Patterns of Disease Recurrence

Integral to designing appropriate adjuvant or neoadjuvant strategies is a basic understanding of which patient and tumor characteristics are associated with the risk of local or systemic recurrence. For patients with local or locally advanced RCC, the risk of recurrence following partial or radical nephrectomy largely depends on tumor size, stage, grade, histology, completeness of resection, presence of symptoms, and performance status, and ranges from 15% to 27% at 5 years. Local or contralateral recurrence following surgical resection is rare, but it is more common following partial nephrectomy. The most frequent sites of distant recurrence include the lung, lymph nodes, liver, bone, and brain; prognosis for patients with mRCC is very poor, with a 5-year survival of less than 10%. Although the majority of recurrences are discovered during routine radiographic surveillance, most patients who relapse have distant metastases and are rarely cured. Although patients are at a lifelong risk, the greatest recurrence risk is in the first 3 to 5 years following surgery, with approximately 10% recurring after 5 years. Patients with a higher tumor grade and stage at initial presentation seem to be at a higher risk for early recurrence, whereas recurrent lesions detected at delayed intervals are more likely to be incidentally diagnosed, have primary tumors with less aggressive features, and are associated with improved overall survival. Because targeted therapy rarely results in complete tumor response, surgical resection (metastasectomy) has been proposed as the only potentially curable therapy in appropriately selected patients, with the exception of the extremely uncommon durable complete responses following high-does interleukin (IL)-2. Five-year survival rates ranging from 30% to 45% have been reported following metastasectomy for patients with isolated renal fossa recurrences or lung metastases. Currently, the role of systemic therapy in a neoadjuvant or adjuvant setting to metastasectomy remains unclear.

Prognostic Variables

Characteristics that have associated with cancer-specific outcomes following surgical resection include presence of clinical symptoms, laboratory values, anatomic variables, histologic subtype, and molecular features ( Table 1 ). Patients presenting with clinical symptoms at presentation attributable to the primary tumor, including pain, hematuria, hematuria, cachexia, or paraneoplastic symptoms, appear to have a significantly worse prognosis than those presenting with incidentally diagnosed lesions. Interestingly, as a greater number of clinically localized lesions are diagnosed incidentally, these prognostic differences become less apparent when controlling for pathologic stage because far fewer patients are now diagnosed from imaging prompted by clinical symptoms. Overall health status, quantitated by scoring systems designed to assess competing medical risk and performance status, including the Karnofsky scale, the Eastern Cooperative Oncology Group performance status (ECOG-PS), and the Charlson comorbidity index, have been closely correlated with survival in patients with RCC and have been integrated into the inclusion criteria for contemporary clinical trials. Likewise, laboratory parameters indicating systemic disease involvement, including anemia, elevated liver function tests, thrombocytosis, elevated C-reactive protein, hypercalcemia, and elevated erythrocyte sedimentation rate, have been investigated in large institutional series that have demonstrated an association with poor disease-specific and overall survival.

Currently, pathologic tumor stage is the single most important prognostic indicator in resected RCC, which incorporates important anatomic variables, including tumor size; local tumor extension; adrenal, venous, or lymphatic involvement; and distant metastases. Originally developed in 1997, the American Joint Committee on Cancer tumor, nodes, metastases (TNM) staging system was modified in 2002 and again in 2010 to improve predictive accuracy. Recent adaptations include the stratification of pT1 tumors by tumor size (<4, 4–7 cm) and pT3 disease by venous involvement above or below the diaphragm. Estimated disease-specific survival by 2002 TNM classification is 97% (pT1a), 87% (pT1b), 71% (pT2), 53% (pT3a), 44% (pT3b), 37% (pT3c), and 20% (pT4), respectively. In 2010, the TNM system was again revised, reclassifying pT2 disease by tumor size (≥7–<10 cm, ≥10 cm) and recategorizing adrenal involvement to pT4 disease. T3 disease was reclassified as renal vein involvement or involving Gerota fascia (pT3a), inferior vena cava (IVC) involvement below the level of the diaphragm (pT3b), and tumor thrombus growing into the chest or invading the wall of the IVC (pT3c).

