Exposure–Response Analysis to Support Nivolumab Once Every 4 Weeks Dosing in Combination with Cabozantinib in Renal Cell Carcinoma

Renal cell carcinoma (RCC) occurs globally in more than 400,000 people per year with approximately 40% succumbing to the disease annually (1). Therapies to treat RCC that focus on targets involved in cell growth/survival, angiogenesis, and tumor immunosuppression include mTOR inhibitors (everolimus and temsirolimus), VEGF pathway inhibitors [bevacizumab (a VEGF-ligand inhibitor); pazopanib, axitinib, sorafenib, sunitinib, lenvatinib, and cabozantinib (multitargeted tyrosine kinase inhibitors; TKI)], and the checkpoint inhibitor nivolumab (2). Recent advancements in RCC treatment that have had significant impact include combination therapies using dual checkpoint inhibitors (nivolumab + ipilimumab) and checkpoint inhibitors with multitargeted VEGF-pathway TKIs (nivolumab + cabozantinib, pembrolizumab + axitinib, avelumab + axitinib, pembrolizumab + lenvatinib; ref. 3–6).

Nivolumab is a fully human IgG4 programmed death-1 immune checkpoint inhibitor antibody that acts to reverse immune cell exhaustion and immunosuppression in the tumor microenvironment (7). An initial nivolumab monotherapy dosing regimen of 3 mg/kg every 2 weeks was approved in the United States and several other regions based on clinical data from several different clinical trials in advanced melanoma, RCC, non–small cell lung cancer, squamous cell cancer of the head and neck, colorectal cancer, classical Hodgkin's lymphoma, urothelial carcinoma, and hepatocellular carcinoma (8, 9). A model-based analysis was conducted post approval to demonstrate clinically comparable pharmacokinetics, safety, and efficacy of the 3 mg/kg bodyweight-based dose to 240 mg every 2 weeks or 480 mg every 4 weeks dosing, allowing flexibility and convenience to patients and physicians (10–12).

Cabozantinib is a potent inhibitor of multiple receptor tyrosine kinases known to play a key role in tumor cell proliferation and/or tumor vascularization, including VEGF receptors, mesenchymal–epithelial transition kinase, and the TAM (TYRO-3, AXL, and MER) family of receptor kinases (13, 14). Inactivation of the von Hippel-Lindau tumor suppressor protein, which occurs in slightly over half of patients with clear cell RCC, results in upregulation of cabozantinib targets: VEGF, mesenchymal–epithelial transition kinase, and AXL (2, 15). Targets of cabozantinib are also implicated in promoting immunosuppressive effects and represent an opportunity for synergistic effects with immune checkpoint inhibitors, such as nivolumab (16, 17). Cabozantinib is approved as a single agent at 60 mg once daily in previously untreated patients with RCC and patients with hepatocellular carcinoma who have been previously treated with sorafenib (13). Cabozantinib dose selection for CheckMate 9ER was based on a phase I dose-ranging study of nivolumab + cabozantinib in patients with advanced genitourinary cancers. A lower cabozantinib dose of 40 mg every day, when combined with nivolumab, had fewer cabozantinib dose reductions than the 60 mg every day dose (33% vs. 75% dose reductions, respectively) and a more tolerable safety profile with comparable efficacy (18), and thus was selected as the cabozantinib dose for CheckMate 9ER.

CheckMate 9ER is an open-label phase III trial evaluating previously untreated patients with clear cell, advanced RCC who were randomized to receive nivolumab 240 mg every 2 weeks + cabozantinib 40 mg every day or sunitinib 50 mg every day for 4 weeks of each 6-week cycle. Nivolumab + cabozantinib had significant progression-free survival (PFS; primary efficacy endpoint) and overall survival (OS; secondary efficacy endpoint) benefits over sunitinib, with a PFS HR of 0.51 [95% confidence interval (CI), 0.41–0.64; P < 0.001] and an OS HR of 0.60 (98.89% CI, 0.40–0.89; P = 0.001) with a manageable safety profile (19). The 240 mg every 2 weeks regimen was selected for the trial as the nivolumab 480 mg every 4 weeks dosing regimen was not approved for RCC at the time of initiation of CheckMate 9ER. Given prior experience with nivolumab single-agent therapy in RCC and the extensive dose-ranging (0.3–10 mg/kg) data available, it was reasoned that model-informed analysis to predict the efficacy and safety of nivolumab 480 mg every 4 weeks + cabozantinib 40 mg every day could be conducted to support health authority approval of this alternative dosing regimen.

