Efficacy and Safety of Chemotherapy Regimens in Advanced or Metastatic Bladder and Urothelial Carcinomas: An Updated Network Meta-Analysis

Background: Gemcitabine plus cisplatin (GC) and methotrexate, vinblastine, adriamycin, and cisplatin (MVAC) have been the first-line treatments for advanced or metastatic urothelial carcinoma (AMUC). However, their effects are unsatisfactory, and more drugs and regimens still need to be explored. Objective: We aimed to comprehensively compare all possible regimens with GC or MVAC in randomized controlled trials (RCTs) by network meta-analysis. Methods: We searched the PubMed, Embase, and Cochrane databases for RCTs that evaluated regimens compared to GC or MVAC on AMUC patients. The major outcomes were progression-free survival (PFS), overall survival (OS), and objective response rate (ORR). A network meta-analysis was used to assess the effectiveness and safety of the included treatment regimens, and the regimens were then clustered by the average linkage method. Results: A total of 19 trials that assessed 3,363 AMUC patients were included. For PFS, paclitaxel plus GC (PGC) was significantly superior to GC (log hazard ratio (HR): −0.16; 95% confidence interval (CI): −0.32, 0.00) with a moderate level of reliability. However, there was no significant difference between PGC and MVAC (log HR: −0.03; 95% CI: −0.27, 0.20). For OS, PGC was significantly superior to GC (log HR:−0.17; 95% CI: −0.33, −0.00) with a moderate reliability level but not significantly different from MVAC (log HR: −0.10; 95% CI: −0.35, 0.15). Analysis of ORR showed that PGC was superior to MVAC (log odds ratio (OR): 0.59; 95% CI: 0.02, 1.16) with a low reliability level and GC (log OR: 0.41; 95% CI: 0.12, 0.71) with a moderate reliability level. In the cluster results, PGC and sorafenib plus GC (GCS) exhibited relative advantages in efficiency, followed by MVAC and apatorsen plus GC (GCA); however, PGC, gemcitabine plus carboplatin (GP), and MVAC had more serious side effects. Conclusions: In our analysis, PGC was superior to MVAC and GC in only the ORR results and superior to GC in the OS and PFS results but was not significantly different from MVAC. More individualized therapies with targeted drugs need to be studied.


INTRODUCTION
Urothelial carcinoma is the most common type of bladder cancer and can also occur in other parts of the urinary system, such as the renal pelvis, ureter, and urethra (Bianchi et al., 2014;Venyo, 2014). The stages of UC include nonmuscle-invasive urothelial carcinoma (NMIUC), which has a high recurrence and progression rate (30-65%) (Jallad et al., 2014) and can develop into muscle-invasive urothelial carcinoma (MIUC) (Resnick et al., 2013) and advanced or metastatic urothelial carcinoma (AMUC) .
NMIUC of the bladder is usually treated by instillation. Bacillus Calmette-Guerin is more effective than other chemotherapy methods (Boehm et al., 2017;Huang et al., 2017) and can be combined with epirubicin (EPI) and mitomycin C (MMC) to prevent recurrence (Zhuo et al., 2016;Wu et al., 2017). Radical cystectomy with bilateral pelvic lymphadenectomy is usually used to prevent recurrence in MIUC patients and is usually supplemented with the chemotherapeutic strategy of methotrexate, vinblastine, adriamycin, and cisplatin (MVAC) (Kim et al., 2016).
For AMUC, gemcitabine plus cisplatin (GC) or MVAC is generally used as a first-line chemotherapy regimen (Ismaili et al., 2011;Racioppi et al., 2012). However, their effects are unsatisfactory, and more drugs and regimens still need to be explored. Taxanes, vinflunine, and immunotherapy are often used as second-line treatments (Narayanan et al., 2015;Holmsten et al., 2016;. In 2016, the FDA rapid review approved atezolizumab as a treatment for AMUC, but subsequent studies confirmed that the drug did not meet the original survival target (Ning et al., 2017).
There are many meta-analysis studies on chemotherapy strategies for AMUC. The combination of multiple chemotherapy drugs Raggi et al., 2016;Wu et al., 2016; and immune checkpoint inhibitors is believed to have survival benefits in patients (Li and Wang, 2018;Di Nunno et al., 2018). However, there is still a lack of direct and indirect comparisons among various chemotherapeutic strategies. Therefore, their clinical application is still unclear.
A Bayesian network meta-analysis assessed the safety and efficacy of various therapeutic strategies for AMUC patients with advanced urothelial cancer who underwent chemotherapy and showed that paclitaxel or sorafenib combined with GC had better survival advantages but serious side effects. However, bladder site tumors were neglected in this study, and the results of the direct and indirect comparisons between the chemotherapeutic strategies were also not clearly listed . In our study, we updated the above work and used the frequentist method to analyze randomized controlled trials (RCTs) comparing MVAC or GC with other strategies to provide guidance for the clinical treatment of AMUC.