To explain disparities in survival among stages, additional histologic characteristics have been investigated. Fuhrman nuclear grade, histologic subtype, and presence of sarcomatoid or necrotic features have been associated with increased risk of disease progression and poor cancer-specific survival. Large series have demonstrated a significant correlation between tumor grade and disease-specific survival that is independent of tumor stage. Likewise, biologic aggressiveness varies by histologic classification, with chromophobe and papillary type I tumors often demonstrating indolent clinical courses, whereas papillary type II and clear cell RCCs or more uncommon variants, such as collecting duct carcinomas that demonstrate more aggressive behavior, and are associated with a worse prognosis. Specific features regardless of cell type that have been associated with the increased likelihood of cancer-specific death include tumor necrosis and sarcomatoid features. Prognostic value appears to be related to percent involvement rather than simply the presence or absence of a given feature. It is important to note that interobserver and intraobserver variation in nuclear grading for RCC has been demonstrated, and the presence/absence or percent involvement of sarcomatoid features and tumor necrosis is not universally mentioned in pathology reports.

The treatment of advanced RCC has evolved significantly following the identification of the von Hippel-Lindau (VHL) gene and increasing delineation of its tumor suppressor function and role in angiogenesis. Inactivation of the VHL gene leads to the failure of proteolytic regulation of the α subunits of hypoxia-inducible factor (HIF) and constitutive upregulation of the HIF complex. The resulting overexpression of HIF target genes, including VEGF and platelet-derived growth factor (PDGF), has been implicated in tumorigenesis and has provided novel targets for molecular directed targeted therapy in advanced RCC. With the development of effective targeted agents, molecular markers are being actively sought to both assess risk (predict response to therapy) and to identify other novel molecular pathways worthy of targeting. Although the description of mechanisms and accumulation of scientific evidence for each marker currently being studied is beyond the scope of this review, the breadth of molecular markers under investigation is described in Table 1 . Carbonic anhydrase (CA) IX is a transmembrane enzyme that is thought to play a role in the adaptation of tumors to hypoxic conditions by regulating the pH of the intracellular and extracellular compartment, facilitating the proliferation of malignant cells and tumor metastasis. Overexpressed in greater than 95% of clear cell RCC tumors and rarely expressed in non–clear cell variants, fetal tissues, or adult benign kidney specimens, the degree of CA IX expression has been inversely correlated with survival in patients with high-risk clinically localized and mRCC. The specificity of CA IX for clear cell RCC makes it an excellent candidate for use as a prognostic marker as well as therapeutic applications, and a monoclonal antibody (Girentuximab or G250) targeting CA IX is currently under clinical trial investigation as a primary therapy and in the adjuvant setting for advanced RCC. In a novel histologic-specific diagnostic application with important implications for guiding management decisions, G250 labeled with 124-iodine (immuno-positron emission tomography) was able to discriminate clear cell renal cell carcinoma (ccRCC) from non-ccRCC with a specificity of 86% and a positive predictive value of 95% in a recent phase III trial (REDECT, Munich, Germany) of 202 patients with solid renal masses undergoing planned surgical excision.

Other biomarkers that have been linked to prognosis in RCC include p53, gelsolin, Ki67, vimentin, phosphatase and tensin homolog (PTEN), epithelial cell adhesion molecule (EpCAM), and CA XII. Increased p53 (cell cycle regulator), gelsolin (cell motility), Ki67 (proliferation marker), and vimentin (eukaryotic cell structural filament) expression and decreased staining of PTEN (regulator of cellular migration, proliferation, and apoptosis), EpCAM (epithelial cell adhesion molecule), and CA XII have been shown to correlate with poor prognosis. Likewise, an association between CA IX, vimentin, and p53 expression and RCC-specific survival has been demonstrated independent of the pathologic stage, presence of metastasis, performance status, and histologic grade. Assessment of hypermethylation of CpG islands in the promoter regions of genes has been associated with transcriptional silencing and has been documented in several malignancies. Quantitative gene methylation profiling has been used to identify unique patterns of gene methylation among RCC histologic subtypes and lesions of differing pathologic stages. In future investigations, the degree of aberrant methylation present may show utility in determining the risk of recurrence or the response to targeted therapy. Currently, contemporary biomarker studies offer insight into molecular tumor biology and malignant potential. However, the strength of current evidence is not sufficiently robust to guide clinical decision making in patients with RCC. As our understanding of the genetics and molecular pathways driving renal cell carcinoma grows, it is hoped that future clinical decisions will be made on an individual patient basis tailored by molecular phenotype.

