Prognostic Factors in Renal Cell Carcinoma: Association of Preoperative Sodium Concentration with Survival

Conventional renal cell carcinoma (RCC) accounts for >100,000 deaths worldwide each year. It is a disease characterized both by its insidious onset and highly variable natural history. At presentation, over a quarter of patients have metastatic disease and a further 30% relapse after nephrectomy with curative intent. The prognosis of patients with metastatic RCC is poor, with a 5-year survival of <10% (1).

The identification of host and disease characteristics that may predict outcome for individual patients is highly desirable. In localized disease, such information could be used to guide the intensity of follow-up and identify high-risk patients who can be targeted for adjuvant therapy trials. This is important given that, to date, trials of adjuvant cytokine therapy have failed to show any benefit (2). The recent introduction of efficacious but costly treatments, such as sunitinib and sorafenib, highlights further the need to be able to define and target specific patient groups.

Several prognostic models have been proposed to date (3–7). The University of California Los Angeles Integrated Staging System stratifies patients based on tumor stage, Eastern Cooperative Oncology Group performance status, and Fuhrman grade into three risk groups (8). Its applicability to an international population has been shown for patients with localized, low, and intermediate risk disease, but not in patients with metastatic disease (9). In 2005, the Memorial Sloan-Kettering Cancer Center proposed a postoperative nomogram for patients with localized RCC based on tumor size, stage, Fuhrman grade, necrosis, vascular invasion (VI), and clinical presentation (7). A second prognostic model in localized disease was introduced by the Mayo Clinic and uses tumor stage, size, grade, and necrosis (SSIGN) to divide patients into 10 different prognostic groups (6, 10). Other models that are limited to patients with metastatic disease have also been proposed and partially validated (3, 11, 12).

Existing models are therefore limited by the fact that they may only apply to patients with metastatic or localized disease or include variables, such as performance status, that are not routinely recorded. The identification of additional validated prognostic factors may strengthen these models, and serum calcium, alkaline phosphatase, hemoglobin (Hb), and erythrocyte sedimentation rate have all been acknowledged as predictors of survival in RCC (13). In this cross-sectional study, we have examined a large number of variables for their potential prognostic value in patients with RCC. In particular, we have looked at routine biochemical and hematologic factors, as well as comprehensive immunophenotyping data.

After approval from Leeds Research Ethics Committee, patients attending an out-patient clinic for assessment before nephrectomy for newly diagnosed renal cancer at St. James's University Hospital between September 1998 and March 2005 were approached to participate in the study and consents were obtained. None of these patients were in-patients receiving i.v. fluids, and none had received systemic pharmacologic agents for their cancer. Clinical data and routine preoperative laboratory measurements were recorded in all patients (not >14 d before nephrectomy), and immunophenotyping was carried out on all patients with sufficient serum sample for analysis.

Clinical features that were recorded included gender, smoking history, and symptoms (local/systemic/asymptomatic) at presentation. Local symptoms were defined as flank pain/mass or macroscopic hematuria. Systemic symptoms included loss of appetite, weight loss, sweats, and fatigue. Pathologic features assessed included histologic subtype, tumor-node-metastasis (TNM) stage (14), Fuhrman grade, tumor size, and presence or absence of sarcomatoid change, histologic necrosis (any area of necrosis within the tumor independent of size), and micro-VI or macro-VI. All pathology was centrally reviewed by a specialist urological pathologist. Routine laboratory variables were measured from preoperative bloods, including full blood count, Hb, and packed cell volume (PCV). In addition, serum sodium, potassium, urea, creatinine, calcium, alkaline phosphatase, alanine transferase, bilirubin, albumin, C-reactive protein, and lactate dehydrogenase were recorded.