METHODS
This network meta-analysis was pereformed according to the Preferred Reporting Items for Systematic Reviews Network Meta-Analyses (PRISMA-NMA) guidelines (Hutton et al., 2015).

DATA SELECTION
Studies were eligible if the following criteria were met: 1: included AMUC patients; 2: used RCTs; 3: had one arm with either GC, GP, or MVAC treatment; and 4: examined either progression-free survival (PFS), overall survival (OS), or objective response rate (ORR). PFS and OS could be in the form of a hazard ratio (HR) from a Cox regression model or predictable survival curve. The exclusion criteria included the following: 1: studies that did not include AMUC patients or did not report AMUC subset results; 2: non-RCTs; 3: RCTs that did not have a GC or MVAC arm or those that compared GC or MVAC with a placebo or blank control; 4: studies that compared the same drugs but different application strategies; and 5: studies that did not obtain any PFS, OS, or ORR results. Conference summaries, commentaries, and editorials were also excluded.

DATA EXTRACTION
The extracted contents included the first author name, publication year, type of patients, sample size, age, experimental intervention, control intervention, and follow-up period. The major outcomes were PFS, OS, and ORR. The secondary outcome was severe adverse events (SAEs) that had a grade> = 3 according to the National Cancer Institute (NCI) Common Terminology Criteria for Adverse Events. We assessed the methodological quality of the included trials using a risk of bias approach according to the methods described by the Cochrane Collaboration, which include seven specified domains (Higgins et al., 2011). In addition, we also applied Grading of Recommendation Assessment, Development and Evaluation (GRADE) guidance to evaluate the quality of the network analysis results with four levels graded from high (best) to very low (worst) (Yan and Xu, 2018). This method takes into account the design level of direct and indirect comparisons, the inconsistency of direct and indirect comparison results, the imprecision of results, and the large effect results that can improve the level of evidence.

STATISTICAL ANALYSIS
The PFS and OS results are represented by HRs and their 95% confidence intervals (CIs). For studies with survival curve results, the results were gathered from the curve. The ORRs and adverse events (AEs) are represented as binary data. The odds ratios (ORs) and their 95% CIs were calculated by extracting the frequencies of events. We used a frequentist framework random-effects model for mixed multiple treatment comparisons (Greco et al., 2015). Global and local inconsistencies between direct and indirect sources of evidence were assessed by the fit of consistency and inconsistency models and the difference between direct and indirect estimates in all closed loops, respectively. To rank the treatments for each outcome, we used sureface under the cumulative ranking (SUCRA) probabilities (Li et al., 2015). After obtaining SUCRA values, the major outcomes and SAEs were clustered by the average linkage clustering method. This exploratory clustering method avoids the intereference of individual deviation samples on the overall results. Comparison-adjusted funnel plots were used to determine small-study effects in the analysis (Trinquart et al., 2012). Data analyses were pereformed using STATA software (version 14.0; STATA Corporation, College Station, TX, USA).
The publication time of the included studies had a long time span, ranging from 1990 to 2018. A total of 3363 AMUC patients were enrolled in the study. The median age was 60-70 years old. Three articles included only advanced bladder cancer patients. Two studies reported different outcomes from the same cohort, and we combined the outcomes of these two studies (von der Maase et al., 2000;von der Maase et al., 2005) (Table 1). All included studies were of RCT design, but in some studies, the generation of random sequences and random masking were not clearly described. Three studies were blinded Krege et al., 2014;. Because the main evaluation results are objective, the quality of the research was generally acceptable (Figure 2).

