SEARCH

SEARCH BY CITATION

Keywords:

  • concordance index;
  • Kattan nomogram;
  • predictive accuracy;
  • prognostic factors;
  • renal cell carcinoma

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Objectives:  The aim of the current study was to establish the predictive accuracy of the Kattan postoperative nomogram for non-metastatic renal cell carcinoma (RCC) in a Japanese population.

Methods:  A total of 211 patients with stage T1–3N0M0 clear cell RCC who underwent radical nephrectomy or nephron-sparing surgery between 1991 and 2004 were included in this analysis. Median follow up was 81 months (range: 4–208). Univariate and multivariate analyses were performed, and the influence of age, sex, clinical presentation, T stage, histological tumor size, grade, and microvascular invasion on disease recurrence-free survival (RFS) was determined. For each patient, the prognostic score for 5-year RFS was calculated using the Kattan nomogram. The discriminating ability of this model was assessed by the concordance index, and bootstrapping was used to evaluate confidence intervals.

Results:  The 5-year RFS rate for all patients calculated using the Kaplan–Meier method was 80.6%. In multivariate analysis, the statistically significant prognostic factors for 5-year RFS were high-grade tumors (P = 0.019) and symptomatic disease (P = 0.017). The concordance index for RFS predicted by the Kattan nomogram was 0.735 (95% confidence interval: 0.734–0.736). There was a slight discrepancy between the RFS predicted by the Kattan nomogram and the likelihood of being recurrence-free at 5 years according to the Cox analysis in the current patient population.

Conclusion:  These findings suggest the necessity of constructing a more useful nomogram for predicting the prognosis of Japanese patients with non-metastatic RCC.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

RCC is the most frequent malignancy of the kidney, with approximately 57 760 new cases expected per year in the USA1, and 20 000 cases expected in the European Union, with a relative increase of 30% over the past 2 decades.2 The main therapy for non-metastatic RCC is radical nephrectomy or nephron-sparing surgery (NSS). However, it is estimated that 30% of patients will eventually develop metastases during the course of the disease.3 Traditionally, interleukin-2 and α-interferon have been administered to RCC patients, with limited improvement in cancer-specific survival seen in North America and Europe. However, newer agents, such as sunitinib, sorafenib, bevacizumab, everolimus, and temsirolimus, have demonstrated great potential and represent a new frontier in the management of high-risk RCC. The role, efficacy, and toxicity of adjuvant and neoadjuvant targeted small-molecule inhibitors in high-risk RCC remain to be delineated. Ideally, clinicians would be able to identify high-risk patients after surgery and offer treatment to those who would benefit most from adjuvant or neoadjuvant therapy, while minimizing toxicity in low-risk patients.4 Furthermore, the median survival time in Japanese metastatic RCC patients in the cytokine era who had undergone nephrectomy (80.5%) and metastasectomy (20.8%) was approximately twice as long as that found in previous studies from North America or Europe. These findings indicate that early diagnosis of metastasis, nephrectomy, metastasectomy, and cytokine-based therapy improves the prognosis of Japanese RCC patients.5 In this context, the use of reliable prognostic indicators for distinguishing patient prognosis might be able to play a crucial role in predicting outcome and enrolling subjects in trials of new adjuvant agents.

Nomograms are charts that provide outcome probabilities for individual patients that are mainly used to inform poor-risk cases about the benefits of a particular treatment or diagnostic procedure.6 Multiple studies have demonstrated the superior accuracy of more complex predictive modeling compared with risk group assignment techniques.7 Nomograms have the ability to account for multiple prognostic factors simultaneously and make unbiased predictions based on objective data. Their use is increasingly common, especially in oncology for counseling patients with kidney, prostate, or bladder neoplasms. Several prognostic models have been developed to predict disease recurrence after surgical treatment for localized RCC using various nomograms with different end-points, such as overall survival, cancer-specific survival, and recurrence-free survival.8–13 The included variables were also different, such as the differences between the Zisman model9 (University of California – Los Angeles Integrated Staging System [UISS], which included TNM stage, Eastern Cooperative Oncology Group performance status, and nuclear grading), the Leibovich model10 (Mayo Clinic Stage, Size, Grade, and Necrosis [SSIGN] score, which included pT, pN, M, tumor size, nuclear grading, and necrosis), and the Sorbellini model11 (which included pT, tumor size, nuclear grading, necrosis, vascular invasion, and symptoms). Currently, the postoperative nomogram proposed by Kattan et al.,8 which includes histological type, tumor size, 1997 TNM classification, and clinical presentation, is widely used to predict tumor recurrence after treatment for RCC.

