Patient-Reported Outcomes from the Phase III Randomized IMmotion151 Trial: Atezolizumab + Bevacizumab versus Sunitinib in Treatment-Naïve Metastatic Renal Cell Carcinoma

This article is featured in Highlights of This Issue, p. 2437

For patients with metastatic renal cell carcinoma (mRCC) with predominant clear cell histology, tyrosine kinase inhibitors (TKI) have been a standard first-line treatment option for the past decade (1). However, few patients achieve complete or durable responses with these agents, and most patients eventually experience disease progression within 5 to 11 months (2). Several phase III trials have recently shown that combination immunotherapy has efficacy in certain mRCC patient populations, and this approach is now considered standard of care in the first-line setting, with other new antiangiogenic plus immunotherapy combinations emerging (3–6).

The PD-1 checkpoint inhibitor nivolumab, in combination with the anticytotoxic T-lymphocyte–associated antigen 4 antibody, ipilimumab, showed significantly longer overall survival (OS) compared with sunitinib (HR, 0.63; 99.8% CI, 0.44–0.89; P < 0.001; refs. 3, 7, 8). Grade 3/4 treatment-related adverse events (TRAE) occurred in 46% of patients treated with nivolumab plus ipilimumab versus 63% in those treated with sunitinib; discontinuation occurred in 22% and 12% of patients, respectively (3). Progression-free survival (PFS) was significantly longer with the anti–PD-L1 inhibitor avelumab in combination with axitinib (TKI) versus sunitinib in patients with mRCC with programmed death-ligand 1 (PD-L1)–positive tumors who received these agents in the first-line setting (HR 0.61; P < 0.001; refs. 4, 9). The rate of grade 3 or higher TRAEs was similar between treatment arms (approximately 70%); adverse events (AE) that occurred during treatment led to discontinuation of both avelumab and axitinib in 8% of patients and of sunitinib in 13% of patients (4). In addition, treatment with pembrolizumab (anti–PD-1) plus axitinib resulted in significantly longer PFS (HR, 0.69; P < 0.001) as well as prolonged OS (HR, 0.53; P < 0.0001) versus sunitinib as first-line treatment for mRCC across risk groups (6, 10). TRAEs that were grade ≥ 3 occurred in 63% versus 58% of patients, respectively, and led to discontinuation of treatment in 8% of patients versus 0% (6). In the phase III IMmotion151 study (NCT02420821), atezolizumab (anti–PD-L1) combined with bevacizumab (TKI) prolonged PFS in patients across all risk groups with untreated mRCC who had PD-L1⁺ disease [≥1% of tumor-infiltrating immune cells (IC) expressing PD-L1; HR, 0.74; P = 0.0217; ref. 5]. The combination of atezolizumab plus bevacizumab had a tolerable safety profile that was consistent with results from the phase II IMmotion150 study (NCT01984242) and previous data for each drug alone (5, 11, 12). Forty percent of patients treated with atezolizumab plus bevacizumab had grade 3/4 TRAEs versus 54% who were treated with sunitinib; 5% and 8%, respectively, had all-grade TRAEs leading to discontinuation of the regimen (5).

Patient-reported outcomes (PRO) supplement the assessment of treatment benefit and help to characterize the tolerability and efficacy of new therapies (13, 14) by allowing patients to provide their unique perspective on disease- and treatment-related symptoms, the impacts of those symptoms on daily life, and overall burden of side effects (15). Patients treated with sunitinib, for instance, have reported more toxicity versus placebo but no clinically meaningful deterioration in most, but not all, QOL measures (16). Additionally, improvements in efficacy made with targeted therapies do not typically coincide with improved QOL for patients with mRCC (17–19). Because poor prognosis and toxicity have been frequently associated with mRCC therapies, it is critical to understand the effects of new treatment agents on HRQOL in this patient population (18) and to identify new therapeutic combinations and treatment sequences that can reduce any potential decrement to patients’ functioning and QOL.

