Less is More? First Impressions from COSMIC-313

Abstract

The COSMIC-313 phase 3 randomized controlled trial (RCT) tested the triplet combination of cabozantinib with nivolumab and ipilimumab in comparison with nivolumab plus ipilimumab control as fist-line systemic therapy in metastatic clear cell renal cell carcinoma. The first results presented at the 2022 European Society of Medical Oncology (ESMO) Congress are a milestone for the renal cell carcinoma field because they signal the advent of triplet combinations as potential treatment options for our patients. The present commentary highlights some considerations and potential next steps based on these first impressions.

The first results of COSMIC-313 presented at the European Society of Medical Oncology (ESMO) 2022 Congress in Paris are an exciting development for renal cell carcinoma as they represent the first phase 3 readout of a triplet systemic therapy combination for patients with advanced or metastatic clear cell renal cell carcinoma (mccRCC) (1). COSMIC-313 is a phase 3 randomized controlled trial (RCT) that enrolled a total of 855 systemic treatment-naive patients with IMDC intermediate or poor risk mccRCC (2) and randomized them to either the triplet combination of cabozantinib + nivolumab + ipilimumab (n=428 patients) or placebo + nivolumab + ipilimumab (n=427). The rationale for testing this triplet was based on the potential immunomodulatory properties of cabozantinib (3,4). Furthermore, cabozantinib has already received regulatory approval in combination with nivolumab after demonstrating improved overall survival (OS), progression-free survival (PFS) and objective response rates (ORR) compared with sunitinib monotherapy in treatment-naïve patients with IMDC favorable, intermediate or poor risk mccRCC in the CheckMate 9ER phase 3 RCT (5). Interestingly, the primary endpoint of COSMIC-313 was PFS by blinded independent radiology review per Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 only in the first 550 randomized patients. This will reduce the precision of the primary endpoint estimates (6). All randomized patients will instead be used for estimating the secondary endpoint of OS. Additional endpoints of interest include ORR and safety (1).

In this first readout of COSMIC-313, the median patient follow-up is less than 24 months and the study has yielded a hazard ratio (HR) = 0.73 (95% CI 0.57 – 0.94, p = 0.013) for its primary endpoint of PFS (1). This relative treatment effect efficacy signal is intriguing but nevertheless underwhelming. The first thing to consider is that the relative treatment effect reported in single RCTs, such as COSMIC-313, is typically very likely to be exaggerated (7,8). To account for this exaggeration, we can shrink the unbiased reported frequentist HR estimator of COSMIC-313 using a Bayesian shrinkage prior distribution derived from the primary results of 23,551 medical RCTs of relative treatment efficacy that are available in the Cochrane Database of Systematic Reviews (9). This yields posterior mean HR = 0.81 with 95% credible intervals of 0.63 – 1.01 for improved PFS with the triplet combination compared with the control arm in COSMIC-313. Moreover, the posterior probability that the placebo control group in COSMIC-313 had better PFS than the triplet combination would be 3.1%, a small but not negligible possibility. Another way to gauge frailty of the reported PFS signal is to convert its p-value into binary surprisal thus providing a measure of the information supplied by the data (10). COSMIC-313 supplied for its primary endpoint only 6 bits of information against the tested model which includes the null hypothesis of no PFS difference between triplet combination vs control. Furthermore, the reported marginal ORR values of 43% vs 36% in the first 550 randomized patients are also consistent with weaker than expected response benefit for the triplet combination compared with the control group (1). Prior experience in oncology RCTs, including RCTs of immune checkpoint therapy combination with multi-targeted tyrosine kinase inhibitors (TKIs) such as cabozantinib, consistently indicates that the underlying mechanism behind the observed relative treatment efficacy in COSMIC-313 is independent drug action and not synergy or additivity (11,12). This independent drug action means that each patient being treated with a combination therapy such as the triplet of cabozantinib + nivolumab + ipilimumab would be expected to respond only to one of the drugs in the combination and not to any of the other drugs (11–13). Accordingly, the lower starting dose of cabozantinib 40 mg daily in the triplet combination, compared with typical monotherapy starting dose of cabozantinib 60 mg daily, may at least partly explain the limited efficacy signal of the triplet combination compared with the control arm. Additional dose reductions due to toxicity from the triplet therapy may have further impacted efficacy. Another explanation is that COSMIC-313 is the first trial in mccRCC to use a combination immunotherapy regimen, such as the nivolumab + ipilimumab doublet, as the control arm. Previous pivotal phase 3 RCTs of immunotherapy combinations in mccRCC typically used the earlier generation TKI sunitinib as the single-agent control arm (5,14–17). The superior relative treatment efficacy in PFS, OS, and ORR of the nivolumab + ipilimumab doublet over sunitinib is well established in treatment-naive patients with IMDC intermediate or poor risk mccRCC (18). For context, per the same Bayesian shrinkage approach we used for COSMIC-313, the posterior mean HR is 0.80 with 95% credible intervals of 0.66 – 0.95 for improved PFS with the nivolumab plus ipilimumab doublet compared with sunitinib. Moreover, the posterior probability that the sunitinib control group had better PFS than the doublet combination would be only 0.3%. However, it should be noted that these estimates are based on a far more mature follow-up of at least 4 years (18).

