The Food and Drug Administration pooled analysis of overall survival according to depth of response in frontline advanced immune-oncology renal cell carcinoma trials

Abstract

Background

Retrospective analyses of studies of IO–containing combinations for advanced renal cell carcinoma (RCC) suggest that depth of response is associated with overall survival but have methodological limitations. We investigated the relationship of week 12 depth of response as a continuous variable with overall survival.

Methods

Pooling data from patients with treatment-naïve advanced RCC enrolled in randomized IO–containing frontline advanced RCC trials submitted to the US Food and Drug Administration that included week 12 imaging assessment, we developed 36-month overall survival prediction models based on week 12 depth of response (reduction from baseline in target lesion diameter) using Cox proportional hazards with natural spline in an IO combination group and a sunitinib group. To avoid guarantee-time bias, only patients in follow-up at the week 12 scan were included. Overall survival was defined from the week 12 imaging date.

Results

Among the 4 trials that met our inclusion criteria, 1364 patients in the IO combination group and 1267 patients in the sunitinib group had week 12 imaging. Depth of response and 36-month overall survival were correlated throughout the entire range of depth of response in both treatment groups, with no plateau in overall survival as depth of response approached complete response. Across this range, estimated 36-month overall survival was higher in the IO combination group.

Conclusions

Deeper response was associated with better 36-month overall survival in this pooled exploratory analysis of treatment-naïve patients with advanced RCC treated with IO combination or sunitinib. Further work characterizing the relationship between depth of response and overall survival at the trial level may advance understanding of the utility of depth of response as a pharmacodynamic response biomarker or early endpoint in signal-seeking trials and to facilitate efficient drug development.

Introduction

Response Evaluation Criteria in Solid Tumors (RECIST) define 4 discrete categories (complete response, partial response, stable disease, and progressive disease) of change in overall tumor burden. These categories, however, were not defined based on biological or clinical data. Several retrospective analyses of individual studies of IO–containing combinations for advanced renal cell carcinoma (RCC) suggest that more granularly defined depth-of-response (ie, percentage reduction from baseline in the sum of target lesion diameter) categories are associated with overall survival.^1-3 These analyses are limited by the use of non–data-driven category definitions, each with a small number of patients, and guarantee-time bias.

The potential for guarantee-time bias in a depth-of-response analysis is apparent when considering that deeper response is generally achieved by longer treatment duration. The longer a patient remains on therapy, the longer that patient’s outcome is considered favorable by virtue of remaining on the same treatment regimen, regardless of depth of response. Thus, the clinical relevance of depth of response is reduced at analyses of depth of response at late time points or at analyses of maximal depth of response.

Our objective was to investigate the probability of long-term survival based on an early depth-of-response assessment to enhance understanding of the early depth-of-response continuum in relation to a high probability of favorable long-term prognosis. These data may inform the use of depth of response in early-phase trials as a pharmacodynamic response biomarker and in future novel trial designs. We therefore sought to investigate the relationship of week 12 depth of response as a continuous variable with overall survival, hypothesizing that increasing depth of response is associated with better long-term survival.

Methods

We pooled data from patients with treatment-naïve advanced RCC enrolled in randomized IO combination frontline advanced RCC trials submitted to the US Food and Drug Administration that included a week 12 imaging assessment. Depth of response was defined, in the subset of patients who had target lesions at baseline, as the percentage change in the sum of target lesion diameters between baseline and week 12 per independent review committee measurement.

We identified a total of 4 trials (CheckMate 9ER,⁴ JAVELIN Renal 101,⁵ CheckMate-214,⁶ and KEYNOTE-426⁷) that met the inclusion criteria (Table 1), then pooled individual patient–level data. We analyzed data by treatment group, allowing for the possibility that the distribution of early depth of response and its association with late overall survival may differ for patients treated with IO combination regimens compared with sunitinib.

Table 1.

List of trials included in analysis and key design features.