PRO analyses have been reported for the nivolumab plus ipilimumab combination, showing fewer symptoms and better HRQOL than sunitinib in patients with intermediate- or poor-risk mRCC (20). In light of the increasing importance that patient perspectives play in drug development, PRO measures were included in the phase II IMmotion150 study to inform the PRO assessment in the phase III IMmotion151 trial. Results from IMmotion150 suggested that patients receiving atezolizumab alone or with bevacizumab maintained daily function with minimal symptom interference versus patients receiving sunitinib (21). These measures were subsequently evaluated as secondary and exploratory endpoints in IMmotion151 to determine key aspects of the patient experience of their disease and treatment. We hypothesized that the combination of atezolizumab plus bevacizumab would not significantly increase overall treatment or symptom burden from the patient's perspective versus sunitinib. Here, we report a comprehensive analysis of PROs from IMmotion151 to inform overall treatment burden in patients with mRCC receiving atezolizumab plus bevacizumab.

Details of the study design for the phase III, global, open-label, randomized IMmotion151 trial have been described previously (ref. 5; Supplementary Fig. S1). Briefly, patients with mRCC were stratified by PD-L1 expression (< 1% vs. ≥ 1% IC expressing PD-L1 as assessed by immunohistochemistry; VENTANA PD-L1 SP142 assay; Ventana Medical Systems), presence of liver metastases (yes vs. no), and Memorial Sloan Kettering Cancer Center (MSKCC) prognostic risk score (0, 1–2, ≥ 3). Patients received atezolizumab 1,200 mg intravenous (i.v.) infusions every 3 weeks (q3w) plus bevacizumab 15 mg/kg i.v. q3w or sunitinib 50 mg/day orally (4 weeks on, 2 weeks off). Patients could continue treatment beyond disease progression per Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 if evidence of clinical benefit was observed per investigator discretion; no crossover from sunitinib to atezolizumab plus bevacizumab was allowed.

Patients had scheduled tumor assessments at baseline, week 12, and every 6 weeks through week 78 followed by every 12 weeks thereafter. Tumor assessments continued until disease progression per RECIST 1.1 or loss of clinical benefit, regardless of whether treatment was discontinued (e.g., for toxicity). The clinical data cutoff was September 29, 2017. The study protocol was approved by the institutional review board or independent ethics committee for each study site and was conducted in full accordance with the Guideline for Good Clinical Practice and the Declaration of Helsinki. All patients gave written informed consent.

Patients’ perspectives regarding treatment and disease burden were captured by three instruments: MD Anderson Symptom Inventory (MDASI), National Comprehensive Cancer Network Functional Assessment of Cancer Therapy (FACT)-Kidney Symptom Index 19 (FKSI-19), and Brief Fatigue Inventory (BFI; Supplementary Table S1 and Supplementary Fig. S2).

The MDASI is a validated and reliable instrument developed for clinical and research use in patients with cancer (22–24). Symptom severity items for the MDASI were scored individually or as a multi-item scale. Seventeen individual item scores, a 13-item core symptom severity scale score, and a 4-item RCC symptom severity scale score were evaluated. The core symptom severity scale asked patients to rate how severe their symptoms were “at their worst” in the last 24 hours: pain, fatigue, nausea, disturbed sleep, distress, shortness of breath, difficulty remembering things, lack of appetite, drowsiness, dry mouth, sadness, vomiting, and numbness or tingling. Patients rated the severity of four additional symptoms specific to RCC and its treatment: mouth/throat sores, rash or skin change, headache, and diarrhea. The symptom interference with daily life scale included six items asking patients to rate how much their symptoms interfered in the last 24 hours with general activity, walking, work, mood, relations with other people, and enjoyment of life. The range for each symptom severity and symptom interference score was 0 to 10, where higher scores indicated greater symptom severity and interference.

