- Email: [email protected]
Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy
EUF 294 1–7 E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
available at www.sciencedirect.com journal homepage: www.europeanurology.com/eufocus
Kidney Cancer
Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy Emil Scosyrev a,b, Edward M. Messing a,*, Richard Sylvester c, Hendrik Van Poppel d a
Department of Urology, University of Rochester Medical Center, Rochester, New York, USA; b Quantitative Safety and Epidemiology, Novartis Pharmaceu-
ticals Corporation, East Hanover, New Jersey, USA; c EORTC Headquarters, Brussels, Belgium;
d
Department of Urology, University Hospital K.U. Leuven,
Leuven, Belgium
Article info
Abstract
Article history: Accepted February 21, 2017
Background: In the European Organization for Research and Treatment of Cancer (EORTC) randomized trial 30904, nephron-sparing surgery (NSS) reduced the risk of renal dysfunction compared with radical nephrectomy (RN); however, overall survival was better in the RN arm. Objective: To determine whether treatment effect on the risk of renal dysfunction and all-cause mortality differed in magnitude across levels of baseline variables. Design, setting, and participants: This was an exploratory subgroup analysis of EORTC 30904, a phase 3 randomized trial conducted in patients with a small (5 cm) renal mass and normal contralateral kidney. Intervention: Patients were randomized to RN (n = 273) or NSS (n = 268). Outcome measurements and statistical analysis: End points included follow-up estimated glomerular filtration rate (eGFR) <60 ml/min/1.73 m2, eGFR <45 ml/min/1.73 m2, eGFR <30 ml/min/1.73 m2, and all-cause mortality. Treatment effect was examined within baseline variables: age (<62 vs 62 yr), sex, chronic disease (any vs none), performance status (0 vs 1), and serum creatinine 1.25 vs >1.25 upper limit of normal (ULN). Logistic and Cox regression models were used for analysis of renal dysfunction and all-cause mortality, respectively. Results and limitations: The median follow-up periods were 6.7 yr for eGFR and 9.3 yr for survival. No variable-by-treatment interactions were significant at alpha = 0.05. For patients with baseline creatinine >1.25 ULN (n = 36), estimated mortality hazard ratio (HR) for NSS versus RN reversed its direction (HR = 0.76, 95% confidence interval [CI]: 0.17–3.39) relative to the rest of the study cohort (HR = 1.56, 95% CI: 1.06–2.29), although this reversal was not statistically significant (interaction p = 0.25). This analysis was limited by low power. Conclusions: This exploratory analysis did not reveal strong evidence of treatment effect modification in EORTC 30904, but it was limited by low power. Patient summary: We aimed to determine whether the effect of partial versus radical nephrectomy on kidney function and overall survival depended on age, sex, and baseline health of patients enrolled in a large clinical trial. Such dependence could not be demonstrated in this analysis. © 2017 Published by Elsevier B.V. on behalf of European Association of Urology.
Associate Editor :James Catto Keywords: Kidney cancer Renal cell carcinoma Nephron-sparing surgery Partial nephrectomy Radical nephrectomy
* Corresponding author. Department of Urology, University of Rochester Medical Center, 601 Elmwood Ave., Box 656, Rochester, New York 14642, USA. Tel.: +1 585 275 1321; Fax: +1 585 273 1068. E-mail address: [email protected] (E.M. Messing). http://dx.doi.org/10.1016/j.euf.2017.02.015 2405-4569/© 2017 Published by Elsevier B.V. on behalf of European Association of Urology.
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 2
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
1.
Randomized (N = 541)
Introduction
In European Organization for Research and Treatment of Cancer (EORTC) randomized controlled trial (RCT) 30904, a total of 541 patients with a small (5 cm) renal mass were randomized to either radical nephrectomy (RN, n = 273) or nephron-sparing surgery (NSS, n = 268), and followed for disease-specific mortality, renal function, and overall survival [1–3]. During a median follow-up of 9.3 yr, renal cancer–related mortality was uncommon in both intervention arms (RN = 1.5%, NSS = 3.0%, p = 0.23) [2]. The incidence of at least moderate renal dysfunction, as determined by an estimated glomerular filtration rate (eGFR) of <60 ml/min/ 1.73 m2, was significantly reduced in the NSS arm compared with that in the RN arm [3]. With a median of 6.7 yr to last eGFR measurement, an eGFR of <60 ml/min/1.73 m2 was reached by 85.7% of patients randomized to RN and 64.7% of those randomized to NSS, with a difference of 21.0% (95% confidence interval [CI]: 13.8–28.3%). Progression to severe renal dysfunction (eGFR <30 ml/min/1.73 m2) was relatively uncommon, and occurred in 10.0% of patients in the RN arm and 6.3% of patients in the NSS arm, with a difference of 3.7% (95% CI: 1.0% to 8.5%) [3]. Despite a significantly lower incidence of at least moderate renal dysfunction in the NSS arm, all-cause mortality was lower in the RN arm of this trial. With a median followup of 9.3 yr for overall survival, 18% of the patients in the RN arm and 25% of those in the NSS arm had died (hazard ratio 1.50, 95% CI 1.03, 2.16, p = 0.03) [2]. These findings are in disagreement with the results of most observational studies of NSS versus RN, which suggest better overall survival after NSS [4]. The findings from the EORTC RCT should not, however, be disregarded simply on the basis of disagreement with observational data because observational studies are subject to patient selection bias. While the increased mortality in the NSS arm of this trial may realistically represent a type I error (because the p value was 0.03), the lower bound of the 95% CI for the hazard ratio, 1.03–2.16, virtually ruled out any substantial survival benefit of NSS relative to RN, at least in patients similar to those enrolled in this trial. Given that no other randomized trial of NSS versus RN is currently ongoing and EORTC trial 30904 will likely remain the only source of level 1 evidence on this subject for years to come, we have performed an exploratory subgroup analysis of available data from this study to determine whether the magnitude of the treatment effect (NSS vs RN) on the incidence of moderate and severe renal dysfunction and all-cause mortality varied as a function of baseline covariates, such as age, sex, presence of chronic disease, the World Health Organization (WHO) performance status, and baseline renal function.