Trial name	ClinicalTrials.gov  ID	Accrual period	Primary endpoint	Eligibility criteria
CheckMate 214	NCT02231749	October 2014 to February 2016	Overall survival, progression-free survival, objective response rate in intermediate-risk and poor-risk disease	Histologically confirmed advanced or metastatic RCC with a clear cell component, age ≥18 y, Karnofsky performance status ≥70%, no prior systemic anticancer therapy for advanced RCC
CheckMate 9ER	NCT03141177	September 2017 to May 2019	Progression-free survival	Histologically confirmed advanced or metastatic RCC with a clear cell component, age ≥18 y, Karnofsky performance status ≥70%, no prior systemic anticancer therapy for advanced RCC
KEYNOTE-426	NCT02853331	October 2016 to January 2018	Progression-free survival, overall survival	Histologically confirmed advanced or metastatic RCC with a clear cell component, age ≥18 y, Karnofsky performance status ≥70%, no prior systemic anticancer therapy for advanced RCC
JAVELIN Renal 101	NCT02684006	March 2016 to December 2017	Progression-free survival in programmed cell death 1 ligand 1–positive disease, overall survival in programmed cell death 1 ligand 1–positive diseasea	Histologically confirmed advanced or metastatic RCC with a clear cell component, age ≥18 y, ECOG-ACRIN performance status 0-1, no prior systemic anticancer therapy for advanced RCC

Abbreviations: RCC = renal cell carcinoma.

^a≥1% of immune cells staining positive using Ventana PD-L1 SP263 assay.

Endpoints and statistical considerations

We developed 36-month overall survival prediction models based on week 12 depth of response using Cox proportional hazards with natural spline and 2 df in both the IO combination and sunitinib groups. We used a bootstrap approach to construct the CI of the predicted survival probabilities. To avoid guarantee-time bias, we included only the subset of patients who were alive and in follow-up at the week 12 scan in the model.⁸ We defined overall survival from date of 12-week imaging until death from any cause; thus, the 36-month overall survival is relative to imaging rather than random assignment. For patients alive at the data cutoff date, overall survival was censored at the last follow-up date. We chose 36-month overall survival based on the maturity of survival data and week 12 depth of response because it was a common time point for scans across trials, and the idea that early depth of response is more useful clinically than a later one.

The study was exempted from institutional review board approval because it was necessary for protection of public health under the Common Rule.

Results

The 4 studies that met the inclusion criteria accrued between October 2014 and May 2019 (Figure 1). The data cutoff dates ranged from August 2019 to June 2021, and median follow-up ranged from 28.4 to 63.9 months (median overall = 37.9 months). In total, 3449 patients had measurable disease at baseline (1730 in the IO combination group and 1719 in the sunitinib group). In the IO combination group, 24.3% had International mRCC Database Consortium favorable-risk disease, 58.7% had intermediate-risk disease, and 16.6% had poor-risk disease; in the sunitinib group, 23.5% had International mRCC Database Consortium favorable-risk disease, 59.7% had intermediate-risk disease, and 16.3% had poor-risk disease. Trial design, including eligibility criteria and primary endpoints, as well as baseline patient characteristics were similar between the trials and similar between the IO combination and sunitinib groups (Tables 1 and 2). Among patients who had undergone 12-week imaging, baseline patient characteristics were similar between the IO combination and sunitinib groups, as well (Table 3). At 12 weeks, 1364 of 1730 (78.8%) patients and 1267 of 1719 (73.7%) patients assigned to any IO combination and to sunitinib, respectively, were alive and in follow-up with imaging.

Figure 1.

CONSORT diagram.

Table 2.

Baseline characteristics of patients in pooled trials.

Trial name	Immuno-oncology combination, a  No.	Sunitinib,a No.	Male sex, %	Age, median (range), y	International mRCC Database Consortium risk: favorable/intermediate/poor, %	Prior nephrectomy, %	Sarcomatoid histology, %	Liver metastases, %
Total	1730	1719	74	62 (21-90)	24/59/17	80	12	18b
CheckMate 214	547	535	74	62 (21-85)	21/61/18	82	13	19
CheckMate 9ER	320	320	74	61 (28-90)	23/58/20	70	12	19
KEYNOTE-426	429	425	73	62 (26-89)	31/56/13	83	12	16
JAVELIN Renal 101	434	439	74	61 (27-88)	21/62/16	81	12	Not reported

^aNumber of patients actually treated; includes patients without week 12 imaging (patients excluded from depth-of-response analysis).