Preoperative Algorithms for Stratifying Risk in Patients with Suspected Renal Malignancies

Preoperative nomograms have been developed to predict the likelihood of benign or malignant pathology, high- versus low-grade disease, overall survival, and recurrence-free survival. Using a large institutional cohort, Lane and colleagues constructed a nomogram based on the findings that gender, tumor size, and smoking history were predictive of malignant versus benign disease. However, although the CI of this model was 0.64, additional efforts to differentiate indolent from aggressive cancers were less successful (CI 0.56). Utilizing a multi-institutional dataset, Jeldres and colleagues developed a tool to accurately predict high-grade (Fuhrman grade III–IV) features at nephrectomy using 4 covariates (age at diagnosis, gender, tumor size, and symptom classification). Of these factors, only tumor size was significantly associated with high-grade disease on univariate analysis, and their most accurate multivariate nomogram for high-grade disease prediction was only 58.3%. Using a large prospectively maintained institutional cohort, Kutikov and colleagues evaluated the relationship between nephrometry score (a reproducible standardized classification system designed to quantitate the salient anatomy of renal masses) and malignant or high-grade pathologic features at the time of surgical resection. They found that the total nephrometry score and all individual anatomic descriptor components significantly differed between tumor histology groups with the exception of the anterior/posterior designation. Based on these data, predictive nomograms integrating anatomic tumor attributes with patients’ age and gender were constructed for the preoperative prediction of tumor malignant histology (area under the curve [AUC] 0.76) and high-grade features (AUC 0.73). This model represents the most accurate preoperative predictive model for tumor grade or malignant features to date, with accuracy rates that rival the results of contemporary percutaneous core biopsy series. In an effort to develop a clinical tool to stratify the competing risks of comorbidity and tumor malignant potential, Kutikov and colleagues developed a comprehensive nomogram to predict the 5-year risk of kidney cancer death, death from other malignancy, and non–cancer death using select preoperative clinical and demographic variables. This nomogram is currently being refined to incorporate comorbidity assessment stratified by the Charlson comorbidity index. Although the preoperative determination of aggressive tumor features and competing risks of death may ultimately provide useful data in identifying candidates for neoadjuvant protocols, the predictive information gleaned from these nomograms is currently more applicable in determining the need for initial treatment versus active surveillance in patients with significant competing risks.

In the first model using purely clinical variables to assess postoperative prognosis in RCC, Yaycioglu and colleagues developed a preoperative scoring system incorporating clinical presentation and tumor size, using recurrence-free survival as an end point. Using this system, the investigators were able to demonstrate a significant difference in the 5-year disease-free survival (92% vs 57%, P <.001) when patients were stratified into low-risk (R _rec ≤3) and high-risk (R _rec >3) groups. Similarly, Cindolo and colleagues constructed a recurrence risk formula (RRF) using tumor size and clinical presentation. From this series of 660 patients, 2- and 5-year survival was 96% and 93% for patients with a calculated RRF of less than or equal to 1.2, compared with 83% and 68% with a calculated RRF of greater than 1.2. In a multi-institutional dataset of 2517 patients undergoing surgical resection, Raj and colleagues developed an accurate nomogram (CI 0.80) predicting freedom from metastatic recurrence at 12 years incorporating gender, mode of presentation, evidence of lymphadenopathy or necrosis on imaging, and tumor size. Although encouraging initial data has been reported, the clinical utility of these nomograms remains undefined because no individual preoperative nomogram has been developed that performs as well as algorithms incorporating pathologic data obtained at the time of surgical resection.

Postoperative Algorithms for the Prediction of Disease Recurrence

The first nomogram designed to predict freedom from recurrence after surgical resection for RCC was developed by investigators from the Memorial Sloan Kettering Cancer Center. From a cohort of 601 patients with localized RCC, their algorithm incorporating pathologic tumor stage, tumor size, histologic subtype, and symptoms at the time of presentation. The 5-year probability of freedom from failure in this cohort was 86% (95% confidence interval 82%–89%), and their model was able to accurately predict disease recurrence with an AUC of 0.74. This nomogram was subsequently updated in 2005 in a cohort of 833 patients with localized conventional ccRCC undergoing resection, using stage, Fuhrman grade, tumor size, necrosis, vascular invasion and clinical presentation. The 5-year probability of freedom from failure in this cohort was 80.9% (95% confidence interval 75.7%–85.1%), and the accuracy of their model improved evidenced by a CI of 0.82. Also using cancer-specific survival as their primary outcome, investigators from the Mayo Clinic devised the state, size, grade, and necrosis (SSIGN) score from a cohort of 1801 patients with ccRCC. Using this system, points are assigned based on 1997 TNM stage, tumor size, nuclear grade, and presence of histologic necrosis, which are then used to estimate CSS at 1-, 5-, and 10-year intervals (CI 0.84). This scoring algorithm has recently been externally validated with a high degree of prognostic accuracy in an Italian series of 388 patients (CI 0.88). From the same institution, a separate scoring algorithm was developed to predict progression to metastatic disease in patients with clinically localized ccRCC. From a cohort of 1671 patients undergoing surgical resection, this novel system is based on points assigned to pathologic tumor stage, regional lymph node status, tumor size (≥10, <10 cm), nuclear grade, and presence of histologic necrosis. With good predictive accuracy (CI 0.82), patients further stratified as high risk (≥6) had a 42% and 63% chance of developing progressive disease at 1 and 3 years, respectively.