The FKSI-19 is a 19-item instrument that assesses symptoms and QOL in kidney cancer (25, 26) and is composed of 4 domains: physical disease-related symptoms, emotional disease-related symptoms, treatment side effects, and function/well-being. Each item was scored on a five-point scale with response categories of “not at all,” “a little bit,” “somewhat,” “quite a bit,” and “very much.” The total scale score, which included all 19 items, ranged from 0 to 76, where 76 indicated best possible HRQOL. The patient's perspective on overall side-effect burden was captured by the GP5 item, where patients self-reported how bothered they were by their treatment side effects. This standalone item is a valid summary measure of the overall impact of treatment-related toxicities in cancer and complements safety reporting by clinicians (27).

The BFI was used to assess the severity and impact of cancer-related fatigue on patients’ daily life (28). The first three items assess patients’ fatigue at present, their usual level of fatigue in the past 24 hours, and fatigue at its worst in the past 24 hours; in this study, we focused on the item of fatigue at its worst for conceptual efficiency (fatigue severity). Additional items assess the impact of fatigue on six global domains in the last 24 hours (i.e., general activity, mood, walking ability, normal work, relations with other people, enjoyment of life). Similar to the MDASI, each BFI item was rated from 0 to 10, with 0 indicating “no fatigue” or “does not interfere” with the patient's daily life and 10 indicating that fatigue was “as bad as you can imagine” or “completely interferes” with the patient's life.

Patients completed these PRO assessments on an electronic device at scheduled clinic visits until loss of clinical benefit. Specifically, assessments were completed at baseline, on days 1 and 22 of each 6-week cycle, at end of treatment (EoT), and during survival follow-up (6, 12, 24, and 36 weeks after EoT; Supplementary Table S1). PROs were also required to be collected prior to administration of study treatment (while on treatment) and/or prior to any other study assessment(s) at each PRO visit. The PRO assessment schedule was also aligned with the study visit schedule to minimize patient completion burden. Due to the two different routes of administration and dosing schedules, patients in the sunitinib arm were without treatment for 2 weeks at day 1 assessment visits; they received their 28-day treatment at day 22 assessment visits, which were required for sunitinib patients during the first year only. BFI data were collected weekly during the first 12 weeks to capture and better characterize the subtle changes after initiation of study treatment; patients completed BFI assessments at home if they did not have a scheduled visit (e.g., days 8, 15, 29, and 36). PRO instruments were translated as required in the local language, distributed by the investigator staff, and completed in their entirety by the patient. Site staff reviewed assessments for completeness only.

Full details for statistical analyses of the primary and secondary endpoints have been previously reported (5). PROs were prespecified as secondary and exploratory endpoints in the intent-to-treat (ITT) population without type I error control. P values were not adjusted for multiplicity and are presented for descriptive purposes. Completion rates were calculated as the number of patients who completed assessments divided by the number of patients expected to complete assessments (i.e., still on study) at each scheduled visit. Per developers’ user manuals, for scales with > 50% of the constituent items completed, a pro-rated score was computed. For scales with ≤ 50% of the items completed, the scale score was considered missing.

Descriptive summaries of scores and score change from baseline for each visit by treatment arm were examined. A patient's last PRO assessment within the 30 days prior to disease progression per RECIST 1.1 was also identified.

