Continue Reading

Search strategy and selection criteria

This meta-analysis was conducted according to the preferred reporting items for systematic reviews and meta-analyses guidelines.29

We searched PubMed, Embase, and Web of Science for published studies that analyzed the prognostic value of pretreatment serum albumin in RCC patients up to November 3, 2015. Search terms used were: “kidney cancer or renal cancer or renal carcinoma or renal cell carcinoma” (all fields) and “albumin or hypoalbuminemia” (all fields) and “prognosis or prognostic or survival or outcome”. We checked the titles and abstracts of the papers retrieved, excluded irrelevant studies, screened the full text for the remaining papers, included satisfactory studies, and extracted the required data. In addition, we manually screened the references of related papers, including all the identified studies, reviews, and editorials, in order to retrieve unpublished but relevant investigations. Considering overall survival (OS), cancer-specific survival (CSS), disease-free survival (DFS), and progression-free survival (PFS) were the frequently used outcomes of RCC, we chose them as the primary outcomes of the studies that were selected for this meta-analysis. Inclusion criteria: 1) diagnosis of RCC was histopathologically confirmed; 2) treatment was limited to surgery, targeted therapy, or immunotherapy; 3) pretreatment albumin values were measured and their potential association with the prognosis of RCC patients was analyzed; 4) retrospective or prospective study design; and 5) studies that directly offered the hazard ratio (HR) and 95% confidence interval (CI) or cases in which the presented data were available for reconstruction of HR and 95% CI.

Exclusion criteria: 1) studies that were not written in English; 2) letters, review papers, meeting records, commentaries, case reports, or clinical guidelines; 3) lack of critical data, such as HR or 95% CI; 4) studies on cancer cells and experimental animal studies; 5) sample sizes smaller than 40; 6) in cases where data overlapped across several different articles, only the study with the largest sample size was reviewed.

Quality assessment

According to the Newcastle–Ottawa quality assessment scale, two researchers independently assessed the quality of each study.30 For quality assessment, scores ranged from 0 (lowest) to 9 (highest), and studies with scores of 6 or more were rated as being of high quality.

Data extraction and conversion

All data were extracted from the literature by two independent reviewers. When divergence appeared in the data-extraction process, a consensus would be reached after discussion. We extracted the following data: 1) basic information on the study: first author’s last name, publication year, country; 2) basic features of the patients, ie, case number, age, TNM staging, pathological grading, follow-up time; 3) cut-off value of pretreatment serum albumin; and 4) HR of albumin for OS, CSS, PFS, or recurrence-free survival (RFS), as well as 95% CI. If both the results of univariate and multivariate analysis were reported, only the result of multivariate analysis was extracted because this is more accurate, as it accounts for confounding factors.

If an article provided HRs and 95% CIs, they were extracted directly. However, if a paper did not provide HR and 95% CI, they were calculated using the data provided in the paper. If only the Kaplan–Meier curves of pretreatment albumin were available, we reconstructed the HRs and 95% CIs from the data extracted from the survival plots. We also sent emails to the corresponding authors to request any additional data needed for our meta-analysis. All the calculation methods mentioned were provided by Parmar et al31 and Tierney et al.32

Statistical analysis

HRs with 95% CIs were used to describe the relationship between pretreatment serum albumin and survival of RCC patients. An HR >1 suggested a worse prognosis in patients with a low concentration of pretreatment serum albumin, and an HR <1 indicated a better prognosis. We used Cochran’s Q test and Higgins I-squared statistic to conduct the test of heterogeneity of the combined HRs. If the P-value was <0.1 and/or I2 was >50%, the heterogeneity of the combined HRs was considered statistically significant, and a random effects model (the DerSimonian–Laird method) was then applied; otherwise, a fixed effects model (the Mantel–Haenszel method) was applied. The factors contributing to heterogeneity were analyzed by subgroup analysis. By assessing the asymmetry of an inverted funnel plot, publication bias was evaluated. Furthermore, we performed Begg’s and Egger’s tests to provide quantitative evidence of publication bias. Data analyses were performed using STATA version 12.0 (Stata Corporation, College Station, TX, USA). All statistical tests were two-sided, and when P<0.05, differences were considered statistically significant.