A nomogram is an equation in which variables are given different weights; therefore, the verified concordance index (c-index) for a nomogram might differ between datasets. However, if the variables are weighted correctly, the high values in a validation dataset should not have a strong adverse effect on the predictive accuracy of the nomogram, as the nomogram should perform proper case-mix adjustment. Several studies have investigated the predictive accuracy of the Kattan nomogram and demonstrated various c-index values for this nomogram. A six-center European study7 consistently found the Kattan model (c-index: 0.807) to be the most accurate in a comparison with the UISS9, SSIGN10, and Yaycioglu12 models. Other studies have suggested that the Sorbellini model (c-index: 0.82) is the most suitable for use in patients with conventional RCC and that the Kattan nomogram (c-index: 0.76–0.86) might have more limited application to patients with papillary or chromophobe RCC.14 On the other hand, Hupertan et al. indicated that the Kattan nomogram showed poor performance in predicting overall RCC recurrence (c-index: 0.607) even though their patient population was nearly as large as that used in the above-mentioned European study.15 In this way, the predictive accuracy of the Kattan nomogram differed when it was tested using various datasets. Moreover, the predictive accuracy of this nomogram was investigated using Western datasets in previous studies. The aim of the current study was to establish the accuracy of the Kattan nomogram for predicting RCC recurrence in a representative patient population who underwent surgery in a single Japanese center.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Between January 1991 and December 2004, a total of 296 patients underwent surgical treatment for RCC at the Department of Urology of Niigata University Medical and Dental Hospital. Among them, 211 patients had N0M0 clear cell tumors and were followed up. The clinical characteristics of the cases are shown in Table 1. Radical nephrectomy was performed in 173 patients, and NSS was performed in the remaining 38 patients. Among the 211 patients, 90 received adjuvant immunotherapy, mainly treated with α-interferon. Fifty-seven patients suffered a relapse. Thirty-six of the patients who underwent NSS had T1 tumors, and six patients suffered recurrence. The male to female ratio was about 7:3. The median age was 59 years (range: 16–87). The 211 patients were divided into 155 without symptoms (73.5%), 43 with local symptoms (20.4%), and 13 with systemic symptoms (6.1%) at the initial presentation. As there were very few patients with tumors of other histological types, we only investigated patients with the clear cell histological type. The tumor size was 7 cm or less in 175 cases (82.9%) and 4 cm or less in 107 (50.7%) cases. NSS was performed in almost 18% of patients, which can be explained by the fact that we only performed NSS in cases involving tumors of 4 cm or smaller with a favorable anatomical location in the kidney. Thirty-two (15.2%) patients had tumors that demonstrated microvascular invasion. The median postoperative follow-up period in this series was 81 months (range: 4–208 months).

Table 1.  Characteristics of the patients in the current and reference studies8
CharacteristicsCurrent study n (%)Kattan et al. study n (%)
  1. NA, not available.

Sex  
 Male152 (72.0)363 (60.4)
 Female59 (28.0)238 (39.6)
Median age at diagnosis59.163.5
Symptoms  
 Incidental155 (73.5)384 (63.9)
 Local43 (20.4)204 (33.9)
 Systemic13 (6.1)13 (2.2)
Nephrectomy type  
 Radical173 (82.0)NA
 Nephron-sparing38 (18.0)NA
Histology type  
 Clear cell211430 (71.5)
 PapillaryNA108 (18.0)
 ChromophobeNA63 (10.5)
Tumor size, cm1.3–20 (median 4.0)0.5–20 (median 4.5)
T stage  
 1140 (66.4)349 (58.1)
 226 (12.3)71 (11.8)
 3a24 (11.4)119 (19.8)
 3b/c21 (9.9)62 (10.3)
Grade  
 168 (32.2)NA
 2119 (56.4)NA
 324 (11.4)NA
Microvascular invasion  
 Positive32 (15.2)NA
 Negative179 (84.8)NA

In our institution, a scheduled follow-up program for RCC patients was used, which consisted of a clinical examination, laboratory tests, chest–abdominal computed tomography every 6 months during the first 5 years, and yearly thereafter. The clinical parameters examined in this study were the presence or absence of local or systematic symptoms at diagnosis. Postoperatively, the pathological parameters analyzed were tumor size (less than 7 cm and more than 7.1 cm), T stage (T1 and more than T2), tumor grade (1 and more than 2), and microvascular tumor invasion. Classification of the tumor grade was based on The General Rules for Clinical and Pathological Studies on Renal Cell Carcinoma formulated by the Japanese Society of Urology, the Japanese Society of Pathology, and the Japanese Radiological Society (pathological grade was classified as Grade 1–3).16

Disease-free survival curves were calculated for all clinical and pathological parameters according to Kaplan–Meier plots and compared using the log–rank test. In multivariate analysis, the Cox proportional hazards regression model was used for 5-year recurrence-free survival (RFS). Differences between the two groups were compared using the Wald test, and a two-sided test with a significance level of P = 0.05 (P < 0.05) was used. Statistical analysis was performed using spss 10.0 J software for Windows (SPSS Inc, Chicago, IL, USA).