Nivolumab pharmacokinetics have been well characterized in several tumor types and treatment settings, including RCC (20–22). Nivolumab exposure–response (E–R) relationships for efficacy and safety have been established in advanced metastatic tumor types where dose-ranging clinical efficacy data were available (11, 23). In general, nivolumab has a flat E–R for efficacy over a broad dose range (0.3–10 mg/kg) in several tumor types, including RCC, for PFS, OS, and tumor objective response rate (ORR). Nivolumab exposure had the strongest association with grade ≥2 immune-mediated adverse events (IMAE) of the safety endpoints evaluated, including grade ≥3 drug-related adverse events (AE) and treatment-related AEs leading to discontinuation, and deaths (11). This is to be expected given the mechanism of action of nivolumab. The observation that nivolumab clearance demonstrates reduction over time highlighted the need to use early measures of nivolumab exposure (e.g., after the first dose) in order to avoid false-positive bias in the assessment of E–R efficacy relationships (21, 23, 24). Also, baseline nivolumab clearance—independent of exposure, which appears to be a surrogate measure of disease status—is often added as a covariate in E–R efficacy analyses when dose-ranging data are available due to its strong association with efficacy independent of dose (10, 23).

The nivolumab E–R analyses of efficacy (PFS/OS) and safety (grade ≥2 IMAEs) reported here were conducted to assess the relationships between nivolumab exposure and PFS/OS and grade ≥2 IMAEs. Prespecified covariates and potential interactions between nivolumab exposure and cabozantinib coadministration were assessed to inform on the contribution of cabozantinib to efficacy and safety and to predict PFS/OS and grade ≥2 IMAEs for the nivolumab 480 mg every 4 weeks + cabozantinib 40 mg every day posology. This work confirmed a flat nivolumab E–R in the context of cabozantinib coadministration and provided a framework to evaluate the benefit:risk of nivolumab 480 mg every 4 weeks + cabozantinib that enabled regulatory approval in the United States and the European Union without the need to conduct an additional clinical study.

CheckMate 9ER (CA2099ER; NCT03141177) is a pivotal phase III clinical trial in which nivolumab was administered (240 mg every 2 weeks by intravenous infusion) in combination with cabozantinib (40 mg every day orally) in previously untreated patients with clear cell, advanced RCC. For E–R efficacy (PFS/OS) analysis, CheckMate 9ER data were combined with data from four previous nivolumab monotherapy clinical trials in the previously treated RCC setting: CA209003 (NCT00730639; phase I dose-ranging study in solid tumors), CA209009 (NCT01358721; phase I dose-ranging biomarker study in previously treated RCC), CA209010 (NCT01354431; phase II dose-ranging trial in previously treated RCC), and CheckMate 025 (CA209025; NCT01668784; pivotal phase III trial comparing nivolumab vs. everolimus in previously treated patients with RCC); clinical trial details are provided in Supplementary Table S1. The nivolumab monotherapy studies covered exposure ranges of 0.3 to 10 mg/kg every 2 weeks or every 3 weeks. The same study data used for E–R efficacy were used for E–R safety (grade ≥2 IMAEs), apart from CA209009, which was excluded as the composite safety endpoint grade ≥2 IMAEs was not derived in the study.

For the E–R efficacy analysis, nivolumab time-averaged concentration over the first dosing interval (C_avg1) was used as the exposure measure and was defined as the area under the curve after the first dose divided by the dosing interval. Nivolumab C_avg1 was selected to avoid the exposure bias due to nivolumab time-varying clearance and serves as a surrogate measure of efficacy throughout the treatment duration. Nivolumab C_avg1 was obtained by predicting the nivolumab concentration-time profiles using individual empirical Bayes estimates of pharmacokinetic parameters for each patient, which were derived from the population pharmacokinetic analysis based on a previous population pharmacokinetic model (20).

