Background
Each year, approximately 214,000 patients are diagnosed with renal cell carcinoma (RCC) worldwide [
1]. In 25–30% of these patients, the disease is already at the metastatic stage at presentation [
2], which may be explained, in part, by lack of early symptoms for metastatic RCC (mRCC). The 5-year survival rate by American Joint Committee on Cancer (AJCC) tumour, lymph nodes, and metastasis (TNM) staging is 8% for Stage IV RCC [
3]. The treatment landscape for patients with mRCC has evolved substantially in the past decade with the introduction of targeted therapies, which has led to significant improvements in patient outcomes.
For patients with mRCC who progress on prior targeted therapy, mostly with sunitinib, treatment options available as second-line and/or subsequent systemic therapies include axitinib, everolimus, lenvatinib in combination with everolimus, cabozantinib, and nivolumab. With an increasing number of targeted agents being approved for the treatment of mRCC, data from comparative studies of these agents would help attending physicians and patients to make decisions based on individualized treatment algorithms. Due to the limited number of head-to-head clinical trials that directly evaluate these targeted agents as second-line and/or subsequent therapy for mRCC, an indirect treatment comparison analysis is needed. Axitinib was a well-established targeted agent in the second-line setting before other targeted therapies were introduced. Axitinib has been compared with sorafenib in the Axitinib as Second Line Therapy for Metastatic Renal Cell Carcinoma (AXIS) trial [
4,
5]. Everolimus has been compared with cabozantinib in the Cabozantinib versus Everolimus in Advanced Renal Cell Carcinoma (METEOR) trial [
6,
7], and with placebo in the Renal Cell Cancer Treatment with Oral RAD001 given Daily (RECORD-1) trial [
8]. A network meta-analysis (NMA) is an indirect treatment comparison technique commonly used to estimate relative treatment effects based on published data from different trials. This technique relies on the availability of a connected network to contrast relative effects between treatments, and assumes homogeneity of trials included in the network. An NMA is not a suitable method to perform a comparison of axitinib with everolimus or cabozantinib, since there is no comparator that links axitinib to either of these two agents in a population of patients who received prior sunitinib. Furthermore, there are important differences in parameters, such as patient baseline characteristics, observed between the trials that are available to create a connected network. In such a case, an alternative indirect treatment comparison approach, such as a matching-adjusted indirect comparison (MAIC) [
9], may be more appropriate.
The MAIC technique has been acknowledged in health technology assessments in oncology [
10,
11] and applied to generate comparative evidence in several diseases [
12‐
17]. Unlike a naïve indirect comparison that is based on the observed outcomes of two arms across different trials without adjustment for baseline differences (and therefore subject to confounding by both the observed and unobserved baseline differences between the trials), MAIC analyses compare treatments using information from compatible studies while adjusting for differences in the population characteristics across trials. Patient-level data from one trial (the index trial) are adjusted to the baseline characteristics of the comparator trial, thereby making it possible to compare outcomes across trials.
The aim of this study was to compare progression-free survival (PFS) and overall survival (OS) in prior sunitinib-treated patients who received axitinib in the AXIS trial [
4] with prior sunitinib-treated patients who received either everolimus or cabozantinib in the METEOR trial [
6,
7], using MAIC analyses.
Discussion
For patients with mRCC who progressed on first-line sunitinib, several treatment options with other targeted agents are available as second-line or subsequent therapies. However, evidence on their relative efficacy is limited due to a lack of head-to-head trials to guide the optimal choice of therapy. The current study was conducted with the aim of comparing PFS and OS for axitinib versus cabozantinib or everolimus in patients with mRCC who were previously treated with sunitinib. To date, the AXIS trial is the only head-to-head phase 3 trial that evaluated axitinib against a comparator, sorafenib, which was the standard of care in the second-line setting at the time of the trial [
4]. Other phase 3 trials conducted in the second-line setting, such as RECORD-1 (everolimus vs. placebo) [
8], METEOR (cabozantinib vs. everolimus) [
6,
7], and CHECKMATE025 (nivolumab vs. everolimus) [
6], used everolimus as a comparator. Since a standard mixed treatment comparison was not possible due to a disjointed network in the prior sunitinib subgroup, an MAIC comparison was necessary to determine the comparative efficacy between axitinib and everolimus, and between axitinib and cabozantinib.
Using the MAIC analyses [
9], prior sunitinib-treated patient subgroups were compared across the AXIS and METEOR trials, adjusting for differences in baseline patient characteristics. In the base-case analysis, all patient characteristics, including MSKCC score derived in AXIS assuming ECOG PS 1 was a risk factor, were used as matching variables. In the sensitivity analysis, the same patient characteristics were used, except MSKCC score in AXIS was derived assuming ECOG PS 1 was not a risk factor. When comparing axitinib with everolimus, a statistically significant advantage in PFS for axitinib versus everolimus was observed in both the base-case (
p = 0.002) and sensitivity (
p = 0.007) analyses. The benefit of axitinib compared with everolimus was also seen for OS in the base-case analysis (
p = 0.032), but not in the sensitivity analysis (
p = 0.309). When comparing axitinib with cabozantinib, the base-case analysis suggested no difference in PFS (
p = 0.423) or OS (
p = 0.983) between the two treatments. The sensitivity analysis suggested a marginal benefit for cabozantinib versus axitinib for PFS (
p = 0.050); however, the differences in OS did not reach statistical significance (
p = 0.096). Different conclusions in the base-case and sensitivity analyses for comparison of axitinib with cabozantinib arise from differences in how the MSKCC score was calculated in AXIS in these two analyses. The disparities between the base-case and sensitivity analyses in this study underscore the importance of adjusting for the differences in baseline characteristics, and that the naïve indirect comparisons are potentially problematic and should be avoided.
