- Top of page
- PATIENTS AND METHODS
- FUNDING SOURCES
- CONFLICT OF INTEREST DISCLOSURES
Radical nephrectomy (RN) was initially described a half-century ago and involves the complete removal of the kidney, Gerota's fascia, and the ipsilateral adrenal gland. However, en bloc excision of whole organs has been associated with serious adverse effects. In many fields, properly selected patients have been offered a functional lumpectomy, sparing adjacent normal tissue while providing equivalent cancer outcomes. For the past few decades, RN has been the standard treatment for patients who have a small (≤4 cm) renal mass and 2 functional kidneys. However, after multiple retrospective series demonstrated oncologic equipoise, partial nephrectomy (PN) became an option, allowing for the preservation of normal renal parenchyma.[1-3]
Whereas adverse health-related quality of life outcomes associated with procedures such as radical mastectomy are immediately visible, those associated with renal cell carcinoma (RCC) treatment are more subtle due to the long-term consequences of nephron loss. Although it has long been recognized that rendering a patient anephric has life-altering consequences, RN was not believed to pose significant health consequences in the presence of a normal contralateral kidney. In recent years, important studies have demonstrated a correlation between the extent of chronic kidney disease (CKD) and increased rates of cardiac events, hospitalizations, and infections.[4, 5]
Although data on adverse health outcomes have come largely from patients with medical CKD, they have been considered applicable to surgically induced CKD. Several retrospective series have demonstrated that PN could limit the occurrence of CKD compared with RN.[12, 13] Whether decreasing the incidence of CKD with PN translates into an improvement in overall survival (OS) has been intensely debated. Recently, the results of a randomized phase 3 trial by the European Organization for Research and Treatment of Cancer (EORTC 30904) demonstrated an OS advantage with RN compared with PN. Critics of this trial point out that it was underpowered and had significant crossover between treatments; however, it is the only level 1 evidence addressing this question.
Recently, Tan et al applied an instrumental variable analysis (IVA) to minimize selection bias in the comparison between PN and RN. Using SEER-Medicare data, they found that PN provided superior OS. Other studies by Kowalczyk et al and Huang et al have demonstrated similar improvements in OS for small renal tumors treated with PN. Kim et al recently reported the results of a meta-analysis comparing PN and RN and found that PN was associated not only with improved OS, but also with cancer-specific survival (CSS). As there is no oncologic explanation for this finding, it raises the question of whether this is due to selection bias rather than treatment effect.
Due to an uncertain survival advantage with PN and concerns a direct comparison to RN may be fraught with selection bias, we conducted a SEER-Medicare matched cohort study comparing RCC surgical groups with separate controls. By matching PN and RN with separate cancer and noncancer controls, we evaluated how each procedure affected OS. We hypothesized that if RN worsened OS compared with PN, survival would be worse than controls. Similarly, if improvement in OS observed with PN is due to selection bias, then survival would be improved compared with controls.
- Top of page
- PATIENTS AND METHODS
- FUNDING SOURCES
- CONFLICT OF INTEREST DISCLOSURES
Our study has several significant findings relevant to the continued uncertainty over the optimal surgical management of the SRM. First, it confirms previous observations based on SEER-Medicare data showing that PN has a significant survival advantage compared with RN. This expected finding of PN having better outcomes than RN highlights how observational studies can inform the debate about optimal management of SRM, yet may be misguided due to the inherent limitations of such data. Second, we demonstrate that RN may not compromise OS based on comparisons with matched BCC and NCC groups. Third, patients who underwent PN had significantly improved OS compared with the matched BCC and NCC control groups (HR of 1.36 and 1.26, respectively). Finally, we explored why OS would be improved with PN and found that thoracic/lung and cardiovascular causes of death were less frequent in the PN group compared with the BCC group.
SEER-Medicare studies repeatedly demonstrate an improvement in survival with PN (Fig. 1), which conflicts with the results of the recent randomized phase 3 EORTC 30904 trial.[6, 7] Because the contrasting studies raise uncertainty over the optimal surgical treatment of the SRM, physicians managing patients diagnosed with SRM must decide whether to rely on level 1 evidence, which may have design flaws, or low-quality evidence from either single institutional or population-based observational data. With continued uncertainty, many academic clinicians have promoted PN to limit adverse renal outcomes associated with RN demonstrated in single institutional and claims-based comparisons.[12, 23] With these concerns in mind, the American Urologic Association issued the “Guideline for the Management of the Clinical T1 Renal Mass” and declared PN the standard of care for healthy patients with a cT1a tumor. In this report, RN was considered an “alternative standard of care if PN is not technically feasible as determined by the urologic surgeon.” Increasing evidence from observational studies reporting the benefits of PN relative to RN and current clinical guidelines have led to increased use of PN over the past decade.