For E–R safety (grade ≥2 IMAEs) analysis, nivolumab time-varying daily C_avg (time-averaged concentration over 24 hours) was used as the exposure measure and obtained by predicting nivolumab concentration-time profiles using actual dosing records and empirical Bayes estimates of pharmacokinetic parameter estimates. Daily C_avg was selected for E–R safety analysis since it captures both peak and overall exposure differences produced by 480 mg every 4 weeks versus 240 mg every 2 weeks. Previous nivolumab E–R safety analyses showed a high degree of concordance between daily C_avg and maximum concentration (C_max; ref. 11). Nivolumab daily C_avg was calculated from day 1 to the day of the event/censor for each patient. Censor was defined as the minimum of 100 days after end of treatment or date last known alive.

The E–R efficacy and safety analysis was characterized with respect to time to PFS/OS event and time to first grade ≥2 IMAEs, respectively, using semiparametric Cox proportional hazards models. The functional form of nivolumab exposure (linear or log-linear) was assessed in a full model that included prespecified covariate effects of interest to account for confounding effects of covariates on E–R, and the most parsimonious model was selected based on the lowest Bayesian information criterion (BIC) value. Multiplicative interactions between nivolumab exposure and significant predictors in the full model were evaluated univariately. Covariate effects were interpreted in the full model and lack of a statistically significant effect was concluded if unity was within the 95% CI of the HR. A parsimonious final model was obtained by backward elimination of covariates not significant by BIC criteria, and the final model was used for model predictions. Full model development was designed to account for the potential modulatory effect of preselected covariates on the E–R relationship of nivolumab C_avg1 on PFS, and to estimate the effect of these covariates on PFS. Goodness-of-fit and model performance was assessed by visual predictive check and five-fold cross-validation (details provided in the Supplementary Methods).

The analysis included 1,009 patients with RCC who received nivolumab monotherapy or nivolumab 240 mg every 2 weeks + cabozantinib (N = 315 patients from CheckMate 9ER) for whom estimates of nivolumab exposure (C_avg1) were available. The percentage of treated patients included in the analysis was 98.4% (315/320) from CheckMate 9ER and 99.4% (694/698) from nivolumab monotherapy trials. Five patients from CheckMate 9ER and 4 patients from nivolumab monotherapy studies were excluded from the analysis due to missing nivolumab exposures. The response variable in the E–R efficacy analysis was OS and blinded independent central review (BICR)-assessed PFS for CheckMate 9ER and investigator-assessed PFS for all nivolumab monotherapy studies. BICR and investigator-assessed PFS were similar for CheckMate 9ER, with PFS HRs (95% CI) of 0.51 (0.41–0.64) and 0.46 (0.36–0.57), respectively, providing justification for combining CheckMate 9ER with previous nivolumab monotherapy data.

The exposure variable was nivolumab C_avg1 (continuous). The response variable was time to disease progression or death event and time to death event. Baseline nivolumab clearance (continuous) was included. Baseline demographic variables were age (continuous), sex [male, female (reference)], bodyweight (continuous), and geographic region [rest of world, Western Europe, United States/Canada (reference)]. The baseline clinical laboratory variable was albumin (continuous). Baseline disease characteristics were International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) score [poor, intermediate, favorable (reference)], programmed death ligand 1 (PD-L1) expression [PD-L1≥1% positive, PD-L1 ≥1% negative (reference)], and performance status [PS; derived from Karnofsky PS score (PS 0 = KPS 100, PS>0 = KPS<100; ref. 25; PS > 0, PS = 0 (reference)]. Cabozantinib coadministration [yes, no (reference)] was also evaluated. HRs were calculated using the final E–R PFS/OS models for nivolumab 240 mg every 2 weeks + cabozantinib at the median, 5th, and 95th percentile of nivolumab exposure relative to the median exposure of nivolumab 3 mg/kg every 2 weeks.

The analysis included 919 patients with RCC who received nivolumab monotherapy or nivolumab in combination with cabozantinib (N = 315 patients from CheckMate 9ER) for whom estimates of nivolumab exposure (time-varying daily C_avg) were available. The percentage of treated patients included in the analysis was 98.4% (315/320) from CheckMate 9ER and 99.5% (604/607) from nivolumab monotherapy trials. Five patients from CheckMate 9ER and 3 patients from nivolumab monotherapy studies were excluded from the analysis due to missing nivolumab exposures. The E–R safety dataset included daily records of exposure to account for time-varying daily C_avg (234,562 records from 919 patients).