Radical nephrectomy (N = 273)
Nephron-sparing surgery (N = 268)
Excluded: No follow-up eGFR (N = 13)
Excluded: No follow-up eGFR (N = 14)
eGFR analysis (N = 259)
Survival analysis (N = 273)
Survival analysis (N = 268)
eGFR analysis (N = 255)
Fig. 1 – CONSORT diagram. eGFR = estimated glomerular filtration rate.
performance status of 0–2. In the current exploratory post hoc analysis, four end points were examined: (1) at least one follow-up eGFR <60 ml/ min/1.73 m2, (2) at least one follow-up eGFR <45 ml/min/1.73 m2, (3) at least one follow-up eGFR <30 ml/min/1.73 m2, and (4) death from any cause (Fig. 1). The effect of randomized intervention (NSS vs RN) on these end points was examined within the strata of five baseline variables: (1) age, dichotomized at the median (<62 vs 62 yr), (2) sex, (3) chronic disease (any vs none), (4) WHO performance status (0 vs 1), and (5) baseline serum creatinine, classified as 1.25 upper limit of normal (ULN) versus >1.25 ULN. This classification of baseline renal function was necessitated by the absence of continuous subject-level creatinine or eGFR measurements at baseline, with baseline creatinine recorded only as 1.25 ULN or >1.25ULN. No other data on baseline renal function were available in this trial. By contrast, continuous eGFR measurements were available during study follow-up for 259 of 273 participants (95%) in the RN arm and 255 of 268 (95%) in the NSS arm, with a median of nine eGFR measurements per participant in each arm, and a median of 6.7 yr to last eGFR measurement [3]. Subgroup analyses of the effect of randomized treatment on each end point were performed by fitting a regression model with treatment as the only covariate within each level of the respective baseline variable. Logistic regression models were used for analysis of the incidence of renal dysfunction, while Cox regression was used for analysis of the overall duration of survival. Tests of baseline variable-by-treatment interactions were performed by including the randomized treatment and the baseline variable of interest as covariates in the model, along with their product term. A small p value for the product term would represent evidence for a difference in the magnitude of the treatment effect across levels of the baseline variable in question. Patients with missing values for a given baseline variable were excluded from subgroup analysis involving the variable but were included in other subgroup analyses. In addition to the subgroup analyses, multivariable analyses were performed to identify independent predictors of progression to eGFR <60 ml/min/1.73 m2, eGFR <45 ml/min/1.73 m2, and eGFR <30 ml/min/ 1.73 m2, as well as independent predictors of time to death due to any cause. A separate multivariable model was fit for each of these four end points, with randomized treatment and the five baseline variables as covariates. Multivariable models were based on patients with available information on all five baseline variables. All analyses were performed in SAS version 9.3. All reported p values are two sided.
2.
Patients and methods
This study was a randomized trial of RN versus NSS, with all-cause mortality as the primary end point (Supplementary material). Details of the study design were reported elsewhere [1–3]. Eligibility criteria included a solitary renal mass suspicious for renal cell carcinoma 5 cm, a radiographically normal contralateral kidney, and a WHO
3.
Results
Baseline characteristics are shown in Table 1 according to randomized treatment. The median age was 62 yr in each intervention arm, about two-thirds of all participants were
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
Table 1 – Baseline characteristics by randomized treatment. RN (N = 273) N Age (dichotomized at the <62 62 Sex Male Female Missing Chronic disease No Cardiovascular Pulmonary Other Missing WHO performance status 0 1 2 Missing Serum creatinine 1.25 ULN 1.26–2.5 ULN 2.6–5.0 ULN Missing
NSS (N = 268)
%
N
%
48.0 52.0
125 143
46.6 53.4
178 91 4
65.2 33.3 1.5
178 87 3
66.4 32.5 1.1
174 61 13 21 4
63.7 22.3 4.8 7.7 1.5
166 57 8 34 3
61.9 21.3 3.0 12.7 1.1
227 37 6 3
83.2 13.6 2.2 1.1
229 35 1 3
85.4 13.1 0.4 1.1
251 17 1 4
91.9 6.2 0.4 1.5
246 18 0 4
91.8 6.7 0.0 1.5
median), yr 131 142
NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal; WHO = World Health Organization.
male, the majority had no chronic disease with a WHO performance status score of 0 (fully active), and over 90% of all participants in each intervention arm had baseline serum creatinine 1.25 ULN.
3
Subgroup analyses of the incidence of eGFR decline below the specified thresholds over a median follow-up of 6.7 yr for eGFR, and tests for interaction are summarized in Figs. 2–4. None of the interaction p values were statistically significant at the conventional level of significance (alpha = 0.05). The incidence of severe renal dysfunction was rare in younger patients regardless of assigned treatment (Fig. 4). Subgroup analyses of all-cause mortality and tests for interactions are shown in Fig. 5 (median follow-up 9.3 yr). Here again, none of the interaction p values were below the conventional level of significance. It is noteworthy that for patients with baseline creatinine >1.25 ULN, the estimated hazard ratio reversed its direction (favoring the NSS over RN) relative to the rest of the study cohort (where survival was better with RN). This could, however, be a chance finding since there were only seven deaths in this subgroup of patients and the test for interaction was not statistically significant (p = 0.25). Multivariable analyses of independent predictors of eGFR decline below the specified thresholds and all-cause mortality are summarized in Table 2. In general, variables associated with lower follow-up eGFR included RN (vs NSS), older age, female sex, presence of chronic disease, and baseline serum creatinine >1.25 ULN, although not all these associations were statistically significant for all eGFR end points (Table 2). Association of chronic disease with lower eGFR seemed to increase in strength with decreasing eGFR cut-off. Baseline creatinine >1.25 ULN was associated with lower follow-up eGFR for all three eGFR cut-offs based on the point estimates, but this association did not reach statistical significance. There was only a very small
Fig. 2 – Subgroup analysis: progression to eGFR <60 ml/min/1.73 m2. CI = confidence interval; eGFR = estimated glomerular filtration rate; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 4
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
Fig. 3 – Subgroup analysis: progression to eGFR <45 ml/min/1.73 m2. CI = confidence interval; eGFR = estimated glomerular filtration rate; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
Fig. 4 – Subgroup analysis: progression to eGFR <30 ml/min/1.73 m2. CI = confidence interval; eGFR = estimated glomerular filtration rate; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
number of patients with baseline creatinine >1.25 ULN. Independent predictors of all-cause mortality were age, performance status, and (marginally) NSS. Association of chronic disease with mortality did not reach statistical
significance after adjusting for performance status and other variables, but it was estimated with a relatively low precision (wide confidence interval). In fact, based on location of the upper 95% confidence limit, observed data were
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 5
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
Fig. 5 – Subgroup analysis: all-cause mortality. CI = confidence interval; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
Table 2 – Multivariable analysis of progression to eGFR <60 ml/min/1.73 m2, eGFR <45 ml/min/1.73 m2, eGFR <30 ml/min/1.73 m2, and allcause mortality. Outcome
Variable
RR
eGFR < 60
Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN) Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN) Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN) Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN)
0.25 4.99 2.53 1.08 0.67 2.55 0.32 4.19 2.30 1.41 0.70 2.03 0.56 8.49 1.51 2.48 1.03 2.59 1.46 3.78 0.79 1.40 2.50 1.00
eGFR < 45
eGFR < 30
Mortality
95% LCL
95% UCL
p value
0.16 3.04 1.50 0.65 0.32 0.87 0.21 2.75 1.52 0.92 0.39 0.91 0.28 2.93 0.76 1.22 0.46 0.95 1.01 2.35 0.53 0.95 1.63 0.46
0.40 8.19 4.28 1.79 1.40 7.46 0.48 6.40 3.49 2.15 1.26 4.55 1.11 24.61 3.01 5.05 2.30 7.04 2.12 6.10 1.18 2.06 3.82 2.18
<0.0001 <0.0001 0.001 0.77 0.28 0.09 <0.0001 <0.0001 <0.0001 0.11 0.24 0.09 0.10 <0.0001 0.24 0.01 0.95 0.06 0.05 <0.0001 0.25 0.09 <0001 0.99
eGFR = estimated glomerular filtration rate; LCL = lower confidence limit; NSS = nephron-sparing surgery; RN = radical nephrectomy; RR = relative risk; UCL = upper confidence limit; ULN = upper limit of normal. RR was measured by the odds ratio in the analysis of renal function, or the hazard ratio in the analysis of all-cause mortality. The logistic models were based on 513 patients with recorded values of all covariates and at least one follow-up eGFR measurement. The Cox model was based on 533 patients with recorded values of all covariates. The total number of randomized patients was 541. Note: RR < 1 favors the first level of the variable, for example NSS; RR > 1 favors the second level of the variable, for example RN.
