WHAT IS ALREADY KNOWN ON THIS TOPIC

Overall survival in patients with advanced renal cell carcinoma treated with first-line nivolumab plus ipilimumab versus sunitinib at 5 years of follow-up on the CheckMate 214 clinical trial has been reported. Treatment-free survival (TFS) as a novel endpoint was previously described for patients with advanced renal cell carcinoma on the CheckMate 214 clinical trial at 42 months of follow-up. This study reports TFS as a part of partitioned survival analysis in the intention-to-treat population as well as in specific patient subgroups with extended follow-up of 5 years.

WHAT THIS STUDY ADDS

The extension of the TFS analysis in CheckMate 214 with 5 years of follow-up shows ongoing greater TFS for immunotherapy versus vascular endothelial growth factor receptor tyrosine kinase inhibitor therapy agents. The application of a partitioned survival analysis approach to this randomized clinical trial provides a more detailed description of how patients initiating different therapies spend survival time over 5 years. Breaking down partitioned survival analysis by patient subsets provides for more individualized patient decision-making. Providing additional information about patient experience during survival time supports treatment decision-making in situations where efficacy parameters are similar.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

TFS as a component of a partitioned survival analysis can inform clinical practice. In the setting of two therapies with similar survival efficacy, providers can use TFS and partitioned survival analysis to understand how patients spend their survival time, which may inform treatment choice. TFS and partitioned survival analysis should be incorporated into future clinical trials as novel and complementary outcome measures that highlight the patient survival experience.

Background

Over the past 5 years, five new treatment combinations have been approved in the front-line management of patients with advanced renal cell carcinoma (aRCC).1 Four of these treatment combinations were shown to improve overall survival (OS) and are considered preferred options for patients with International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) intermediate or poor-risk disease.2–7 As new treatment options with similar efficacy are developed, it is increasingly important to characterize how patient survival time is spent, both on and off treatment, with and without side effects, to help counsel patients and guide front-line treatment selection. Treatment with immunotherapy can be associated with prolonged disease control after cessation of therapy without the need for further anticancer treatment.8 Immunotherapy treatment-related adverse events (TRAEs) can also persist even after therapy discontinuation.9 A novel outcome measure, treatment-free survival (TFS), illustrates the average amount of survival time patients spend treatment-free with or without toxicity.10–13 TFS was developed as a component of a partitioned survival model to characterize how patients spend their survival time whether on or off therapy and with or without TRAEs. These important aspects of patient survival time are not captured by standard outcome measures and can be highlighted by partitioned survival models using TFS.11 13

We have previously reported results of TFS from the phase 3 CheckMate 214 clinical trial of nivolumab plus ipilimumab (NIVO+IPI) versus sunitinib (SUN) as first-line treatment in patients with aRCC.13 CheckMate 214 continues to show an OS benefit in the intent-to-treat (ITT) population of NIVO+IPI compared with SUN (median 55.7 months vs median 38.4 months) with a minimum follow-up of 5 years.6 14–16 We now provide a comprehensive survival analysis of these 60-month follow-up data, showing time spent on therapy with and without TRAEs, TFS with and without TRAEs, and survival after initiation of subsequent therapy. We also report differences in mean TFS between NIVO+IPI versus SUN for patient subgroups based on various baseline characteristics.

Methods

The CheckMate 214 study population included 1096 patients with aRCC who were randomized 1:1 to receive NIVO+IPI (n=550) or SUN (n=546) as first-line treatment.6 TFS was defined as the area between two Kaplan-Meier curves of time from randomization to protocol therapy cessation and time from randomization to subsequent systemic anticancer therapy initiation or death (online supplemental figure 1). Time on protocol therapy and TFS were further characterized as time with and without grade 2+ and grade 3+TRAEs. For subgroup comparisons, differences in mean survival times were estimated for IMDC risk category, previous nephrectomy status, high versus low neutrophil-to-lymphocyte ratio (NLR), presence of bone/liver metastases, tumor programmed death ligand 1 (PD-L1) expression ≥1% or <1%, and presence of sarcomatoid features (based on local review and only among IMDC intermediate/poor-risk patients).

