To compare the outcomes of patients who had a elective partial nephrectomy (PN) or radical nephrectomy (RN) for clear cell renal cell carcinoma (RCC) of 4–7 cm.
To compare the outcomes of patients who had a elective partial nephrectomy (PN) or radical nephrectomy (RN) for clear cell renal cell carcinoma (RCC) of 4–7 cm.
From March 1998 to July 2004, 45 and 151 patients underwent PN and RN, respectively, for clear cell RCC. A multivariate Cox model was constructed for disease-free survival with adjustment for markers of disease severity, and a propensity-score approach used as a confirmatory analysis.
In the PN and RN cohorts the treatment failed in one and 20 patients, respectively; the median follow-up was 21 months. The hazard ratio (95% confidence interval) for PN after adjusting for disease severity was 0.36 (0.05–2.82; P = 0.3). Using planned PN as a predictor (intent-to-treat analysis) the hazard ratio was 1.06 (0.32–3.53; P = 0.9). In the propensity-score model, planned PN was associated with a hazard ratio of 1.75 (0.50–6.14; P = 0.4). The serum creatinine level 3 months after surgery was significantly lower in patients who had PN, with a difference between the means of 0.36 (0.23–0.48; P < 0.001).
Renal function was preserved after PN for 4–7 cm clear cell RCC tumours. When comparing the outcomes of PN and RN it is important to consider the intended operation as an independent variable. There was no clear evidence that PN was associated with worse cancer control, although a continued follow-up of this and other cohorts is warranted.
partial, radical nephrectomy
University of California Los Angeles
Elective partial nephrectomy (PN) is developing into the standard of care in the USA for renal cortical tumours of <4 cm. A large minority of these lesions at pathological examination of the surgical specimen have benign or indolent histology that could not have been predicted before surgery [1–3], and PN tends to preserve renal function better than radical nephrectomy (RN) [4–7]. PN has been shown to control cancer with few complications for patients with RCC tumours of ≤ 4 cm [8,9]. Although earlier studies suggested that 4 cm be established as the maximum size for PN , recent data suggest that it might be possible to extend similar success and benefits to the treatment of tumours of ≥ 4 cm [11–13].
The important factors affecting the outcome of nephrectomy are tumour size, stage, Fuhrman grade and RCC histological subtype, with the clear cell subtype conferring the worst prognosis [14,15]. However, none of these factors are fully known until the surgical specimen has been examined. Therefore surgeons attempting and patients undergoing PN must understand the implications of oncological control in the worst-case scenario before proceeding. In this context, our objective was to determine whether the type of intervention, elective PN vs RN, affected the prognosis for clear cell RCC.
With Institutional Review Board approval, we reviewed our departmental renal cancer database, which is updated prospectively, but some missing data were collected retrospectively. At the time of analysis (July 2004) the database included 1194 patients. The first PN for a 4–7 cm clear cell RCC tumour at our institution was done in March 1998; from that date until July 2004, 196 patients with a normal contralateral kidney had either elective PN (45) or RN (151) for suspicious solitary renal lesions detected on imaging studies. The tumour presentation, either incidental or symptomatic, was noted at the initial evaluation; both local and/or systemic symptoms were considered as symptomatic presentations.
Our protocol was to initially offer elective PN to most patients with tumours of ≤ 7 cm if not visibly invading the vascular hilum on preoperative radiographic studies. There was an understanding with the patient that if, after a visual, manual and ultrasonographic examination of the kidney during surgery, the surgeon judged that a PN was not technically feasible, then RN would follow. Factors involved in the surgical judgement to perform RN included multifocality, invasion into segmental or main vessels, and the likelihood of minimal residual parenchyma after PN without compromising the margins. Central location alone was not a sufficient reason to perform RN. Ultimately, the treatment rendered was at the surgeon's discretion. The surgeon noted at the initial consultation, or we extracted from the medical record or surgical consents, whether a PN or RN was planned. If any mention was made of performing PN, we interpreted this as a planned PN.