Immunophenotyping. Immunophenotyping was done in a subset of patients (n = 163) measuring the following variables: CD3, CD4, CD8, CD19, CD14, CD56, and CD16 cells as a percentage of lymphocytes, with CD4/CD8 ratio being calculated from these figures. In addition the percentage of CD3, CD4, CD8, CD14, and CD19 cells expressing CD25 or CD54 were also examined as were the percentage of CD3+ and CD3− cells expressing CD16/CD56. The following monoclonal antibodies at optimized concentrations were used: CD3, CD4, CD8, CD14, CD19, CD25 (BD Biosciences), CD16, CD54, CD56 (Serotec). Briefly, an appropriate volume of antibody, conjugated with either FITC, PE, or PercP, was added to 100 μL of whole blood, mixed and incubated at 4°C for 30 min. Erythrocytes were lysed, and leukocytes were fixed by the addition of 2 mL of FACS Lysing Solution (Becton Dickinson), then vortexed, and left at room temperature for 10 min. After centrifugation and washing, 20,000 cells within the peripheral blood mononuclear cells gate were acquired using a FACSCalibur. Two variable dot plots and quadrant statistics were generated using Cellquest software.

Statistical methods. Patients with RCC were analyzed together, with primary end points including overall survival (OS) and disease-free survival (DFS). Given that the predictive value of the variables examined may vary with tumor stage, we carried out a second planned analysis of the subset of patients presenting with nonmetastatic (N0M0) disease with the same end points.

OS and DFS were calculated using the Kaplan-Meier method. Survival was calculated as the time from date of nephrectomy to the date of death from any cause (OS) and the date of disease recurrence or death from any cause (DFS). Patients who were still alive (OS) or alive and disease-free (DFS) or who were lost to follow-up were censored at the last time they were known to be alive or disease-free.

Univariate and multivariate Cox proportional hazards models were used to assess the predictive ability of the baseline characteristics and hematologic, biochemical, and immunophenotyping data on OS and DFS. The distribution of the continuous variables was assessed and transformed using log transformation, if appropriate, to reduce skewness. Due to the large number of patients with C-reactive protein levels below the limit of detection, the distribution of this variable was highly skewed and was not altered when transformed. We therefore chose to dichotomize this variable into above and below the limit of detection (5.0 mg/L). All other variables were initially analyzed as continuous variables.

Model building using forward stepwise selection was used to identify statistical models that incorporate only those factors that are independently predictive of OS and DFS, building a separate model for each. The 10% significance level was set as the boundary for inclusion into the model as the analysis is exploratory and is designed to generate hypotheses for further confirmation rather than to formally test the variables (15). The relationship between survival and each variable was summarized using hazard ratios (HR) and 95% confidence intervals (95% CI). The following variables were considered in the multivariate model building analysis: sex, age at operation, smoking history, symptoms, sarcomatoid change, maximum tumor diameter, Fuhrman grade, TNM Stage (T Stage when considering N0 patients), VI, necrosis, RBC count, Hb, PCV, WBC count, neutrophil count, lymphocyte count, monocyte, platelet number, sodium concentration, potassium concentration, urea concentration, creatinine concentration, and C-reactive protein concentration. Complete case analysis was considered in analysis of the variables detailed above. All other variables had at least 22% of patients with missing data. Comparison of all patients with those patients who were not included in the multivariate analysis due to missing data showed similar baseline characteristics.

To consider the application of the SSIGN scoring algorithm in our patient group, cancer-specific survival (CSS) was calculated and compared with the survival probabilities up to 5 y according to the scoring algorithm (6).

Of those approached, only a very small number (<1%) of patients declined participation in the study. Two hundred sixty-five patients were consented in total. We limited this study to the 212 patients (80%) with newly diagnosed clear cell RCC who formed our study population. Patients with other histologic subtypes were not included. At the time of data analysis, median length of follow up from nephrectomy of those patients still alive was 32 months (range, 0-73 months). Comparison of their characteristics (Table 1) with a larger population-based register of renal cancer in Leeds showed our cross-section of patients to be a highly representative sample (data not shown). Considering the subgroup of patients presenting with nonmetastatic (N0M0) RCC (n = 141; 66.5%), median length of follow up from nephrectomy among patients still alive was 36 months (range, 0-73 months).

