SEARCH

SEARCH BY CITATION

Keywords:

  • C-reactive protein;
  • mortality;
  • prognosis;
  • renal cell carcinoma

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

BACKGROUND.

C-reactive protein (CRP) represents a promising prognostic variable in patients with sporadic renal cell carcinoma (RCC). It was hypothesized that CRP can improve the prognostic ability of standard RCC-specific mortality (RCC-SM) predictors in patients treated with nephrectomy for all stages of RCC.

METHODS.

Radical nephrectomy was performed in 314 patients from 2 European centers. Life table, Kaplan-Meier, and Cox regression analyses addressed RCC-SM. Covariates included age, gender, TNM stage, tumor size, Fuhrman grade, and histologic subtype.

RESULTS.

The median survival of the cohort was 19.9 years. Age ranged from 10 to 77 years. Most patients were male (69%). T-stages were distributed as follows: T1-121 (38.7%), T2-45 (14.4%), T3-140 (44.7%), T4-7 (2.2%). CRP values ranged from 1.0 to 358.0 mg/L (mean 40.9, median 11.0 mg/L). In multivariable analyses, CRP was an independent predictor of RCC-SM (P = .003). The consideration of CRP in the multivariable model increased the predictive accuracy by 3.7% (P < .001). Moreover, the model with CRP performed 2.4% and 4.6% better than the UCLA Integrated Staging System (UISS) at, respectively, 2 and 5 years.

CONCLUSIONS.

CRP represents an informative predictor of RCC-SM. Its routine use could allow better risk stratification and risk-adjusted follow-up of RCC patients. Cancer 2007. © 2007 American Cancer Society.

Investigators have recently demonstrated a remarkable association between systemic inflammatory response to tumor proliferation and renal cell carcinoma-specific mortality (RCC-SM). The results have so far shown that markers of inflammation, such as C-reactive protein (CRP) and erythrocyte sedimentation rate (ESR), demonstrated promise in predicting survival of RCC patients.1–8 CRP is an indicator of systemic inflammatory response. Previous reports indicate that CRP levels are correlated with the production of proinflammatory cytokines, such as interleukin-6 (IL-6),9 and with tumor progression.4, 10, 11 Finally, preliminary data indicate that elevated CRP levels predict poor survival in patients with metastatic RCC, as well as in patients with localized RCC.2–6 These studies suggest a promising role for CRP as a prognostic marker in RCC. Despite this preliminary evidence, the ability of CRP to improve the prediction of RCC-SM has never been quantified.

On the basis of these considerations, we decided to formally test the prognostic ability of CRP in RCC patients of all stages treated with nephrectomy. Moreover, we compared the predictive accuracy of an established staging scheme, the UCLA Integrated Staging System (UISS),12 to that of a model relying on CRP and other routine predictors of RCC-SM.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Patient Population

Data were retrieved from 2 individual surgeon prospective databases, which totaled 314 consecutive patients treated with either partial or radical nephrectomy for RCC of all stages between 1984 and 2005 (Table 1).

Table 1. Descriptive Characteristics of 313 Patients Treated With Either Partial or Radical Nephrectomy for Renal Cell Carcinoma
VariablesNo.
  1. ECOG indicates Eastern Cooperative Oncology Group.

Total313 (100.0%)
Age, y
 Mean (median)62.6 (74.0)
 Range10–77
Sex
 Women97 (31.0%)
 Men216 (69.0%)
T Classification
 T1121 (38.7%)
 T245 (14.4%)
 T3140 (44.7%)
 T47 (2.2%)
Tumor size, cm
 Mean (median)6.9 (6.0)
 Range1.0–22.0
Presence of nodal metastases, N1-N233 (10.5%)
Presence of distant metastases, M166 (21.1%)
Histological subtype
 Conventional clear cell268 (85.6%)
 Papillary33 (10.5%)
 Chromophobe12 (3.8%)
Fuhrman grade
 I17 (5.4%)
 II124 (39.6%)
 III126 (40.3%)
 IV46 (14.7%)
ECOG performance status 1101 (32.3%)
Continuously coded C-reactive protein level, mg/L
 Mean (median)40.9 (11)
 Range1.0–358.0
Categorically coded C-reactive protein level, mg/L
 ≤4.0111 (35.5%)
 4.1–23.084 (26.8%)
 >23.0118 (37.7%)
Follow-up, y
 Mean (median)3.8 (2.3)
 Range0.1–20.8
Renal cell carcinoma-specific mortality54 (17.3%)
Actuarial survival, y
 Mean (median)13.9 (19.9)