A standard technique was used for RN and PN, and intraoperative ultrasonography was used routinely if PN was planned. Fourteen patients had a laparoscopic RN. Tumours were assessed by the pathologist for the following features: size was reported as the single greatest dimension at gross examination; tumours were assigned a histological subtype according to the Heidelberg classification scheme ; clear cell carcinomas were assigned a Fuhrman grade; grades 1 and 2 RCC were considered low-grade, and grades 3 and 4 as high-grade; tumours were staged according to the 2002 TNM staging system.
Patients were followed after surgery with a physical examination, chest radiography and serum creatinine measurement every 6 months, and abdominal CT annually. All-cause deaths, cancer-specific deaths, local or distant recurrences, or development of a contralateral tumour were defined as treatment failure; these events were used to estimate disease-free survival as the only survival endpoint in all models.
The analytical plan was to compare disease-free survival after PN and RN using multivariate Cox proportional hazards models with adjustment for disease severity. We initially used Kaplan–Meier and univariate Cox analysis, but the immediate problem was that there were only 21 events in the study cohort, compromising the validity of any direct adjustment. We considered two separate approaches: a ‘multivariate confounder score’ and propensity-score method . The latter is a more common approach, but would not take full advantage of our database, which includes many hundreds of patients treated by RN before the first PN was conducted. We therefore conducted the multivariate confounder score as the primary analysis, and used propensity scores as a secondary analysis, to determine whether our findings were affected by the methodological approach. The multivariate confounder score and the propensity score are both two-stage approaches. In the first stage, a predictive model is constructed using variables related to disease severity, and then a prediction made on the basis of this model. This prediction relates to the hazard of failure for the multivariate confounder score and the probability of receiving a PN in the propensity score approach. In the second stage, the prediction is used as the covariate in a Cox model with treatment as the predictor. Hence the multivariate model has only two terms, and can therefore be used for a dataset with 21 events.
For the confounder-score approach, we first created a multivariate model for treatment failure events using patients in the database receiving RN, including those treated before PN was first used on 4–7 cm tumours. We used two different data sets to create confounder scores: the 292 patients in our database with a 4–7 cm clear cell tumour who were treated by RN, and all 700 patients receiving RN, irrespective of tumour size and histology. The results of each approach were almost identical and so we present the data only for the former analysis. Predictors included age, stage, grade, size, date of surgery, vascular invasion, and whether the patient was symptomatic at presentation. Fuhrman grades 1 and 2 were considered low-grade and grades 3 and 4 high-grade; stage was classified as T1 or T3. In the first stage of the confounder score approach, we created a Cox model using the clinicopathological predictors described above. This model was then applied to the 196 patients in the present study and a linear predictor of hazard calculated. The linear predictor for patient i is Σβy xy, where βy is the coefficient (i.e. log hazard ratio) for the yth variable and xy is the value of the yth variable for patient i. The linear hazard therefore summarizes the prognostic information from numerous variables related to disease severity into a single variable. This predicted linear hazard was used as a covariate in the second stage Cox model. Such an approach is sound if the correlation between the covariate score and the treatment is <0.5–0.6 ; in the present study it was close to 0.3.
For the propensity approach, we used the 196 patients in our main comparison to create a logistic model to predict which patient would receive PN, using the predictor variables described above. The predicted probability of treatment (‘propensity score’) was then used as a covariate in a Cox model to determine the effects of surgery type.
Differences in creatinine levels were tested by analysis of covariance, with baseline creatinine as the covariate. All analyses were conducted on Stata 8 (Stata Corp., College Station, TX, USA).
From March 1998 until July 2004, 45 patients had PN for clear cell RCC of 4–7 cm; the demographic and pathological features of these patients are compared to the 151 who had RN for similar RCC during the same period (Table 1). The PN cohort tended to be younger and more frequently male and Caucasian. The tumours in the PN cohort also appeared to have more favourable pathology, including being smaller and of lower stage and grade. All 45 PN cases had negative surgical margins. The vascular invasion status was often unknown in both cohorts, but more frequently unknown in the PN patients. There were one and 20 treatment failures in the PN and RN cohorts, respectively. The median follow-up for survivors was 21 months.