Longitudinal models included PRO data collected at study treatment visits up to but not including the EoT visit and assumed that data were missing at random (29, 30). The primary longitudinal analysis to estimate least-squares mean change in each PRO score from baseline to each visit was based on repeated-measures models. Each model assumed a first-order autoregressive covariance structure and included covariates for visit (categorical), treatment, a treatment-by-visit interaction, baseline score, and stratification factors. As supportive analyses, linear mixed-effects models estimated least-squares mean change in each PRO score from baseline up to EoT. Each linear model assumed an unstructured covariance matrix and random effects of intercepts and slopes and included covariates for time (continuous), treatment, baseline score, and stratification factors. The difference in change between treatment arms (i.e., atezolizumab + bevacizumab versus sunitinib) was summarized at each visit, at visits through cycle 10 day 1 (i.e., week 54), and over the entire study treatment period. Effect sizes (ES) supported interpretation of differences between treatment arms, where the absolute value of ES ≥ 0.20 likely represented a clinically important difference (31). ES was calculated as the difference in score change divided by the pooled standard deviation. For the MDASI and BFI scales, negative ES values indicated favor toward atezolizumab + bevacizumab, and positive ES values indicated favor toward sunitinib. Conversely, for the FKSI-19 scale, positive ES values denoted favor toward atezolizumab + bevacizumab, and negative ES values denoted favor toward sunitinib.

Time-to-event analyses were evaluated as time to clinically meaningful deterioration, defined as a patient's first ≥ 2-point score increase above baseline on the MDASI (core symptom scale, RCC symptom scale, symptom interference scale) and BFI (fatigue severity item and fatigue interference scale) or a patient's first ≥ 5-point score decrease from baseline on the FKSI-19 total score. The hazard ratios and 95% confidence intervals comparing atezolizumab plus bevacizumab with sunitinib were estimated using a stratified Cox regression model where the stratification factors were the same as those used in the repeated-measures models. Kaplan–Meier methodology was used to estimate the probability of deterioration. Patients with a missing baseline PRO or post-baseline assessment were censored at randomization, and patients without a deterioration event were censored at the date of the last nonmissing PRO assessment.

Post hoc analyses of associations between PFS and baseline or change from baseline PRO were also performed in the ITT population. Time-dependent stratified Cox proportional hazards regression models for PFS included a term for baseline PRO score and a term for PRO score change from baseline as a time-dependent covariate and adjusted for treatment arm (for the overall model only). Kaplan–Meier estimates of PFS were grouped by median PRO score at baseline (i.e., greater than or equal to median score vs. less than median PRO score).

Descriptive summaries and longitudinal model analyses were performed on patients with a nonmissing baseline PRO assessment and ≥ 1 post-baseline PRO assessment. Time to deterioration (TTD) analyses were performed on all randomized patients. Analyses were conducted using SAS version 9.4.

The study enrolled 915 patients with mRCC between May 20, 2015, and October 12, 2016, at 152 sites across 21 countries, with 454 patients randomized to receive atezolizumab plus bevacizumab and 461 patients randomized to receive sunitinib alone. Patient characteristics were well balanced across arms prior to study treatment (ref. 5; Table 1). At baseline, 386 patients (86%) in the atezolizumab plus bevacizumab arm and 369 patients (83%) in the sunitinib arm completed the MDASI and FKSI-19; 389 patients (86%) and 370 patients (83%) completed the BFI, respectively. In both arms, prior to receiving study treatment, patients reported mild symptom severity and mild symptom interference with daily life (Table 2). Additionally, baseline FKSI-19 total scores (Table 2) were comparable to those of the US adult general population (32).

PRO data were collected during study treatment until week 111 (atezolizumab plus bevacizumab) or week 114 (sunitinib). Completion rates for each instrument were similar between arms at on-study assessment points through the week 54 assessment (each ≥ 70%; Supplementary Fig. S3). After week 54, completion rates at day 22 visits were lower in the sunitinib arm than in the atezolizumab plus bevacizumab arm, likely because day 22 clinic visits for patients randomized to the sunitinib arm were not required after the first year. Given differential completion rates between arms at day 22 visits after week 54, we focused on PRO data collected through week 54 (inclusive).