OS rates (95% CI) from nephrectomy for all patients were 84.3% (79.3-89.4) for 1 year, 63.1% (55.7-70.6) for 3 years, and 54.5% (45.5-63.5) for 5 years. Equivalent figures for the subset of patients with N0M0 disease were 94.6% (90.7-98.5), 79.6% (72.0-87.1), and 69.4% (58.5-80.4). DFS rates at 1, 3, and 5 years were 86.5% (80.8-92.2), 67.4% (59.0-75.8), and 59.1% (48.8-69.4), respectively, among all patients and 91.2% (86.3-96.2), 73.6% (65.2-82.0), and 64.0% (52.9-75.1) among those with N0M0 disease. Median OS and DFS were not reached in either group of patients.

OS. Considering all patients with conventional RCC, the following were statistically significantly associated with OS individually (P < 0.10): presence of symptoms, tumor diameter, Fuhrman grade, sarcomatoid change, TNM stage, VI, tumor necrosis, RBC count, Hb, PCV, WBC count, neutrophil count, lymphocyte count, monocyte count, platelet count, serum sodium, potassium, calcium, alkaline phosphatase, albumin, C-reactive protein, and lactate dehydrogenase. Additionally the immunophenotyping variables %CD56, %CD19, and CD56 (%CD16) were also significant (Table 2), with an HR of <1 implying improved survival as lymphocyte percentage increases.

Among patients with N0M0 disease, significant factors were tumor necrosis, T stage, TNM stage, Hb, PCV, monocyte count, serum sodium, urea, creatinine, calcium, alkaline phosphatase, and albumin. The %CD19 positive lymphocytes remained significant among immunophenotyping variables.

DFS. Only patients with M0 disease (any N stage; n = 157) were considered in DFS analyses. With the exception of the presence of symptoms, RBC, PCV, lymphocyte count, lactate dehydrogenase, and CD56 (%CD16), clinical and laboratory variables predictive of OS were also significant for DFS among all patients, and additionally, sex and %CD16 were significant.

Factors predictive of OS among patients with N0M0 disease were also predictive of DFS, with the exception of PCV, urea, creatinine, and %CD19 cells and the addition of tumor diameter, VI, and platelet count.

There were 158 patients with complete data for the variables considered for inclusion within the multivariate analysis. Univariate analysis of this subgroup of patients for all survival variables showed highly similar results to those considering the group as a whole (data not shown). For the subgroup analysis of patients with N0M0 disease, there were 107 with complete data for OS, four of whom were missing DFS data.

OS. Among all patients with RCC, factors found to be independent statistically significant predictors of OS in the multivariate setting were sarcomatoid change, Fuhrman grade, TNM stage, VI, neutrophil count, and sodium concentration (Table 3). For patients with N0M0 disease, T stage and serum sodium were the only two factors significantly associated with OS (Table 4).

DFS. For patients free of metastatic disease and with complete data for all variables under consideration for multivariate analysis of DFS (n = 116), VI, necrosis, Fuhrman grade, and sodium concentration were independent significant predictive factors of DFS (Table 3). Considering patients with N0M0 disease only and complete data for all variables, T stage, necrosis, and sodium concentration were found to be significantly associated with DFS among this subgroup of patients (Table 4).

The range of values for serum sodium in our study population was 127 to 146 mmol/L. Only one patient (<0.5%) had serum sodium of <130 mmol/L and would therefore be defined as being clinically hyponatremic. Sixteen (7.5%) and one (0.5%) patients had sodium of <135 and >145 mmol/L, respectively—the upper and lower limit of our laboratory reference range. Because serum sodium concentration was identified as a novel, independent statistically significant predictor of OS and DFS when considered in its continuous form, further analyses were carried out with sodium concentration dichotomized to above and equal to or below the median value (139 mmol/L; reference range, 135-145 mmol/L) to allow a more meaningful clinical interpretation. Figure 1 shows Kaplan-Meier curves for OS and DFS according to sodium concentration. This shows that survival is increased for patients with sodium concentrations above 139 mmol/L at all times after nephrectomy considering all patients and the N0M0 subgroup. The 1-year OS estimates above and below or equal to the median (95% CI) were 92.6% (87.4-97.9) and 77.3% (69.3-85.3). Corresponding 5-year OS estimates were 67.6% (54.2-80.9) and 44.3% (32.8-55.8). Among the subset of patients with N0M0 disease, 1-year OS estimates above and below or equal to the median were 97.3% (93.5-100) and 91.5% (84.3-98.6). Corresponding figures at 5 years were 77.8% (63.7-92.0) and 60.7% (44.8-76.6).