|Number of patients||45||151|
|Age, years||56.7 (13.0)||63.1 (11.5)|
|Tumour size, cm||4.85 (0.94)||5.42 (0.89)|
|Serum creatinine, mg/dL:|
|before surgery||1.16 (0.33)||1.09 (0.36)|
|≥ 3 months after surgery||1.39 (0.60)||1.60 (0.76)|
|Male||32 (71)||99 (66)|
|Year of surgery|
|1998||1 (2)||17 (11)|
|1999||3 (7)||21 (14)|
|2000||6 (13)||25 (17)|
|2001||7 (16)||27 (18)|
|2002||8 (18)||22 (15)|
|2003||10 (22)||28 (19)|
|2004||10 (22)||11 (7)|
|Race (Caucasian)||42 (93)||127 (84)|
|pT1||41 (91)||124 (82)|
|pT3||4 (9)||27 (18)|
|Incidental presentation||29 (64)||111 (73.5)|
|Symptomatic (local or systemic)||16 (36)||40 (27)|
|Low grade||35 (78)||107 (71)|
|High grade||9 (20)||43 (28)|
|Grade unknown||1 (2)||1 (1)|
|absent||19 (42)||97 (64)|
|present||1 (2)||6 (4)|
|unknown||25 (56)||48 (32)|
Kaplan–Meier curves for disease-free survival are presented by operation (Fig. 1). In the univariate analysis, PN appeared to be protective over RN for disease-free survival (hazard ratio 0.22; 95% CI, 0.03–1.66; P = 0.14), but this difference was not statistically significant. Table 2 shows the multivariate model used for the confounder score approach, which was based on the 292 RN patients with 4–7 cm clear cell RCC. As would be expected, the hazard ratios associated with tumour size and vascular invasion were large and statistically significant. There was a small effect of age; the effects of grade and stage were equivocal. The concordance index for the model was 0.672. Using the confounder score approach to adjust for multiple covariates simultaneously in the final Cox model, we calculated a hazard ratio for PN that still indicated a protective effect for disease-free survival over RN (0.36; 95% CI, 0.05–2.82; P = 0.3); however, this difference was not statistically significant.
|Characteristic (n)||Hazard ratio (95% CI)||P|
|Age, years||1.03 (1.00–1.05)||0.017|
|pT3 (62)||1.03 (0.58–1.83)||0.9|
|low (181)||1.23 (0.65–2.33)||0.5|
|high (58)||0.77 (0.31–1.92)||0.6|
|Symptomatic presentation||0.64 (0.35–1.17)||0.15|
|Tumour size, cm||1.49 (1.18–1.89)||0.001|
|present (9)||4.59 (1.70–12.40)||0.003|
|unknown (53)||1.03 (0.49–2.14)||0.9|
|Date of surgery, years||1.01 (0.91–1.13)||0.8|
The protective effect of PN was obviously counter-intuitive. Factors which the surgeon recognized during surgery that dissuaded the use of a PN, whether consciously recognized or not, may have been associated with a poorer prognosis. We therefore sought to understand these factors, using an intent-to-treat analysis with the confounder score approach as above, but using ‘planned operation’ as the predictor. The linear predictions of hazard obtained from the initial multivariable Cox model and clinical features by treatment group are summarized in Table 3. The lowest hazard was associated with patients who had a PN, intermediate in those who had a planned PN but received RN, and highest in those who had a planned RN. The association between type of surgery and linear prediction of hazard from the multivariate confounder score was highly significant (P < 0.001). However, after adjusting for covariates a planned PN was associated with a hazard ratio of 1.06 (95% CI, 0.32–3.53; P = 0.9). Given these findings, we used the planned PN in our propensity-score model. Increasing age and size were the variables most strongly predictive of a planned RN. The area under the curve for the model was 0.740. The risk factors were approximately balanced across treatment groups after stratifying for propensity score (Tables 4 and 5). In a Cox model adjusting for propensity score, planned PN was associated with a hazard ratio of 1.75 (95% CI, 0.50–6.14; P = 0.4). Although this hazard ratio is higher than in the multivariate model using the confounder score approach, we consider the estimates similar in the context of the wide CIs associated with each.
