From left, top to bottom: Nicholas Vogelzang, MD, FASCO, FACP1; Thomas Hutson, DO, PharmD, FACP3; Marissa Blieden, MS: Corresponding Author4; Kyle Fahrbach, PhD5; Rachel Huelin, BA6; Adriana Valderrama, PhD, MBA7; Lonnie Wen, RPh, PhD8; Svetlana Babajanyan, MS, MD9 Not pictured: Thomas Powles, MBBS, MRCP, MD2 Affiliations are listed at the end of this article.
Previously, the standard of care for advanced/metastatic renal cell carcinoma (RCC) consisted primarily of interleukin-2 (IL-2) or interferon-alfa (IFN-a).1 Since 2005, new targeted therapies have proven superior to cytokines for the first- and second-line treatment of RCC, including sorafenib, sunitinib, bevacizumab (in combination with IFN-a), pazopanib, temsirolimus, everolimus, and axitinib.2 Additionally, since November of 2015, the United States (US) Food and Drug Administration (FDA) has approved three new treatments for RCC: nivolumab (a programmed death 1 [PD-1] inhibitor),3,4 lenvatinib (a tyrosine kinase inhibitor [TKI], as either monotherapy and in combination with everolimus),5,6 and cabozantinib (another TKI).7,8
According to the European Society for Medical Oncology (ESMO) 2014 guidelines for RCC, commonly recommended first-line treatments for patients with a good/intermediate prognosis include bevacizumab + IFN-a, sunitinib, and pazopanib.9,10 Other ESMO-recommended first-line treatments include sorafenib, high-dose IL-2, and low-dose IFN-α+ bevacizumab. RCC patients with a poor prognosis typically receive temsirolimus, although sunitinib, pazopanib, and sorafenib are potential options.9 Treatment options for patients in the second line include axitinib, sorafenib, and pazopanib following cytokine treatment, and axitinib and everolimus following treatment with a TKI; sorafenib is also an option following TKI treatment.9,10 Updated 2016 treatment guidelines from the National Comprehensive Cancer Network (NCCN)11 recommend nivolumab, lenvatinib, or cabo-zantinib as subsequent treatment after first-line therapy, based on positive results comparing the three treatments to everolimus in Phase II/III trials.6,8,12
There is a lack of head-to-head trial data comparing these second- or later-line treatment options, especially since the few trials that addressed this used a variety of comparators, such as placebo, everolimus, and sorafenib. Therefore, it is difficult to understand the comparative efficacy and safety of these newer drugs (e.g., nivolumab, lenvatinib, and cabozantinib), and how these compare to more established treatments.
To provide practice guidance into the rapidly expanding treatment landscape for advanced RCC, a network meta-analysis (NMA) was conducted to determine the efficacy and safety of sorafenib in relation to other established and investigational agents in second- and later- lines of therapy.
Materials and Methods
Literature Identification We systematically searched for randomized controlled trials (RCTs) indexed in Embase and PubMed. These bibliographic databases were accessed on January 5, 2016 to identify English-language studies published since January 1, 2004. A search strategy was developed using terms to specifically identify trials in advanced RCC. The search algorithm was not limited to any particular treatments, so that new investigational therapies could also be captured. The search was validated by cross-referencing the results against the bibliographies of published systematic reviews. Recent abstracts from major clinical oncology and urology meetings were reviewed to identify any Phase II/III trials that had not yet been published (i.e., “grey” literature).
Study Selection and Data Extraction All identified abstracts were reviewed against the inclusion and exclusion criteria described in Table 1. RCTs of patients with advanced or metastatic RCC that compared targeted therapy and/or chemotherapy alone or in combinations were included. The full texts of studies that passed abstract screening were evaluated using the same criteria. To limit the results to studies with more precise estimates of treatment effects and higher statistical power, trials were excluded if they enrolled fewer than 50 patients per treatment arm. Data elements extracted from each included study were study design, patient and treatment characteristics, and clinical efficacy and safety outcomes.
Feasibility Assessment A feasibility assessment was conducted to determine whether the identified trials provided data suitable for an NMA. This assessment sought to determine the selection of trials that presented sufficient data for generating an evidence network for each outcome of interest. The assessment also ensured that there were minimal systematic differences in patient or disease characteristics (clinical heterogeneity) for factors that may influence the comparability of relative results (treatment effect modifiers).