Under univariate analysis, sodium concentration as a binary variable (above/below median) was found to be statistically significantly associated with OS (HR 0.44, 95%CI 0.26-0.72, P = 0.001) and DFS (HR 0.47, 95% CI 0.26-0.84, P = 0.012). When we incorporated sodium concentration as a binary variable into the already established multivariate model for OS containing neutrophil count, sarcomatoid change, Fuhrman grade, TNM stage, and VI, sodium concentration was still statistically significantly associated with OS (HR 0.44, 95% CI 0.22-0.88, P = 0.014). However, VI became nonsignificant. When sodium concentration was incorporated as a binary variable in the multivariate model for DFS, all variables remained significant, including sodium concentration (HR 0.39, 95% CI 0.18-0.84, P = 0.012). Both univariate and multivariate analyses therefore indicate that patients with a sodium concentration less than or equal to the median value of 139 mmol/L have worse survival than patients with levels above the median value.

Further exploration of these findings by subgrouping patients by serum sodium into quartiles (Q1, ≤138; Q2, 138<Na≤139; Q3 139<Na≤141; Q4, >141) was done. Eighty percent of patients in the lower quartile (Q1) had a serum sodium of ≥135 mmol/L (i.e., within laboratory reference range) and 68% had M0 disease. HRs showed that survival was distinctly different between Q1 and the other groups. We therefore further considered survival for the two groups, ≤Q1 and >Q1, and the corresponding HRs for all patients were 0.45 (95% CI, 0.28-0.72) for OS and 0.54 (95% CI, 0.30-0.95) for DFS. To address whether any further groupings of serum sodium should be considered, we conducted univariate analysis on those patients with serum sodium of >138 mmol/L and found that serum sodium was no longer statistically significantly associated with survival, indicating no further categorizations necessary. As the lower quartile and median values of serum sodium are so similar (138 and 139 mmol/L), these results indicate that, in our population, patients with a sodium concentration below either the lower quartile or the median value have a worse prognosis than patients with sodium concentration above these values.

Finally, the value of serum sodium in terms of predicting CSS was also assessed but, although significant on univariate testing, was found not to be significant on multivariate testing.

Concomitant medications. Because serum sodium levels can be influenced by a number of medications, we looked at the drugs patients were already taking at the time of sampling. The 88 patients with a history of hypertension, ischemic heart disease, cerebrovascular disease, or heart failure were considered, documenting those taking diuretic, ACE inhibitor, or angiotensin II receptor antagonist. The results are shown in Table 5. Of the 109 patients with sodium values equal to or below the median, 83 were on none of these medications at baseline, with 26 patients on one or more. For those 103 patients above the median sodium value, 76 were on no such medication at baseline, with 27 patients on one or more. Therefore, Table 5 shows that there was no difference in the percentage of patients with such comorbidities or the number taking such medications when considered by median serum sodium value.

To assess how well our patients fit into existing RCC prognostic models, we used the existing SSIGN algorithm (6), which is used to assess patients' prognosis in terms of CSS from the time of nephrectomy, with a higher score equating to a poorer outcome. Whereas the large number of possible scores meant patient numbers were small in some cases (e.g., 0 patient scored 3, one scored 8), in general, our patient population was consistent with the trend of decreasing CSS as the SSIGN score increases (Table 6).