The exposure variable was nivolumab daily C_avg (continuous). Response variable was time to first occurrence of grade ≥2 IMAEs (continuous). Baseline demographic variables were age (continuous), sex [male, female (reference)], and bodyweight (continuous). The baseline clinical laboratory variable was estimated glomerular filtration rate (eGFR; continuous). Baseline nivolumab clearance (continuous) was included. Cabozantinib coadministration [yes, no (reference)] was also evaluated. The composite safety endpoint of grade ≥2 IMAEs was defined as specific events (or groups of Medical Dictionary for Regulatory Activities preferred terms describing specific events) that included diarrhea/colitis, hepatitis, pneumonitis, nephritis and renal dysfunction, rash, hypersensitivity/infusion reactions, and endocrine-related events (adrenal insufficiency, hypophysitis, hypothyroidism/thyroiditis, hyperthyroidism, and diabetes mellitus), regardless of treatment with immune-modulating medications. HRs comparing nivolumab + cabozantinib versus nivolumab monotherapy were calculated as described for E–R efficacy analysis.

The final E–R efficacy models were used to predict the cumulative probability of PFS/OS over time (Kaplan–Meier curve) and mean PFS/OS at 6 months, 9 months, 1 year, and 2 years. The final E–R safety model was used to predict the cumulative probability of grade ≥2 IMAEs over time in patients in CheckMate 9ER receiving nivolumab 240 mg every 2 weeks + cabozantinib 40 mg every day and nivolumab 480 mg every 4 weeks + cabozantinib 40 mg every day. The model-predicted mean probability of grade ≥2 IMAEs at 6 months, 9 months, 1 year, and 2 years for nivolumab 240 mg every 2 weeks + cabozantinib 40 mg every day and nivolumab 480 mg every 4 weeks + cabozantinib 40 mg every day were compared.

The E–R efficacy and safety analyses, data presentations, and modeling were performed using SAS v.9.4 (26) or R software v.3.4.3 (27). Pharmacokinetic analyses were performed using NONMEM v.7.4.0 (28).

The CheckMate 9ER trial was approved by the institutional review board or ethics committee at each site and was conducted according to Good Clinical Practice guidelines, defined by the International Conference on Harmonisation. All enrolled patients provided written informed consent that was based on the Declaration of Helsinki principles.

Bristol Myers Squibb's policy on data sharing may be found at https://www.bms.com/researchers-and-partners/independent-research/data-sharing-request-process.html.

Nivolumab pharmacokinetics were described by modifying a previously developed pharmacokinetic model (20) to incorporate the effect of cabozantinib coadministration on nivolumab clearance (Supplementary Methods). Nivolumab serum concentration-time data were well described by a linear, two-compartment model with zero-order intravenous infusion, first-order elimination, and time-varying clearance. Nivolumab clearance was higher in patients with higher baseline bodyweight in comparison with lower baseline bodyweight, lower baseline albumin in comparison with higher baseline albumin, higher eGFR in comparison with lower eGFR, and PS >0 in comparison with PS = 0 (Supplementary Fig. S1). Nivolumab clearance was lower in female in comparison with male patients and Asian patients in comparison with White patients (Supplementary Fig. S1); the magnitude of these clearance differences is not considered to be clinically relevant. Nivolumab baseline clearance was estimated to be 17% lower in patients who received nivolumab + cabozantinib versus nivolumab monotherapy. However, the lower nivolumab clearance with cabozantinib coadministration did not result in meaningful differences in nivolumab exposures between nivolumab monotherapy and nivolumab + cabozantinib with <20% difference observed across different nivolumab exposure measures (Supplementary Table S2).

Nivolumab C_max after the first dose (C_max1) was 100% higher, C_avg after the first dose (C_avg1) 57.5% higher, and minimum concentration (C_min) on day 28 (C_mind28) 20.8% lower for 480 mg every 4 weeks + cabozantinib versus 240 mg every 2 weeks + cabozantinib (Supplementary Table S3). At steady state, these differences reduced to C_maxss 36.4% higher, C_avgss no difference, and C_minss 19.5% lower. Nivolumab E–R safety and efficacy analyses were conducted to predict the impact of the exposure differences. Cabozantinib exposures were not included in the E–R analysis, but rather included as a categorical covariate to simplify the nivolumab E–R analyses as the focus of the E–R analyses was to predict impact of changes in nivolumab exposure on efficacy and safety response.