^bExcluding the JAVELIN Renal 101 trial.

Table 3.

Baseline characteristics of patients with 12-week imaging.

Treatment group	Immuno-oncology combination, No.	Male sex, %	Age, median (range), y	International mRCC Database Consortium risk: favorable/intermediate/poor, %	Prior nephrectomy, %	Sarcomatoid histology, %	Liver metastases,a %
Immuno-oncology combination	1364	73	62 (29-90)	26/59/15	80	12	22
Sunitinib	1267	75	61 (21-88)	25/62/13	80	11	23

^aExcluding the JAVELIN Renal 101 trial.

The distributions of 12-week depth of response by treatment assignment are shown on the x-axis in Figure 2. Depth of response ranged from ‒100% to +326% across all patients. The median depth of response was a reduction of 27.6% (IQR = 8.7-43.4%) in the IO combination group and a reduction of 13.7% (IQR = 2.5-26.3%) in the sunitinib group. In total, 34.7% of patients had a 12-week depth of response of at least a 30% reduction, corresponding to the threshold for RECIST time point objective response.

Figure 2.

Estimated 36-month overall survival since 12-week imaging, according to week 12 depth of response in patients with advanced renal cell carcinoma treated with (A) any frontline IO combination or sunitinib, (B) IO–tyrosine kinase inhibitor combination or sunitinib, or (C) IO–IO combination or sunitinib. Shaded bands represent 95% CIs. The histogram represents the total percentage of patients in both treatment groups across depths of response in bins of 5% (eg, first bin includes ‒100% to ‒95%).

At 36 months since 12-week imaging, the estimates of overall survival were 61.2% (95% CI = 58.6% to 64.0%) and 52.5% (95% CI = 49.6% to 55.5%) of patients surviving in the IO combination and sunitinib groups, respectively. The increasing probability of 36-month overall survival with greater 12-week depth of response reduction is shown according to treatment group in Figure 2. For example, in the any IO combination group (Figure 2, A), patients with a 12-week depth-of-response reduction of at least 60% were estimated to have at least a 75% probability of surviving to 36 months (at 60% depth of response, estimated 36-month overall survival was 75.0% [95% CI = 71.0% to 78.1%]). The relation of 12-week imaging with 24-month overall survival was similar (Figure S1).

Across the range of depth of response, the IO combination group had slightly higher 36-month overall survival than the sunitinib group (Figure 2, A), with greater differences between the 2 treatment groups at deeper responses. In subgroup analyses by combination drug class, we observed potential heterogeneity of depth of response and its relationship with overall survival according to type of regimen (Figure 2, B and 2, C).

Discussion

Objective response rate, defined as the percentage of patients who have at least a 30% reduction in tumor diameters as best response (confirmed or unconfirmed depending on protocol definition), has been used as a measure of drug efficacy for regulatory decision making, supported by duration of response. Objective response rate itself is, however, an imperfect pharmacodynamic marker. For example, in certain disease settings and in some trials with IO agents, the magnitude of objective response rate difference between groups has been discordant with overall survival benefit, suggesting that the mechanism of survival benefit from IO may not be directly related to tumor shrinkage or cytotoxicity.⁹

The relationship between objective response rate and overall survival may vary with disease type and drug class. Given the pace of drug development and the unmet need for patients with life-threatening diseases, it is important to better characterize the relationship between reduction in tumor burden and overall survival in individual disease types to explore the potential uses of novel endpoints. Heterogeneity in this relationship across drug classes was suggested by our exploratory subgroup analyses. Acknowledging that the IO–IO data are from a single trial, 36-month overall survival appeared to be better maintained at less robust depths of response in the IO–IO therapy trial than in the IO–tyrosine kinase inhibitor or IO–tyrosine kinase inhibitor. The better overall survival seen in patients with modest partial response or stable disease in the IO–IO trial could be due to the different mechanism of action of immune-only IO regimens (eg, the antitumor effect of an all-immune regimen may not be fully reflected by changes in measurable tumor burden). We hypothesized that patients with immediate tumor growth on an IO–tyrosine kinase inhibitor regimen may have biology inherently resistant to 2 mechanisms of action and would likely predict poor survival given the available drugs in subsequent lines of therapy compared with patients who received frontline IO–IO therapy. Given that these are the dominant mechanisms of existing approved drugs in RCC, the likelihood of response to subsequent therapy and maintenance of overall survival is low. The IO–tyrosine kinase inhibitor group also has inferior 36-month overall survival, with tumor burden increases at 12 weeks compared with the sunitinib group. This finding again may reflect the ability of patients in the sunitinib group to receive salvage nivolumab, which has a proven overall survival benefit in this setting¹⁰; salvage therapy was available in select countries during the time frame of these studies, and 19% to 38% of patients in the sunitinib treatment group in each of the trials received nivolumab as subsequent therapy. Caution should be used, however, when interpreting depth of response at the tails due to sparse data, as evidenced by wide CIs.