During study treatment, patients in the atezolizumab plus bevacizumab arm reported numerically milder symptoms for the 17 symptoms assessed by MDASI. The difference between arms based on linear mixed-effects models for 16 symptoms was each P < 0.05, with the exception of headache (ref. 5; Fig. 1). ES ≤ −0.20 favoring atezolizumab plus bevacizumab versus sunitinib was reported for 12 symptoms: mouth/throat sores, rash or skin change, diarrhea, nausea, lack of appetite, vomiting, dry mouth, shortness of breath, fatigue, sadness, distress, and drowsiness. When evaluated as composite scores using the core symptom scale and RCC symptom scale, symptoms were reported as less severe in the atezolizumab plus bevacizumab arm versus the sunitinib arm (Fig. 2A and B). Based on repeated-measures models, symptom severity score changes from baseline also indicated significantly milder symptoms with atezolizumab plus bevacizumab versus sunitinib (P < 0.05) at visits through week 54, with the exception of week 6 for core symptoms. The average difference in least-squares mean score changes at visits through week 54 was −0.63, with a mean ES of −0.40 (ES range, −0.66 to −0.12) for the core symptom scale and −0.75 with a mean ES of −0.52 (ES range, −0.83 to −0.23) for the RCC symptom scale.

Patients receiving atezolizumab plus bevacizumab also reported less interference of symptoms with day-to-day life versus patients receiving sunitinib (P < 0.05) at most visits through week 54 (Fig. 2C). The average differences in least-squares mean score changes at visits through week 54 was −0.61, with a mean ES of −0.29 (range, −0.51 to 0.10). Repeated-measures model estimates were consistent with linear mixed-effects model estimates: the difference in least-squares mean change from baseline up to EoT was −0.62 (95% CI: −0.87, −0.37; P < 0.0001), with an ES of −0.28 favoring atezolizumab plus bevacizumab versus sunitinib.

Patients treated with atezolizumab plus bevacizumab reported less worsening in HRQOL compared with patients treated with sunitinib (Fig. 2D). Differences in least-squares mean score changes from baseline favored atezolizumab plus bevacizumab versus sunitinib (P < 0.05) at each visit through week 54, with the exception of week 6. The average difference in least-squares mean score changes for atezolizumab plus bevacizumab versus sunitinib at visits through week 54 was 3.67, and the mean ES was 0.42 (range, 0.16–0.67). The linear mixed-effects model of least-squares mean change from baseline up to EoT estimated a difference in change of 3.39 (95% CI: 2.37–4.42; P < 0.0001) and a corresponding ES of 0.35 favoring atezolizumab plus bevacizumab versus sunitinib. Patients in the atezolizumab plus bevacizumab arm also reported considerably less bother from side effects throughout study treatment compared with patients in the sunitinib arm (Fig. 3). Differences in proportions of patients reporting “not at all” or “a little bit” of bother between arms ranged from 14 to 35 percentage points.

Descriptive summaries by visit were consistent with repeated-measures model estimates (data not shown), including lower symptom burden (milder symptoms and less functional impairment, as measured by the MDASI scales) and better HRQOL at disease progression (as measured by the FKSI-19 total scale) with atezolizumab plus bevacizumab versus sunitinib. BFI results generally supported the fatigue results measured by the MDASI (Supplementary Fig. S4).

Delayed time to symptom deterioration, interference of symptoms with patients’ day-to-day life, and HRQOL deterioration was observed with atezolizumab plus bevacizumab versus sunitinib (Fig. 4). The median time to core symptom deterioration was not estimable (NE; 95% CI: 16.4–NE) in the atezolizumab plus bevacizumab arm and was 5.6 months (95% CI: 4.3–6.9) in the sunitinib arm, with a stratified HR of 0.50 (95% CI: 0.40–0.62; Fig. 4A). The median time to RCC symptom deterioration was 13.9 months (95% CI: 10.0–NE) with atezolizumab plus bevacizumab and 3.3 months (95% CI: 2.8–4.3) with sunitinib, with a stratified HR of 0.45 (95% CI: 0.37–0.55; Fig. 4B). As previously reported (5), the median time to symptom interference with daily life in the atezolizumab plus bevacizumab and sunitinib arms was 11.3 months (95% CI: 8.3–17.5) and 4.3 months (95% CI: 3.1–5.6), respectively, with a stratified HR of 0.56 (95% CI: 0.46–0.68; Fig. 4C). Delayed deterioration of HRQOL as measured by the FKSI-19 total scale was also observed for the atezolizumab plus bevacizumab arm versus the sunitinib arm, with a median of 2.8 months (95% CI: 2.1–3.0) versus 1.5 months (95% CI: 1.4–2.1; HR 0.68; 95% CI: 0.58–0.81; Fig. 4D).