Baseline characteristics are summarized for studies in the E–R analyses (Table 1). The model with nivolumab C_avg1 as a log-linear function for PFS and linear function for OS had a lower BIC relative to the linear and log-linear functions, respectively. The effect of nivolumab exposure (C_avg1) on PFS or OS was not statistically significant (95% CI for the HR included 1; Supplementary Table S4). Estimated effects of exposure and covariates on the HR of PFS/OS in the full model are presented in Fig. 1A and B. Baseline nivolumab clearance, sex, baseline bodyweight, PS, and cabozantinib coadministration were identified as significant PFS/OS predictors in the full model. Cabozantinib coadministration was the strongest predictor of PFS/OS out of all covariates evaluated (PFS HR, 0.38; 95% CI, 0.31–0.47; OS HR, 0.63; 95% CI, 0.46–0.85) versus nivolumab monotherapy. The subgroups of patients with lower baseline clearance (<10.4 mL/hour), higher baseline bodyweight (>81.8 kg), who were male, and had PS = 0 (KPS = 100) each had reductions up to approximately 20% in the risk of disease progression or death and had reductions up to approximately 50% in the risk of death. Importantly, age, baseline albumin, and tumor PD-L1 status (1% cutoff) were not significant predictors of PFS/OS for nivolumab monotherapy or nivolumab + cabozantinib combination therapy. Region and IMDC score were not significant PFS predictors, but regions outside of the United States and Canada and poor IMDC risk were negative OS predictors. Subsequent analysis of the multiplicative interaction between C_avg1 and all significant covariates in the model, including cabozantinib coadministration, did not indicate a significant interaction, suggesting that the nivolumab C_avg1 covariate effect on response is not influenced by cabozantinib or the other significant predictors of PFS/OS. This indicated a flat nivolumab E–R for nivolumab monotherapy and nivolumab + cabozantinib. The predicted PFS/OS HRs comparing nivolumab 240 mg every 2 weeks + cabozantinib versus nivolumab 3 mg/kg every 3 weeks indicated a significant OS/PFS improvement with cabozantinib across the median, 5th, and 95th percentiles of nivolumab exposures (Fig 2A and B).

The impact of line of therapy on the full model was also evaluated, since the previous nivolumab monotherapy studies were for the most part in previously treated RCC (with the exception of 23 patients in CA209009) while nivolumab + cabozantinib from CheckMate 9ER was in the first-line setting. The analysis indicated that line of therapy was not a significant predictor in the full model.

The final model was developed by backward elimination of the covariates based on BIC. The covariates included in the final model were sex, baseline nivolumab clearance, and cabozantinib coadministration for PFS, and sex, baseline nivolumab clearance, performance score, baseline bodyweight, and cabozantinib coadministration for OS, with model parameter estimates similar to those in the full model. IMDC score and region were not significant in the E–R PFS or E–R OS final models. The model appropriateness for predicting the observed data was evaluated using a visual predictive check by performing 1,000 simulations of PFS/OS versus time and then comparing a summary of those results to the observed PFS/OS over time. The model-predicted median (90% prediction interval) was in good agreement with the observed Kaplan–Meier plot of PFS/OS up to 25 months, indicating adequate model performance. The low number of patients with data available after 25 months (<35) likely contributed to the model's overprediction of PFS/OS after this time (Supplementary Fig. S2A and S2B).

Additional validation of the E–R PFS model was performed by five-fold cross-validation, whereby model predictions of PFS for patients not included in the training dataset were compared with their observed values and repeated five times, each time selecting a different training dataset. Overall, the difference (bias) in the observed and mean predicted probability of PFS using the original model at 1 and 2 years (−0.05481 and −0.08483, respectively) was similar to the difference in the observed and predicted probability of PFS for the average of the five cross-validation models at 1 and 2 years (−0.05477 and −0.08469, respectively). This confirms the robustness of the original model evaluation analyses to support the predictions of the E–R efficacy model.