Postoperative Nomograms for the Prediction of Survival

Several algorithms using pathologic features and specific laboratory parameters have been designed to predict overall survival in patients with mRCC before or following systemic therapy, including the University of California Los Angeles Integrated Staging System (UISS), the Survival after Nephrectomy and Immunotherapy, and the Memorial Sloan Kettering Cancer Center Motzer criteria. Although specific discussion regarding these algorithms are deferred to chapters discussing metastatic disease, the pertinent details are summarized in Table 2 . A significant limitation of these algorithms is that they do not predict the probability of an adverse event on an individual basis but instead place patients into risk-stratified groups. Although useful for stratifying patients for clinical protocols, this has less clinical applicability on the individual patient level. Other limitations include significant variety in patient selection criteria and lack of internal or external validation. The discriminating ability of 4 prognostic algorithms was compared using an independent multi-institutional dataset of 2404 patients with nonmetastatic RCC. Calculations of CIs and 95% bootstrap confidence intervals for overall survival, cancer-specific survival, and recurrence-free survival at 5 years consistently revealed that postoperative algorithms performed with higher accuracy than preoperative models. Of the 4 algorithms, the Kattan model was consistently found to be the most accurate (CI 0.71), although the UISS model was only slightly less well performing (CI 0.68). Further comparative evaluation and validation of existing prognostic algorithms is necessary before any contemporary models can be rigorously employed.

Preoperative Algorithms for Stratifying Risk in Patients with Suspected Renal Malignancies

Preoperative nomograms have been developed to predict the likelihood of benign or malignant pathology, high- versus low-grade disease, overall survival, and recurrence-free survival. Using a large institutional cohort, Lane and colleagues constructed a nomogram based on the findings that gender, tumor size, and smoking history were predictive of malignant versus benign disease. However, although the CI of this model was 0.64, additional efforts to differentiate indolent from aggressive cancers were less successful (CI 0.56). Utilizing a multi-institutional dataset, Jeldres and colleagues developed a tool to accurately predict high-grade (Fuhrman grade III–IV) features at nephrectomy using 4 covariates (age at diagnosis, gender, tumor size, and symptom classification). Of these factors, only tumor size was significantly associated with high-grade disease on univariate analysis, and their most accurate multivariate nomogram for high-grade disease prediction was only 58.3%. Using a large prospectively maintained institutional cohort, Kutikov and colleagues evaluated the relationship between nephrometry score (a reproducible standardized classification system designed to quantitate the salient anatomy of renal masses) and malignant or high-grade pathologic features at the time of surgical resection. They found that the total nephrometry score and all individual anatomic descriptor components significantly differed between tumor histology groups with the exception of the anterior/posterior designation. Based on these data, predictive nomograms integrating anatomic tumor attributes with patients’ age and gender were constructed for the preoperative prediction of tumor malignant histology (area under the curve [AUC] 0.76) and high-grade features (AUC 0.73). This model represents the most accurate preoperative predictive model for tumor grade or malignant features to date, with accuracy rates that rival the results of contemporary percutaneous core biopsy series. In an effort to develop a clinical tool to stratify the competing risks of comorbidity and tumor malignant potential, Kutikov and colleagues developed a comprehensive nomogram to predict the 5-year risk of kidney cancer death, death from other malignancy, and non–cancer death using select preoperative clinical and demographic variables. This nomogram is currently being refined to incorporate comorbidity assessment stratified by the Charlson comorbidity index. Although the preoperative determination of aggressive tumor features and competing risks of death may ultimately provide useful data in identifying candidates for neoadjuvant protocols, the predictive information gleaned from these nomograms is currently more applicable in determining the need for initial treatment versus active surveillance in patients with significant competing risks.