Current prognostic models in RCC are largely based on tumor-related factors, such as tumor stage, size, grade, necrosis, and VI. Although such factors are consistent predictors of outcome at a population level, they are not always reliable in individual patients (16). Incorporation of factors that may be induced in the host by the presence of the tumor, as investigated in this study, may be valuable. Our study population characteristics are largely comparable with those from previous reports as exemplified in a study of >4,000 patients with RCC from eight centers across the United States and Europe, where mean (±SD) patient age was 59.8 ± 12.2 years compared with 62.2 ± 10.5 years in our population, distribution of TNM stages I, II III, and IV was 39%, 12%, 26%, and 23%, respectively, and the distribution of Fuhrman grades I, II, III, and IV was 20%, 36%, 35%, and 9%, respectively, with the only differences being in the distribution of grades I and IV disease. However, the spread of grade I disease was large (4-61%) across the eight centers, and patients with histology other than conventional RCC were included in the study (9). Further evidence of the representative nature of our sample population comes from an existing prognostic model, the SSIGN algorithm, which seems to generally be applicable to our patient population. The SSIGN algorithm was developed based on a study in the United States of 1,801 patients with RCC (6), and we selected this as it does not require knowledge of the performance status of patients, which is often not recorded.

The tumor-related factors conventionally considered to be of prognostic value in RCC were consistently predictive of survival in our patient population as a whole. The slight anomaly was that tumor grade was not significant for the N0M0 subset of patients, although this was previously reported as an independent predictor of both OS and DFS among patients with N0M0 disease (7, 17). This may be due to the relatively small number of patients in our subset analysis.

Our results show for the first time that preoperative serum sodium level is an additional factor highly predictive of OS and DFS for RCC, irrespective of whether all patients or only patients with nonmetastatic disease are considered. The majority of patients (92%) had a serum sodium value within the normal laboratory reference range, but patients with values above the median value had significantly increased survival compared with those patients with levels equal to or below the median.

The control of sodium and ECF occurs principally in the distal tubules and collecting ducts of the nephron, where the capacity to reabsorb sodium is regulated by the renin-angiotensin-aldosterone system. Serum sodium is a recognized prognostic marker in patients with small cell lung cancer (18), where the syndrome of inappropriate ADH is often the underlying cause of hyponatremia. However, there is little to suggest this is the case in patients with RCC, and one alternative possibility is that renal tumors may be inducing a subtle disturbance of the renin-angiotensin-aldosterone axis.

Hyponatremia is associated with a higher mortality (19–21). In a study of 106 patients on a general cancer ward, mortality was 19.5% versus 6.3% in the hyponatremic (<130 mmol/L) and nonhyponatremic groups, respectively. Few patients died of the electrolyte imbalance itself, suggesting that hyponatremia may simply reflect the gravity of the underlying disease (22). Sodium is unlikely to be serving as a similar marker of poor general health or performance status in our study population as among those patients with a serum sodium below the median value, most would not conventionally be termed hyponatremic. Additionally, the effect persisted in the subset of patients with N0M0 disease who would generally be expected to be of good health and performance status. We also considered the role of comorbidities, such as hypertension, ischemic heart disease, and cardiac failure, for which patients often take medication that can affect electrolyte balance. The proportion of patients taking such medications was not found to be different between patient groups dichotomized by serum sodium concentration.

Several of our findings on univariate analysis confirm previous preliminary associations. For example high neutrophil counts have been associated with a more rapid progression and poorer outcome in patients with metastatic RCC on cytokine treatment (23, 24). Similarly, variables, such as calcium, lactate dehydrogenase, and alkaline phosphatase, were significant on univariate testing for survival in our patient population, in agreement with previous reports (11, 25, 26). Platelet count (27) and Hb (3, 11, 28) have previously been associated with poorer survival in patients with RCC and were significant on univariate testing, but not multivariate analysis in this study.

Few data exist regarding the prognostic value of peripheral blood lymphocyte subsets in patients with RCC. A small number of variables examined were consistently found to be significant for OS and DFS in the univariate setting, including %CD19, %CD56, and %CD16, and these may warrant further investigation in larger studies allowing multivariate analysis.

In conclusion, we have shown that lower preoperative serum sodium is associated with reduced OS and DFS in patients with RCC. A further, larger study would be desirable to confirm, extend, and validate these findings in a multicenter population, ideally designed to determine a cutoff value for risk stratification and the value of adding serum sodium into existing traditional predictive models for renal cancer. Finally, our immunophenotyping data are encouraging and suggest that this is also an area worth examining further.