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 6
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
still compatible with as much as a two-fold increase in mortality due to the presence of chronic disease (Table 2). 4.
Discussion
The main objective of subgroup analyses in clinical trials is to explore patient characteristics, which may modify the magnitude of the treatment effect, thus helping to identify patients who benefit the most or the least from the treatment under investigation [5,6]. While a subgroup analysis of clinical trials is a common practice, it has a number of limitations [5,6]. To show that treatment effect varies in magnitude across the subsets (this variation is known as treatment effect modification or covariate-by-treatment interaction), such an analysis first and foremost requires estimation of the treatment effect in subsets of the original study cohort, which results in a smaller sample size, lower precision of estimation (wide confidence intervals), and low power to reject the null hypothesis of no treatment effect in any given subset. In addition, subgroup analyses usually involve screening a large number of treatment–outcome associations within levels of baseline covariates, which increases the overall probability of a type I error (ie, finding a significant association by chance, where none actually exists). While there are different methods to control the analysis-wide type I error rate, they involve a further loss of power and precision, which is already low in subgroup analyses even without adjustment for multiplicity [7]. In other words, subgroup analyses generally have low ability to find true associations and are not unlikely to turn up false-positive associations. Despite these limitations, post hoc subgroup analyses, when appropriately regarded as exploratory and hypothesis generating, may produce valuable information [5–7]. In particular, in the case of EORTC trial 30904, it could be conjectured that a survival benefit of NSS, while not seen in the overall cohort of patients (mostly healthy with no baseline renal impairment and a normal contralateral kidney), could potentially be present in a small subset of patients with baseline renal impairment. However, because the number of patients with baseline creatinine >1.25 ULN was so small, the observed reversal of estimated hazard ratios (Fig. 5) cannot be interpreted as a strong or even a modest evidence of effect modification. Other variables such as age, sex, presence of chronic disease, and performance status could also potentially modify the magnitude of the treatment effect on the incidence of renal dysfunction and all-cause mortality. The present analysis, however, did not reveal strong evidence of such effect modification, although it was limited by low power. The most important clinical issue remains that most observational studies of NSS versus RN favor NSS on the basis of better kidney function, lower incidence of cardiovascular events, better overall survival, and good oncological outcomes. The benefit of NSS in terms of preservation of kidney function was demonstrated in EORTC 30904 as well as in meta-analyses of observational studies, and is generally not disputed [3,8–10]. However, the magnitude of this effect may depend on the baseline eGFR. In one
observational study of NSS (n = 540) versus RN (n = 766), regardless of surgery type, progression to stage IIIB or higher chronic kidney disease (CKD) was very rare in patients with preoperative eGFR >90 ml/min/1.73 m2 and was very common in patients with preoperative eGFR 45– 59 ml/min/1.73 m2. As a result, the benefit of NSS relative to RN was most noticeable in patients with baseline eGFR 60– 89 ml/min/1.73 m2 (CKD stage II) [11]. However, as noted in the editorial that accompanied this article [12], that analysis was limited by low precision of estimation within the subsets of the study cohort defined by baseline eGFR (due to sample size reduction by sub-setting) and by short average duration of follow-up (median 12 mo). Our current subgroup analysis was similarly limited by a small number of patients with baseline creatinine >1.25 ULN (Fig. 2–4). In meta-analyses of observational studies, the rates of kidney cancer recurrence and mortality tend to be lower with NSS than with RN, which is certainly a manifestation of selection bias [8–10]. As observational designs in general do not guarantee valid causal inference, the benefit of NSS relative to RN in terms of cardiovascular events and overall survival seen in observational studies may also potentially represent uncontrolled selection bias persisting even after adjustment for multiple baseline covariates. For example, in an observational study of cardiovascular events after NSS (n = 869) versus RN (n = 462) with detailed assessment and control of baseline cardiovascular risk factors, NSS was associated with a reduced rate of cardiovascular events (hazard ratio = 0.57, 95% CI: 0.34, 0.96) [13]. However, this analysis seemed to include adjustment for postoperative eGFR, which may be in the causal pathway from the type of surgery to the outcome. Furthermore, as was noted in an editorial, the unadjusted cumulative incidence curves for NSS versus RN separated nearly instantly after surgery in that study, while the true benefit of renal parenchymal preservation on cardiovascular risk would be expected to accumulate gradually over time. Similar trends were seen in propensitymatched analysis of overall survival after NSS versus RN, possibly representing uncontrolled selection bias, with healthier patients selected for the NSS [14]. In fact, current European Association of Urology guidelines recommend NSS rather than RN for most patients with a small renal mass, but explicitly recommend RN over NSS for patients who also have a “significant deterioration of a patient’s general health” [15]. In a large SEER-Medicare study, overall survival of patients treated with NSS (n = 1436) or RN (n = 4247) was compared with that of subsets of the general Medicare population matched separately to NSS and RN on demographic factors and various comorbidity indicators [16]. In that analysis, RN patients had the same overall survival as their matched controls, but NSS patients had better overall survival than their matched controls. Of course, the reasons for this survival advantage after NSS could not be related to nephron sparing because the control group did not undergo kidney surgery. These findings confirm that confounding mechanisms can operate in a very subtle manner. Given the limitations of observational data and the current widespread use of elective NSS, compared with when the EORTC 30904 trial began and was conducted, another
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 7
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
randomized trial of RN versus NSS is warranted. Such a trial could be limited to T1b–T2a tumors where oncological outcomes, regardless of types of surgery, are less favorable than they are for smaller masses and where less parenchyma is likely to be spared, on average, thanwith smaller renal masses [17]. Additionally, detailed information on baseline comorbidities and renal function should be recorded, baseline creatinine should be < 1.4 mg/dl, and nephrometry score, as a surrogate for degree of difficulty of NSS, should be recorded. In summary, this exploratory subgroup analysis of EORTC 30904 did not reveal strong evidence of treatment effect modification by baseline variables, although it was limited by low power. It is possible that in patients with baseline renal impairment, NSS improves overall survival relative to RN, but we could not prove this with any reasonable certainty due to the small sample size in this subgroup. For patients with no baseline CKD and a normal contralateral kidney, we feel that the original conclusion of EORTC 30904 still holds—if a partial nephrectomy is likely to be risky and very complex, a safe, straightforward, and uncomplicated RN may well be preferable.