RESULTS OF THE NETWORK META-ANALYSIS
In the network meta-analysis of PFS, 15 treatment strategies were analyzed ( Figure 3A). There were nine GC-related comparisons and 6 MVAC-related comparisons. An inconsistency plot was used to assume loop-specific heterogeneity, and one quadratic loop showed that there was no significant difference between the direct and indirect comparisons (inconsistency factor: 0.33, 95% CI: 01.12) (Supplementary Figure 1). Global inconsistency was also not detected in the analysis (p = 0.412). In the network comparisons, paclitaxel plus GC (PGC) was significantly superior to GC (log HR: −0.16; 95% CI: −0.32, 0.00) with a moderate level of reliability. However, there was no significant difference between PGC and MVAC (log HR:  Table 2). A comparisonadjusted funnel plot did not suggest that there was any publication bias ( Figure 4A).
No publication bias was detected ( Figure 4C). The cluster method was generally used to classify and analyze the validity and security of the results in the network analysis. We used the average linkage clustering method to classify the treatment regimens by clustering the efficiency and safety outcomes separately. Fourteen treatment regimens were included in the efficiency cluster analysis, while others were excluded due to a lack of SUCRA results. PGC and GCS had relative advantages in efficiency, followed by MVAC and GCA. In contrast, cisplatin, DC, PC, LC, GP, and MCAVI were considered relatively inefficient ( Figure 5).
The cluster results of safety outcomes and SAE (grade> = 3) included analyses of neutropenia, anemia, thrombocytopenia, infection, mucositis, and nausea/vomiting, which are frequently reported. The cluster analysis showed that regimens such as cisplatin, MCAVI, CP, LC, and DC had fewer SAEs but were also less effective. The cytotoxic drugs that have a weak effect on cancer cells may also have a weaker effect on normal cells. GCrelated treatment strategies, such as GC, GCCET, GCS, GCA, and gemcitabine, cisplatin, and gefitinib (GCG), had similar SAE clusters. However, PGC, GP, and MVAC exhibited more serious side effects. In general, PGC and MVAC had better therapeutic effects than other regimens but had more serious adverse effects as well. GCS and GCA exhibited a similar efficacy to MVAC and had relatively mild SAEs, similar to GC ( Figure 6).