The OS results in the current analyses should nevertheless be interpreted with caution due to the difference in maturity of the OS data across the trials. In addition, OS is impacted by study treatments and by post-study treatments. Indeed, a higher percentage of axitinib-treated patients in the AXIS trial received post-study therapies compared with everolimus- or cabozantinib-treated patients in the METEOR trial (60, 55, and 48%, respectively); this may be explained, in part, by the difference in the maturity of the trials. The most common post-study systemic treatments in axitinib-treated patients were everolimus (43%) and sorafenib (20%). In the METEOR trial, the most common post-study treatments were axitinib (39%), followed by either pazopanib (7%) or sorafenib (7%) in everolimus-treated patients; and everolimus (25%) and axitinib (18%) in cabozantinib-treated patients. The OS results would have been confounded by the observed imbalance in the composition of subsequent systemic therapies between the trials, which could not be corrected for, and, more importantly, the fact that 39% of everolimus-treated patients and 18% of cabozantinib-treated patients in the METEOR trial received post-study axitinib, and 43% of axitinib-treated patients in the AXIS trial received post-study everolimus (none received cabozantinib).
The study results are not in agreement with the findings of three studies that compared survival between axitinib, everolimus, and cabozantinib in the second-line setting using different analytical methods [
22‐
24]. In an NMA study, Amzal et al. [
23] reported HRs in favour of cabozantinib versus axitinib for PFS (2.13 [95% CI 1.32–3.43]) and OS (1.96 [95% CI 0.68–5.7]). Sherman et al. [
24] conducted a weight-adjusted indirect comparison of prior sunitinib-treated patients with second-line mRCC treated with everolimus from the RECORD-1 trial and axitinib-treated patients from the AXIS trial, and found no statistically significant differences between axitinib and everolimus in PFS; however, the sample size was small (
n = 43) and included patients who were sunitinib-intolerant. Similarly, in a retrospective chart review of the axitinib versus everolimus cohort study by Vogelzang et al. [
22], no differences in OS and PFS were observed between axitinib and everolimus in the overall study population. However, subgroup analyses suggested a significant OS benefit with axitinib among patients who had received sunitinib or sorafenib as first-line treatment for < 6 months.
There are inherent limitations to our study. Although the clinical trial design, inclusion criteria, and outcomes definitions were comparable between the AXIS and METEOR trials, some differences were noted that could have potentially impacted the comparison. Firstly, differences in timing of PFS assessments by an independent review committee may have led to overestimation of PFS for everolimus and cabozantinib (METEOR: 8 weeks for the first 12 months, and every 12 weeks thereafter) compared with axitinib (AXIS: after 6 and 12 weeks of therapy, and every 8 weeks thereafter). Secondly, the difference between the MSKCC scores derived in the AXIS versus METEOR trial could not be fully adjusted for in our analyses. Thirdly, there was a potential for residual confounding due to omitting some patient characteristics from analyses (e.g., number of metastases, duration of the prior sunitinib treatment, and time since diagnosis), since these were not available for the prior sunitinib-treated patient subgroups in the METEOR trial. Although the current study adjusted for the main prognostic factors [
25‐
27], the impact of excluding other characteristics from the analyses is unclear. Finally, differences in subsequent therapies, including the high percentages of everolimus- and cabozantinib-treated patients who received axitinib post-study in the METEOR trial, could not be accounted for when analysing the OS data.
Despite these limitations, the MAIC analysis offers advantages over a naïve comparison since it attempts to adjust for differences in baseline patient characteristics, and thus allows for a more comprehensive comparison of the treatment effects.
Conclusions
Although assumptions were required, an indirect comparison using MAIC based on the AXIS and METEOR trials suggested no differences in PFS and OS in the base-case analysis between axitinib and cabozantinib in prior sunitinib-treated patients. Sensitivity analyses suggested a significant PFS gain with cabozantinib compared with axitinib; however, no significant difference in OS was observed. For axitinib versus everolimus, a beneficial treatment effect was observed for PFS, and potentially for OS, in patients with mRCC previously treated with sunitinib. Disparities in the base-case and sensitivity analyses between axitinib and cabozantinib highlight the importance of adjusting for differences in trial populations for indirect treatment comparisons. Additionally, these analyses demonstrate that MAIC can improve the reliability of indirect comparisons compared with using aggregate data alone; however, a randomised, head-to-head, controlled trial is needed if definitive conclusions are to be made.