OS is a key secondary endpoint and its estimates will need to be carefully evaluated. Intermediate endpoints such as PFS are nowadays more reliable as primary endpoints for oncology phase 3 RCTs because they more directly benefit from the random treatment assignment whereas OS estimates are more likely to be biased by mediator-outcome confounding (19–22). However, barring extreme scenarios such as crossover (22), we have not yet reached the point in most oncology phase 3 settings whereby the biases in OS estimation due to the effect of subsequent therapies would be expected to reverse the conclusions between PFS and OS signals. Therefore, we should expect the OS estimates of COSMIC-313 to either show a positive signal in favor of the triplet combination or at least be inconclusive. It would certainly be worrisome if the OS signal instead favors the control group. Future pivotal RCTs should carefully collect data not only on the number and type of subsequent therapies but also on mediator-outcome confounders that may influence how these therapies were chosen. This will facilitate the more reliable estimation of OS as part of dynamic treatment regimes tailored towards improving survival, preserving quality of life, and minimizing logistical, financial and other costs for each individual patient (23–25).

It is a common error, known as the Table 1 fallacy, to scan for imbalances between the treatment groups that could explain the observed relative treatment effect estimates in RCTs such as COSMIC-313 (26). The randomization of treatment assignment, i.e., the most powerful inferential feature of RCTs, is in fact expected to generate such imbalances (20,26–28). Another common practice is to look for forest plots illustrating potential subgroup differences in relative treatment efficacy on the HR scale (25). However, such comparisons for “predictive” biomarkers typically yield inconclusive signals (29). Accordingly, PFS estimates on the HR scale in the poor IMDC risk patient subgroup were inconclusive with 95% CIs ranging from 0.65 to 1.69. Conversely, IMDC risk is far more likely to be clinically actionable when used as a prognostic biomarker (25,30–34). Accordingly, should the triplet combination of cabozantinib + nivolumab + ipilimumab obtain regulatory approval based on the results of COSMIC-313, it would be a reasonable choice for treatment-naïve patients with poor prognosis mccRCC that are unlikely to receive a second-line therapy if their cancer does not respond to the first treatment choice (12). Furthermore, the effect of cabozantinib in the triplet combination is more likely to yield earlier responses compared with the nivolumab + ipilimumab doublet, which will be particularly useful for patients with highly symptomatic aggressive disease (35). It is far less clear, however, whether the triplet combination will yield earlier responses compared with TKI-based immunotherapy doublets such as pembrolizumab + lenvatinib, which can be used with lenvatinib starting doses as high as 20 mg daily thus yielding a potent TKI effect (17). Another consideration is that metastatic sites such as brain and bone metastases may particularly benefit from the cabozantinib component in the triplet based on preclinical models and clinical observations (36–38). More data are needed on the relative benefits of cabozantinib as part of a triplet, doublet or monotherapy regimen in these settings. Furthermore, the lack of robust phase 3 data on non-clear cell renal cell carcinoma histologies compels us to rely more on mechanistic considerations using contextual biological knowledge to inform our clinical decisions (31,39,40). We do have evidence indicating that certain renal cell carcinoma histologies are differentially sensitive to each of the cabozantinib and the immune checkpoint inhibitor components in the triplet combination (41,42). However, experimental clinical data from carefully designed clinical trials are irreplaceable and sorely needed.

The adverse events and impact on patient-reported quality of life will also greatly influence treatment choices when considering triplet combinations on mccRCC. Unfortunately, adverse events in oncology RCTs are typically presented as marginal estimates for the whole treatment and control groups, which have questionable value when considering transporting the RCT results to specific patients seen in clinic (43). The advent of triplet therapy combinations for mccRCC, and perhaps even quadruple combinations in the future, necessitates more emphasis on comprehensive adverse event reporting in RCTs, e.g., by presenting the covariate-specific causal risk differences (44,45) or by optimizing visual displays (46). From the currently available information from COSMIC-313, we can at least surmise that the observed marginal grade ≥3 treatment-related adverse event (TRAE) frequency of 73% in the triplet arm is consistent with what has previously been observed with combinations of TKIs with immune checkpoint therapies in mccRCC (5,47). Furthermore, the TRAE frequency of 41% in the control arm is consistent with what has previously been observed with nivolumab plus ipilimumab in patients with mccRCC (47).

In summary, the COSMIC-313 study team and participants should be commended for successfully conducting and reporting the first phase 3 RCT readout of a triplet combination in mccRCC. While there were no unexpected adverse event signals, the PFS and ORR benefit from the triplet combination is not markedly better than the control arm. It will be very interesting to see the first full published results, how the OS and PFS signals mature over time, as well as how regulators respond to these data.