Per the BFI, TTD for fatigue severity was similar between arms (stratified HR, 0.89; 95% CI: 0.75–1.04). A modest delay in time to meaningful fatigue-related interference was observed with atezolizumab plus bevacizumab versus sunitinib, with a stratified HR of 0.75 (95% CI: 0.63–0.89; Supplementary Fig. S5).

Baseline PRO scores were associated with median PFS (Supplementary Table S2). Significant associations (P < 0.05) were observed between PFS and baseline score as well as PFS and score change from baseline for the MDASI core symptom severity, MDASI symptom interference, and FKSI-19 total scales based on time-dependent stratified Cox proportional hazards models (Supplementary Fig. S6). For the three MDASI scales, worse PFS (HR > 1) was associated with worse baseline PROs as well as worsening PROs during study treatment. Similarly, for the FKSI-19 total scale, better PFS (HR < 1) was associated with better baseline HRQOL and improvement of HRQOL during the study. For example, a one-unit increase in MDASI core symptom severity change from baseline (i.e., worsening) is associated with an 18% increase in risk of PFS in the atezolizumab plus bevacizumab arm (based on the HR of 1.18 for one unit of worsening in MDASI core symptom severity). A comparable increase in PFS risk was observed in the sunitinib arm. Similarly, a one-unit increase in FKSI-19 total score change from baseline (i.e., improvement) is associated with 4% and 3% decrease in PFS risk in the atezolizumab plus bevacizumab and sunitinib arms, respectively.

When evaluating new RCC therapies, particularly in a largely noncurative setting, it is important that disease and treatment do not significantly compromise patients' day-to-day function. This analysis, based on high-quality, complete PRO data, represents a comprehensive evaluation of the patient experience while undergoing treatment with atezolizumab plus bevacizumab or sunitinib in the first year. For patients enrolled in the phase III IMmotion151 trial, those receiving atezolizumab plus bevacizumab reported milder symptoms, less functional impairment and a delay in meaningful deterioration of daily functioning, better HRQOL, and less bother from treatment side effects versus those receiving sunitinib. Although direct comparisons were not made to reconcile safety and PRO data, taken together, patient-reported symptom severity, symptom interference, and overall side-effect bother further support the tolerable safety profile of atezolizumab plus bevacizumab.

HRQOL with TKI therapies in the first-line setting for mRCC have been reported, such as in a phase III trial of pazopanib versus sunitinib where pazopanib showed similar efficacy but a more favorable safety profile and better HRQOL scores than sunitinib (19). PRO results have also been reported for checkpoint inhibitor therapy in a phase III trial evaluating nivolumab plus ipilimumab versus sunitinib in patients with untreated mRCC (20). Results from this study showed fewer symptoms and better HRQOL with the combination versus sunitinib in patients with intermediate- or poor-risk mRCC (FKSI-19 total score: HR, 0.54; 95% CI: 0.46–0.63; FACT-G total score: HR, 0.63; 95% CI: 0.52–0.75; EQ-5D-3L visual analogue rating scale score: HR, 0.75; 95% CI: 0.63–0.89; and EQ-5D-3L UK utility score: HR, 0.67; 95% CI: 0.57–0.80; ref. 20). These findings, together with those from IMmotion151, which assessed PROs in patients with mRCC from all prognostic groups, suggest better PROs with checkpoint inhibitor therapy versus sunitinib. Unfortunately, PRO results from other trials evaluating checkpoint inhibitors in first-line mRCC have not been published yet. It is critical that we fully understand treatment impacts on patients’ functioning and ability to pursue day-to-day activities as novel agents and combinations become available. Future research should examine how PRO data could be used to personalize and better support clinician and patient treatment decision-making in practice. The significant association between PFS and PROs provides evidence of the clinical relevance of PROs with respect to PFS outcomes in mRCC. However, further investigation is needed to better understand the prognostic role of PROs in clinical care practice.