The Kaplan-Meier method was used to estimate the 5-year probability of the time-to-event endpoints. Restricted mean times were estimated for time-to-event endpoint curves and survival states, which included OS, survival after subsequent therapy, TFS (with/without grade 2+ and 3+TRAEs), and time on protocol therapy (with/without grade 2+ and 3+TRAEs) (online supplemental figure 1). The choice of restricted mean time was 60 months, which was the minimum follow-up time at the time of the analysis. Times with grade 2+ or 3+TRAEs were estimated based on the number of non-overlapping unique days that patients experienced one or more TRAEs without double-counting (the days during which a patient had more than one qualifying adverse event). Bootstrap sampling with 1,000 samples was used to obtain 95% CIs for the comparison of 60-month restricted mean time between treatment arms overall, and between treatment arms in the different baseline characteristics subgroups.

Patient and public involvement

Although patients made important contributions to this research as study participants, patients and members of the public were not involved with the research study design, recruitment, or conduct of the study presented in this manuscript. Further, they are not involved in the dissemination of study results.

Results

At 5 years from randomization, 48% and 37% of patients in the ITT population assigned to NIVO+IPI and SUN were alive, respectively. Overall, 25% and 7% of patients in the ITT population treated with NIVO+IPI and SUN were surviving free of subsequent systemic therapy, respectively. As 8% of patients treated with NIVO+IPI remained on NIVO therapy at 60 months, 18% were surviving treatment-free. For patients treated with SUN, 2% remained on therapy at 60 months and thus 5% were surviving treatment-free (online supplemental table 1; online supplemental figures 2,3). Over the 60-month follow-up period, the mean OS was 40.7 and 36.1 months in the NIVO+IPI and SUN groups, respectively. In the NIVO+IPI treatment group, the OS time was spent on average with 16.0 months on protocol therapy, 11.1 months treatment-free, and 13.7 months surviving after subsequent therapy initiation. In the SUN treatment group, OS time was spent with an average of 13.6, 4.4, and 18.1 months on protocol therapy, in TFS, and surviving after subsequent therapy initiation, respectively (table 1). The 60-month mean between-treatment group difference in TFS was 6.7 months (95% CI 5.1 to 8.4).

Table 1

60-Month Mean TFS and survival states by IMDC risk score with a minimum 60 months of follow-up

Among IMDC favorable-risk patients, 63% treated with NIVO+IPI and 55% treated with SUN were alive at 60 months. There were 21% of favorable-risk patients assigned to NIVO+IPI who had not initiated second-line therapy, which includes 16% surviving treatment-free and 5% receiving maintenance NIVO, compared with 9% of patients assigned to SUN who did not initiate second-line therapy, which includes 7% surviving treatment-free and 2% continuing on SUN. The 60-month mean time on protocol therapy for favorable-risk patients was 15.1 months for NIVO+IPI, including 4.9 months with grade 2+TRAEs and 0.7 months with grade 3+TRAEs. For favorable-risk patients treated with SUN, the mean time on protocol therapy was 21.6 months, including 13.6 months with grade 2+TRAEs and 2.1 months with grade 3+TRAEs. Among favorable-risk patients, the 60-month mean TFS was 14.4 months, including 5.0 months with grade 2+TRAEs and 1.2 months with grade 3+TRAEs for patients on NIVO+IPI, versus 5.5 months TFS, including 2.1 months with grade 2+TRAEs and 0.3 months with grade 3+TRAEs with SUN (figure 1A,B, table 1). The 60-month mean time surviving after second-line therapy initiation was 18.4 months for the NIVO+IPI arm and 22.1 months for SUN in the favorable-risk population (figure 1A,B, table 1).

Figure 1

Treatment-free survival, and comprehensive survival states with and without grade 2+ toxicity, 60 months since randomization. IMDC, International Metastatic Renal Cell Carcinoma Database Consortium; NIVO+IPI, nivolumab plus ipilimumab; SUN, sunitinib; TFS, treatment-free survival; TRAE, treatment-related adverse event.

