The flow chart of study selection process was depicted in Figure 1. A total of 251 studies were identified from online databases, including PubMed, Cochrane Library, Web of Science and Embase. Based on screening of titles and abstracts, 54 studies were selected for further investigation according to following criteria: repetitive articles, not a human study, not original articles, no clinical parameters, unrelated to TSP-1, unrelated to malignant neoplasms and unrelated to prognosis or survival. Of these 54 studies, 30 were excluded due to insufficient survival data and overlapping data. Ultimately, 24 articles were included for further analysis.
The major characteristics of the eligible data were summarized in Table 1. We collected basic data on 24 articles published between 2000 and 2019. The meta-analysis included 2379 participants from different regions of the United States, France, Japan, China, India, Greece, the United Kingdom and Norway, including tumors such as breast cancer, liver cancer, ovarian cancer, esophagus cancer, lung cancer, gastric cancer, colon cancer, skin cancer, cervical cancer, oral cancer and bladder cancer.
The expression of TSP-1 was measured by Immunohistochemical staining (IHC) in the most of studies. Besides, Quantitative real-time polymerase chain reaction (qRT-PCR) assay and enzyme linked immunosorbent assay (ELISA) was applied to detect TSP-1 in 3 and 2 studies, respectively, and immunoblot analysis and a standard Dextran Polymer Conjugate Two-step Visualization System Envision was applied in 1 study each. The data of HR and 95% CI was extracted from survival curves or literature reports. In all these studies, 17 studies researched OS [17, 22-37], 7 studies investigated DFS/RFS [22, 23, 26, 38-41] and 6 studies focused on progression-free survival (PFS)/ metastasis-free survival (MFS) [28, 31, 32, 42-44] (Table 2).
OS associated with TSP-1 expression
Because of the mild heterogeneity (p=0.016, I2=47.3), the fixed effect model was used for data analysis. The results showed that high level of TSP-1 indicated poor OS, (HR=1.40; 95% CI: 1.17~1.68) and the effect was statistically significant (P＜0.001) (Figure 2A). In order to analyze the source of heterogeneity, we did subgroup analyses according to nationality, dominant ethnicity, main pathological type, disease type, assay method and source of HR. When stratified by ethnicity, we found that the high level of TSP-1 was significantly correlated with the OS of Caucasians (HR=1.74; 95%CI: 1.37-2.22; P＜0.001), while among Asians, there was no significant correlation (HR=1.07, 95%CI: 0.82-1.40; P=0.629) (Figure 3A). In the source of HR analysis, the OS of “reported” group was significantly worse(HR=1.48; 95%CI: 1.18~1.87; P=0.001), while the OS of the other group was also worse, however, with no statistical significance(HR=1.29; 95%CI: 0.97~1.171; P=0.081) (Figure 3B). According to the subgroup analysis of disease type, the pooled HR of breast cancer was 1.78(95%CI: 1.09~2.92; P=0.022) (I2=0.0%, P=0.536), and the pooled HR of gynecological cancer was 1.72(95%CI:1.13-2.64; P=0.012)(I2=0.0%, P=0.511), with no heterogeneity (Figure 3C). Finally, there was a significant relationship between elevated TSP-1 and poor OS in Americans. (HR=1.72; 95%CI: 1.13-2.64; P=0.012) (Figure 3E). Other kinds of diseases had no obvious significance (Figure 3D, 3F).
PFS/MFS and DFS/RFS associated with TSP-1 expression
Six studies were included in the PFS/MFS analysis, in which a random-effect model was used because of the significant heterogeneity (p=0.006, I2=69.2) (Figure 2B). Our outcomes showed that there was no significant correlation between TSP-1 and PFS/MFS (HR=1.35; 95%CI: 0.87-2.10; P=0.176). Likewise, subgroup analyses were stratified for the PFS/MFS group to identify the potential source of heterogeneity and other significant information. In ethnicity subgroup, high expression of TSP-1 was related to unfavorable PFS/MFS in Caucasians (HR = 1.80, 95%CI: 1.34–2.40; P＜0.001) (Figure 4A). Stratifying by the source of HR, high TSP-1 expression revealed a significant relationship with poor PFS/MFS, mainly in the report group (HR = 1.63, 95%CI: 1.24–2.15; P=0.001) but not in the SC group (Figure 4B). The subgroup analysis of cancer type indicated that TSP-1 have a statistically significant association with the breast cancer group (HR = 1.80, 95%CI: 1.20–2.71; P=0.004) and gynecologic cancer group (HR = 1.79, 95%CI: 1.18–2.71; P=0.006) (Figure 4C). when stratified by main pathological type, analysis in the adenocarcinoma group exhibited a significant correlation between up-regulated expression of TSP-1 and PFS/MFS (HR = 1.80, 95%CI: 1.20–2.71; P=0.004) (Figure 4D). Elevated TSP-1 predict poorer PFS/MFS in patients in the USA group (HR = 1.79, 95%CI: 1.18–2.71; P=0.006) (Figure 4E). The subgroup analysis in different assay methods had no obvious significance (Figure 4F).
We analyzed tumor recurrence associated with overexpression of TSP-1 by DFS/RFS. Seven studies focused on DFS/RFS analysis, with a high degree of heterogeneity (P=0.001, I2=73.7) (Figure 2C).There was no correlation between high level of TSP-1 and poor DFS/RFS, (HR = 1.40, 95%CI: 0.77–2.53; P=0.271) by random effect model. Furthermore, through the subgroup analyses, we did not observe statistically significant outcomes (Figure S1). In summary, no relationship was found between DFS/RFS and TSP-1.
The main function of cumulative meta-analysis is reflecting the dynamic trend of the research results and evaluating the impact of each research on the comprehensive results. All the selected studies were sorted according to the year of publication. (Figure 5). The relationship between OS and TSP-1 was first statistically significant in 2001. In addition, the corresponding 95% CIs of OS became narrower with the continuous inclusion of studies, suggesting increasing estimated accuracy. On the contrary, as time goes on, the relationship of TSP-1 and DFS/RFS or PFS/MFS are no longer statistically significant, indicating growing controversy in recent research.
Egger’s test and Begg’s funnel plot were applied to indicate publication bias in the included studies (Figure S2). No obvious asymmetry was noticed in funnel plots and the P value of Egger’s test also revealed no obvious publication bias. (OS: P = 0.066; DFS/RFS: P =0.934; PFS/MFS: P =0.713).
In order to ensure the robustness of the above results and evaluate the stability of results, a sensitivity analysis was carried out by Stata 12.0 software. The analyzed result from a fixed model of OS and two random model of DFS/RFS and PFS/MFS demonstrated that no single study considerably influenced the pooled HRs or 95% CIs, suggesting that the results of the present meta-analysis are credible (Figure 6).