|Surgery performed||Surgery planned|
|Number of patients||129||22|
|Age, years||64 (11.5)||60 (11.0)|
|Tumour size, cm||5.5 (0.96)||5.1 (0.82)|
|Median follow-up for survivors, months||23||21|
|Linear prediction of hazard from covariates||4.38 (0.74)||4.00 (0.67)|
|Recurrences*||17 (13)||3 (14)|
|Number of patients||0||45|
|Age, years||–||57 (13.0)|
|Tumour size, cm||–||4.8 (0.89)|
|Median follow-up for survivors, months||–||14|
|Linear prediction of hazard from covariates||–||3.65 (0.60)|
|Variable||Propensity score quintile|
|N (%)||36 (92)||3 (8)||32 (82)||7 (18)||21 (54)||18 (46)||21 (54)||18 (46)||19 (48)||21 (53)|
|Mean (sd) Age, years||71 (9.6)||73 (7.6)||66 (9.4)||67 (8.3)||63 (11.4)||61 (10.1)||60 (9.5)||58 (13.6)||52 (9.9)||50 (10.3)|
|Tumour size, cm||6.2 (0.7)||6.3 (0.3)||5.6 (0.9)||5.9 (0.6)||5.3 (0.9)||5.1 (0.9)||4.8 (0.7)||5.0 (0.6)||4.7 (0.7)||4.2 (0.4)|
|Stage pT3, %||11||0||31||29||14||17||5||11||16||14|
|Characteristic||PN planned||RN||Unadjusted P||Adjusted P*|
|Number of patients||67||129|
|Age, years||58 (12.4)||64 (11.5)||<0.005||0.9|
|Tumour size, cm||4.94 (0.87)||5.47 (0.96)||<0.005||0.9|
|Year of surgery||0.01||0.9|
|1998||2 (3)||16 (12)|
|1999||5 (7)||19 (15)|
|2000||10 (15)||21 (16)|
|2001||12 (18)||22 (17)|
|2002||11 (16)||19 (15)|
|2003||13 (19)||25 (19)|
|2004||14 (21)||7 (5)|
|Male||48 (72)||83 (64)||0.3||0.4|
|Race (Caucasian)||58 (87)||111 (86)||0.9||0.6|
|pT1||57 (85)||108 (84)|
|pT3||10 (15)||21 (16)|
|incidental||46 (69)||94 (73)|
|symptomatic (local or systemic)||21 (31)||35 (27)|
|low||50 (75)||92 (71)|
|high||16 (24)||36 (28)|
|unknown||1 (1)||1 (1)|
|absent||33 (49)||83 (64)|
|present||1 (1)||6 (5)|
|unknown||33 (49)||40 (31)|
Data on serum creatinine 3 months after surgery were available for 182 patients (93%); the mean values are shown in Table 1 for patients receiving PN and RN, respectively; the values for the intention-to-treat analysis were 1.45 (sd, 1.73) in 64 patients and 1.59 (0.79) in 118. The increase in creatinine level was significantly smaller in the PN group when analysing both treatment received (difference between means 0.36 mg/dL; 95% CI, 0.23–0.48; P < 0.001 by anova) or treatment planned (difference between means 0.23 mg/dL; 95% CI 0.11–0.34; P < 0.001). The results were similar for the 6–12 month creatinine measurements (difference between means for treatment planned: 0.21 mg/dL; 95% CI 0.09–0.33; P = 0.001). There were no important differences between the groups in the delay between surgery and follow-up creatinine estimates, and adding delay as a covariate had no impact on the findings. Thus, PN appeared to diminish the rise in creatinine level after nephrectomy.
