Introduction

Until the advent of targeted therapies, metastatic renal cell carcinoma (mRCC) represented a lethal malignancy, with median survival of 12 months [1]. Before the introduction of targeted and immunotherapy agents, the only available molecules were interleukin-2 and interferon alpha. In 2005 and 2006, two novel targeted agents were Food and Drug Administration (FDA) approved for clear-cell mRCC (ccmRCC): sorafenib and sunitinib. Temsirolimus, an mTOR inhibitor, was introduced in 2007. Three additional agents were introduced in 2009: everolimus, bevacizumab and pazopanib. In 2012, axitinib was made available. Finally, in 2015 and 2016, three most recent agents, nivolumab, cabozantinib and lenvatinib, were introduced [2]. With novel therapies, overall survival (OS) improved according to several pivotal phase III studies [3,4,5].

Survival outcomes were also examined in population-based analyses. These studies showed better survival in more contemporary patients [6,7,8,9,10]. However, most of these studies focused on patients treated until 2010 [6,7,8,9]. Based on ongoing advancements with respect to available novel targeted therapies for mRCC [2], it may be postulated that the previously observed survival improvements may have become even more meaningful. Based on this hypothesis, we decided to re-examine cancer-specific mortality (CSM) in mRCC patients diagnosed until 2014, using the most contemporary version of the Surveillance, Epidemiology, and End Results (SEER) database. Moreover, we relied on competing risks regression (CRR) models to account for other-cause mortality (OCM) that is of importance, especially in the light of longer disease-free survival. Adjustment for OCM allows to maximally reduce the confounding effect of significant comorbidities that compete for OS, with CSM.

Methods

Data source and study population

Within the SEER database, we identified patients diagnosed with mRCC (International Classification of Disease for Oncology C64.9). We only included patients ≥ 18 years old, diagnosed between 2001 and 2014, with histologically confirmed mRCC. Death certificate only or autopsy cases and bilateral tumors were excluded. Histological subtypes were stratified as follows: ccmRCC (n = 7280), non-ccmRCC [papillary (n = 665) and chromophobe (n = 147)] and other mRCC histological variants [sarcomatoid mRCC (n = 1037), cyst-associated RCC (n = 9), collecting duct carcinoma: (n = 132) and mRCC not otherwise classified (NOS) (n = 6174)]. Patients without follow-up data were also excluded.

Variables definition

In the SEER database, detailed information on specific chemotherapy lines directed at mRCC is not available. In consequence, analyses that are stratified according to type or number of sequential treatment lines could not be completed. Despite this limitation, we still attempted to test for better survival in most contemporary patients. To accomplish this, diagnosis year was stratified into three intervals: contemporary (2010–2014), intermediate (2006–2009) and historical (2001–2005). Covariates consisted of age categories (≤ 49, 50–64, 65–74, 75–84, ≥ 85), gender, race (white, black, other), marital status [married, never married, separated/divorced/widowed (SDW), unknown], Fuhrman grade [11, 12] (G1/G2, G3/G4, Gx), treatment type (nephrectomy, no nephrectomy) and histological subtype (ccmRCC, non-ccmRCC, other mRCC). CSM was defined according to SEER mortality code. All other deaths were considered as OCM.

Statistical analysis

Descriptive statistics relied on tests of means and proportions and used the Chi-square for categorical and the t test for continuous variables. Three sets of analyses were performed. First, cumulative incidence plots depicted CSM rates after accounting for OCM [13, 14]. Second, CRR models examined the effect of diagnosis year on CSM [15]. Multivariable CRR models accounted for the effect of OCM to provide the most unbiased estimate of CSM, after controlling for all covariates. Third, subgroup analyses were performed according to histological subtype: ccmRCC and non-ccmRCC.

All statistical tests were two-sided. The level of significance was set at p < 0.05. Analyses were performed using the R software environment for statistical computing and graphics (version 3.3.2; http://www.r-project.org/).