Regardless, we demonstrated that patients undergoing RN did not have worse survival than their matched NCC and BCC controls. This is in contrast to the argument in favor of performing PN to avoid long-term health consequences associated with RN that may worsen survival.[9, 25] Although estimated glomerular filtration rate (GFR) is considered the best indicator of overall renal function, recent evidence suggests that GFR alone may not adequately reflect the extent of kidney disease in a patient with a reduced number of nephrons. A patient with a low GFR after nephrectomy may have fewer long-term cardiovascular risks than a patient with similar GFR, two kidneys, but uncontrolled hypertension and diabetes. This finding is supported by a study of living kidney donors in which patients were not adversely affected by the loss of a kidney. However, one may argue that kidney cancer patients are far different than highly selected renal donors, since many RCC patients have preexisting CKD or are of older age at time of diagnosis, which can be significantly worsened with treatment. Although this may be true, our findings also suggest that RN may pose less harm than believed previously, because OS was similar compared with NCC and BCC controls.
Clinical trials often cannot be undertaken due to ethical, financial, or feasibility concerns. In the absence of a clinical trial, patients and physicians must rely on well-designed observational studies to evaluate the comparative effectiveness of 2 different treatment alternatives for a given disease (in this case, SRM). Multiple studies have shown that PN has improved survival compared with RN.[7, 9, 10, 25] However, observational series are limited by their inability to appropriately control for latent variables that are accounted for in randomized trials. In patients with kidney cancer, multiple unmeasured factors—including body habitus, prior abdominal surgery, and tumor complexity—may influence treatment decisions. Additionally, in observational data, no method can account for the alteration in preoperative planning due to unanticipated, intraoperative findings. This is not an infrequent event, as it occurred in 14.6% of patients randomized to PN who went onto to receive RN in the EORTC 30904 trial. In a randomized trial, this would could be accounted for by cross over, but in observational series in which cross-over data are not available, these patients with more aggressive disease would be identified only in the RN group and could affect both CSS and OS.
Several approaches are available to counter bias that attends analysis of observational data. Propensity scoring approaches and IVA have become more frequent in the recent literature. IVA, which was recently used by Tan et al in their SEER-Medicare comparison of PN versus RN, has the advantage of correcting for both observed and unobserved sources of endogeneity, when the instrument employed is robust. In fact, Tan et al used IVA to test their hypothesis that the survival outcomes reported by EORTC 30904, an actual randomized control trial of PN versus RN, were unreliable due to study limitations. Our data provide another perspective on this puzzle. While greedy algorithm matching does not account for unobserved bias, it is an excellent approach for reducing bias from observed sources. By selecting comparators other than patients undergoing RN, we allow a different inference to be drawn than was possible previously. Because we can identify no biologically plausible reason for patients undergoing PN to have higher OS than their matched counterparts with low-risk bladder cancer or those without cancer, we must conclude that findings regarding OS after PN from prior work using propensity scoring approaches or standard multivariable regression analysis are flawed. The implications of our findings for studies of this topic using IVA are less clear and rest on an appraisal of the performance of the instruments used in such studies.
We performed an exploratory analysis to investigate differences in the cause of death reported by SEER between the PN and BCC groups. Both thoracic/lung cavity and cardiovascular deaths were increased in the BCC groups. Because smoking, the main risk factor for bladder cancer, is unable to be characterized with SEER-Medicare data, it is possible that this greatly influences the results of this comparison and may bias comparisons with RN as well.
Some critics have dismissed the EORTC 30904 trial for its various flaws in favor of observational data supporting PN. We caution against ignoring the level 1 evidence demonstrating that OS is at least equivalent (and perhaps improved) after RN. Although we do not advocate abandoning PN, our data suggest that the survival benefit of PN in observational studies may be based on selection bias. If so, the relegation of RN, with its attendant reduction in perioperative risks compared with PN, to “alternative standard of care” status may merit revisitation by guideline-producing organizations. This matter should be considered before embarking on a planned randomized control trial for clinical stage T1b/2a renal tumors in which OS is considered the primary end point.
Our study has some limitations, including the fact that the SEER-Medicare database is restricted to the elderly population. Because the median age of onset is approximately 64 years, half of the patients with kidney cancer are younger than the patients in our cohort. It is possible that age influences the benefit of nephron-sparing surgery, as younger patients may have more time to develop long-term sequelae from RN. This has been suggested in prior studies demonstrating that improvement in OS may be more pronounced in younger patients and that increased morbidity of PN in the elderly may offset survival advantage.[25, 31] Finally, although we excluded patients with preexisting ESRD at diagnosis, some patients included in the analysis may not have had a normal contralateral kidney.
Although the optimal surgical management of the SRM is frequently debated, both PN and RN options provide excellent CSS. The results of the EORTC 30904 trial and several observational datasets differ in their conclusion regarding the most effective surgical treatment modality. In our analysis of elderly patients diagnosed with RCC in SEER-Medicare data, we demonstrate that RN does not worsen survival compared with matched NCC and BCC groups. Additionally, because PN significantly improved survival compared with similarly matched groups, this finding indicates that the SEER-Medicare observational data are fraught with selection bias that may be impossible to overcome despite the use of analytic methods. Although preservation of renal function is an important concept and we generally advocate for PN, surgical treatment should be individualized on a case-by-case basis. Based on the EORTC 30904 trial and concerns over selection bias with observational data, both RN and PN should be considered a standard of care in the setting of normal overall renal function.