The process of searching and screening the literature is shown in Figure 1. We identified a total of 822 publications in accordance with the search method described. When scanning the titles, abstracts, publication types, and full texts of the 822 publications, it was found that only 35 articles mentioned the correlation between pretreatment serum albumin and the outcome of RCC patients. Among these, 18 articles were excluded (three lacked some important data, seven used continuous or two cutoffs, and eight only reported odds ratios or relative risks). Thus, a total of 17 studies comprising 6,447 cases were included in our meta-analysis.14–28
Characteristics of included studies
The basic information on the studies conforming to the inclusion criteria is collected in Table 1. The 17 included articles were published from 2000 to 2015, and most of them came from Asia (n=4), Europe (n=4), and North America (n=8). Our meta-analysis included a total of 6,447 patients, with a median number of 209 patients per study (range: 70–2,119). Of these studies, nine explored the effect of pretreatment serum albumin on the prognosis of metastatic renal cell carcinoma, three on localized RCC, and four on clear cell renal cell carcinoma. Meanwhile, ten studies were evaluated using multivariate analysis, and seven were evaluated using univariate analysis. OS was evaluated in eleven studies, CSS was reported in eight studies, PFS was reported in two studies, and RFS was reported in one study. Considering there are more studies about OS, we chose OS as the primary outcome of the studies selected for this meta-analysis. HRs were directly reported for 15 studies, and estimated indirectly for the remaining two studies. The cutoff value of pretreatment serum albumin used in most studies was 3.5 g/dL (the lower limit of normal).
(To view a larger version of Table 1, click here.)
According to the Newcastle–Ottawa scale, we assessed the quality of the 17 eligible studies included in our meta-analysis. The quality scores of the studies varied from 4 to 9, with a mean of 6.5. A higher score showed better methodology. None of the studies mentioned were excluded from subsequent analyses.
Pretreatment serum albumin and survival of RCC patients
The main results of this meta-analysis are shown in Table 2. Considering the similarities between RFS and PFS, we merged them together to conduct this analysis. A decreased pretreatment serum albumin level yielded a worse OS (I2=54.7%; P=0.015, random-effects model; HR=2.46, 95% CI 1.92–3.13, P<0.001) (Figure 2A); CSS (I2=39.6%, P=0.128, fixed-effects model; HR=2.22, 95% CI 1.87–2.64, P<0.001); and RFS/PFS (I2=39.4%, P=0.192, fixed-effects model; HR=1.75, 95% CI 1.28–2.38, P<0.001) (Figure 2B). To explore the heterogeneity between these studies, the significance of pretreatment serum albumin was evaluated further via subgroup analysis based on the main features, including tumor type, analysis type, cutoff value, and HR-obtaining method. Considering that the number of studies that evaluated RFS and PFS was relatively small, we only conducted subgroup analysis for OS and CSS. In the tumor type subgroup, a decreased pretreatment serum albumin level was closely associated with the poor prognosis of RCC patients, except for OS in the all-stage RCC group (Figure 3). As for the analysis type, cutoff value, and HR-obtaining method subgroups, previous findings were powerful, as shown in Table 2.
(To view a larger version of table 2, click here.)
(To view a larger version of Figure 2, click here.)
(To view a larger version of Figure 3, click here.)
Each single cohort included in our meta-analysis was deleted, in turn, to check if individual studies influenced the results. Results of sensitivity analyses indicated the strength of our findings (data not shown).
Publication bias was assessed for both OS and CSS. The publication bias of all enrolled studies was evaluated using funnel plots, and Egger’s and Begg’s tests. The funnel plots were almost symmetrical (Figure 4). The P-values of the Egger’s and Begg’s tests were all greater than 0.05 (OS, P=0.938 for the Begg’s test, P=0.967 for the Egger’s test; CSS, P=0.230 for the Begg’s test, P=0.620 for the Egger’s test). Therefore, there was no significant publication bias in our meta-analysis.
(To view a larger version of Figure 4, click here.)