Results

We identified 15,444 mRCC patients. Of those, 41.0, 28.7 and 30.3% were diagnosed, respectively, in contemporary, intermediate and historical diagnosis year intervals. The median follow-up was 8 (IQR 3–20) months. The median age at diagnosis was 63 years (IQR 55–72). The majority were male (68.6%), married (60.9%) and Caucasian (83.9%). Of all, 14.4, 37.8 and 47.8%, respectively, harbored Fuhrman grade G1/G2, G3/G4 and Gx. Overall, 47.1, 5.3 and 47.6% of patients had ccmRCC, non-ccmRCC and other mRCC histological variants. Overall, 47.6% underwent nephrectomy (Table 1).

Table 1 Clinical and pathological characteristics of patients with metastatic renal cell carcinoma, stratified according to diagnosis year
Full size table

During the study period, a total of 10,896 deaths were recorded. Of those, 6.9% were OCM events and accounted for, respectively, 6.9, 6.9 and 7.0% of all recorded deaths, respectively, in contemporary, intermediate and historical diagnosis years. These observations validate the importance of OCM and the need for OCM adjustment.

At 24 months, CSM rates were, respectively, 61.0, 63.7 and 67.3% in contemporary, intermediate and historical patients (Fig. 1). In multivariable CRR models, intermediate (HR 1.11, CI 1.06–1.17; p < 0.001) and historical patients (HR 1.24, CI 1.18–1.30; p < 0.001) had higher CSM rates than their contemporary counterparts (Table 2).

Fig. 1
figure 1

Cumulative incidence plots depict CSM rates in the entire cohort as well as in the clear-cell metastatic renal cell carcinoma and in non-clear-cell metastatic renal cell carcinoma subgroups

Full size image
Table 2 Multivariable competing risk regression analyses predicting cancer-specific mortality in patients with metastatic renal cell carcinoma
Full size table

In ccmRCC subgroup analyses, the 24-month CSM rates were, respectively, 51.8, 54.7 and 57.5% in contemporary, intermediate and historical patients (Fig. 1). Here, in multivariable CRR models, intermediate (HR 1.12, CI 1.04–1.20; p < 0.001) and historical (HR 1.29, CI 1.19–1.39; p < 0.001) patients had higher CSM rates than their contemporary counterparts, to an even greater extent than recorded in all mRCC patients (Table 2).

In non-ccmRCC subgroup analyses, cumulative incidence and multivariable CRR models failed to reveal statistically significant differences between the three diagnosis year intervals (Fig. 1 and Table 2).

Discussion

Over the past decade, numerous novel therapies were introduced in the management of metastatic renal cell carcinoma and several novel molecules showed an improved progression-free survival [3, 4, 16,17,18,19,20,21]. Three molecules (cabozantinib, temsirolimus and nivolumab) demonstrated an improved OS in pivotal phase III trials [3,4,5].

These milestones resulted in an improved survival that was also recorded in population-based registry analyses [6,7,8,9,10]. However, virtually all such analyses only included patients diagnosed up to 2010 [6,7,8,9]. Based on this consideration, we hypothesized that more contemporary data might be capable of showing that survival continues to improve. To test our hypothesis, we relied on the most contemporary SEER database that provides patient follow-up until 2014. Our results showed two key observations.

First, we documented improved CSM-free survival according to three diagnosis year intervals. Specifically, we recorded a significant reduction in mortality rates over time. To the best of our knowledge, this is the first report of a dose–response effect according to diagnosis year. Specifically, we report highest survival in contemporary patients (2010–2014) that were not included in most of previous analyses [6,7,8,9]. The survival of contemporary patients exceeded that of intermediate diagnosis year patients, in a clinically meaningful and statistically significant fashion.

Second, our results corroborate an improved survival in contemporary and intermediate patients relative to their historical counterparts in subgroup analyses that focused on ccmRCC patients, but not in patients with non-ccmRCC. This observation corroborates treatment efficacy in ccmRCC but shows not improved survival trends in non-ccmRCC [22, 23].

Our results corroborate observations made in more historical series that also reported statistically significant survival improvements among patients with advanced or metastatic RCC, over the time. However, all such previous studies have limitations, of which the foremost is their historical nature [6,7,8,9,10]. Indeed, most studies relied on patients diagnosed up to 2010 [6,7,8,9]. In consequence, the effect of some of the FDA-approved agents, such as bevacizumab, everolimus, pazopanib and axitinib, could not be truly investigated, since these molecules were FDA approved as of 2009 or later [2]. This limitation was circumvented in our study since follow-up was available until December 2014.