work was performed under a prior affiliation of the author (University of Rochester). Funding/Support and role of the sponsor: This study was supported by Fonds Cancer (FOCA) from Belgium. The funding source had no role in the design, conduct, or analysis of this study, interpretation of the data, writing of this report, or decision to submit it for publication.
References [1] Van Poppel H, Da Pozzo L, Albrecht W, et al. A prospective randomized EORTC intergroup phase 3 study comparing the complications of elective nephron-sparing surgery and radical nephrectomy for low-stage renal cell carcinoma. Eur Urol 2007;51:1606–15. [2] Van Poppel H, Da Pozzo L, Albrecht W, et al. A prospective randomized EORTC intergroup phase 3 study comparing the oncologic outcome of elective nephron-sparing surgery and radical nephrectomy for lowstage renal cell carcinoma. Eur Urol 2011;59:543–52. [3] Scosyrev E, Messing EM, Sylvester R, Campbell S, Van Poppel H. Renal function after nephron-sparing surgery versus radical nephrectomy: results from EORTC randomized trial 30904. Eur Urol 2014;65:372–7. [4] Scosyrev E, Wu K, Levey HR, et al. Overall survival after partial versus radical nephrectomy for a small renal mass: systematic
5.
Conclusions
review of observational studies. Urol Pract 2014;1:27–34. [5] Zhang S, Liang F, Li W, Hu X. Subgroup analyses in reporting of phase
In this exploratory subgroup analysis of the EORTC trial 30904 of NSS versus RN, we aimed to determine whether treatment effect on the risk of renal dysfunction and all-cause mortality differed in magnitude across levels of baseline variables. This analysis did not reveal strong evidence of treatment effect modification, but it was limited by low power. Registration EORTC trial 30904 ClinicalTrials.gov identifier: NCT00002473 http://www.clinicaltrials.gov/ct2/results? term=NCT00002473
III clinical trials in solid tumors. J Clin Oncol 2015;33:1697–702. [6] Rothwell PM. Treating individuals 2. Subgroup analysis in randomised controlled trials: importance, indications, and interpretation. Lancet 2005;365:176–86. [7] Rothman K. No adjustments are needed for multiple comparisons. Epidemiology 1990;1:43–6. [8] Mir MC, Derweesh I, Porpiglia F, Zargar H, Mottrie A, Autorino R. Partial nephrectomy versus radical nephrectomy for clinical T1b and T2 renal tumors: a systematic review and meta-analysis of comparative studies. Eur Urol 2017;71(4):606–17. [9] Kim SP, Thompson RH, Boorjian SA, et al. Comparative effectiveness for survival and renal function of partial and radical nephrectomy for localized renal tumors: a systematic review and meta-analysis. J Urol 2012;188:51–7. [10] Pierorazio PM, Johnson MH, Patel HD, et al. Management of renal masses and localized renal cancer: systematic review and meta-
Author contributions: Edward M. Messing had full access to all the data in
analysis. J Urol 2016;196:989–99.
the study and takes responsibility for the integrity of the data and the
[11] Woldu SL, Weinberg AC, Korets R, et al. Who really benefits from
accuracy of the data analysis. Study concept and design: Van Poppel, Sylvester.
[12] Campbell SC, Demirjian S, Ercole C. Editorial comment: who really
Acquisition of data: Van Poppel, Sylvester. Analysis and interpretation of data: Scosyrev, Messing, Sylvester, Van
nephron-sparing surgery? Urology 2014;84:860–7. benefits from nephron-sparing surgery? Urology 2014;84:867–8. [13] Capitanio U, Terrone C, Antonelli A, et al. Nephron-sparing techni-
Poppel.
ques independently decrease the risk of cardiovascular events
Drafting of the manuscript: Scosyrev.
relative to radical nephrectomy in patients with a T1a–T1b renal
Critical revision of the manuscript for important intellectual content: Mes-
mass
sing, Sylvester, Van Poppel.
2015;67:683–9.
Statistical analysis: Scosyrev, Sylvester. Obtaining funding: Van Poppel, Sylvester.
and
normal
preoperative
renal
function.
Eur
Urol
[14] Tomaszewski JJ, Kutikov A. Retrospective comparison of cardiovascular risk in preselected patients undergoing kidney cancer surgery:
Administrative, technical, or material support: None.
reflection of reality or simply what we want to hear? Eur Urol
Supervision: Van Poppel, Sylvester.
2015;67:690–1.
Other: None.
[15] Ljungberg B, Bensalah K, Bex A, et al. Guidelines on renal cell
Financial disclosures: Edward M. Messing certifies that all conflicts of
[16] Shuch B, Hanley J, Lai J, et al. Overall survival advantage with partial
carcinoma. European Association of Urology 2014. p. 36. interest, including specific financial interests and relationships and affilia-
nephrectomy:
tions relevant to the subject matter or materials discussed in the manu-
2013;119:29812989.
a
bias
of
observational
data?