Of intermediate/poor-risk patients, 43% and 31% treated with NIVO+IPI and SUN were alive at 60 months, respectively. There were 27% of patients in the intermediate/poor-risk NIVO+IPI group who had not initiated second-line therapy at 60 months, including 18% who were surviving treatment-free and 9% who continued on NIVO therapy, versus 6% of patients in the SUN group who had not initiated second-line therapy, including 4% surviving treatment-free and 2% continuing on SUN. The 60-month mean time on protocol therapy for intermediate/poor-risk patients was 16.2 months for NIVO+IPI with 4.6 months with grade 2+TRAEs and 0.5 months with grade 3+TRAEs. For patients treated with SUN, the mean time on protocol therapy was 11.2 months, with 6.4 months with grade 2+TRAEs and 1.2 months with grade 3+TRAEs. The 60-month mean TFS was 10.1 months for NIVO+IPI with 4.0 months with grade 2+TRAEs and 0.6 months with grade 3+TRAEs. For patients treated with SUN, the 60-month mean TFS was 4.1 months with 2.0 months with grade 2+TRAEs and 0.3 months with grade 3+TRAEs. The 60-month mean survival time after the start of second-line therapy in the intermediate/poor-risk population was 12.3 and 16.9 months for NIVO+IPI and SUN, respectively (figure 1C,D, table 1).

Of the 513 patients in the ITT NIVO+IPI group who discontinued treatment, 52% discontinued due to disease progression, 37% due to related and unrelated adverse events, and 11% for other reasons, as listed in online supplemental table 2. Of the 526 patients in the ITT SUN group who discontinued treatment, 68% discontinued due to disease progression, 20% due to related and unrelated adverse events, and 12% for other reasons (online supplemental table 2). In the NIVO+IPI treatment group, the most TFS was seen in patients who discontinued due to adverse events or other reasons as opposed to disease progression. In the SUN treatment group, short TFS was seen regardless of the reason for discontinuation (figure 2, figure 3).

Figure 2

Treatment-free survival, survival after subsequent systemic therapy and prior protocol therapy durations, relative to the time of protocol therapy cessation, according to categories of reasons off treatment. X-axis is truncated at 24 months prior and 60 months after protocol therapy cessation. AE, adverse events; NIVO+IPI, nivolumab plus ipilimumab.

Figure 3

Treatment-free survival, survival after subsequent systemic therapy and prior protocol therapy durations, relative to the time of protocol therapy cessation, according to categories of reasons off treatment. X-axis is truncated at 24 months prior and 60 months after protocol therapy cessation. AE, adverse events; SUN, sunitinib.

TFS by baseline variables

The 60-month mean TFS varies across different baseline variables ranging from 6.5 to 14.4 months for patients treated with NIVO+IPI (figure 4). For patients treated with SUN, there was little variability in TFS across baseline variables, ranging from 2.5 to 5.5 months, with the least TFS seen among the poor-risk patients. For all baseline characteristic variables analyzed, TFS was longer with NIVO+IPI treatment compared with SUN, with the difference ranging from 4.0 to 8.9 months longer (figure 5). Longer TFS was observed with more favorable IMDC risk category score in both treatment groups with favorable-risk, intermediate-risk, and poor-risk patients having 14.4, 11.1, and 6.5 months TFS, respectively, after treatment with NIVO+IPI and 5.5, 4.5, and 2.5 months TFS after treatment with SUN (online supplemental table 3).

Figure 4

60-Month Mean TFS with minimum 60-month follow-up by baseline variables for NIVO+IPI versus SUN. IMDC, International Metastatic Renal Cell Carcinoma Database Consortium; NIVO+IPI, nivolumab plus ipilimumab; NLR, neutrophil-to-lymphocyte ratio; PD-L1, programmed death ligand 1; SUN, sunitinib.

Figure 5

Mean treatment-free survival differences over 60-month follow-up for NIVO+IPI versus SUN by baseline variables. IMDC, International Metastatic Renal Cell Carcinoma Database Consortium; NIVO+IPI, nivolumab plus ipilimumab; NLR, neutrophil-to-lymphocyte ratio; PD-L1, programmed death ligand 1; SUN, sunitinib.

Overall, in patients treated with NIVO+IPI, the longest TFS was seen in patients with IMDC favorable risk (14.4 months), followed by sarcomatoid differentiation (13.0 months), absence of bone/liver metastases (12.7 months), low NLR (12.7 months), and previous nephrectomy (11.5 months) (figure 4). TFS was similar for PD-L1≥1% at 11.6 months and PD-L1<1% at 11.4 months after treatment with NIVO+IPI. The shortest 60-month mean TFS after treatment with NIVO+IPI was 6.5 months for the poor-risk patients. For patients treated with SUN, TFS was longest for favorable-risk patients at 5.5 months and shortest for poor-risk patients at 2.5 months (figure 4).

