1.Search strategy
The two studies independently searched the Medline, Embase, PubMed, Cochrane Libraries,and Web of science databases for all relevant articles. Medical subject heading terms used in the search included “liver cancer ", “Hepatocellular carcinoma (HCC)”, “Hepatic carcinoma”, “hepatoma”, "sorafenib", "microvascular invasion". The study's title and abstract were independently reviewed by the two authors, excluding studies that did not answer relevant research questions. The full text of the remaining articles, including references, was examined to determine if they contained relevant information. When incomplete information was available, attempts were made to contact the corresponding author of the study for more information. Figure 1 shows the literature screening process..
2.Selection criteria
If the cohort and case-control studies or randomized controlled trial of sorafenib therapy associated with microvascular invasion of hepatocellular carcinoma (HCC), in which the experimental group for the use of sorafenib therapy, treatment group does not contain sorafenib, and study the related risk ratio (hr)/relative risk (RR)/odds ratio (or) and corresponding 95% confidence intervals (CIs), or provides ample research data will be incorporated in the analysis. Inclusion is not limited by language, time, nationality and other factors. If the study object is the same population, select a more complete study to be included in the analysis.
3.Data extraction
In the Endnote version X9, references are merged together for ease of administration. Two authors independently completed the literature search and evaluated the full text. For the selected literature, two authors read the title, abstract and full text of the article. Authors in each eligible studies to extract the relevant data, to extract the study's lead author, publication date, country, research methods, data sources, research time, fixed number of year of follow-up, the study of the characteristics of the object (age, gender, exposure, the definition of a test dose and duration of the relevant risk assessment (including hr, RR or) and 95% confidence interval and relevant confounding factors. The authors did all this work independently. Because subjects in some studies used combination therapy, we defined the experimental group as treatment with sorafenib and the control group as non-treatment with sorafenib.
4.Quality assessment
The Newcastle Ottawa scale[19] was used to evaluate the quality of the cohort study. In this scale, the study was divided into three categories: subject selection (4 stars), study group comparability (2 stars), result / exposure assessment (3 stars). The quality adopts star system, the highest 9 stars, 0–5 stars are low quality, 6–9 stars are high quality. The characteristics of the study are shown in Table 1.
5.Statistical analysis
The statistical analysis in this article used the 15th generation of STATA software (STATA, College Station, TX, USA). The probability values (P values) of all statistical results were bilateral, and P < 0.05 was considered statistically significant. Higgins and Thompson's I2 was used to assess heterogeneity between studies. When I2 > 50% is represented as large heterogeneity, I2 < 50% indicates small heterogeneity. [20], DerSimonian and Laird random effect model (REM) is used as pooling method, otherwise, fixed effect model (FEM) is used as pooling method.
6.Results
Four related studies were reviewed[21–24], with a total of about 950 HCC patients with microvascular invasion. Sorafenib significantly improved survival in HCC patients with microvascular invasion compared to those without sorafenib treatment (RR = 1.369, 95% confidence interval (CI) 1.193,1.570; P < 0.001). There was no significant heterogeneity between the 4 included studies (P = 0.352 > 0.01; I2 = 8.2%) The relevant results are shown in Fig. 2. The results of publication bias show that there is no publication bias(z = 1.02 > 0.05, Pr > |z| = 0.308), The funnel is shown in Fig. 3. Through the sensitivity analysis, it can be concluded that there is no significant difference among the four studies, with good consistency (Fig. 4).