It is well established that clear cell RCC is the worst histological subtype of RCC. In the Mayo Clinic series of 2385 patients, the 5-year cancer-specific survival for clear cell tumours was 68.9%. This is in contrast to 87.4% and 86.7% 5-year cancer-specific survival for papillary and chromophobe subtypes, respectively. Based on the success of PN in tumours of ≤ 4 cm, the technique is increasingly being applied to tumours of ≥ 4 cm, but the effect of larger tumours remains a concern. Investigators from the Mayo Clinic examined 840 patients with pT1 tumours who had RN ; in their multivariate analysis of patients with clear cell RCC (81.2% of the cases), both Fuhrman grade ≥ 3 and tumour size were significant variables for predicting cancer-specific and metastasis-free survival. Although size is important, it appears that it should not be used as a sole determinant in pursuing PN rather than RN. Two recent comprehensive studies examined the outcomes after PN for tumours of ≤ 7 cm. A large multi-institutional international study including data from the University of California Los Angeles (UCLA) compared 379 cases of PN with 1075 of RN for pT1 N0 M0 tumours of <7 cm . In the subset of pT1b tumours (4–7 cm), 65 patients had PN (17.1% of overall PN cases) and 576 had RN (54% of overall RN cases). There was a similar distribution of clear cell and non-clear cell histology between the PN and RN patients. These investigators reported no differences in local recurrence, distant recurrence, or cancer-specific survival between patients who had PN or RN for pT1b tumours; however, the rate of distant recurrences was nearly double for the RN patients (15.6% vs 7.1%). In a univariate analysis of cancer-specific survival, Eastern Cooperative Oncology Group status and Fuhrman grade were significant, whereas histological type, surgery (PN vs RN), and tumour size (pT1a vs pT1b) were not. Patients who had RN for pT1b tumours had a statistically significantly higher rate of local and distant recurrences than did pT1a patients, but this pattern did not similarly affect PN patients.
A recent study from the Mayo Clinic compared 91 patients treated by PN for sporadic, unilateral, nonmetastatic tumours of <7 cm, with 841 treated by RN . Two-thirds of the PN and RN patients had pT1b tumours of 4–7 cm. Notably, only one patient treated by PN (1.1%), vs 101 (12.1%) treated by RN had ≥ pT3 stage tumours. Clear cell histology was more frequent in the RN than in the PN patients (83% vs 64%). In univariate analyses of cancer-specific and distant metastasis-free survival, the hazard ratios were >3 and statistically significant for RN, whereas the hazard ratio for recurrence-free survival was greater and statistically significant for PN. Histological subtype was considered as a single variable in multivariate analyses, therefore the risk associated with clear cell tumours alone was not assessed; however, there were no statistically significant differences in the three outcome possibilities between PN and RN. An interesting subgroup analysis of patients with CT findings available for review showed that patients treated by PN were more likely (63% vs 20%) than those treated by RN to have exophytic tumours identified on preoperative CT. Despite these studies showing favourable outcomes in patients undergoing PN for tumours of ≤ 7 cm, they did not specifically analyse clear cell RCC alone as a variable in the analyses. However, these studies provide a context for the successful treatment of clear cell RCC of 4–7 cm with PN.
Because clear cell histology has the worst potential outcome, our objective was to determine whether patients who had PN and were found to have clear cell histology fared worse than the corresponding RN patients. This is important because the histology of renal tumours cannot be reliably assessed until pathological confirmation, but physicians and patients need to be aware of differences in outcome with the worst possible histology of the tumour before choosing the appropriate operation. We considered only clear cell tumours in our analysis, but retrospective observational studies such as ours attempting to distinguish difference in outcome suffer from inherent selection bias that may obscure actual differences in treatment. If the intent-to-treat and disease severity are not carefully considered, the selection may obscure actual differences in treatment even in a multivariate analysis. For example, in the present disease-free survival multivariate model not considering intent-to-treat, PN was protective compared to RN; the hazard ratio associated with PN relative to RN was 0.36 (95% CI, 0.05–2.82; P = 0.3). The Mayo Clinic also reported risk ratios for RN that were worse relative to PN in a multivariate analysis . The risk ratios of RN relative to PN were 1.60 (95% CI, 0.50–5.12; P = 0.430) for cancer-specific and 1.76 (95% CI, 0.64–4.83; P = 0.273) for distant metastasis-free survival, respectively. The UCLA also reported a hazard ratio of 0.8 (not significant) for PN relative to RN in a univariate analysis predicting cancer-specific survival . Neither the Mayo nor UCLA study considered intent-to-treat and this may partly explain why these studies also showed a protective effect of PN. Because our objective was to most accurately determine if outcomes for 4–7 cm clear cell tumours were different based on treatment, we were compelled to consider intent-to-treat and control for disease severity in the models. Our propensity score model indicated a bias for treating older patients with larger clear cell RCC tumours by RN, and our initial analysis implied that PN offered a survival benefit over RN. However, the apparent benefit disappeared when we used an intent-to-treat analysis, analysing the procedure that was planned rather than the procedure that was used. This suggested that some factors became more readily apparent during the operation that technically precluded a PN. We noted that patients in whom PN was planned but actually had RN were intermediate in age and tumour size between the PN and RN cohorts (Table 3), but these were unlikely to be the sole factors that contributed to performing a RN. More likely was tumour too close to the hilum to allow a safe attempt at PN, or that performing a PN in some other way exceeded the technical capabilities of the operating surgeon