The presence of sarcomatoid features was associated with one of the largest differences in TFS among the baseline variables assessed at 8.5 months longer with NIVO+IPI versus SUN, compared with a 5.6-month difference for tumors without sarcomatoid features. TFS was 7.3 months longer in patients with a previous nephrectomy and 4.1 months longer in patients without a previous nephrectomy with NIVO+IPI versus SUN (figure 5). Patients with a low NLR had 7.9 months longer TFS with NIVO+IPI versus SUN while patients with a high NLR had 5.6 months longer TFS. Patients without bone or liver metastases had 8.3 months longer TFS with NIVO+IPI versus SUN and patients with bone or liver metastases had 4.5 months longer TFS. The difference in TFS with tumor PD-L1 expression ≥1% or <1% was similar, at 6.7 and 7.3 months longer, respectively, with NIVO+IPI versus SUN.

Discussion

After a minimum of 5 years of follow-up, OS continues to be longer for NIVO+IPI compared with SUN in the ITT population of CheckMate 214 (HR 0.72; 95% CI 0.62 to 0.85).16 Median OS has remained longer for intermediate/poor-risk patients for NIVO+IPI versus SUN at 46.9 versus 26.6 months (HR 0.68; 95% CI 0.58 to 0.81). In favorable-risk patients, the median OS is now 74.1 months for NIVO+IPI versus 68.4 months for SUN (HR 0.94; 95% CI 0.65 to 1.37).16 After the approval of NIVO+IPI in aRCC, there have been several vascular endothelial growth factor receptor (VEGFR)/tyrosine kinase inhibitor (TKI) and immune checkpoint inhibitor combination therapies approved in the first-line setting.17 As more treatment options with similar efficacy become available, it is important to consider the characteristics of patient survival time, including time spent on therapy with or without TRAEs, time spent off therapy with or without TRAEs, and survival time after subsequent systemic therapy in choosing a first-line treatment approach. Partitioning patient survival time helps to further characterize patient experience with different front-line treatment regimens beyond standard outcome measures.

On average, based on 60-month means, patients treated with NIVO+IPI had similar time on protocol therapy regardless of IMDC risk with 15.1, 16.7, and 14.5 months on therapy for patients with favorable, intermediate, and poor risk, respectively. However, for patients treated with SUN, mean time on protocol therapy decreased with higher IMDC risk category at 21.6, 12.5, and 6.5 months for favorable, intermediate, and poor risk, respectively (table 1, online supplemental table 3). Mean time spent on protocol therapy with grade 2+ or grade 3+toxicity was greater for SUN compared with NIVO+IPI across all risk categories (figure 1, table 1). This illustrates that the amount of time spent with toxicity while on therapy is greater for patients receiving SUN than those receiving NIVO+IPI with NIVO maintenance. Factors contributing to this may include that immunotherapy is often held or stopped for toxicity while VEGFR TKI therapy is often continued at a dose reduction, and immunotherapy TRAEs can continue once treatment is stopped while VEGFR TKI TRAEs often resolve quickly. Therefore, time on therapy with toxicity is greater with VEGFR TKIs and time off therapy with toxicity is greater with immunotherapy than with VEGFR TKI therapy.

TFS was consistently longer with NIVO+IPI compared with SUN across all subgroups examined. In both treatment arms, the higher IMDC risk category was associated with shorter TFS, yet the mean TFS was consistently 2.5 times longer after NIVO+IPI than SUN for all three risk categories. As TRAEs can persist even after immunotherapy cessation, TFS with ongoing TRAEs is an important aspect of characterizing patient survival time.13 Overall, TFS with adverse events made up a small percentage of the mean TFS for NIVO+IPI. TFS with grade 2+ and 3+TRAEs was longer with NIVO+IPI than SUN; however, TFS without any grade 2+ or 3+TRAEs was more than 2.5 times longer with NIVO+IPI at 6.9 months (11.1 months TFS minus the 4.2 months spent with grade 2+TRAEs) versus SUN at 2.4 months (4.4 months TFS minus the 2.0 months spent with grade 2+TRAEs) in the ITT population, and this was consistent across both the favorable and intermediate/poor risk groups. The presence of sarcomatoid differentiation was associated with TFS that was three times longer for patients with aRCC treated with NIVO+IPI versus SUN. This is consistent with previous research suggesting that sarcomatoid differentiation has improved outcomes with immunotherapy and worse outcomes with single-agent VEGFR TKIs than in patients with other tumor-related risk factors.1 2 18 Other baseline characteristics associated with a greater difference in TFS with NIVO+IPI versus SUN include previous nephrectomy, low NLR, and no bone/liver metastases. These variables are associated with an improved prognosis, suggesting that patients who have tumors with certain favorable prognostic features may have longer TFS due to underlying disease biology.19–21 Similar improvement in TFS was seen for treatment with NIVO+IPI compared with SUN regardless of tumor PD-L1 expression. This suggests that PD-L1 status did not predict longer TFS with NIVO+IPI and is consistent with a previous study showing PD-L1 is an imperfect predictive marker of immunotherapy outcomes in aRCC.22 This testing is at times done on the primary tumor as opposed to metastases, which may in part explain a lack of correlation.23