Analysis Fixed-effects Bayesian NMAs were used to conduct indirect comparisons of the included trials for each of the outcomes of interest. Efficacy outcomes were progression-free survival (PFS), overall survival (OS), and the proportion of patients with clinical benefit, which is defined as complete response, partial response, or stable disease per the Response Evaluation Criteria in Solid
Tumors (RECIST). Safety outcomes included the proportion of patients with treatment discontinuations due to any cause, and the proportion of patients with one or more dose reductions due to adverse events. PFS and OS were analyzed as hazard ratios (HRs) with Bayesian 95% credible intervals (CrI). The Bayesian 95% CrI has the interpretation that, given the clinical and statistical assumptions made, there is a 95% chance that the mean falls within the CrI noted. CBR and the safety outcomes evaluated proportions of patients and were analyzed as odds ratios (ORs). For discussion purposes, 95% CrIs that do not include ‘1’ are considered “statistically significant”. Analyses were conducted in OpenBUGS 3.2.2 with a 50,000 run-in iteration phase and a 50,000-iteration phase for parameter estimation. Because no two trials in the network evaluated the same two treatments, statistical heterogeneity could not be assessed and random-effects models were not employed.
Two sensitivity analyses with modified networks were conducted to explore the impact of prior therapies and study design on the results. These analyses included: (1) network modification to evaluate the mammalian target of rapamycin (mTOR) inhibitors (everolimus and temsirolimus) as single therapies (i.e., introducing the assumption that these two therapies have equivalent treatment effects); and (2) within this network from the first sensitivity analysis, the exclusion of studies that enrolled only patients treated previously with only cytokines—to capture a population more representative of the current treatment landscape.
Study Eligibility and Characteristics Figure 1 depicts the flow diagram of the systematic literature search and RCT selection. The search identified 2,230 abstracts, of which 157 publications—representing 52 unique RCTs—investigated treatment of advanced or metastatic RCC, reported outcomes of interest, and were eligible for inclusion in the NMA. These 52 trials were then evaluated in the feasibility assessment for clinical heterogeneity and their ability to contribute to quantitative analyses.
The feasibility assessment determined that the included studies were heterogeneous with regards to previous treatment, so analyses were limited to trials evaluating therapies in second- or later-line settings; studies involving patients who were treatment-naïve for advanced or metastatic disease were excluded. Given that the therapeutic landscape has been changing rapidly over the last decade, we considered the potential impact of the date/year of patient enrollment as a source of heterogeneity between trials, as different targeted therapies came to market. Notably, the addition of new agents to the therapeutic landscape as well as increasing experience of the treating oncologists from 2005 to the present (the TKI era), has led to better clinical outcomes.13-17 We observed that the number and type of prior therapies (cytokine therapy, vascular endothelial growth factor [VEGF] and mTOR inhibitors) were sources of potential heterogeneity across trials. Similarly, we observed variations in trial enrollment by Memorial Sloan Kettering Cancer Center (MSKCC) risk score. The analysis results should be interpreted in the context of such differences among trials (Table 2. To view a larger version of this table. click here).
Overall, the feasibility assessment concluded that nine trials could contribute data for the outcomes of interest to an NMA of second- or later-line therapies for advanced or metastatic RCC. The nine RCTs consisted of a total of 5,147 patients in 21 different treatment arms, including sorafenib, axitinib, cabozantinib, dovitinib, everolimus, temsirolimus, lenvatinib, lenvatinib + everolimus, nivo-lumab, pazopanib, and placebo. Table 3 (to view a larger version of this table, click here) summarizes the base-case results of the NMA. Table 4 (to view a larger version, click here) summarizes the results of the different sensitivity analyses.
Clinical Benefit Rate Treatment with sorafenib resulted in a CBR that was significantly higher than that for placebo, and very similar to rates for dovitinib, everolimus, nivolumab, and temsirolimus (Table 3). The odds of clinical benefit were approximately four times greater for sorafenib compared to placebo; however, axitinib and cabozantinib were associated with a significantly greater likelihood of clinical benefit than sorafenib—approximately two and three times greater, respectively. The width of the CrIs associated with these comparisons suggests that the estimation of comparative effect is acceptably precise.