Strengths of the PRO analyses conducted in this study include the large number of patients evaluated in a randomized study. Further, study procedures concerning the administration of PRO assessments were consistent with published guidelines for ensuring high-quality PRO data. Per protocol, patients completed PRO assessments alone, before administration of study treatment or any other assessments, and without interactions that could bias their responses.

Potential limitations of these analyses include the open-label design of IMmotion151, as it may have influenced how patients perceived their symptoms and HRQOL. Additionally, as sunitinib-related toxicities tend to be worse toward the end of the 28-day treatment cycle (33), PRO data captured at day 22 may not represent the worst toxicities experienced by patients in this treatment arm. Additionally, it is not unusual to have lower completion rates as the study progresses and more patients drop out, which may lead to biased estimates. Unfortunately, reasons for noncompletion were not captured in this study. Lastly, the PRO instruments included in the study were developed before the era of checkpoint inhibitors. Still, they do capture relevant symptoms such as fatigue, rash, cough, musculoskeletal pain, diarrhea, fever, and chills (associated with atezolizumab) as well as fatigue and rash (associated with bevacizumab). These PRO instruments also measure seven of the eight most frequently reported symptomatic AEs associated with anti–PD-1/PD-L1 inhibitor immunotherapies according to an FDA review, including shortness of breath, fatigue, cough, musculoskeletal (bone) pain, fever, diarrhea, and rash (34).

Together with the previously reported efficacy data and extensive safety data in patients with mRCC (5, 11, 12), PROs from IMmotion151 suggest that, overall, atezolizumab in combination with bevacizumab does not significantly increase symptom or treatment burden compared with sunitinib.

M.B. Atkins holds ownership interest (including patents) in Werewolf and Pyxis Oncology, and is an advisory board member/unpaid consultant for BMS, Merck, Pfizer, Novartis, Roche, Exelixis, Eisai, Aveo, Array, ImmunoCore, Iovance, Cota, Arrowhead, and Leads. B.I. Rini is an employee/paid consultant for Roche, Pfizer, Merck, and BMS, and reports receiving commercial research grants from Roche, AstraZeneca, Pfizer, Aveo, Merck, and BMS. R.J. Motzer is an employee/paid consultant for Pfizer, Genentech/Roche, Eisai, Novartis, Exelixis, and Merck, and reports receiving commercial research grants from Pfizer, Merck, Genentech/Roche, Bristol-Myers Squibb, Novartis, Eisai, and Exelixis. T. Powles is an employee/paid consultant for AstraZeneca, BMS, Exelixis, Roche, Ipsen, Merck/MSD, Novartis, Pfizer, and Seattle Genetics, and reports receiving commercial research grants from AstraZeneca and Roche. D.F. McDermott is an employee/paid consultant for Genentech. C. Suarez reports receiving speakers bureau honoraria from Bristol-Myers Squibb, Ipsen, Pfizer, Roche/Genentech, and AstraZeneca; is an advisory board member/unpaid consultant for Bristol-Myers Squibb, Ipsen, Sanofi, Pfizer, EUSA Pharma, Astellas Pharma, and Novartis; and reports receiving other remuneration from Bristol-Myers Squibb and Roche. S. Bracarda is an advisory board member/unpaid consultant for Roche, Pfizer, BMS, and Ipsen. W.M. Stadler is an employee/paid consultant for Roche-Genentech, Pfizer, Caremark/CVS, AstraZeneca, Bayer, Eisai, and Merck. H. Gurney reports receiving speakers bureau honoraria from MSD, Pfizer, and Ipsen, and is an advisory board member/unpaid consultant for BMS, Pfizer, MSD, AstraZeneca, and Ipsen. S. Oudard is an advisory board member/unpaid consultant for Bayer, Novartis, BMS, Merck, Pfizer, and Ipsen. E.T. Lam reports receiving commercial research grants from Roche/Genentech. C. Quach is an employee/paid consultant for and has ownership interest in F. Hoffman-La Roche. S. Carroll is an employee/paid consultant for Genentech/Roche, Calithera Biosciences, and Ultragenyx, and holds ownership interest in Genentech/Roche and Calithera Biosciences. B. Ding is an employee/paid consultant for Genentech/Roche Inc. Q. Zhu is an employee/paid consultant for Experies. C. Schiff is an employee/paid consultant for and holds ownership interest (including patents) in Roche/Genentech. No potential conflicts of interest were disclosed by the other authors.