Clinical and Pathologic Evaluation

Serum CRP levels were assayed by a rate immunonephelometric technique on an Analyzer Protein System (Beckman-Coulter, Chaska, Minn). Levels were obtained preoperatively for all patients in our cohort and were expressed in mg/L. Tumors were classified according to the 2002 TNM staging system and according to Fuhrman grade. Tumor size was based on pathological specimens and was defined as the greatest diameter in centimeters. Before study inclusion, histologic subtypes were stratified according to the 2002 American Joint Commission on Cancer (AJCC/UICC) classifications.13 One patient with collecting duct histology was excluded from analyses, resulting in 313 evaluable cases. Eastern Cooperative Oncology Group (ECOG) performance status was recorded prospectively and was classified in institutional databases on a scale from 0 (no symptoms, fully active, able to work) to 4 (completely bedbound).14 In this analysis only patients with ECOG 0 or 1 were included, according to previous recommendations for nephrectomy in the context of metastatic disease.15, 16 Finally, patients were staged preoperatively with computed tomography (CT) of the abdomen and pelvis, chest CT or chest X-ray, serum electrolytes, and liver function tests.

The presence of nodal metastases was defined according to lymphadenectomy findings. In all cases a hilar lymphadenectomy was performed and included all lymph nodes on the ipsilateral side of the great vessels. In select cases, based on surgeon preference, more extensive lymphadenectomies were performed that included interaortocaval lymph nodes. In all cases the presence of nodal metastases was confirmed pathologically. The presence of distant metastases was confirmed based on radiographic and/or histologic findings.

Follow-up consisted of 1 postoperative baseline visit and was then performed every 6 months for a minimum of 2 years. Subsequently, minimum follow-up consisted of annual visits. At each visit CT of the chest or chest radiography accompanied a CT of the abdomen.

Statistical Analyses

The analyses of RCC-SM after nephrectomy were performed on 313 evaluable patients. Assessment of mortality and determination of the cause of death were performed by the treating physician, who relied on chart review and/or death certificate. In cancer-specific survival analyses, perioperative deaths (within 30 days of surgery) were censored.

Kaplan-Meier plots were used to graphically illustrate the RCC-specific survival (RCC-SS) for the entire cohort, and to show the effect of categorically coded CRP on RCC-SS. Univariable and multivariable Cox regression models addressed the effect of all predictors on RCC-SS. The predictors included age, gender, TNM stages, tumor size, Fuhrman grade, histologic type, ECOG performance status as well as CRP level. CRP cut-off values were systematically tested and the most informative cutoffs were defined according to univariable P-values. Proportional hazards assumptions were systematically verified for all proposed models by using the Grambsch-Therneau residual-based test.

Univariable predictive accuracy was determined for each predictor, including CRP. In all analyses predictive accuracy was defined as the ability of the model to discriminate between those who died of RCC from those who did not. Predictive accuracy was expressed as a percentage using the Harrell concordance index. A value of 100% indicates perfect prediction, whereas 50% is equivalent to a toss of a coin. The combined predictive accuracy of all RCC-SM predictors was quantified in the multivariable models. The base model consisted of all predictors except for CRP. The predictive accuracy difference between the base model and the full model with CRP was used to quantify the accuracy gain related to consideration of CRP. Conversely, the decrease in predictive accuracy related to sequential exclusion of 1 variable at a time was used to quantify the multivariable contribution of other variables to a set of predictors that include CRP. To reduce overfit bias, all univariable and multivariable models were subjected to 200 bootstrap resamples. The Mantel-Haenszel statistic was used to test the significance of predictive accuracy differences.