A similar approach was used to characterize the E–R safety as was used for E–R efficacy. Nivolumab daily C_avg from day 1 to the event/censor was used to represent nivolumab exposure in the full model for grade ≥2 IMAEs. The model with nivolumab daily C_avg as a linear function had a lower BIC relative to the log-linear model of C_avg, and this effect was included in the full model. The effect of nivolumab exposure (daily C_avg) on the risk of grade ≥2 IMAEs was not statistically significant since the 95% CI for the HR included 1 (Supplementary Table S5). In the full model, only cabozantinib coadministration was identified as a significant predictor of grade ≥2 IMAEs (Fig. 1C). Patients administered nivolumab + cabozantinib had higher risk of grade ≥2 IMAEs versus nivolumab monotherapy (HR, 2.19; 95% CI, 1.79–2.67). The interaction effect between nivolumab daily C_avg and cabozantinib coadministration did not decrease the BIC, indicating that the nivolumab daily C_avg covariate effect on response is not influenced by cabozantinib. The final model after backward elimination included nivolumab daily C_avg and the covariate effect of cabozantinib coadministration. The predicted IMAE HR comparing nivolumab 240 mg every 2 weeks + cabozantinib versus nivolumab 3 mg/kg every 3 weeks indicates cabozantinib effects on IMAEs are similar across the nivolumab exposure range (Fig 2C).

To determine whether there was a broad increased incidence of IMAEs with nivolumab + cabozantinib versus nivolumab monotherapy across event types, or if this was driven by specific event types, grade ≥2 IMAEs were compared by category. The percentage of patients who experienced each category of grade ≥2 IMAEs (out of total patients within the study) was determined for each RCC study included in the nivolumab E–R analysis (Supplementary Fig. S3A). Diarrhea-related AEs represented the highest percentage of grade ≥2 IMAEs for nivolumab + cabozantinib, followed by hepatic-related AEs at 23.8 and 18.4%, respectively (Supplementary Fig. S3A). The percentage of diarrhea-related grade ≥2 IMAEs was higher for nivolumab + cabozantinib versus nivolumab monotherapy studies (2.9%–8.4%), followed by hepatic-related events (3.6%–11.8%), with the percentage of the other grouped IMAEs (pneumonitis, renal, rash/hypersensitivity, and endocrine) showing similar or lower proportions with nivolumab + cabozantinib as in the nivolumab monotherapy studies (Supplementary Fig. S3A).

The final model visual predictive check indicated that the model adequately predicted the cumulative probability of grade ≥2 IMAEs over time (Supplementary Fig. S3B) and results of the five-fold cross-fold validation support a robust model for prediction of grade ≥2 IMAEs.

To assess the effects of alternative nivolumab dosing regimens on PFS/OS and grade ≥2 IMAEs, the E–R efficacy and safety models were used to predict the cumulative probabilities of PFS/OS and time to first grade ≥2 IMAE over time for nivolumab 240 mg every 2 weeks + cabozantinib 40 mg every day (dosed in CheckMate 9ER) and nivolumab 480 mg + cabozantinib 40 mg every day (model-informed bridge).

The model-predicted cumulative probabilities of PFS/OS for nivolumab 240 mg every 2 weeks + cabozantinib 40 mg every day and nivolumab 480 mg every 4 weeks + cabozantinib 40 mg every day were similar and superior to the observed probability of PFS/OS in the sunitinib comparator arm (Fig. 3A and B). The model-predicted cumulative probabilities of grade ≥2 IMAEs were also very similar for these posologies (Fig. 3C). Model-predicted PFS/OS (Table 2) and grade ≥2 IMAE (Table 3) rates for the nivolumab posologies up to 2 years were similar (≤2.5% difference).

Accelerated drug development approaches can pose challenges to dose optimization, some of which can potentially be addressed using model-informed drug development (29, 30). Global health authority guidance acknowledges that a thorough understanding of pharmacokinetic and dose E–R relationships for efficacy and safety can increase the robustness of dosing-related decision making (31–35). Recent examples in oncology have highlighted the use of model-informed drug development approaches to optimize dosing regimens post-approval, where knowledge of pharmacokinetics and E–R relationships were used to demonstrate comparable benefit:risk of proposed dosing regimens versus the regimens evaluated in the pivotal trial (10–12, 36–39).