Qualified researchers may request access to individual patient-level data through the clinical study data request platform (http://www.clinicalstudydatarequest.com). Further details on Roche's criteria for eligible studies are available here (https://clinicalstudydatarequest.com/Study-Sponsors/Study-Sponsors-Roche. aspx). For further details on Roche's Global Policy on the Sharing of Clinical Information and how to request access to related clinical study documents, see here (http://www.roche.com/research_and_development/who_we_are_how_we_work/clinical_trials/our_commitment_to_data_sharing.htm).

Conception and design: M.B. Atkins, B.I. Rini, R.J. Motzer, T. Powles, D.F. McDermott, C. Suarez, C. Quach, E. Piault-Louis, C. Schiff, B. Escudier

Development of methodology: M.B. Atkins, B.I. Rini, T. Powles, C. Quach, B. Ding, E. Piault-Louis, C. Schiff, B. Escudier

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.B. Atkins, B.I. Rini, R.J. Motzer, T. Powles, D.F. McDermott, C. Suarez, S. Bracarda, W.M. Stadler, F. Donskov, H. Gurney, M. Uemura, E.T. Lam, C. Grüllich, E. Piault-Louis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.I. Rini, R.J. Motzer, T. Powles, D.F. McDermott, S. Bracarda, F. Donskov, S. Oudard, M. Uemura, C. Grüllich, C. Quach, S. Carroll, B. Ding, Q. Zhu, E. Piault-Louis, C. Schiff, B. Escudier

Writing, review, and/or revision of the manuscript: M.B. Atkins, B.I. Rini, R.J. Motzer, T. Powles, D.F. McDermott, C. Suarez, S. Bracarda, W.M. Stadler, F. Donskov, H. Gurney, S. Oudard, M. Uemura, E.T. Lam, C. Grüllich, C. Quach, S. Carroll, B. Ding, E. Piault-Louis, C. Schiff, B. Escudier

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Quach

Study supervision: M.B. Atkins, B.I. Rini, R.J. Motzer, C. Quach, C. Schiff

We thank the patients participating in this trial and their families, the nurses, research coordinators, data managers, and clinical study site investigators. The authors would like to acknowledge Yong Wang for statistical analyses. Patients treated at the Memorial Sloan Kettering Cancer Center were supported in part by Memorial Sloan Kettering Cancer Center Support Grant/Core Grant (P30 CA008748). Medical writing assistance for this manuscript was provided by Paige S. Davies, PhD, of Health Interactions, Inc, and funded by F. Hoffmann-La Roche, Ltd. This study was sponsored by F. Hoffmann-La Roche Ltd/Genentech, Inc, a member of the Roche Group. F. Hoffmann-La Roche Ltd. was involved in the design and conduct of the study; management, analysis and interpretation of the data; and preparation, review, and approval of the manuscript.

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.