The postoperative nomogram developed by Kattan et al.8 was used to calculate the probability that a patient would be recurrence-free at 5 years of follow up. A scatterplot was used to compare for each patient the probability of remaining free of RCC at 5 years as estimated by the Kattan nomogram and as given by the Cox proportional hazards analysis. The predictive accuracy of the nomogram was evaluated on the basis of the c-index reported by Harrell et al.17 (c-index of 0.5: no discrimination; c-index of 1.0: perfect discrimination). The 95% confidence intervals (95%CI) of the c-index were calculated by bootstrapping; that is, by testing all data using 1000 replicate models obtained from samples selected with replacement from the original dataset.17

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

A total of 211 N0M0 clear cell RCC patients were studied. The 5-year RFS rate for all 211 patients calculated using the Kaplan–Meier method was 80.37% (data not shown). We evaluated the significance of several clinicopathological factors as predictors of 5-year RFS after surgery using univariate and multivariate analysis. As shown in Table 2, univariate analysis showed that five factors – symptomatic disease (P < 0.001), T-stage (P < 0.001), a tumor size greater than 7.1 cm (P < 0.001), microvascular invasion (P < 0.001), and high-grade tumors (P = 0.002) – were significant risk factors for disease recurrence. Although old age and male sex tended to be associated with a high risk of recurrence, age and sex showed no such association. In multivariate analysis using the Cox proportional hazards model, high-grade tumors (P = 0.019) and symptomatic disease (P = 0.017) were found to be statistically significant prognostic factors of RFS after 5 years. When only the three variables included in the Kattan nomogram – symptoms, T stage, and tumor size – were tested, symptoms (P = 0.002) and T stage (P = 0.047) were found to be associated with recurrence after 5 years, although only tumor size was not recognized as an independent predictor.

Table 2.  Univariate and multivariate Cox regression of clinicopathological features that might be useful for predicting postoperative 5-year recurrence in patients with renal cell carcinoma
Univariate analysis of prognostic factors
VariableP-value
Age (<60 vs≥60)0.239
Sex (male vs female)0.122
Symptoms (incidental vs local or systemic)<0.001
T stage (1 vs 2, 3)<0.001
Size (≤7 cm vs >7.0)<0.001
Grade (1 vs 2, 3)0.002
Microvascular invasion (negative vs positive)<0.001
Multivariate analysis of prognostic factors
VariableHazard ratio (95%CI)P-value
Age (<60 vs≥61)1.058 (0.519–2.158)0.877
Sex (male vs female)0.552 (0.235–1.296)0.172
Symptoms (incidental vs local or systemic)2.734 (1.198–6.239)0.017
T stage (1 vs 2, 3)2.269 (0.876–5.877)0.092
Size (≤7 cm vs >7.0)1.252 (0.543–2.884)0.598
Grade (1 vs 2, 3)4.147 (1.259–13.663)0.019
Microvascular invasion (negative vs positive)1.913 (0.879–4.163)0.102
Variable (only the variables entered into the Kattan nomogram were included)
  1. CI, confidence interval.

Symptoms, Local 0.002
 Incidental0.385 (0.185–0.805) 
 Systemic1.882 (0.773–4.586) 
T stage, T1 0.047
 T21.330 (0.447–3.953) 
 T3a2.279 (0.942–5.511) 
 T3bc3.280 (1.181–9.110) 
Size (cm)1.057 (0.954–1.171)0.287

The c-index, which describes predictive accuracy, of the Kattan nomogram was 0.7345 with 95%CI of 0.7341 and 0.7359, which were obtained by bootstrapping. Figure 1 illustrates the probability of recurrence-free survival after surgery. The scatter plot illustrates the difference between recurrence as predicted by the Kattan nomogram and that predicted by Cox proportional hazards analysis. The probability of recurrence predicted by the Kattan nomogram tended to be lower than that given by Cox analysis in our dataset. Furthermore, there tended to be a more marked difference between the reference lines for 100% agreement in patients with a lower likelihood of being recurrence-free.

image

Figure 1. Scatter plot showing the agreement between the renal cell carcinoma recurrence predicted by the Kattan nomogram and that given by Cox proportional hazards analysis.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

The present data show that pathological tumor grade and symptoms are strong and independent predictors of 5-year RFS in patients with non-metastatic RCC after surgery. Our results also indicate that the Kattan nomogram is useful in clinical practice for Japanese single-center datasets to some degree, although it could not precisely estimate disease-free survival for each patient undergoing radical nephrectomy or NSS. This is the first report to confirm the predictive accuracy of the Kattan nomogram in an Asian dataset.