In subsequent analyses, we compared the full multivariable model predicted RCC-SM to the observed RCC-SM at 2 and 5 years. Model-derived mortality predictions were computed with the model formula. Subsequently, the model-predicted probabilities of RCC-SM were compared with the observed rates of RCC-SM at 2 and 5 years after nephrectomy and the accuracy of time-specific predictions was quantified using the predicted probability validation method (val.prob) from the SPlus design library. Moreover, the relation between predicted and observed rates was graphically explored using the val.surv function from the R statistical package. Finally, the UCLA Integrated Staging System (UISS), which predicts RCC-specific survival at 2 and 5 years, was used as a comparison benchmark for the newly developed model.12 The Mantel-Haenszel test was used to compare the difference between the accuracy of our model and that of the UISS prognostic scheme. All statistical tests were performed using S-PLUS Professional, v. 1 (MathSoft, Seattle, Wash) and the R statistical package. Moreover, all tests were 2-sided with a significance level set at .05.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Patient characteristics are shown in Table 1. The majority (n = 216, 69.0%) were men and their age ranged from 10 to 77 years (mean, 62.6; median, 74.0). Of 313 RCC patients treated with nephrectomy, pT1, pT2, pT3, and pT4, respectively, accounted for 121 (38.7%), 45 (14.4%), 140 (44.7%), and 7 (2.2%) cases. The mean tumor size was 6.9 cm (range, 1.0–22.0, median 6.0 cm). ECOG 1 was recorded in 101 patients (32.3%) and of these 41 had metastatic disease. Clear cell histology was present in 268 cases (85.6%). Fuhrman II (39.6%) and III (40.3%) represented the most frequent tumor grades. Node-positive disease was diagnosed in 10.5% of cases (n = 33), whereas 21.1% had systemic metastases (n = 66). Overall, CRP levels ranged from 1.0 to 358.0 mg/L (mean, 40.9; median, 11.0 mg/L).

The overall follow-up time ranged from 0.1 to 20.8 years (mean, 3.8; median, 2.3). Of all patients, 54 (17.3%) died of RCC. The actuarial mean and median survival were, respectively, 13.9 and 19.9 years.

Figure 1A shows RCC-SS for the entire cohort. Figure 1B shows RCC-SS stratified according to the most informative CRP cutoff. According to life table analyses, the use of these cutoffs resulted in 3 specific strata where RCC-SS was, respectively, 96.4%, 84.1, and 57.1% at 5 years. These RCC-SS differences were statistically significant in overall (P < .001) as well as in intergroup comparisons (all P-values ≤.004).

thumbnail image

Figure 1. (A) Renal cell carcinoma-specific survival in the study cohort of 313 patients treated with nephrectomy. (B) Renal cell carcinoma-specific survival stratified according to measured serum C-reactive protein.

Download figure to PowerPoint

Table 2 shows the univariable analyses addressing RCC-SM. Except for age, gender, and histologic subtype, all predictors were statistically significantly associated with RCC-SM in univariable analyses. When univariable predictive accuracy was analyzed, categorically coded CRP represented the most informative predictor of RCC-SS (75.6%), followed by T-stage (70.6%) and ECOG performance status (70.1%), where 50% accuracy represents a random event.

Table 2. Univariate and Multivariate Cox Regression Model for Prediction of Renal Cell Carcinoma-Specific Mortality
VariablesUnivariate modelMultivariate model
Base modelFull model% Decrease in predictive accuracy if full model variable removed
Rate ratioP% Predictive accuracyRate ratioPRate ratioP
  1. ECOG indicates Eastern Cooperative Oncology Group.

  2. Rate ratios indicate the increase in the rate of cancer-specific mortality, relative to reference categories for which the rate ratio is 1.0. An overall statistic (P-value) is provided for categorical variables. For these variables the rate ratios are provided for each variable level.