In addition to PD-L1 expression, there are a multitude of tumor-related and host-related predictive factors that may inform the treatment of patients with RCC.24 Biomarkers other than PD-L1 were not evaluated in CheckMate 214, and in our analysis, no differences were observed according to PD-L1 expression. Further exploration of biomarkers and their potential association with TFS may be considered in the future. Time to next treatment has emerged as an intermediate endpoint for oncology trials to further characterize the clinical benefit of therapy.25–28 Time to subsequent therapy initiation or death is the upper bound of the TFS survival state. This partitioned survival model not only shows time to next treatment but further characterizes how patients spend their time until initiating second-line treatment as being either on protocol therapy or in TFS with or without TRAEs as well as time spent surviving after second-line therapy initiation. In CheckMate 214, patients treated with SUN spent most of the 60-month mean time on therapy, either remaining on protocol therapy or surviving after subsequent therapy initiation, and very little time treatment-free.

Compared with the 42-month follow-up of CheckMate 214, in the analysis of 60-month follow-up, TFS increased from 7.8 to 11.1 months for patients treated with NIVO+IPI and from 3.3 to 4.4 months for patients treated with SUN with the longer follow-up time.13 As a percentage of the follow-up time period, TFS remained relatively stable at 18.6% of the 42-month follow-up and 18.5% of the 60-month follow-up for NIVO+IPI and 7.9% of the 42-month follow-up and 7.3% of the 60-month follow-up for SUN.13 With longer follow-up in the intermediate/poor-risk population, mean TFS increased the most for patients treated with NIVO+IPI while survival after subsequent therapy initiation increased the most for patients treated with SUN. As the minimum follow-up was 60 months in this study, survival after subsequent therapy initiation can continue to be assessed with longer follow-up. For favorable-risk patients, more of the additional survival time was spent as survival after subsequent therapy initiation in both treatment groups.

TFS is a component of a comprehensive survival analysis to further characterize patient experiences with different treatment regimens. TFS complements standard outcome measures to provide additional prognostic information. This comprehensive survival analysis with TFS is an adaptation of the quality-adjusted time without symptoms or toxicity methodology to incorporate quality-of-life aspects of patient survival time into the analysis of OS.26 28 A particular strength of this analysis is that it includes all study patients from the initiation of first-line therapy, as opposed to highlighting a select group who only later can be identified to have fared well on their first-line therapy. A comprehensive survival analysis including TFS complements standard clinical endpoints and quality-of-life analyses by illustrating how patients spend their time after starting first-line therapy, whether continuing on first-line treatment, on second-line treatment or off all treatment, and whether they are experiencing toxicity or not. A recent survey by the Kidney Cancer Research Alliance on which aspects are most important to patients in choosing treatment options showed that the ability to go off therapy was highly ranked by patients29 and should be considered when making clinical decisions. The current analysis has some limitations. The CheckMate 214 trial did not have a defined treatment end date. Treatment was stopped for disease progression, intolerable side effects, or rarely for patient/physician decision (online supplemental table 2). Therefore, TFS might have been longer in the current trial if patients stopped treatment at a predefined time point or based on their treatment response, which may lead to longer TFS. It is also possible that TFS may have been longer for some patients if they were able to receive more therapy leading to improved disease control and subsequently, longer TFS. Disease response to treatment is not incorporated in TFS. Therefore, TFS may include the time that a patient spends alive with disease progression before starting subsequent therapy or until death. For most patients treated with NIVO+IPI, TFS after disease progression was short, and the most durable TFS was seen after treatment cessation for adverse events or other reasons (figures 2,3). The reason for stopping therapy did not influence TFS for sunitinib. Measuring TFS would ideally be done in a trial with a defined treatment discontinuation such as was recently reported for HCRN GU16-260.23 27 For purposes of this analysis, any grade 2+ or grade 3+TRAE was included whether it caused the patient clinical symptoms or impacted quality of life. For example, many TRAEs included laboratory abnormalities such as lipase elevation, amylase elevation, and thyroid function abnormalities, which often do not impact patient quality of life. Lastly, there was censoring due to patients lost to follow-up, although this was minimal and similar between treatment arms (online supplemental figures 2,3). This analysis compares first-line NIVO+IPI to a now outdated first-line regimen of SUN. Analysis of TFS in other first-line treatment regimens with immunotherapy and VEGFR TKI combination therapy is ongoing. A recent descriptive, exploratory analysis found that TFS and TFS without toxicity were similar in an immunotherapy and VEGFR TKI combination therapy group compared with a SUN group, in contrast to improved TFS in NIVO+IPI versus SUN seen in published literature.30 The analysis did not account for potential differences in individual immunotherapy–TKI regimens.