Durable complete responses generally have greater clinical meaningfulness than partial responses, which naturally have smaller impacts on the overall disease burden. For instance, durable complete response results were critical to the benefit:risk assessment leading to regulatory approval of high-dose interleukin 2. Complete responses are also observed in approximately 10% of patients with advanced RCC treated with IO combinations.^4-7 Patients with a large percentage of tumor burden reduction but who do not meet criteria for complete response, however, may also do well in the long term, in part because residual radiographic abnormalities (eg, small lung nodules or lymph nodes) may not contain viable tumor. In the absence of functional imaging in RCC, such residual abnormalities preclude placing a given patient in the RECIST complete response category, but these patients may have a long-term outcome similar to those patients with RECIST complete response.

In our analyzed studies, the deepest responses in both treatment groups occurred at approximately 7 months after random assignment (median = 31 weeks [IQR = 18-55 weeks] in the IO combination group; median = 29 weeks [IQR = 17-48 weeks] in the sunitinib group). We chose to estimate survival based on tumor response at an earlier time point to avoid guarantee-time bias and because we believe that the model can provide the most clinically useful information. By the time patients with advanced RCC have been treated with the same regimen for 7 months, their clinical outcomes may already be considered favorable, and a predictive overall survival model may not be as useful. In addition, depth of response as a pharmacodynamic response biomarker to facilitate efficient drug development and dose optimization is more useful at an earlier time point.

We saw no plateau in overall survival as depth of response approached complete response (ie, although complete response predicts favorable overall survival, deep partial responses also predict favorable overall survival, with no clear depth-of-response cutoff that would delineate a high probability of favorable overall survival from an unfavorable overall survival). When pooling all IO combination trials, the 36-month overall survival for any given week 12 depth of response was higher for IO combinations than for sunitinib, particularly at deeper responses. This finding likely reflects that the mechanism of tumor shrinkage in an IO–containing regimen is different from tyrosine kinase inhibitor monotherapy and more likely to lead to durable effects on longer-term overall survival.

Our large, pooled exploratory analysis is consistent with prior reports from individual studies in advanced RCC based on less mature data, suggesting that deeper response is associated with better overall survival in patients treated with IO combinations or sunitinib.^1-3 Our patient-level analysis supports the prognostic value of 12-week depth of response in patients with advanced RCC treated with IO combinations but does not demonstrate its role as a potential surrogate for overall survival, which would require a trial-level analysis with a larger number of trials and therapeutic classes.

We acknowledge the limitations of this analysis. Tumor measurements may not capture all metastatic sites (eg, bone lesions are considered nontarget, and RECIST 1.1 guidelines state that a maximum of 2 lesions in each organ should be considered measurable). Further, we used percentage change in measured tumor burden, not absolute change. It is likely that the clinical outcome of a given percentage change is different for patients with low tumor burden compared with patients with a high tumor burden. We also did not include the CLEAR trial (ClinicalTrials.gov ID NCT02811861) in our pool because it lacked week 12 imaging. Our analysis does not include patients who died before week 12 or did not have week 12 imaging; however, we consider this approach appropriate because these patients generally have a poor prognosis and thus 36-month overall survival prediction is less clinically relevant for this subgroup. We also did not adjust for any covariates in this analysis. In addition, this is a post hoc analysis of an endpoint that was not prespecified in these trials. Finally, results may be primarily attributed to the relationship between depth of response and overall survival with IO–tyrosine kinase inhibitor combination because 3 of the 4 trials evaluated IO–tyrosine kinase inhibitor and only 1 trial evaluated IO–IO combinations.