Clinical data with nivolumab 480 mg every 4 weeks + cabozantinib 40 mg every day are not available. However, nivolumab monotherapy regimens of 240 mg every 2 weeks or 480 mg every 4 weeks have been approved for patients with advanced RCC who have received prior antiangiogenic therapy (9). A detailed understanding of nivolumab monotherapy pharmacokinetic and E–R relationships over a broad dose range of 0.3 to 10 mg/kg every 2 weeks or every 3 weeks combined with data from CheckMate 9ER (nivolumab + cabozantinib) enabled the effective prediction of a similar benefit:risk profile for nivolumab 240 mg every 2 weeks + cabozantinib 40 mg every day (dose used in CheckMate 9ER) and 480 mg every 4 weeks + cabozantinib 40 mg every day (model-informed analysis). This analysis led to health authority approval of nivolumab 480 mg every 4 weeks + cabozantinib.

There were no significant PK interactions between nivolumab and cabozantinib. The approval of the clinically untested 480 mg every 4 weeks + cabozantinib regimen required understanding of any potential impact of cabozantinib on nivolumab pharmacokinetics and nivolumab impact on cabozantinib pharmacokinetics. As would be expected for a small molecule that is primarily metabolized by cytochrome P450s, cabozantinib did not significantly impact the pharmacokinetics of nivolumab (exposure differences <20%), a monoclonal antibody eliminated by proteolytic degradation. Likewise, nivolumab did not impact cabozantinib pharmacokinetics. While not significant, the mechanism for 17% lower nivolumab clearance when cabozantinib is coadministered is unknown. The clearance reduction is independent of the other known covariate effects on clearance. One potential explanation is that an early cabozantinib treatment benefit may improve patients' health status more over the first month of dosing compared with nivolumab monotherapy, resulting in lower clearance. Health status and subsequent improvement while on treatment were previously associated with lower baseline nivolumab clearance and time-varying clearance (20, 21, 24).

Cabozantinib coadministration with nivolumab improved PFS/OS compared with sunitinib and nivolumab monotherapy. Improvement to PFS/OS was independent of nivolumab exposure across a broad dose range. Since the effect of nivolumab C_avg1 on PFS/OS was not modified by cabozantinib coadministration, this indicated a flat nivolumab E–R for nivolumab monotherapy and nivolumab + cabozantinib. This was consistent with previous observations of nivolumab monotherapy in RCC, where a nivolumab early exposure measure (time-averaged concentration over the first 28 days) was not a significant predictor of OS or tumor ORR in RCC (10, 11).

Identifying cabozantinib coadministration as a significant covariate in the E–R PFS/OS analyses provides supportive evidence of the contribution of cabozantinib to PFS/OS for the nivolumab + cabozantinib combination evaluated in CheckMate 9ER. Cabozantinib coadministration contributes to PFS/OS beyond what nivolumab monotherapy contributes across a broad dose range. The PFS/OS benefit for nivolumab + cabozantinib is not exclusively coming from cabozantinib, because median PFS/OS from CheckMate 9ER (PFS 16.6 months, 95% CI, 12.5–24.9; OS not reached; N = 323) were improved versus results from CABOSUN, a phase III trial evaluating cabozantinib versus sunitinib in patients with previously untreated aRCC (PFS 8.6 months, 95% CI, 6.8–14.0; OS 26.6 months, 14.6 months to not estimable; N = 79; refs. 19, 40). Therefore, the combined mechanisms of both nivolumab and cabozantinib contribute to the PFS/OS benefit.

One limitation of the E–R efficacy analyses is that nivolumab monotherapy studies were conducted in previously treated patients (second line or higher), with the exception of 23 patients in CA209009 treated in the first-line setting. Therefore, almost all of the first-line patients (N = 315) included in the dataset are from CheckMate 9ER, and line of therapy could confound the interpretation. As part of a sensitivity analysis, line of therapy was included in the current analysis as a covariate and was not found to be a significant predictor of PFS, although coefficients in the model for line of therapy and cabozantinib coadministration were highly correlated (r = 0.922). However, line of therapy is unlikely to have contributed significantly to the cabozantinib effects observed in this analysis. The recent HCRN GU16–260 phase II trial in first-line RCC patients treated with nivolumab monotherapy showed an ORR of 29% (34/117) and a median PFS of 7.4 (95% CI, 5.5–10.9) months relative to ORR of 25% (103/410) and a median PFS of 4.6 (95% CI, 3.7–5.4) months in previously treated RCC patients in the phase III pivotal trial of nivolumab versus everolimus (CheckMate 025; refs. 41, 42).