Cancer
script (eg, employment/affiliation, grants or funding, consultancies, hono-
[17] Campbell SC, Novick AC, Bukowski RM. Renal tumors. In: Walsh PC,
raria, stock ownership or options, expert testimony, royalties, or patents
Vaughan Jr ED, Retik AB, Wein AJ, editors. Campbell’s urology. ed. 9.
filed, received, or pending), are the following: None. Emil Scosyrev is
Philadelphia, PA: WB Saunders Co.; 2007.
currently employed by Novartis Pharmaceuticals Corporation. The present
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
available at www.sciencedirect.com journal homepage: www.europeanurology.com/eufocus
Kidney Cancer
Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy Emil Scosyrev a,b, Edward M. Messing a,*, Richard Sylvester c, Hendrik Van Poppel d a
Department of Urology, University of Rochester Medical Center, Rochester, New York, USA; b Quantitative Safety and Epidemiology, Novartis Pharmaceu-
ticals Corporation, East Hanover, New Jersey, USA; c EORTC Headquarters, Brussels, Belgium;
d
Department of Urology, University Hospital K.U. Leuven,
Leuven, Belgium
Article info
Abstract
Article history: Accepted February 21, 2017
Background: In the European Organization for Research and Treatment of Cancer (EORTC) randomized trial 30904, nephron-sparing surgery (NSS) reduced the risk of renal dysfunction compared with radical nephrectomy (RN); however, overall survival was better in the RN arm. Objective: To determine whether treatment effect on the risk of renal dysfunction and all-cause mortality differed in magnitude across levels of baseline variables. Design, setting, and participants: This was an exploratory subgroup analysis of EORTC 30904, a phase 3 randomized trial conducted in patients with a small (5 cm) renal mass and normal contralateral kidney. Intervention: Patients were randomized to RN (n = 273) or NSS (n = 268). Outcome measurements and statistical analysis: End points included follow-up estimated glomerular filtration rate (eGFR) <60 ml/min/1.73 m2, eGFR <45 ml/min/1.73 m2, eGFR <30 ml/min/1.73 m2, and all-cause mortality. Treatment effect was examined within baseline variables: age (<62 vs 62 yr), sex, chronic disease (any vs none), performance status (0 vs 1), and serum creatinine 1.25 vs >1.25 upper limit of normal (ULN). Logistic and Cox regression models were used for analysis of renal dysfunction and all-cause mortality, respectively. Results and limitations: The median follow-up periods were 6.7 yr for eGFR and 9.3 yr for survival. No variable-by-treatment interactions were significant at alpha = 0.05. For patients with baseline creatinine >1.25 ULN (n = 36), estimated mortality hazard ratio (HR) for NSS versus RN reversed its direction (HR = 0.76, 95% confidence interval [CI]: 0.17–3.39) relative to the rest of the study cohort (HR = 1.56, 95% CI: 1.06–2.29), although this reversal was not statistically significant (interaction p = 0.25). This analysis was limited by low power. Conclusions: This exploratory analysis did not reveal strong evidence of treatment effect modification in EORTC 30904, but it was limited by low power. Patient summary: We aimed to determine whether the effect of partial versus radical nephrectomy on kidney function and overall survival depended on age, sex, and baseline health of patients enrolled in a large clinical trial. Such dependence could not be demonstrated in this analysis. © 2017 Published by Elsevier B.V. on behalf of European Association of Urology.
Associate Editor :James Catto Keywords: Kidney cancer Renal cell carcinoma Nephron-sparing surgery Partial nephrectomy Radical nephrectomy
* Corresponding author. Department of Urology, University of Rochester Medical Center, 601 Elmwood Ave., Box 656, Rochester, New York 14642, USA. Tel.: +1 585 275 1321; Fax: +1 585 273 1068. E-mail address: [email protected] (E.M. Messing). http://dx.doi.org/10.1016/j.euf.2017.02.015 2405-4569/© 2017 Published by Elsevier B.V. on behalf of European Association of Urology.
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 2
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
1.
Randomized (N = 541)
Introduction
In European Organization for Research and Treatment of Cancer (EORTC) randomized controlled trial (RCT) 30904, a total of 541 patients with a small (5 cm) renal mass were randomized to either radical nephrectomy (RN, n = 273) or nephron-sparing surgery (NSS, n = 268), and followed for disease-specific mortality, renal function, and overall survival [1–3]. During a median follow-up of 9.3 yr, renal cancer–related mortality was uncommon in both intervention arms (RN = 1.5%, NSS = 3.0%, p = 0.23) [2]. The incidence of at least moderate renal dysfunction, as determined by an estimated glomerular filtration rate (eGFR) of <60 ml/min/ 1.73 m2, was significantly reduced in the NSS arm compared with that in the RN arm [3]. With a median of 6.7 yr to last eGFR measurement, an eGFR of <60 ml/min/1.73 m2 was reached by 85.7% of patients randomized to RN and 64.7% of those randomized to NSS, with a difference of 21.0% (95% confidence interval [CI]: 13.8–28.3%). Progression to severe renal dysfunction (eGFR <30 ml/min/1.73 m2) was relatively uncommon, and occurred in 10.0% of patients in the RN arm and 6.3% of patients in the NSS arm, with a difference of 3.7% (95% CI: 1.0% to 8.5%) [3]. Despite a significantly lower incidence of at least moderate renal dysfunction in the NSS arm, all-cause mortality was lower in the RN arm of this trial. With a median followup of 9.3 yr for overall survival, 18% of the patients in the RN arm and 25% of those in the NSS arm had died (hazard ratio 1.50, 95% CI 1.03, 2.16, p = 0.03) [2]. These findings are in disagreement with the results of most observational studies of NSS versus RN, which suggest better overall survival after NSS [4]. The findings from the EORTC RCT should not, however, be disregarded simply on the basis of disagreement with observational data because observational studies are subject to patient selection bias. While the increased mortality in the NSS arm of this trial may realistically represent a type I error (because the p value was 0.03), the lower bound of the 95% CI for the hazard ratio, 1.03–2.16, virtually ruled out any substantial survival benefit of NSS relative to RN, at least in patients similar to those enrolled in this trial. Given that no other randomized trial of NSS versus RN is currently ongoing and EORTC trial 30904 will likely remain the only source of level 1 evidence on this subject for years to come, we have performed an exploratory subgroup analysis of available data from this study to determine whether the magnitude of the treatment effect (NSS vs RN) on the incidence of moderate and severe renal dysfunction and all-cause mortality varied as a function of baseline covariates, such as age, sex, presence of chronic disease, the World Health Organization (WHO) performance status, and baseline renal function.
Radical nephrectomy (N = 273)
Nephron-sparing surgery (N = 268)
Excluded: No follow-up eGFR (N = 13)
Excluded: No follow-up eGFR (N = 14)
eGFR analysis (N = 259)
Survival analysis (N = 273)
Survival analysis (N = 268)
eGFR analysis (N = 255)
Fig. 1 – CONSORT diagram. eGFR = estimated glomerular filtration rate.