Further work characterizing the relationship between depth of response and overall survival at the trial level may advance understanding of the utility of depth of response as a pharmacodynamic response biomarker or early signal-seeking endpoint to facilitate efficient drug development. Additional prospective studies would also be needed to investigate the role of depth of response in clinical decision making as potentially more effective therapies become available for advanced RCC with the potential for cure. These studies may use an adaptive design, as has been done in other disease settings. For example, in the treatment of multiple myeloma, particularly after autologous hematopoietic stem cell transplantation, clinical trials are being conducted to assess whether minimal residual disease status is appropriate to guide further treatment and intervention (ie, escalation or de-escalation of therapy, such as in clinical trials NCT05231629, NCT04140162, NCT04934475, NCT04876248, NCT03697655, and NCT05192122). In addition, given the lack of plateau seen in our depth-of-response analyses, more effective therapies may require functional imaging and disease measurement modalities, such as blood-based biomarkers, to measure relevant differences in antitumor effect with precision.

In summary, this patient-level analysis of patients with advanced RCC suggests that 12-week depth of response as a continuous variable is associated with 36-month survival across the full range of RECIST response categories in patients treated with either an IO–containing combination or sunitinib. Additional studies are needed to characterize trial-level associations between depth of response and survival. The role of novel endpoints such as depth of response may become increasingly important in future adaptive trials with escalation or de-escalation strategies.

Author contributions

Elaine Chang, MD (Conceptualization; Methodology; Project administration; Writing—original draft; Writing—review & editing), Haley Gittleman, PhD (Data curation; Formal analysis; Visualization; Writing—review & editing), Chi Song, PhD (Data curation; Formal analysis; Visualization), Erik Bloomquist, PhD (Methodology; Supervision), Laura Fernandes, PhD (Data curation; Formal analysis), Chana Weinstock, MD (Conceptualization; Writing—review & editing), Sundeep Agrawal, MD (Writing—review & editing), Nicole Gormley, MD (Supervision), Shenghui Tang, PhD (Supervision; Writing—review & editing), Daniel L. Suzman, MD (Supervision; Writing—review & editing), Laleh Amiri-Kordestani, MD (Supervision; Writing—review & editing), Richard Pazdur, MD (Resources; Supervision), Paul G. Kluetz, MD (Resources; Supervision; Writing—review & editing), David F. McDermott, MD (Conceptualization; Methodology; Writing—review & editing), Meredith M. Regan, ScD (Conceptualization; Methodology; Visualization; Writing—original draft; Writing—review & editing), and Brian I. Rini, MD (Conceptualization; Methodology; Writing—original draft; Writing—review & editing)

Supplementary material

Supplementary material is available at JNCI Cancer Spectrum online.

Funding

None declared.

Conflicts of interest

David F. McDermott reports grants and personal fees from BMS, Merck, Eisai, Cullinan, Pfizer, and Exelixis during the conduct of the study. Meredith M. Regan reports grants, personal fees, and nonfinancial support from AstraZeneca, BMS, TerSera; grants from Bayer; grants and nonfinancial support from Biotheranostics, Novartis, Pfizer, and Roche; and personal fees from Tolmar outside the submitted work. Brian I. Rini reports grants and personal fees from Pfizer, Merck, BMS, Eisai, and Exelixis during the conduct of the study. Elaine Chang, Haley Gittleman, Chi Song, Chana Weinstock, Sundeep Agrawal, Nicole Gormley, Shenghui Tang, Daniel L. Suzman, Laleh Amiri-Kordestani, Richard Pazdur, and Paul G. Kluetz have no disclosures.

Data availability

The data generated in this study are not publicly available due to federal disclosure laws and regulations protecting confidential commercial information, personal privacy information of clinical trial participants, and data and information in regulatory applications submitted to the Food and Drug Administration. Summary data are available in both the article and supplementary file.