The results of the sensitivity analyses for CBR did not differ substantively from the base-case analysis (Table 4). There were no changes in significance, and the overall magnitude of difference was minimal in each sensitivity analysis. The similarity between the base-case and these sensitivity analyses is likely due to the similar effect of temsirolimus and everolimus on CBR. As we see in the base-case analyses, temsirolimus and sorafenib were equally likely to produce clinical benefit (OR [95% CrI]: 1.00 [0.69, 1.45]), and the results for everolimus were nearly identical (OR [95% CrI]: 0.93 [0.53, 1.57]). These results are in line with the assumption that mTOR inhibitors should be considered functionally and clinically equivalent to each other, with respect to CBR.
Progression-free Survival Sorafenib was associated with significantly longer PFS compared with placebo and exhibited similar PFS compared with dovitinib, pazopanib, and temsirolimus. Although this was not statistically significant, sorafenib therapy demonstrated a net improvement in PFS compared to pazopanib (HR [95% CrI]: 0.82 [0.50, 1.34]). Everolimus showed somewhat longer PFS than sorafenib, but the analysis was not statistically significant.
Several TKIs—axitinib, cabozantinib, lenvatinib (as well as lenvatinib in combination with everolimus), and nivolumab—showed significantly longer PFS compared with sorafenib. The PFS hazard rates for sorafenib ranged from 50% to almost three-and-a-half times larger than the rates of these comparators. No substantive differences were seen in the sensitivity analyses for PFS.
Overall Survival Base-case OS comparisons are reported in Table 3. Comparisons of sorafenib with other agents yielded hazard ratios that ranged from 0.76–1.84; however, none of these comparisons were statistically significant, except for sorafenib resulting in statistically significantly longer OS compared with temsirolimus. OS for sorafenib and everolimus was nearly identical (OR [95% CrI]: 1.01 [0.72, 1.41]), but the sensitivity analyses that combined mTOR inhibitors and restricted to studies with prior angiogenesis inhibitors exhibited results that were almost the same as the results seen for temsirolimus in the base case.
In the sensitivity analyses (Table 4), there was no change in the direction, statistical significance, or material difference in the magnitude of the relative estimates of comparison between sorafenib and other therapies with regard to OS. When analyzed as a class, mTORs had a slightly shorter OS compared with sorafenib (HR: 0.83), falling between the estimates for temsirolimus (HR: 0.76) and everolimus (HR: 1.01) in the base case.
Dose Reductions Sorafenib was associated with significantly greater odds of dose reduction due to adverse events versus placebo, axitinib, and temsirolimus (Table 3). Similarly, the odds of dose reduction were higher for sorafenib when compared to cabozantinib, and this finding became significant with a modified network. In contrast, the OR for dose reduction was numerically lower for sorafenib when compared with everolimus (OR [95% CrI]: 0.34 [0.01, 2.38]).
When evaluating mTORs as a class with a modified network (Table 4), the OR for mTORs (2.32), fell between the base-case estimates for everolimus (0.34) and temsirolimus (2.58). When the trials evaluating only cytokine-pretreated patients were removed (the second sensitivity analysis), the connection between sorafenib and mTOR was based solely on the INTORSECT trial22 of temsirolimus; thus, the estimate reverted back to the base-case estimate for temsirolimus (2.58). Because the METEOR trial8 compared cabozantinib to everolimus, the comparison of cabozantinib to sorafenib in the sensitivity analyses was influenced by the performance of temsirolimus. As the OR for sorafenib versus the mTORs increased in the sensitivity analyses, the OR of sorafenib versus cabozantinib correspondingly increased; in the base case, the OR was 1.60, and in the sensitivity analyses rose to statistically significant ORs of 10.87 (#1) and 12.07 (#2).