In the first model using purely clinical variables to assess postoperative prognosis in RCC, Yaycioglu and colleagues developed a preoperative scoring system incorporating clinical presentation and tumor size, using recurrence-free survival as an end point. Using this system, the investigators were able to demonstrate a significant difference in the 5-year disease-free survival (92% vs 57%, P <.001) when patients were stratified into low-risk (R _rec ≤3) and high-risk (R _rec >3) groups. Similarly, Cindolo and colleagues constructed a recurrence risk formula (RRF) using tumor size and clinical presentation. From this series of 660 patients, 2- and 5-year survival was 96% and 93% for patients with a calculated RRF of less than or equal to 1.2, compared with 83% and 68% with a calculated RRF of greater than 1.2. In a multi-institutional dataset of 2517 patients undergoing surgical resection, Raj and colleagues developed an accurate nomogram (CI 0.80) predicting freedom from metastatic recurrence at 12 years incorporating gender, mode of presentation, evidence of lymphadenopathy or necrosis on imaging, and tumor size. Although encouraging initial data has been reported, the clinical utility of these nomograms remains undefined because no individual preoperative nomogram has been developed that performs as well as algorithms incorporating pathologic data obtained at the time of surgical resection.

Postoperative Algorithms for the Prediction of Disease Recurrence

The first nomogram designed to predict freedom from recurrence after surgical resection for RCC was developed by investigators from the Memorial Sloan Kettering Cancer Center. From a cohort of 601 patients with localized RCC, their algorithm incorporating pathologic tumor stage, tumor size, histologic subtype, and symptoms at the time of presentation. The 5-year probability of freedom from failure in this cohort was 86% (95% confidence interval 82%–89%), and their model was able to accurately predict disease recurrence with an AUC of 0.74. This nomogram was subsequently updated in 2005 in a cohort of 833 patients with localized conventional ccRCC undergoing resection, using stage, Fuhrman grade, tumor size, necrosis, vascular invasion and clinical presentation. The 5-year probability of freedom from failure in this cohort was 80.9% (95% confidence interval 75.7%–85.1%), and the accuracy of their model improved evidenced by a CI of 0.82. Also using cancer-specific survival as their primary outcome, investigators from the Mayo Clinic devised the state, size, grade, and necrosis (SSIGN) score from a cohort of 1801 patients with ccRCC. Using this system, points are assigned based on 1997 TNM stage, tumor size, nuclear grade, and presence of histologic necrosis, which are then used to estimate CSS at 1-, 5-, and 10-year intervals (CI 0.84). This scoring algorithm has recently been externally validated with a high degree of prognostic accuracy in an Italian series of 388 patients (CI 0.88). From the same institution, a separate scoring algorithm was developed to predict progression to metastatic disease in patients with clinically localized ccRCC. From a cohort of 1671 patients undergoing surgical resection, this novel system is based on points assigned to pathologic tumor stage, regional lymph node status, tumor size (≥10, <10 cm), nuclear grade, and presence of histologic necrosis. With good predictive accuracy (CI 0.82), patients further stratified as high risk (≥6) had a 42% and 63% chance of developing progressive disease at 1 and 3 years, respectively.

Postoperative Nomograms for the Prediction of Survival

Several algorithms using pathologic features and specific laboratory parameters have been designed to predict overall survival in patients with mRCC before or following systemic therapy, including the University of California Los Angeles Integrated Staging System (UISS), the Survival after Nephrectomy and Immunotherapy, and the Memorial Sloan Kettering Cancer Center Motzer criteria. Although specific discussion regarding these algorithms are deferred to chapters discussing metastatic disease, the pertinent details are summarized in Table 2 . A significant limitation of these algorithms is that they do not predict the probability of an adverse event on an individual basis but instead place patients into risk-stratified groups. Although useful for stratifying patients for clinical protocols, this has less clinical applicability on the individual patient level. Other limitations include significant variety in patient selection criteria and lack of internal or external validation. The discriminating ability of 4 prognostic algorithms was compared using an independent multi-institutional dataset of 2404 patients with nonmetastatic RCC. Calculations of CIs and 95% bootstrap confidence intervals for overall survival, cancer-specific survival, and recurrence-free survival at 5 years consistently revealed that postoperative algorithms performed with higher accuracy than preoperative models. Of the 4 algorithms, the Kattan model was consistently found to be the most accurate (CI 0.71), although the UISS model was only slightly less well performing (CI 0.68). Further comparative evaluation and validation of existing prognostic algorithms is necessary before any contemporary models can be rigorously employed.