Furthermore, two studies relied on January 2005 to define patients as contemporary [6, 8]. Such definition may not be ideal, since the first two targeted therapy molecules, sorafenib and sunitinib, were only FDA approved as of December 2005 [2]. In consequence, the vast majority of patients with 2005 diagnosis year truly represent pre-targeted therapy patients. In our study, we circumvented this potential misclassification error and relied on the date of January 2006 for inclusion into the intermediate diagnosis year category that defined early targeted therapy patients.

Additional limitations of previous studies consisted of broad inclusion criteria. Specifically, patients with locally advanced and patients with mRCC were included in three of such studies [7, 8, 10]. This may have resulted in inclusion of patients with T4M0 RCC. Such patients may show better survival than their M1 counterparts, and they represent a specific highly selected patient category [24, 25]. We circumvented this limitation and only included M1 patients in all our analyses.

Additionally, unlike previous studies, our study is the first to rely on CRR models that account for OCM. The latter allows to control for the confounding effect of OCM and obviates the need for adjustment for comorbidities that may culminate in OCM. Indeed, not all mRCC patients die of their disease. During the study period, 6.9% of deaths were OCM events, which accounted for, respectively, 6.9, 6.9 and 7.0% of all recorded deaths, respectively, in contemporary, intermediate and historical diagnosis year. In the presence of non-negligible OCM rate, the use of competing risks is strongly indicated [14, 15], as was done in the current study with the intent of maximally reducing the confounding effect of OCM that originates from significant comorbidities.

It is also noteworthy that we were able to identify a dose–response effect according to diagnosis year, where contemporary and intermediate diagnosis year intervals included patients who could be treated with targeted therapies. Our study is the first to confirm statistical significance of the dose–response effect, even in multivariable CRR models. In the most recent report, Pal et al. relied on the SEER-Medicare population (2006–2012) of 1245 patients versus 15,444 included in our study. In their report, lack of statistical significance was reported in OS, when contemporary patients were compared to historical patients in multivariable analysis (2006–2011 vs. 1994–2005). However, statistically significant differences in OS were identified, when patients diagnosed between 2010 and 2012 were compared to those diagnosed between 2006 and 2009, but only in univariable models that do not account for OCM [10]. Conversely, in our study we were able to show statistically significant survival differences also in multivariable CRR models.

It also warrants emphasis that our analysis only showed an improved survival over time in patients with ccmRCC. This finding is in agreement with prospective phase III trials that exclusively or predominantly focused on ccmRCC [26]. In phase III trials that included non-ccmRCC patients, the absolute survival duration was shorter in non-ccmRCC patients than in those with ccmRCC [22, 23].

Finally, a large proportion of our patients, classified in the SEER database as mRCC NOS, were included in other mRCC histological variants. In these patients, the exact histological subtype could have represented more aggressive variants or could have been misclassified. Moreover, if a central review had been performed, patients with mRCC NOS histology may have been classified as ccmRCC or non-ccmRCC, resulting in inclusion in those specific categories [27].

Several additional limitations apply to this study. First, the SEER database does not allow to ascertain the exact type of systemic therapy that was used. Moreover, it is unknown how many therapy lines were used and in which sequences such therapies were administered. Furthermore, most recent systemic agents, such as nivolumab, cabozantinib and lenvatinib, that were FDA approved after 2014 could not have affected the observed rates [2]. Additionally, adjustment could not be performed according to established stratification variables such as performance status. Also we could not stratify according to risk stratification tools used in reference publications, such as Motzer or Heng score [28, 29]. Furthermore, the variable defining the site of metastases was only made available in the SEER database as of 2010. For this reason, it was not possible to adjust for this covariate, despite the established effect of different metastatic sites on cancer survival, as well as on the known role of single versus multiple metastases in single versus multiple organs [30, 31]. Finally, all the limitations related to the retrospective nature of the SEER database apply to this, as well as all other SEER database or population-based analyses.

Conclusion

The introduction of new therapeutic agents resulted in CSM-free survival improvement over study time. However, this effect exclusively applies to patients with ccmRCC, but not to those with non-ccmRCC. This observation is in agreement with established efficacy of systemic therapies for ccmRCC but lesser efficacy of these agents for non-ccmRCC.