In some cases, the CrIs for the analyses of dose reductions were notably wide (i.e., indicating a low level of precision for the estimates), due to a low number of patients reducing doses in some trials. This was particularly true for the placebo arms of trials, where, for example, only one patient who received placebo in the RECORD-1 trial12 had a dose reduction. In the second sensitivity analysis (which removes trials that enrolled only cytokine-pretreated patients), the comparison of sorafenib and placebo relied on the RECORD-1 trial12 rather than the TARGET trial,19 leaving the resulting estimate highly unstable (OR [95% CrI]: 33.41 [5.17, 591.70]).
Treatment Discontinuations Sorafenib was associated with significantly decreased odds of treatment discontinuation compared with placebo. Sorafenib was also associated with greater odds of treatment discontinuation compared to axitinib, cabozantinib, everolimus, temsirolimus, and nivolumab; however, the results of the comparison to temsirolimus were not statistically significant.
As with the analysis of dose reductions, the network modifications relied more heavily on temsirolimus for the estimated effect of sorafenib compared to the mTORs and, by extension, to cabozantinib and nivolumab, which were compared to (and connect to the network through) everolimus. The sensitivity analyses consistently found that the differences between therapies were reduced, particularly for the comparisons of sorafenib with cabozantinib and nivolumab.
For CBR and PFS, sorafenib performed better than placebo and similar to dovitinib and the two mTORs, everolimus and temsirolimus. In addition, while there was an advantage for nivolumab for PFS, there was no significant or substantive difference between nivolumab and sorafenib with regard to CBR. Sorafenib performed worse than axitinib, cabozantinib, and both monotherapy and combination therapy involving lenvatinib.
For OS, many of the differences between sorafenib and other treatments were attenuated or disappeared. Sorafenib performed better than placebo and was substantively equivalent (i.e., an HR close to 1) to all other treatments with the exception of the newer agents, cabozantinib, nivolumab, and lenvatinib + everolimus. This may be an inherent bias in the time-dependent analysis favoring the three new agents since OS has been improving since 2005 as more drugs become available; patients were generally poorer risk in the early TKI era and physicians have become more skilled at using all the drugs at their disposal. In addition, in the sensitivity ana-lysis in which the two mTORs were assumed equivalent and involving studies excluding patients exposed to only cytokines, the substantive differences with cabozantinib and nivolumab disappeared (HRs = 1.14 and 1.05, respectively).
For safety outcomes, there was more variation across the base-case and sensitivity analyses. The only general conclusion to be made is that dose reductions and treatment discontinuations were more common on sorafenib than most other treatments, which may be related to stomatitis and hand-foot syndrome.
While an association between PFS and OS has been observed for some treatments for advanced RCC, researchers have cautioned against using PFS as a surrogate endpoint for OS. Becker et al. observed that the net benefit of PFS is not often seen in OS and that more careful examination of subsequent therapies (both in those studies showing an association and those not) is warranted.23 Given that PFS and CBR are less clinically relevant to patients and providers than OS, that OS is a bedrock of regulatory approval, and that comparisons of sorafenib with other agents yielded non-significant differences in OS (hazard ratios 0.76–1.84), the role of sorafenib in the treatment sequencing of metastatic RCC should be re-considered. These results underscore what many clinicians believe, namely that sorafenib is active yet widely underused mostly due to low-grade chronic toxicities and a perception of poorer survival outcomes. We here demon- strate that OS endpoints with sorafenib are not dramatically different from the newer agents. The bias against sorafenib may well be time-dependent. It was the first agent approved for metastatic RCC, and physicians were inexperienced in managing toxicities and had few options to use to improve survival when sorafenib failed. Moreover, the prolonged OS results seen with the newer agents reflect in part the benefit of subsequent therapies, including in some cases sorafenib.24,25
There were two main limitations for the analysis. The first is the sparsity of the network; most connections between treatments were based on the results of only one study; thus, it was impossible to estimate statistical heterogeneity. Based on differences across studies, potential sources of clinical heterogeneity were the number and type of prior therapies (cytokine therapy, VEGF, and mTOR inhibitors) and baseline MSKCC risk score. While these differences were generally minor, the older studies (which were also studies that enrolled patients with only prior cytokine therapy) such as TARGET and VEG105192, enrolled more patients with poorer risk; assuming that these differences impact the relative effects of treatment, this could have introduced bias into the base-case analyses and is explored in the second sensitivity analysis. Between the clinical heterogeneity present and the lack of ability to estimate statistical heterogeneity, generalizations from the extant literature on these questions are limited.