In the present study, only two tumor features, namely grade and symptomatic disease, stood out as independent 5-year tumor-free survival predictive factors in multivariate analysis. Earlier studies have also suggested that nuclear grade is an important prognostic factor. An examination of the correlation between tumor grade and recurrence rate showed that recurrence was significantly high in patients with grade 3 disease,18 and the recurrence rate was also found to be significantly higher in pT1 and pT3 patients with grade 3 disease.19 An examination of the association of recurrence rate with symptoms at the first presentation showed a significantly higher recurrence rate in patients with symptoms, as was found in a previous study.20 Other data have indicated that microvascular invasion is a significant prognostic value in RCC patients because systemic progression depends on tumor access to the microvasculature, although it was not proven to be a significant prognostic factor in this study.

A c-index between 0.50 and 0.70 expresses a low predictive accuracy, between 0.71 and 0.90 a moderate accuracy, and >0.90 a high accuracy.21 The c-index of the Kattan nomogram found in this study suggested that the Kattan model is moderately accurate, and useful to some degree for the evaluation of RFS in Japanese non-metastatic RCC patients. However, the numerical value of the c-index found in the current study was a little lower than that reported previously. As one reason for this, there were differences in the backgrounds of Kattan's dataset and our dataset. The fact that only clear cell carcinoma patients were present in our dataset might also have reduced the predictive accuracy of the Kattan nomogram in this study, as this model might not be suitable for use in patients with conventional RCC, as mentioned above.14 In addition, the differences in disease stage between the two groups, as shown by the fact that the percentage of patients with stage T1 disease in our dataset was about 10% higher than that of Kattan's dataset, might have affected the results. Moreover, there might be some regional or racial differences in the epidemiology and/or histology of RCC. For example, the global age-standardized incidence rates of RCC patients per 100 000 ranges from 1.5 to 6.6 in Japan, which is about one-third of the rate in the USA. Therefore, it seems that the predictive accuracy of the nomogram differs for datasets from different areas due to lifestyle and racial differences.

The risk of recurrence in T3a patients was lower than that for T2 patients and the weight given to T stage in the Kattan nomogram was not so high, whereas that given to size was high. Furthermore, patients with local symptoms, such as hematuria, abdominal pain, and palpable abdominal mass, had no point in this nomogram. In this study, local and systemic symptoms were found to be significant (P = 0.002) predictors of recurrence, and the hazard ratios of T2, T3a, and T3bc tumors compared with T1 tumors were 1.330, 2.279, and 3.280, respectively, according to the Cox proportional hazards analysis of the three prognostic factors of the Kattan nomogram, namely symptoms, T stage, and tumor size. These findings are other possible reasons why the numerical value of the c-index was decreased in this study. However, Kattan et al. mentioned that they included pathological stage because omitting it generally decreased the predictive accuracy of the nomogram by overstating the effects of the remaining variables, even though the effect was not statistically significant.8 The inclusion of non-significant variables in the model artificially enhances its predictive performance with regard to the initial dataset, but hinders subsequent application of the model to other datasets.

Here, other limitations affecting our current findings must be considered. The present study was retrospective and included a relatively small number of patients whose population differed from that of the original dataset, although it was strengthened by the fact that the data was obtained at a single center. Therefore, prospective multicenter studies, conducted under a reasonable follow-up protocol, are required to overcome these problems.

In conclusion, the use of prognostic indicators, such as the Kattan nomogram, might play a crucial role in predicting outcome and tailoring new adjuvant treatments to the needs of individual patients. Using such indicators and estimating prognosis as an object of comparison, we would be able to confirm the effects of new adjuvant therapies even when there were no groups available to make comparisons with, and such trials could act as substitutes for randomized clinical trials. Our results also indicate the possibility of constructing a useful nomogram composed of a combination of variables, such as grade, symptoms, and others, for instance T stage. Such probability estimates would be useful for various purposes, as mentioned previously.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References