performance status of 0–2. In the current exploratory post hoc analysis, four end points were examined: (1) at least one follow-up eGFR <60 ml/ min/1.73 m2, (2) at least one follow-up eGFR <45 ml/min/1.73 m2, (3) at least one follow-up eGFR <30 ml/min/1.73 m2, and (4) death from any cause (Fig. 1). The effect of randomized intervention (NSS vs RN) on these end points was examined within the strata of five baseline variables: (1) age, dichotomized at the median (<62 vs 62 yr), (2) sex, (3) chronic disease (any vs none), (4) WHO performance status (0 vs 1), and (5) baseline serum creatinine, classified as 1.25 upper limit of normal (ULN) versus >1.25 ULN. This classification of baseline renal function was necessitated by the absence of continuous subject-level creatinine or eGFR measurements at baseline, with baseline creatinine recorded only as 1.25 ULN or >1.25ULN. No other data on baseline renal function were available in this trial. By contrast, continuous eGFR measurements were available during study follow-up for 259 of 273 participants (95%) in the RN arm and 255 of 268 (95%) in the NSS arm, with a median of nine eGFR measurements per participant in each arm, and a median of 6.7 yr to last eGFR measurement [3]. Subgroup analyses of the effect of randomized treatment on each end point were performed by fitting a regression model with treatment as the only covariate within each level of the respective baseline variable. Logistic regression models were used for analysis of the incidence of renal dysfunction, while Cox regression was used for analysis of the overall duration of survival. Tests of baseline variable-by-treatment interactions were performed by including the randomized treatment and the baseline variable of interest as covariates in the model, along with their product term. A small p value for the product term would represent evidence for a difference in the magnitude of the treatment effect across levels of the baseline variable in question. Patients with missing values for a given baseline variable were excluded from subgroup analysis involving the variable but were included in other subgroup analyses. In addition to the subgroup analyses, multivariable analyses were performed to identify independent predictors of progression to eGFR <60 ml/min/1.73 m2, eGFR <45 ml/min/1.73 m2, and eGFR <30 ml/min/ 1.73 m2, as well as independent predictors of time to death due to any cause. A separate multivariable model was fit for each of these four end points, with randomized treatment and the five baseline variables as covariates. Multivariable models were based on patients with available information on all five baseline variables. All analyses were performed in SAS version 9.3. All reported p values are two sided.
2.
Patients and methods
This study was a randomized trial of RN versus NSS, with all-cause mortality as the primary end point (Supplementary material). Details of the study design were reported elsewhere [1–3]. Eligibility criteria included a solitary renal mass suspicious for renal cell carcinoma 5 cm, a radiographically normal contralateral kidney, and a WHO
3.
Results
Baseline characteristics are shown in Table 1 according to randomized treatment. The median age was 62 yr in each intervention arm, about two-thirds of all participants were
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
Table 1 – Baseline characteristics by randomized treatment. RN (N = 273) N Age (dichotomized at the <62 62 Sex Male Female Missing Chronic disease No Cardiovascular Pulmonary Other Missing WHO performance status 0 1 2 Missing Serum creatinine 1.25 ULN 1.26–2.5 ULN 2.6–5.0 ULN Missing
NSS (N = 268)
%
N
%
48.0 52.0
125 143
46.6 53.4
178 91 4
65.2 33.3 1.5
178 87 3
66.4 32.5 1.1
174 61 13 21 4
63.7 22.3 4.8 7.7 1.5
166 57 8 34 3
61.9 21.3 3.0 12.7 1.1
227 37 6 3
83.2 13.6 2.2 1.1
229 35 1 3
85.4 13.1 0.4 1.1
251 17 1 4
91.9 6.2 0.4 1.5
246 18 0 4
91.8 6.7 0.0 1.5
median), yr 131 142
NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal; WHO = World Health Organization.
male, the majority had no chronic disease with a WHO performance status score of 0 (fully active), and over 90% of all participants in each intervention arm had baseline serum creatinine 1.25 ULN.
3
Subgroup analyses of the incidence of eGFR decline below the specified thresholds over a median follow-up of 6.7 yr for eGFR, and tests for interaction are summarized in Figs. 2–4. None of the interaction p values were statistically significant at the conventional level of significance (alpha = 0.05). The incidence of severe renal dysfunction was rare in younger patients regardless of assigned treatment (Fig. 4). Subgroup analyses of all-cause mortality and tests for interactions are shown in Fig. 5 (median follow-up 9.3 yr). Here again, none of the interaction p values were below the conventional level of significance. It is noteworthy that for patients with baseline creatinine >1.25 ULN, the estimated hazard ratio reversed its direction (favoring the NSS over RN) relative to the rest of the study cohort (where survival was better with RN). This could, however, be a chance finding since there were only seven deaths in this subgroup of patients and the test for interaction was not statistically significant (p = 0.25). Multivariable analyses of independent predictors of eGFR decline below the specified thresholds and all-cause mortality are summarized in Table 2. In general, variables associated with lower follow-up eGFR included RN (vs NSS), older age, female sex, presence of chronic disease, and baseline serum creatinine >1.25 ULN, although not all these associations were statistically significant for all eGFR end points (Table 2). Association of chronic disease with lower eGFR seemed to increase in strength with decreasing eGFR cut-off. Baseline creatinine >1.25 ULN was associated with lower follow-up eGFR for all three eGFR cut-offs based on the point estimates, but this association did not reach statistical significance. There was only a very small
Fig. 2 – Subgroup analysis: progression to eGFR <60 ml/min/1.73 m2. CI = confidence interval; eGFR = estimated glomerular filtration rate; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 4
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
Fig. 3 – Subgroup analysis: progression to eGFR <45 ml/min/1.73 m2. CI = confidence interval; eGFR = estimated glomerular filtration rate; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
Fig. 4 – Subgroup analysis: progression to eGFR <30 ml/min/1.73 m2. CI = confidence interval; eGFR = estimated glomerular filtration rate; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
number of patients with baseline creatinine >1.25 ULN. Independent predictors of all-cause mortality were age, performance status, and (marginally) NSS. Association of chronic disease with mortality did not reach statistical
significance after adjusting for performance status and other variables, but it was estimated with a relatively low precision (wide confidence interval). In fact, based on location of the upper 95% confidence limit, observed data were
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 5
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
Fig. 5 – Subgroup analysis: all-cause mortality. CI = confidence interval; NSS = nephron-sparing surgery; RN = radical nephrectomy; ULN = upper limit of normal.