The second limitation, somewhat connected to the first, concerns the equivalence (or not) of mTORs. To explore any bias introduced by including trials enrolling patients with only prior cytokines, the mTORs were assumed equivalent in the sensitivity analyses so as to retain a connected evidence network. While there is some suggestion they may be similar (i.e., that the molecules differ structurally only in an ester moiety) the difference in results we found for dose reduction and discontinuations suggest that this may not be the case. For the safety analyses in particular, we can conclude that there is some heterogeneity; however, because there are at least three sources of bias (receipt of prior cytokines in the base case, combining mTORs in the sensitivity analyses, and time-dependent physician experience with the drugs) it is not possible to fully account for the differences.
The results of this network meta-analysis show that OS is nearly identical for sorafenib and a number of comparators, including axitinib, dovitinib, everolimus, pazopanib, and, in sensitivity analysis, cabozantinib, and nivolumab, as well. This result is supported by another published meta-analysis in which sorafenib showed a significantly better OS than temsirolimus.25 However, sorafenib had more dose reductions and discontinuations compared with newer treatments (usually related to stomatitis and hand foot syndrome, two toxicities that are still resistant to supportive care interventions), but had comparable efficacy to the mTOR inhibitors, everolimus and temsirolimus. These results suggest that in resource-restricted environments, health authorities should require that high-cost newer agents, such as cabozantinib and nivo-lumab, be compared in Phase III trials to sorafenib, a lower-cost older agent, before approving the newer agents.
1. Verzoni E, de Braud F, Fabiani F, Grassi P, Testa I, Procopio G. Patient approach in advanced/metastatic renal cell carcinoma: focus on the elderly population and treatment-related toxicity. Future Oncol. 2013; 9(11):1599-1607.
2. European Association of Urology (EAU). Guidelines: Renal Cell Carcinoma. 2016; http://uroweb.org/guideline/renal-cell-carcinoma/#3. Accessed 16 August 2016.
3. US Food and Drug Administration. Approved drugs: Nivolumab (Opdivo Injection). November 23, 2015; https://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm474092.htm.
4. Motzer RJ, Escudier B, McDermott DF, et al. Nivolumab versus Everolimus in Advanced Renal-Cell Carcinoma. N Engl J Med. 2015;373 (19):1803-1813.
5. US Food and Drug Administration. Approved drugs: Lenvatinib in combination with Everolimus. May 13, 2016; https://www.fda.gov/ Drugs/InformationOnDrugs/ApprovedDrugs/ucm501070.htm.
6. Motzer RJ, Hutson TE, Glen H, et al. Lenvatinib, everolimus, and the combination in patients with metastatic renal cell carcinoma: a randomised, phase 2, open-label, multicentre trial. Lancet Oncol. 2015; 16(15):1473-1482.
7. US Food and Drug Administration. Approved drugs: Cabozantinib (CABOMETYX). April 25, 2016; https://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm497483.htm.
8. Choueiri TK, Escudier B, Powles T, et al. Cabozantinib versus Evero-limus in Advanced Renal-Cell Carcinoma. N Engl J Med. 2015;373 (19):1814-1823.
9. Escudier B, Porta C, Schmidinger M, et al. Renal cell carcinoma: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol. 2014;25 Suppl 3:iii49-56.
10. Sun M, Larcher A, Karakiewicz PI. Optimal first-line and second-line treatments for metastatic renal cell carcinoma: current evidence. Int J Nephrol Renovasc Dis. 2014;2014(7):401-407.
11. National Comprehensive Cancer Network (NCCN). NCCN Clinical Practice Guidelines in Oncology: Kidney Cancer, Version 2.2017. October 2016; https://www.nccn.org/professionals/physician_gls/pdf/kidney.pdf.
12. Motzer RJ, Escudier B, Oudard S, et al. Phase 3 trial of everolimus for metastatic renal cell carcinoma: final results and analysis of prognostic factors. Cancer. 2010;116(18):4256-4265.
13. Wahlgren T, Harmenberg U, Sandstrom P, et al. Treatment and overall survival in renal cell carcinoma: a Swedish population-based study (2000-2008). Br J Cancer. 2013;108(7):1541-1549.
