample sizes ranged from 30 to 157; 14 studies were conducted in Asian populations, while the remaining 3 used European populations [21,25,40]. Fifteen studies investigated survivin by IHC, and two investigated survivin by RT-PCR [25,32]. All of the studies investigated survivin expression using gastric cancer tissues except one study that detected expression using peripheral blood [25]. Eight studies reported survivin as an indicator of poor prognosis, and seven showed no significant impact on overall survival. One study reported survivin as an indicator of good prognosis [40]. In addition, one study reported survivin nuclear staining as a favorable prognosis, while cytoplasmic staining showed no significant impact [33].

Methodological quality of the studies

For included studies, two authors independently extracted data and assessed methodological quality using the Newcastle-Ottawa quality assessment scale. The scores are shown in Table 1. Luckily, the studies included in our meta-analysis all have high levels of methodological quality (>5 stars on the Newcastle-Ottawa scale).

Assessment of heterogeneity

Highly significant heterogeneity was detected when all studies were pooled (chi-squared=45.32, I²=62%, p=0.0002). Then we easily found the source of heterogeneity from the forest plot (Figure 2): One study examined the association of survivin expression with prognosis in patients with gastric cancer, and all the patients had been treated with aneoadjuvant chemotherapy [40]. In this study, the author reported cytoplasmatic survivin expression as an indicator of good prognosis. We excluded this study from meta-analysis for two reasons: (1) this was the only study in which the patients were treated with chemotherapy before surgical resection and (2) the sample size was small (n=30) and the weight was slight (1.9%), so the impact on the combined HR would be negligible. After excluding it, a final analysis of 1365 patients from 16 evaluable studies was performed. Heterogeneity still existed, but it dropped to the level that random-effects meta-analyses could be conducted without great error (chi-squared=30.67, I²=48%, p=0.01).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Forest plot of Hazard ratio (HR) for survival of gastric cancer patients.

Highly significant heterogeneity can be found before Vallbohmer D’ study was excluded.

https://doi.org/10.1371/journal.pone.0071930.g002

Results of meta-analysis

The main results of this meta-analysis are presented in Table 2. Overall, the pooled HR for all evaluable studies on survivin expression in gastric cancer was 1.39 (95% CI: 1.16-1.68). This illustrated that survivin expression was significantly associated with worse overall survival of gastric cancer patients. When grouped according to ethnicity, the combined HR of Asian studies was 1.35 (95% CI: 1.09-1.67). The heterogeneity of the remaining two European studies was highly significant (chi-squared=5.08, I²=80%, p=0.02), so the pool of HR was given up to calculate. In the subgroup analysis according to the method of survivin detection used, the combined HR was 1.41 (95% CI: 1.11-1.78) for IHC and 1.34 (95%CI: 1.15-1.56) for RT-PCR. When stratified according to language of publication, the combined HR of English literature showed an inverse effect on survival (HR=1.40, 95% CI: 1.13-1.75), while the non-English literature did not (HR=1.38, 95% CI: 0.93-2.05). When stratified according to type of sample, the combined HR of gastric cancer tissue showed an inverse effect on survival (HR=1.41, 95% CI: 1.13-1.76). When the HRs directly extracted from the five evaluable studies were pooled, the combined HR was 1.29 (95% CI: 1.00-1.66). When the HRs calculated indirectly from Kaplan-Meier-based survival curves were pooled, the combined HR was 1.5 (95% CI: 1.14-1.97). The combined HR from the subgroup indicated that survivin was an independent prognostic marker in gastric cancer.

To further investigate the relationship between survivin subcelluar location and overall survival, eight studies, which reported survivin located in cytoplasm in 739 patients, were enrolled in the meta-analysis [28,29,31,34–36,38,39]. Due to the absence of heterogeneity (chi-squared=10.6, I²=34%, p=0.16), a fixed effect model was accepted. The combined HR was 1.46 (95% CI: 1.12-1.90), which demonstrated that positive expression of survivin in cytoplasm was significantly associated with poor prognosis of gastric cancer patients. Four studies reported that survivin expression was located in nucleus in 416 patients [26,27,33,34]. Because of heterogeneity (chi-squared=11.5, I²=74%, p=0.009), a random effect model was adopted. The combined HR was 1.29 (95% CI: 0.72-2.31), which illustrated that survivin expression in nucleus was not significantly associated with overall survival of gastric cancer patients. However, the heterogeneity was highly significant (I²>50%), so the result should be accepted with discretion.

Publication bias

Publication bias may exist when no significant findings remain unpublished, thus artificially inflating the apparent magnitude of an effect. Funnel plots of our study results are shown in Figure 3. In all included studies, no obvious funnel plot asymmetry was found. Thus, no evidence of publication bias was detected.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Funnel plots were used to detect publication bias on overall estimate.

Studies are distributed symmetrically, and suggest that publication bias is absence in the meta-analysis.

https://doi.org/10.1371/journal.pone.0071930.g003