T classification<.00170.6.2.30.6
 T2 vs T14.7.013.2.13.6.06
 T3 vs T18.8<.0013.6.042.8.1
 T4 vs T116.9<.0013.10.22.5.3
Presence of nodal metastases, N1–N23.7<.00161.02.1.031.8.081.1
Presence of distant metastases, M15.6<.00170.02.9<.0012.9<.0012.7
Tumor size, cm1.01.00164.91.0.50.9.10.9
Fuhrman grade, III/IV vs I/II3.5<.00162.01.5.41.1.80.6
ECOG performance status, 1 vs 05.5<.00170.13.0.0012.4.0090.9
Age1.0.454.61.004.61.005.60.0
Sex, men vs women1.4.350.41.0.91.02.90.7
Histological subtype.752.7.8.60.9
 Papillary vs clear cell0.7.50.7.50.7.5
 Chromophobe vs clear cell0.006.60.007.70.007.7
Continuously-coded C-reactive protein, mg/L1.006<.00174.0
Categorically-coded C-reactive protein, mg/L<.00175.6.0033.7
 4.1–23.0 vs ≤4.07.0.015.2.03
 >23.0 vs ≤4.023.5<.00111.0.002
Predictive accuracy, %81.685.3

Table 2 also shows the multivariate analyses and the difference in predictive accuracy related to the exclusion of each variable from the multivariate model. In our full model, categorically coded CRP (P = .003), ECOG performance status (P = .009), and M stage (P < .001) achieved independent predictor status. Relative to the baseline CRP stratum (0–4 mg/L), CRP values between 4.1 and 23.0 conferred a 5.2-fold increase in RCC-SM risk versus 11.0-fold for patients with CRP > 23.0 mg/L. The base multivariate model consisting of all variables except for CRP was 81.6% accurate in predicting RCC-SM. Removal of CRP from the model containing all other predictors (full model) resulted in 3.7% accuracy loss (P < .001). Exclusion of other predictors, 1 at a time, from the full multivariable model resulted in predictive accuracy detriments ranging from 0.0 to 2.7%.

Subsequently, we compared the predictive accuracy of the full model, which includes CRP as well as age, gender, TNM stage, histology, Fuhrman grade, and ECOG performance status, to that of the UISS 2- and 5-year predictions.12 In our cohort the accuracy of the full model, which contained categorically coded CRP, at 2 and 5 years was, respectively, 87.7 and 84.4%. Conversely, at 2 and 5 years the UISS was 85.3 and 79.8% accurate. The difference between our model with CRP and the original UISS model corresponds to 2.4 (P < .001) and 4.6% (P < .001) gains, respectively, relative to the UISS. Finally, we tested the ability of CRP to increase the predictive accuracy of the UISS predictors in a Cox regression model. The addition of CRP to that model resulted in 2.5% gain at 2 years (from 85.3%–87.8%, P < .001) and in 3.8% gain at 5 years (from 80.2%–84.0%, P < .001).

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

CRP is an acute phase reactant produced exclusively by the liver in response to cytokines such as IL-6, known mediators of ongoing inflammatory processes. In vivo CRP predominantly reflects IL-6 secretion, which represents 1 of the main immune modulators in patients with metastatic RCC treated with systemic therapy.17–19 Blay et al.9 reported that pretreatment IL-6 is indirectly related to the effect of immunotherapy. Moreover, Fujikawa et al.5 suggested that a decrease in serum CRP as a response to cyto-reductive nephrectomy and immunotherapy predicts a more favorable prognosis. More recently, Ito et al.2 (n = 178), Komai et al.6 (n = 101), Casamassima et al.4 (n = 110), and Lamb et al.3 (n = 100) demonstrated that CRP is an independent predictor of RCC-SM in patients undergoing nephrectomy. Taken together, these studies indicate that CRP represents a promising prognostic marker in patients with various stages of RCC. We corroborate these findings within a large cohort of European patients where we confirmed the independent predictor status of CRP (P = .003).

Despite data indicating the independent predictor status of CRP, no previous study quantified the added value of CRP relative to established predictors of RCC-SM. We addressed this void and used the most stringent methodological approach suggested by Kattan,20 where besides demonstrating the independent predictor status of a novel marker, the candidate marker should significantly enhance the accuracy of established predictors. Previous studies indicated that in several instances independent predictor status did not invariably translate into better predictive accuracy, as not all independent predictors actually translate into a practical benefit if they are considered along with established predictors.21–23 In our analyses, consideration of CRP within a model consisting of baseline predictors contributed to a 3.7% predictive accuracy gain. The consideration of the M stage, describing the presence or absence of distant metastases, represented the second most informative predictor, which contributed 2.7% to the multivariable model's accuracy. The contribution of all other predictors had a negligible effect on predictive accuracy. From a practical perspective, the consideration of CRP may result in correct classification of 37 additional patients of 1000. This is not negligible, especially to those potentially incorrectly ranked patients, who deserve the most accurate survival predictions.