Table 2 – Multivariable analysis of progression to eGFR <60 ml/min/1.73 m2, eGFR <45 ml/min/1.73 m2, eGFR <30 ml/min/1.73 m2, and allcause mortality. Outcome
Variable
RR
eGFR < 60
Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN) Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN) Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN) Randomized treatment (NSS vs RN) Age (62 vs <62 yr) Sex (female vs male) Chronic disease (present vs absent) Performance status (1 vs 0) Baseline creatinine (>1.25 ULN vs 1.25 ULN)
0.25 4.99 2.53 1.08 0.67 2.55 0.32 4.19 2.30 1.41 0.70 2.03 0.56 8.49 1.51 2.48 1.03 2.59 1.46 3.78 0.79 1.40 2.50 1.00
eGFR < 45
eGFR < 30
Mortality
95% LCL
95% UCL
p value
0.16 3.04 1.50 0.65 0.32 0.87 0.21 2.75 1.52 0.92 0.39 0.91 0.28 2.93 0.76 1.22 0.46 0.95 1.01 2.35 0.53 0.95 1.63 0.46
0.40 8.19 4.28 1.79 1.40 7.46 0.48 6.40 3.49 2.15 1.26 4.55 1.11 24.61 3.01 5.05 2.30 7.04 2.12 6.10 1.18 2.06 3.82 2.18
<0.0001 <0.0001 0.001 0.77 0.28 0.09 <0.0001 <0.0001 <0.0001 0.11 0.24 0.09 0.10 <0.0001 0.24 0.01 0.95 0.06 0.05 <0.0001 0.25 0.09 <0001 0.99
eGFR = estimated glomerular filtration rate; LCL = lower confidence limit; NSS = nephron-sparing surgery; RN = radical nephrectomy; RR = relative risk; UCL = upper confidence limit; ULN = upper limit of normal. RR was measured by the odds ratio in the analysis of renal function, or the hazard ratio in the analysis of all-cause mortality. The logistic models were based on 513 patients with recorded values of all covariates and at least one follow-up eGFR measurement. The Cox model was based on 533 patients with recorded values of all covariates. The total number of randomized patients was 541. Note: RR < 1 favors the first level of the variable, for example NSS; RR > 1 favors the second level of the variable, for example RN.
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 6
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
still compatible with as much as a two-fold increase in mortality due to the presence of chronic disease (Table 2). 4.
Discussion
The main objective of subgroup analyses in clinical trials is to explore patient characteristics, which may modify the magnitude of the treatment effect, thus helping to identify patients who benefit the most or the least from the treatment under investigation [5,6]. While a subgroup analysis of clinical trials is a common practice, it has a number of limitations [5,6]. To show that treatment effect varies in magnitude across the subsets (this variation is known as treatment effect modification or covariate-by-treatment interaction), such an analysis first and foremost requires estimation of the treatment effect in subsets of the original study cohort, which results in a smaller sample size, lower precision of estimation (wide confidence intervals), and low power to reject the null hypothesis of no treatment effect in any given subset. In addition, subgroup analyses usually involve screening a large number of treatment–outcome associations within levels of baseline covariates, which increases the overall probability of a type I error (ie, finding a significant association by chance, where none actually exists). While there are different methods to control the analysis-wide type I error rate, they involve a further loss of power and precision, which is already low in subgroup analyses even without adjustment for multiplicity [7]. In other words, subgroup analyses generally have low ability to find true associations and are not unlikely to turn up false-positive associations. Despite these limitations, post hoc subgroup analyses, when appropriately regarded as exploratory and hypothesis generating, may produce valuable information [5–7]. In particular, in the case of EORTC trial 30904, it could be conjectured that a survival benefit of NSS, while not seen in the overall cohort of patients (mostly healthy with no baseline renal impairment and a normal contralateral kidney), could potentially be present in a small subset of patients with baseline renal impairment. However, because the number of patients with baseline creatinine >1.25 ULN was so small, the observed reversal of estimated hazard ratios (Fig. 5) cannot be interpreted as a strong or even a modest evidence of effect modification. Other variables such as age, sex, presence of chronic disease, and performance status could also potentially modify the magnitude of the treatment effect on the incidence of renal dysfunction and all-cause mortality. The present analysis, however, did not reveal strong evidence of such effect modification, although it was limited by low power. The most important clinical issue remains that most observational studies of NSS versus RN favor NSS on the basis of better kidney function, lower incidence of cardiovascular events, better overall survival, and good oncological outcomes. The benefit of NSS in terms of preservation of kidney function was demonstrated in EORTC 30904 as well as in meta-analyses of observational studies, and is generally not disputed [3,8–10]. However, the magnitude of this effect may depend on the baseline eGFR. In one
observational study of NSS (n = 540) versus RN (n = 766), regardless of surgery type, progression to stage IIIB or higher chronic kidney disease (CKD) was very rare in patients with preoperative eGFR >90 ml/min/1.73 m2 and was very common in patients with preoperative eGFR 45– 59 ml/min/1.73 m2. As a result, the benefit of NSS relative to RN was most noticeable in patients with baseline eGFR 60– 89 ml/min/1.73 m2 (CKD stage II) [11]. However, as noted in the editorial that accompanied this article [12], that analysis was limited by low precision of estimation within the subsets of the study cohort defined by baseline eGFR (due to sample size reduction by sub-setting) and by short average duration of follow-up (median 12 mo). Our current subgroup analysis was similarly limited by a small number of patients with baseline creatinine >1.25 ULN (Fig. 2–4). In meta-analyses of observational studies, the rates of kidney cancer recurrence and mortality tend to be lower with NSS than with RN, which is certainly a manifestation of selection bias [8–10]. As observational designs in general do not guarantee valid causal inference, the benefit of NSS relative to RN in terms of cardiovascular events and overall survival seen in observational studies may also potentially represent uncontrolled selection bias persisting even after adjustment for multiple baseline covariates. For example, in an observational study of cardiovascular events after NSS (n = 869) versus RN (n = 462) with detailed assessment and control of baseline cardiovascular risk factors, NSS was associated with a reduced rate of cardiovascular events (hazard ratio = 0.57, 95% CI: 0.34, 0.96) [13]. However, this analysis seemed to include adjustment for postoperative eGFR, which may be in the causal pathway from the type of surgery to the outcome. Furthermore, as was noted in an editorial, the unadjusted cumulative incidence curves for NSS versus RN separated nearly instantly after surgery in that study, while the true benefit of renal parenchymal preservation on cardiovascular risk would be expected to accumulate gradually over time. Similar trends were seen in propensitymatched analysis of overall survival after NSS versus RN, possibly representing uncontrolled selection bias, with healthier patients selected for the NSS [14]. In fact, current European Association of Urology guidelines recommend NSS rather than RN for most patients with a small renal mass, but explicitly recommend RN over NSS for patients who also have a “significant deterioration of a patient’s general health” [15]. In a large SEER-Medicare study, overall survival of patients treated with NSS (n = 1436) or RN (n = 4247) was compared with that of subsets of the general Medicare population matched separately to NSS and RN on demographic factors and various comorbidity indicators [16]. In that analysis, RN patients had the same overall survival as their matched controls, but NSS patients had better overall survival than their matched controls. Of course, the reasons for this survival advantage after NSS could not be related to nephron sparing because the control group did not undergo kidney surgery. These findings confirm that confounding mechanisms can operate in a very subtle manner. Given the limitations of observational data and the current widespread use of elective NSS, compared with when the EORTC 30904 trial began and was conducted, another
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015
EUF 294 1–7 7
E U R O P E A N U R O L O G Y F O C U S X X X ( 2 0 17 ) X X X – X X X
randomized trial of RN versus NSS is warranted. Such a trial could be limited to T1b–T2a tumors where oncological outcomes, regardless of types of surgery, are less favorable than they are for smaller masses and where less parenchyma is likely to be spared, on average, thanwith smaller renal masses [17]. Additionally, detailed information on baseline comorbidities and renal function should be recorded, baseline creatinine should be < 1.4 mg/dl, and nephrometry score, as a surrogate for degree of difficulty of NSS, should be recorded. In summary, this exploratory subgroup analysis of EORTC 30904 did not reveal strong evidence of treatment effect modification by baseline variables, although it was limited by low power. It is possible that in patients with baseline renal impairment, NSS improves overall survival relative to RN, but we could not prove this with any reasonable certainty due to the small sample size in this subgroup. For patients with no baseline CKD and a normal contralateral kidney, we feel that the original conclusion of EORTC 30904 still holds—if a partial nephrectomy is likely to be risky and very complex, a safe, straightforward, and uncomplicated RN may well be preferable.