14. Soerensen AV, Donskov F, Hermann GG, et al. Improved overall survival after implementation of targeted therapy for patients with metastatic renal cell carcinoma: results from the Danish Renal Cancer Group (DARENCA) study-2. Eur J Cancer. 2014;50(3):553-562.
15. Akaza H, Fukuyama T. Axitinib for the treatment of advanced renal cell carcinoma. Expert Opin Pharmacother. 2014;15(2):283-297.
16. Rini BI. Changing Landscape of Refractory Renal Cell Carcinoma. J Oncol Pract. 2016;12(5):422-423.
17. Pal SK, Ghate S, Li N, et al. Real-world survival outcomes and prognostic factors among patients receiving first targeted therapy for advanced renal cell carcinoma: a SEER-Medicare database analysis. Submitted manuscript.
18. Motzer RJ, Porta C, Vogelzang NJ, et al. Dovitinib versus sorafenib for third-line targeted treatment of patients with metastatic renal cell carcinoma: an open-label, randomised phase 3 trial. Lancet Oncol. 2014; 15(3):286-296.
19. Escudier B, Eisen T, Stadler WM, et al. Sorafenib for treatment of renal cell carcinoma: Final efficacy and safety results of the phase III treatment approaches in renal cancer global evaluation trial. J Clin Oncol. 2009;27(20):3312-3318.
20. Sternberg CN, Hawkins RE, Wagstaff J, et al. A randomised, double-blind phase III study of pazopanib in patients with advanced and/or metastatic renal cell carcinoma: final overall survival results and safety update. Eur J Cancer. 2013;49(6):1287-1296.
21. Rini BI, Escudier B, Tomczak P, et al. Comparative effectiveness of axitinib versus sorafenib in advanced renal cell carcinoma (AXIS): a randomised phase 3 trial. Lancet. 2011;378(9807):1931-1939.
22. Hutson TE, Escudier B, Esteban E, et al. Randomized phase III trial of temsirolimus versus sorafenib as second-line therapy after sunitinib in patients with metastatic renal cell carcinoma. J Clin Oncol. 2014; 32(8):760-767.
23. Becker A, Eichelberg C, Sun M. Progression-free survival: does a correlation with survival justify its role as a surrogate clinical endpoint? Cancer. 2014;120(1):7-10.
24. Iacovelli R, Verri E, Cossu Rocca M, et al. Is there still a role for sorafenib in metastatic renal cell carcinoma? A systematic review and meta-analysis of the effectiveness of sorafenib over other targeted agents. Crit Rev Oncol Hematol. 2016;99:324-331.
25. Motzer RJ, Nosov D, Eisen T, et al. Tivozanib versus sorafenib as initial targeted therapy for patients with metastatic renal cell carcinoma: results from a phase III trial. J Clin Oncol. 2013;31(30):3791-3799.
1Medical Oncologist, Comprehensive Cancer Centers of Nevada; Chair, Genitourinary Committee, US Oncology Research; Network Vice Chair, Genitourinary Committee, Southwest Oncology Group Las Vegas, Nevada; 2Professor, Barts Cancer Institute, West Smithfield, London, EC1A 7BE, United Kingdom; 3Director of the Urologic Oncology Program; Co-Chair of the Urologic Cancer Research and Treatment Center at Baylor University Medical Center; Professor of Medicine at Texas A&M College of Medicine; Texas Oncology-Baylor Charles A. Sammons Cancer Center, Dallas, Texas; 4Senior Research Associate, Evidera, Waltham, Massachusetts; 5Principal Biostatistician, Evidera, Waltham, Massachusetts; 6Research Scientist, Evidera, Waltham, Massachusetts 7Deputy Director, USMA–Data Generation and Observational Studies, Bayer Healthcare Pharmaceuticals Whippany, New Jersey; 8Deputy Director, HEOR Field Team, USMA–Data Generation and Observational Studies, Bayer Healthcare Pharmaceuticals, Whippany, New Jersey; 9Medical Director, Oncology, US Medical Affairs, Bayer HealthCare Pharmaceuticals Inc.,Whippany, New Jersey KCJ