Clinical trial outcomes in RCC and other malignancies are typically based on standard efficacy endpoints such as OS, progression-free survival, or objective response rate. Although these endpoints are reflective of how well a treatment prolongs life, delays disease progression, or reduces tumor size, they do not capture the complexity of the day-to-day patient experience such as duration of toxicity and time spent on or off treatment. TFS partitioned survival analysis is a more nuanced endpoint that considers not just the time a patient remains on the initial therapy, but also the duration of subsequent therapy and the length of time during which a patient is not on treatment, which are additional factors to consider, particularly in the context of evolving treatment landscapes.

These analyses highlight the patient’s journey through different therapies, providing insights into the cumulative benefits and burdens of treatment regimens, which aligns with a patient-centered care model, emphasizing the importance of long-term management strategies. Compared with traditional subjective quality-of-life measures that rely on patient self-reporting, TFS analyses provide more objective data. TFS partitioned analyses reflect real-world treatment patterns more accurately than traditional endpoints. This is particularly important as it aligns clinical trial outcomes with everyday clinical practice, enhancing the relevance and applicability of trial results. Adding TFS partitioned survival analyses as endpoints in oncology trials offers a more comprehensive and realistic assessment of treatment efficacy,31 aligning clinical trial outcomes with real-world treatment patterns. It has the potential to enhance decision-making, support health economic evaluations, and prioritize patient-centered outcomes, ultimately contributing to better cancer care and management.

The current TFS analysis helps to highlight important aspects of patient survival time. While OS was similar for favorable-risk patients in both treatment groups, patients treated with NIVO+IPI spent more survival time treatment-free without toxicity compared with SUN during the 60-month follow-up period. Favorable-risk patients treated with SUN spent more time on treatment with grade 2+TRAEs compared with NIVO+IPI. Intermediate/poor-risk patients treated with NIVO+IPI had longer OS, spent longer time on protocol therapy without grade 2+TRAEs, and had a longer time in TFS without grade 2+TRAEs compared with SUN. This characterization of patient survival time should be taken into consideration when choosing a first-line treatment regimen for patients with aRCC.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

Not applicable.

Acknowledgments

The authors would like to acknowledge the patients and families who made this study possible, the clinical study teams who participated in the study, Dako, an Agilent Technologies company, for collaborative development of the PD-L1 IHC 28-8 pharmDx assay (Santa Clara, California, USA), Bristol Myers Squibb (Princeton, New Jersey, USA), and Ono Pharmaceutical Company (Osaka, Japan). Writing and editorial assistance were provided by Jen Reinhold, PharmD, which was funded by Bristol Myers Squibb. This study was supported in part by the National Institutes of Health/National Cancer Institute (NIH/NCI) Cancer Center Support Grant to Memorial Sloan Kettering Cancer Center (P30 CA008748). This study was supported in part by the NIH SPORE Grant P50CA101942 to Dana-Farber/Harvard Cancer Center.