work was performed under a prior affiliation of the author (University of Rochester). Funding/Support and role of the sponsor: This study was supported by Fonds Cancer (FOCA) from Belgium. The funding source had no role in the design, conduct, or analysis of this study, interpretation of the data, writing of this report, or decision to submit it for publication.
References [1] Van Poppel H, Da Pozzo L, Albrecht W, et al. A prospective randomized EORTC intergroup phase 3 study comparing the complications of elective nephron-sparing surgery and radical nephrectomy for low-stage renal cell carcinoma. Eur Urol 2007;51:1606–15. [2] Van Poppel H, Da Pozzo L, Albrecht W, et al. A prospective randomized EORTC intergroup phase 3 study comparing the oncologic outcome of elective nephron-sparing surgery and radical nephrectomy for lowstage renal cell carcinoma. Eur Urol 2011;59:543–52. [3] Scosyrev E, Messing EM, Sylvester R, Campbell S, Van Poppel H. Renal function after nephron-sparing surgery versus radical nephrectomy: results from EORTC randomized trial 30904. Eur Urol 2014;65:372–7. [4] Scosyrev E, Wu K, Levey HR, et al. Overall survival after partial versus radical nephrectomy for a small renal mass: systematic
5.
Conclusions
review of observational studies. Urol Pract 2014;1:27–34. [5] Zhang S, Liang F, Li W, Hu X. Subgroup analyses in reporting of phase
In this exploratory subgroup analysis of the EORTC trial 30904 of NSS versus RN, we aimed to determine whether treatment effect on the risk of renal dysfunction and all-cause mortality differed in magnitude across levels of baseline variables. This analysis did not reveal strong evidence of treatment effect modification, but it was limited by low power. Registration EORTC trial 30904 ClinicalTrials.gov identifier: NCT00002473 http://www.clinicaltrials.gov/ct2/results? term=NCT00002473
III clinical trials in solid tumors. J Clin Oncol 2015;33:1697–702. [6] Rothwell PM. Treating individuals 2. Subgroup analysis in randomised controlled trials: importance, indications, and interpretation. Lancet 2005;365:176–86. [7] Rothman K. No adjustments are needed for multiple comparisons. Epidemiology 1990;1:43–6. [8] Mir MC, Derweesh I, Porpiglia F, Zargar H, Mottrie A, Autorino R. Partial nephrectomy versus radical nephrectomy for clinical T1b and T2 renal tumors: a systematic review and meta-analysis of comparative studies. Eur Urol 2017;71(4):606–17. [9] Kim SP, Thompson RH, Boorjian SA, et al. Comparative effectiveness for survival and renal function of partial and radical nephrectomy for localized renal tumors: a systematic review and meta-analysis. J Urol 2012;188:51–7. [10] Pierorazio PM, Johnson MH, Patel HD, et al. Management of renal masses and localized renal cancer: systematic review and meta-
Author contributions: Edward M. Messing had full access to all the data in
analysis. J Urol 2016;196:989–99.
the study and takes responsibility for the integrity of the data and the
[11] Woldu SL, Weinberg AC, Korets R, et al. Who really benefits from
accuracy of the data analysis. Study concept and design: Van Poppel, Sylvester.
[12] Campbell SC, Demirjian S, Ercole C. Editorial comment: who really
Acquisition of data: Van Poppel, Sylvester. Analysis and interpretation of data: Scosyrev, Messing, Sylvester, Van
nephron-sparing surgery? Urology 2014;84:860–7. benefits from nephron-sparing surgery? Urology 2014;84:867–8. [13] Capitanio U, Terrone C, Antonelli A, et al. Nephron-sparing techni-
Poppel.
ques independently decrease the risk of cardiovascular events
Drafting of the manuscript: Scosyrev.
relative to radical nephrectomy in patients with a T1a–T1b renal
Critical revision of the manuscript for important intellectual content: Mes-
mass
sing, Sylvester, Van Poppel.
2015;67:683–9.
Statistical analysis: Scosyrev, Sylvester. Obtaining funding: Van Poppel, Sylvester.
and
normal
preoperative
renal
function.
Eur
Urol
[14] Tomaszewski JJ, Kutikov A. Retrospective comparison of cardiovascular risk in preselected patients undergoing kidney cancer surgery:
Administrative, technical, or material support: None.
reflection of reality or simply what we want to hear? Eur Urol
Supervision: Van Poppel, Sylvester.
2015;67:690–1.
Other: None.
[15] Ljungberg B, Bensalah K, Bex A, et al. Guidelines on renal cell
Financial disclosures: Edward M. Messing certifies that all conflicts of
[16] Shuch B, Hanley J, Lai J, et al. Overall survival advantage with partial
carcinoma. European Association of Urology 2014. p. 36. interest, including specific financial interests and relationships and affilia-
nephrectomy:
tions relevant to the subject matter or materials discussed in the manu-
2013;119:29812989.
a
bias
of
observational
data?
Cancer
script (eg, employment/affiliation, grants or funding, consultancies, hono-
[17] Campbell SC, Novick AC, Bukowski RM. Renal tumors. In: Walsh PC,
raria, stock ownership or options, expert testimony, royalties, or patents
Vaughan Jr ED, Retik AB, Wein AJ, editors. Campbell’s urology. ed. 9.
filed, received, or pending), are the following: None. Emil Scosyrev is
Philadelphia, PA: WB Saunders Co.; 2007.
currently employed by Novartis Pharmaceuticals Corporation. The present
Please cite this article in press as: Scosyrev E, et al. Exploratory Subgroup Analyses of Renal Function and Overall Survival in European Organization for Research and Treatment of Cancer randomized trial of Nephron-sparing Surgery Versus Radical Nephrectomy. Eur Urol Focus (2017), http://dx.doi.org/10.1016/j.euf.2017.02.015