This meta-analysis did not require the use of an institutional review board or ethical committee. Additionally, the primary data were obtained from previously published research; therefore, there was no direct effect on the participants.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines were used to compile the data for this meta-analysis (22).

We systematically and extensively searched the Cochrane Library, Embase, PubMed, and Web of Science databases. Our exhaustive and targeted search methodology consisted of the following steps: (geriatric nutritional risk index OR GNRI) AND (bladder cancer OR renal cell cancer OR prostate cancer OR urothelial cancer OR urological cancer OR urinary cancer). A new search update was implemented on September 10, 2022. Articles written in languages other than English were also excluded. Furthermore, we also analyzed all the cited sources of the reviews and studies to find other papers that were relevant to our topic.

The inclusion criteria were as follows: (i) patients with upper tract urothelial cancer, bladder cancer, PCa, RCC, and UC were pathologically diagnosed; (ii) patients were divided into subgroups based on their GNRI; (iii) a GNRI cut-off value was determined; (iv) the GNRI was calculated as 14.89 × albumin (mg/dl) + 41.7 × (present/ideal body) weight (kg) before treatment; (v) hazard ratios (HRs) and 95% confidence intervals (CIs) were reported or adequate data were provided to compute them; and (vi) recurrence-free survival (RFS), cancer-specific survival (CSS), overall survival (OS), and progression-free survival (PFS) were reported. The following studies were excluded: animal studies, studies that did not provide enough data for analysis, studies that were duplicated and featured the same patients, reviews and conference abstracts, letters and case reports, and comments.

The literature review was conducted by two scholars working separately (QW and FY). All disagreements were discussed and resolved verbally until agreement was reached. Data from relevant studies included the first author’s name, year of publication, sample size, country, sex, time period, type of cancer, study design, study center (multicenter or single-center), treatment, tumor-node-metastasis (TNM) stage, duration of follow-up, GNRI cut-off value, type of survival analysis, survival outcomes, and HRs and 95% CIs. When both multivariate and univariate HRs and 95% CIs were used, the results of the multivariate analysis (MVA) were employed. In cases where only UVA was available, the HRs and 95% CIs were used instead. Each study included in the list was scored on the Newcastle-Ottawa scale (NOS) (23) to evaluate the research design quality. The final NOS score may range from 0 to 9, with points awarded for comparability (1–2), patient selection (0–4), and outcome (0–3). A high-quality study received a score of ≥ 6.

The predictive significance of the GNRI in patients with UCs was evaluated by calculating the 95% CI and HR for survival outcomes. The I² statistic and Cochrane Q statistic were used to assess statistical heterogeneity between studies. Owing to the low levels of heterogeneity, indicated by an I² value below 50% and a Q-test significance level above 0.10, a fixed-effects model (FEM) was used. Without this information, a random-effects model (REM) was utilized. To determine the origin of the observed variation, a subgroup analysis was performed, stratified by several clinicopathological characteristics. Publication bias was determined using Egger’s test and Begg’s funnel plot. The Stata version 12.0 was used for all statistical analysis (Stata Corporation, College Station, TX, USA) was used for all statistical analyses. Statistical significance was set at p < 0.05.

As shown in Figure 1 , the initial literature search generated a total of 125 items. After filtering out 44 duplicates, the abstracts and titles of 81 papers were read. Thereafter, 66 papers were discarded, leaving only 15 for the full-text analysis. Seven studies were excluded for the following reasons: (1) they did not provide survival data (n = 3), (2) they did not perform a GNRI analysis (n = 2), (3) they did not determine a GNRI cut-off value (n = 1), and (4) they included patients who had already been studied (n = 1). Eight studies with 6,792 patients (14–21) were included in the final meta-analysis ( Figure 1 ).

Table 1 shows the typical characteristics of the included studies. The articles considered were published in full-text format in the English language between 2015 and 2022 (14–21). Four studies were performed in Japan (15, 16, 19, 21), two in China (14, 20), and one each in Korea (17) and Taiwan (18). The sample sizes ranged from 68 to 4,591, with a median of 319.5. Four studies recruited patients with RCC (14, 15, 17, 20), two studies enrolled patients with PCa (16, 18), and two studies included patients with UC (19, 21). Seven studies were retrospective studies (15–21) and one was a prospective trial (14). Five studies recruited patients with TNM stage IV (14, 16, 18, 19, 21) and three studies enrolled patients with TNM stages I–III (15, 17, 20). Three studies included patients receiving surgery (15, 17, 20), two studies recruited patients undergoing chemotherapy (18, 19), and one study used ADT (16), immune checkpoint inhibitor (21), and targeted therapy (14). Seven studies adopted 92 as the cut-off value for the GNRI (14, 16–21) and one study adopted 98 (15). The significance of the GNRI as an OS prognostic factor was revealed in six studies (14, 16, 18–21), three studies presented the association between the GNRI and RFS (15, 17, 20), two studies reported the HR and 95%CI for PFS (18, 19), and three studies demonstrated a correlation between the GNRI and CSS (15–17). Six studies described the HRs and 95% CIs from the MVA (14, 17–21), and two studies reported data from the UVA (15, 16). Five studies were multicenter (14, 16, 17, 19, 21) and three were single-center (15, 18, 20). The NOS score of the considered studies varied from 7 to 9, with a median of 8, showing that the methodology of all considered studies was of a high standard.

The predictive importance of the GNRI for OS in patients with UCs was revealed in six investigations, including a total of 1,769 participants (14, 16, 18–21). In this case, substantial heterogeneity (I²=75.6%, Ph=0.001) necessitated REM deployment. As shown in Table 2 and Figure 2 , the combined results indicated that a low GNRI was significantly associated with poor OS in patients with UCs (HR = 2.62, 95% CI = 1.69–4.09, p < 0.001). The subgroup analysis revealed that regardless of study design, type of survival analysis, or sample size, a low GNRI was a clear indication of worse OS ( Table 2 ). Patients with UC and PCa, but not RCC, had a low GNRI and poor OS ( Table 2 ).

We merged RFS and PFS into the RFS/PFS groups because they were both event-free survival endpoints. Five studies comprising 5,955 patients (15, 17–20) reported the relationship between RFS/PFS and GNRI. The pooled HR and 95% CI were as follows: p < 0.001, HR = 1.77, 95% CI = 1.51–2.08 in the FEM ( Figure 3 , Table 3 ), which suggested that patients with UCs with low GNRI had poor RFS/PFS. The prognostic significance of GNRI for RFS/PFS remained significant in various subgroups of sample size, cancer type, study center, TNM stage, and cut-off value, as shown in Table 3 from the subgroup analysis.

Three studies, consisting of 5,362 patients (15–17) described the HRs and 95% CIs for CSS. REM was used, and the combined outcomes were as follows: HR = 2.32, 95% CI = 1.28–4.20, p = 0.006 ( Figure 4 ). As shown in Table 4 , subgroup analysis revealed that decreased GNRI was an important prognostic marker for poor CSS, regardless of the study center and cut-off value in patients with UCs.

This meta-analysis did not exhibit any significant publication bias according to Egger’s test and Begg’s test ( Figure 5 ).