To identify non-exophytic vs exophytic tumours, one radiologist reviewed the CT data before surgery from a subset of 76 patients in the current study; of these, 11 had PN planned but ultimately had RN, including 10 with non-exophytic tumours and one with an exophytic tumour. Interestingly, nearly all the patients in this small subset had a central tumour. This is consistent with the observation of Leibovich et al. and supports the hypothesis that tumour location probably altered the choice of operation used. Although we were unable to evaluate the effect of tumour location in our models in the current study, it would be interesting to investigate differences in the outcome of 4–7 cm tumours by location
In the present study the benefit of PN over RN was apparent as a marginally better serum creatinine level at 3 and 6 months after PN than after RN; however, this is a short follow-up for assessing renal function. A study of matched patients with a long-term follow-up of renal function at 10 years found that the cumulative incidence of renal insufficiency, defined as a serum creatinine level of >2 mg/dL, after RN was twice that after PN . That study suggests that with a longer follow-up of the present patients we would expect a similar difference. Diminished renal function is associated with a worse quality of life , but it is not yet known whether it also is associated with greater mortality. Our finding of improved renal preservation has important implications for the increasing popularity of laparoscopic RN. It would be interesting for future study to note if the immediate benefits of laparoscopic nephrectomy, e.g. quicker recovery, would outweigh the long-term benefit of renal preservation after open PN of 4–7 cm tumours, but is beyond the scope of the current study. In addition, the vast majority of patients with clear cell tumours will survive and thus remain at risk of developing a contralateral tumour .
Although there were some differences between the cohorts, we attempted to control for these in the design of our model, including size and pathological features of the tumours, and patient age. The present study had other limitations, including small size and relatively short follow-up. The protective effect of PN reflected an operative bias and that planned operative type, either PN or RN, did not determine the likelihood of treatment failure. Given the wide CIs currently calculated with the hazard ratios, an extended follow-up is necessary before conclusively excluding the possibility that patients fare worse after PN than RN.
We continue to recommend PN for patients with 4–7 cm renal tumours if technically feasible. In our experience of 45 patients with a short follow-up it appears that those with clear cell pathology who had PN had no worse an outcome and had better preservation of renal function than comparable patients after RN. This may be a technically demanding operation, but can be safe in skilled and experienced hands. Despite often re-imaging patients with CT with three-dimensional reconstruction, MRI and renal ultrasonography before surgery, planned PN may not be feasible due to the proximity to major vascular structures, the resection of which, to achieve a clear surgical margin, leaves too little salvageable kidney. Extensive resection involving the collecting system, with subsequent reconstruction, is of less concern, particularly with adequate drainage of the retroperitoneum. In our experience, ureteric stents are not routinely used. We recommend using thorough manual and intraoperative ultrasonography to exclude multifocality, invasion of segmental or main vessels, need for extensive collecting system reconstruction, or likelihood of minimal residual parenchyma after PN, to assist in judging if a PN is technically feasible and warranted.
In conclusion, the present study addresses whether PN can be applied to 4–7 cm RCC with the worst subtype, clear cell histology, without adversely affecting oncological outcome. It is critical to consider the intended treatment as an independent variable. There was no clear evidence that PN gave a poorer outcome, although given the moderate sample size, we were unable to exclude clinically relevant harm. PN was clearly better for preserving renal function, being associated with a significantly smaller increase in creatinine levels after surgery. Therefore, we consider that our current practice of offering PN by experienced surgeons is justified, with the understanding that RN may be used because of technical limitations. We intend to continue monitoring this cohort, and future patients, to obtain more precise estimates of the relative oncological effectiveness of PN and RN.
Atreya Dash is supported by a gift from the Tina and Richard V. Carolan Foundation and by a Training Grant T32–82088 from the National Institutes of Health.