DISCUSSION
There is still a need to explore more effective methods for the treatment of AMUC. Because GC and MVAC are widely used and regarded as effective treatments for AMUC, this study comprehensively analyzed all possible regimens compared to GC or MVAC in RCTs by network meta-analysis. In addition, carboplatin and cisplatin are both platinum-based treatments; therefore, we included GP and GC in the analysis. This work aimed to find more advantageous regimens to guide clinical applications.
In this study, seventeen AMUC treatment regimens were included, and fourteen of them were clustered. No suitable clustering scheme was identified when all results were considered at the same level. Therefore, we applied the average linkage method to merge multiple primary and secondary results. Using this approach, analysis of the efficiency revealed that PGC, GCS, MVAC, and GCA were more effective than CIS, DC, PC, LC, GP, and MCAVI. Analysis of the safety outcomes indicated that GC-related treatment strategies, such as GC, GCCET, GCS, GCA, and GCG, had similar side effects whereas PGC, GP, and MVAC had more serious SAEs.
The exclusion of several studies from this work needs to be explained. Vinflunine-related and lapatinib-related RCTs were excluded because both studies conducted a more than four-week drug-eluting phase after chemotherapy (Garcia-Donas et al., 2017;Powles et al., 2017). Due to the eluting phase, the research drug and chemotherapy were not considered to be a combination treatment strategy. Pembrolizumab was demonstrated to have longer OS rates than the investigator's choice of chemotherapy for advanced UC patients in a phase 3 trial . That study was excluded because the chemotherapy regimens were unclear. A granulocyte stimulating factor (GSF)-related RCT comparing GSF-MVAC and MVAC was excluded because GSF is commonly used in AMUC patients with hematologic AEs after chemotherapy (Logothetis et al., 1995). In addition, because GC and MVAC regimens have been proven to be effective for AMUC, blank control studies were not included in the analysis.
Compared with the previous network meta-analysis, the main difference in this study is that advanced bladder cancer patients were included; therefore, more studies and intervention regimens were also included . A retracted study (Roberts et al., 2006) and a nonrandomized study (Hsieh et al., 2016) were excluded, which were included in the previous analysis. Furthermore, a trastuzumab-related RCT was excluded because of an unclear chemotherapy strategy in the intervention group (Oudard et al., 2015). In addition, this work used the frequentist method instead of the Bayesian method for the network meta-analysis, which further provides the results of direct and indirect comparisons among the included regimens and then classifies the reliability of the results according to the GRADE method. Notably, PGC ranked well in SUCRA scores and was significantly better than MVAC and GC regimens in ORR. However, for OS and PFS, PGC was only significantly better than GC and not significantly different from MVAC.
Although there was no significant survival benefit compared to MVAC, PGC is still a potential chemotherapy regimen for AMUC. In this regimen, cisplatin intereferes with DNA replication and mitosis by crosslinking with DNA. Gemcitabine is a derivative of a cytosine nucleoside that stops the synthesis of DNA strands and results in masked chain termination to avoid normal repair, leading to DNA breakage. These two drugs cooperatively prevent DNA replication and cell separation as a first-line treatment. Compared with GC, PGC had a significantly better ORR, OS, and PFS in this study. Paclitaxel is a microtubule-associated inhibitor that stabilizes microtubule polymers and promotes their assembly to arrest cells in the M phase. However, even the PGC regimen did not provide a significant survival benefit compared to MVAC.
In addition to traditional cytotoxic drugs, targeted drugs have also been researched for AMUC treatment. Sorafenib is a multitarget inhibitor of tyrosine kinase. GCS exhibited similar effects as PGC in the cluster analysis, showing SAEs similar to those observed with GC. However, there was still no significant difference between GCS and GC or MVAC regarding PFS, OS, or ORR. The unsatisfactory therapeutic result of GCS may be due to the small number of patients, or GCS may only be effective for specific AMUC patients. In a recent phase I study of metastatic urothelial carcinoma second-line therapy, sorafenib and vinflunine combined had an OS rate of 7 (1.8-41.7) months and ORR of 41%; however, RCTs are needed to confirm this therapeutic effect (Knievel et al., 2014;Shah et al., 2019). Individualized therapy for these targeted and sensitive drugs may further improve the therapeutic effect. However, in in vitro studies, the limited role of sorafenib in the mitogen-activated protein kinase (MAPK) signaling pathway of urothelial cancer cell lines suggests that sorafenib is not very suitable for UC treatment (Knievel et al., 2014).  Table 1. The SUCRA scores are weighted, with white indicating high scores and black indicating low scores.
Gefitinib is an epidermal growth factor receptor tyrosine kinase domain inhibitor. However, in a trial, 13 of 20 metastatic bladder cancer patients had obvious epidermal growth factor receptor (EGFR) expression, and gefitinib did not confer enough survival benefits to patients or improved ORR (Philips et al., 2008). In preclinical studies, gefitinib was demonstrated to reverse the sensitivity of cisplatin and paclitaxel-resistant UC cells, indicating that the potential mechanism of gefitinib requires further research . As EGFR inhibitors, cetuximab combined with GC also did not achieve the desired effect (Hussain et al., 2014). Lapatinib, a target drug for human epidermal growth factor receptor 2 (HER2), did not significantly improve the therapeutic effect in HER1/HER2-expressing metastatic bladder cancer (MBC) patients. Even in the strong HER1/2 position subgroup, lapatinib did not significantly improve the survival benefit . In an excluded study, trastuzumab combined with GC or GP was used to treat metastatic urothelial cancer (Oudard et al., 2015). However, the combination did not significantly improve ORR, OS, or PFS in patients. The low incidence of HER2 overexpression in patients suggests that trastuzumab lacks the means of universal application. Therefore, EGFR may not be a desired therapeutic target for AMUC.
Apatorsen, which inhibits the production of heat shock protein 27 (Hsp27), has also been shown to have significant  survival benefits for AMUC patients. However, after classification by Hsp27 level, subgrouped patients with <5.7 ng/ml and < = 20.5% exhibited an obvious survival benefit after apatorsen treatment (Rosenberg et al., 2018). The above results suggest that individualized treatment is a research direction for improving treatment effects on AMUC patients. For other macromolecule-targeted drugs, such as vascular endothelial growth factor inhibitors (e.g., bevacizumab) and PD-L1 inhibitors (e.g., atezolizumab), no related studies were included because none met the inclusion criteria. Although the therapeutic advantage of atezolizumab has been demonstrated in phase II of a single-arm study , the results of large-scale RCTs are still needed. PD-L1 expression detection was also considered a prediction of the therapeutic effect of atezolizumab (Crist and Balar, 2017). Ultimately, for the application of targeted drugs, the characteristics of AMUC patients, such as target protein expression, need to be a focus of future studies.

LIMITATIONS
There were still several limitations in this work. First, this study included only GC-and MVAC-related RCTs to maintain the continuity of the network analysis and neglected some non-GCor MVAC-related RCTs. This work did not analyze the differences in drug dosage or application duration among treatment regimens. The small sample sizes of single arms may affect the accuracy and reliability of the results. Finally, as more RCTs are reported, the conclusions may change.