The significant baseline predictors identified for PFS/OS were not unexpected and consistent with other nivolumab E–R analyses, particularly the association of nivolumab baseline clearance with PFS/OS (20–23, 43). The clearance association is observed with other checkpoint inhibitors and is independent of dose; the primary hypothesis for this relates to the higher catabolic status, protein turnover, and muscle loss in patients with cancer who have cachexia, which then results in faster turnover and higher clearance of the dosed monoclonal antibody (20, 21, 24, 44–47).

Cabozantinib coadministration with nivolumab increased grade ≥2 IMAEs compared with nivolumab monotherapy. The higher probability of time to first grade ≥2 IMAE for nivolumab + cabozantinib versus nivolumab monotherapy raises the question of whether and how cabozantinib could potentially contribute to the increase in IMAEs. From a mechanistic basis there are potential immunomodulatory properties associated with cabozantinib (16), but these are secondary to the primary effects of cabozantinib as a multi-TKI primarily impacting the VEGF pathway. The apparent higher frequency of IMAEs with nivolumab + cabozantinib versus nivolumab monotherapy appears to mainly reflect certain frequently occurring event types (mainly hepatotoxicity and diarrhea) known to have overlapping toxicity with nivolumab and cabozantinib monotherapy (9, 13). Therefore, a higher rate of these events would be expected overall with combination therapy and less likely due to a broad immune-related synergistic effect, which would present as higher frequencies across all IMAE categories. Although it is difficult to differentiate immune-mediated versus non–immune-mediated causes for AEs such as hepatoxicity and diarrhea, it is reassuring that the increase in grade ≥2 IMAEs for nivolumab + cabozantinib versus nivolumab monotherapy was independent of nivolumab exposure, as no significant interaction was detected between cabozantinib coadministration and nivolumab exposure. This resulted in similar grade ≥2 predictions for nivolumab 240 mg every 2 weeks + cabozantinib and 480 mg every 4 weeks + cabozantinib.

Model-predicted mean probability of PFS/OS and grade ≥2 IMAEs for nivolumab 240 mg every 2 weeks + cabozantinib and nivolumab 480 mg every 4 weeks + cabozantinib were nearly identical. This is consistent with a lack of PFS/OS or grade ≥2 IMAE dependence on nivolumab exposure. The model-predicted PFS/OS rates at 6 months through 2 years decreased while grade ≥2 IMAE rates increased. These trends were similar for both nivolumab + cabozantinib regimens and provide supportive evidence of the comparability of these regimens. These analyses provided the necessary benefit:risk assessment to support health authority approval of nivolumab 240 mg every 2 weeks + cabozantinib 40 mg every day and 480 mg every 4 weeks + cabozantinib 40 mg every day, and provides a modeling framework to support dose optimization for other combination therapies.

L. Hamuro is an employee of Bristol Myers Squibb. Z. Hu is an employee of Bristol Myers Squibb. J. Passarell reports other support from Cognigen during the conduct of the study. H. Barcomb reports other support from Cognigen Corporation during the conduct of the study. J. Zhang is an employee of Bristol Myers Squibb. S. Goldstein is an employee of Bristol Myers Squibb. A. Bello is an employee of Bristol Myers Squibb. A. Roy is an employee of and reports stock ownership in Bristol Myers Squibb. L. Zhu is an employee of Bristol Myers Squibb.

L. Hamuro: Conceptualization, formal analysis, writing–original draft, writing–review and editing. Z. Hu: Conceptualization, formal analysis, writing–review and editing. J. Passarell: Conceptualization, formal analysis, writing–review and editing. H. Barcomb: Conceptualization, formal analysis, writing–review and editing. J. Zhang: Conceptualization, writing–review and editing. S. Goldstein: Conceptualization, writing–review and editing. A. Bello: Conceptualization, writing–review and editing. A. Roy: Conceptualization, validation, writing–review and editing. L. Zhu: Conceptualization, validation, writing–review and editing.

The patients and families who made this study possible. The clinical study teams who participated. Dako, an Agilent Technologies, Inc. company, for collaborative development of the PD-L1 IHC 28-8 pharmDx assay. Bristol Myers Squibb (Princeton, NJ) and Ono Pharmaceutical Company Ltd. (Osaka, Japan). The study was supported by Bristol Myers Squibb. All authors contributed to and approved the article; writing and editorial assistance were provided by Nicolette Belletier, PhD, of Parexel, funded by Bristol Myers Squibb.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.