Besides excellent predictive accuracy, our model also demonstrated virtually perfect performance characteristics. Moreover, when compared with an established staging system, the UISS, our model was 2.4% and 4.6% better at 2 and 5 years, respectively. Therefore, the use of the UISS instead of the current model would incorrectly rank 24 and 46 patients of 1000 at 2 and 5 years, respectively, after nephrectomy. Interestingly, CRP improved UISS predictions to virtually the same extent as it did in our model. This finding demonstrates the robustness of CRP's ability to improve predictive accuracy.

Several limitations may apply to our findings. CRP is a nonspecific marker of systemic inflammatory response. Although its pivotal role in RCC tumoral proliferation has been discussed earlier, CRP is also elevated in several clinical syndromes including infectious diseases, trauma, autoimmune, and malignant disorders.24 It is possible that the presence of 1 or multiple inflammatory conditions might have contaminated our data. Nonetheless, despite this potential contamination, CRP still yielded highly accurate prognostic input. Detailed information regarding adjuvant and/or salvage treatment regimens of some of our patients was not available. Some received adjuvant immunotherapy, whereas others received immunotherapy at recurrence. Finally, some were treated with experimental chemotherapy, whereas others received only the best supportive care. It is unlikely that adjuvant or salvage therapies have contributed to a significantly longer survival, as the majority of historic regimens are associated with virtually no effect on survival.25, 26 Moreover, we only addressed RCC-SM. CRP may have a different effect on RCC recurrence. Therefore, separate analyses will be needed to elucidate the prognostic significance of this variable in recurrence prediction. CRP was analyzed as a categorically coded variable, as suggested by Mazumdar and Glassman.27 However, Royston et al.28 suggested the use of continuously coded variables. Continuously coded variables obviate the issue of lack of standardization of cutoffs, which represents an advantage. Conversely, categorically coded CRP performed better in the current model. Previous investigators also used categorized CRP. Their cutoffs differed from ours, as did their analyses. Therefore, pooled multiinstitutional analyses could shed the most objective light on the relative merits of categorized versus continuously coded CRP in prediction of RCC cancer control outcomes.

Our patients originated from Europe. Individuals of different ethnic or racial backgrounds may exhibit different characteristics with respect to the importance of CRP. Finally, all patients in our cohort underwent nephrectomy, including patients with metastatic RCC. Our results could have been different if we had conducted such a study in a nonsurgical cohort. Although bootstrap resampling represents a valid alternative to external validation, it does not represent a perfect substitute. In consequence, our results await formal external validation before routine use in clinical practice. Finally, different assays may be used to quantify CRP. Those may be related to interassay variability, as reported by Devleeschouwer et al.29 Therefore, our cutoff values may not apply to CRP levels that were measured with different CRP kits. It also has to be stated that no data are available regarding the validity of serum banked CRP levels, which may have affected our results. However, degradation of CRP in the serum would most likely have contributed to a null effect, which was not the case in our study. Nonetheless, stability of CRP over time deserves further attention, especially if this marker becomes a routine predictor of cancer control after nephrectomy. Despite these limitations, our data provide an important insight into RCC-SM of patients with high CRP levels.

Although CRP represents a highly informative predictor of RCC-SM, the postnephrectomy management of high-risk RCC patients with localized disease is still under investigation. Currently, adjuvant therapy is not routinely offered. However, 3 ongoing trials (ASSURE [adjuvant sorafenib or sunitinib in unfavorable renal cell carcinoma; Eastern Cooperative Oncology Group 2805], STAR [sunitinib trial in adjuvant renal cancer], and SORCE [a phase 3, randomized, double-blind, controlled study comparing sorafenib with placebo in patients with resected primary RCC at high or intermediate risk of recurrence]) will hopefully provide answers regarding the benefit of adjuvant molecularly targeted therapies.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Supported by the University of Montreal Health Center Urology Associates, the Quebec Foundation for Health Research (FRSQ), the University of Montreal Department of Surgery and the University of Montreal Health Center (CHUM) Foundation to (P.I.K.).

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES