Predictive ability of the 2002 and 2010 versions of the Tumour-Node-Metastasis classification system regarding metastasis-free, cancer-specific and overall survival in a European renal cell carcinoma single-centre series

Authors


Martin Pichler, Medical University Graz (MUG), Division of Oncology, Auenbruggerplatz 25, A-8036 Graz, Austria. e-mail: martin.pichler@medunigraz.at

Abstract

What's known on the subject? and What does the study add?

As different and heterogenous populations of patients with RCC can be found in different geographic regions, newly proposed cancer staging systems need an independent validation of their clinical usefulness and prognostic significance. Only a few published studies have compared the ‘old’ 2002 version of the TNM classification system for RCC with the recent 2010 version, all of them using cancer-specific survival as their endpoint, and controversial results were reported regarding the potential superiority of the 2010 version over the 2002 version of this cancer staging system.

The aim of the study was to validate and compare the predictive ability of the 2010 with the 2002 version of the TNM classification system regarding metastasis-free, overall and cancer-specific survival in a large central European cohort of patients with RCC. According to our data, the predictive ability of the 2010 version of the TNM classification system regarding the evaluated endpoints is not superior to the 2002 version.

OBJECTIVE

  • • To compare the predictive ability of the Tumour-Node-Metastasis (TNM) classification systems for renal cell carcinoma (RCC) using three different endpoints: metastasis-free (MFS); overall (OS); and cancer-specific survival (CSS).

PATIENTS AND METHODS

  • • Data from 2739 consecutive patients with RCC, who underwent surgery at a single academic centre, were evaluated using multivariate Cox proportional models, Harrell's concordance (c)-index and by applying decision curve analysis (DCA) with regard to MFS, OS and CSS.

RESULTS

  • • According to TNM 2010, significant differences for MFS were observed for pT1a vs pT1b, pT1b vs pT2a, pT3a vs pT3b and pT3b vs pT3c stages, respectively (all P < 0.05).
  • • With regard to OS, significant differences could be observed in pT1a vs pT1b and pT3a vs pT3b stages, respectively (all P < 0.05).
  • • The c-index for CSS, OS and MFS was slightly higher for the 2002 than for the 2010 version of the TNM classification system.
  • • Non-inferiority of the 2002 TNM system is supported by the results of the DCA.

CONCLUSION

  • • According to our data, the predictive ability of the 2010 version of the TNM classification system regarding three different clinical endpoints is not superior to the 2002 version of this staging system.
Abbreviations
MFS

metastasis-free survival

OS

overall survival

CSS

cancer-specific survival

DCA

decision curve analysis

c

concordance.

INTRODUCTION

The global incidence rates of RCC have increased within the last three decades, showing a stage migration towards small and organ-confined tumours [1,2]. Over time, the subdivision of RCC according to size, local spread, lymph node invasion and presence of metastatic lesions resulted in the internationally used and standardized TNM classification system. In addition to its paramount role of worldwide standardized RCC diagnosis, the TNM classification system represents an important tool for patients' risk assessment regarding disease recurrence, survival and epidemiological analyses, as well as for comparison of the results of international clinical multicentre trials [3,4]. Moreover, the TNM classification system for RCC has been, in part, included into novel prognostic risk assessment tools, such as prognostic scores or nomograms, providing the basis for ongoing adjuvant clinical phase III trials [5]. In the recent 7th edition of the Union Internationale Contre le Cancer/American Joint Committee on Cancer TNM classification system for RCC, tumours that are confined to the kidney and >10 cm are subdivided into pathological pT2a and pT2b stages, and stage pT3a tumours include renal vessel infiltration and renal sinus infiltration, whereas infiltration of the ipsilateral adrenal gland represents pT4 stage [6].

Since different and heterogenous populations of patients with RCC can be found in different geographic regions, newly proposed cancer staging systems need an independent validation of their clinical utility and prognostic significance. Currently, only a few published studies have compared the ‘old’ 2002 version of the TNM classification system for RCC with the recent 2010 version, all of them using cancer-specific survival (CSS) as their evaluated endpoint, and controversial results have been reported regarding the potential superiority of the 2010 version over the 2002 version of the TNM classification system [7–9]. To the best of our knowledge, none of these studies validated the predictive ability of the 2010 version of the TNM classification system for RCC with regard to patients' metastasis-free survival (MFS), which is an important clinical endpoint, widely used in ongoing adjuvant clinical phase III trials. Furthermore, the predictive ability of the 2010 version of the TNM classification system has never been explored with regard to overall survival (OS), which has frequently been used as a ‘gold standard’ primary endpoint to evaluate the outcome of any drug, biological intervention, or procedure [10]. The aim of the present study was to validate and compare the predictive ability of the 2002 and 2010 versions of the TNM classification system for RCC using MFS, CSS and OS in a large central European cohort of patients.

MATERIALS AND METHODS

This retrospective analysis included data from a nephrectomy register from 2852 consecutive patients who underwent radical or partial nephrectomy for RCC at the Department of Urology, Medical University of Graz, Austria between January 1984 and December 2010. Patients' clinicopathological data were retrieved from a database established from medical records at the Department of Urology, as well as from pathology reports from the Institute of Pathology at the same institution. Patients with synchronous bilateral tumours (n = 25), von Hippel-Lindau disease (n = 3), and patients lacking follow-up data (n = 85), were excluded from the analyses. In patients with bilateral metachronous RCC, only the first tumour was chosen for analyses. Thus, a total of 2739 patients with sporadic RCC were included in the final data analyses. Histological RCC subtypes were adjusted based on the histopathological description used in the Heidelberg classification system [11]. For the comparison of the 2002 and 2010 versions of the TNM classification system, each tumour was categorized according to these two different systems [6]. For each patient, postoperative surveillance was performed including routine clinical, laboratory examinations and risk stratified radiological surveillance as described elsewhere [12]. No patient received adjuvant or neoadjuvant treatments. MFS was defined as time (in months) from date of surgery to recurrence of radiologically or histologically confirmed distant metastases. CSS was defined as time (in months) from date of surgery to date of cancer-related death. OS was defined as time (in months) from date of surgery to date of death from any cause.

STATISTICAL ANALYSES

Quantitative data are described as medians and interquartile ranges. Estimates of MFS and OS for each pT stage group were performed using the Kaplan–Meier method and compared using pairwise log-rank tests. Associations between OS and clinicopathological variables of the two versions of the TNM classification system were evaluated using univariable and multivariable Cox proportional hazard regression models. Harrell's concordance (c)-index was used for the assessment of the prognostic accuracy of the two versions tested. Decision curve analysis (DCA) was performed to determine the net benefit derived from the use of 2010 TNM as described by Vickers et al. [13] DCA explores the theoretical relationship between the threshold probability of an event (e.g. metastatic disease) and the relative value of false-negative and false-positive results to identify the value (net benefit) of a predictive model. All reported P-values are two-sided. Statistical analyses were performed using SPSS (SPSS Inc., Chicago, IL, USA) and STATA 12 (StataCorp, College Station, TX, USA).

RESULTS

The median (interquartile range) age of the 2739 patients with RCC was 64 (55–71) and 1595 (58.2%) patients were males. Of these patients, 2283 patients (82.6%) had clear-cell, 309 (11.3%) papillary and 88 (3.2%) chromophobe RCC, while 79 patients (2.9%) had other histological subtypes. The median (interquartile range) follow-up was 74 (27–140) months. There were 158 (5.7%) patients with metastatic disease (M1) at date of surgery, 512 patients (18.7%) developed distant metastases during follow-up, and 1132 (41.3%) patients died. Characteristics of the two different versions of the TNM classification system of the entire study cohort are shown in Table 1. For the 2002 version of the TNM classification system, a pairwise comparison (pairwise log-rank tests) for adjacent pT stages with regard to MFS showed significant differences for pT1a vs pT1b (P < 0.001), pT1b vs pT2 (P = 0.01), pT3a vs pT3b (P < 0.001) and pT3b vs pT3c (P = 0.015), whereas no significant difference was found for pT2 vs pT3a (P = 0.194) and pT3c vs pT4 (P = 0.987 [Fig. 1]). Regarding OS, pairwise comparisons showed significant differences between pT1a vs pT1b (P = 0.009), pT3a vs pT3b (P < 0.001) and pT3b vs pT3c (P = 0.006), whereas no significant differences were observed for pT1b vs pT2 (P = 0.244), pT2 vs pT3a (P = 0.050) and pT3c vs pT4 (P = 0.776 [Fig. 2]). After re-classifying the pT stages according to 2010 TNM, we observed significant differences in MFS for pT1a vs pT1b (P < 0.001), pT1b vs pT2a (P = 0.035), pT3a vs pT3b (P < 0.001) and pT3b vs pT3c (P < 0.001), whereas no differences were observed for pT2a vs pT2b (P = 0.671), pT2b vs pT3a (P = 0.125) and pT3c vs pT4 (P = 0.987 [Fig. 3]). Regarding OS, we observed significant differences in pT1a vs pT1b (P = 0.009) and pT3a vs pT3b (P = 0.023), whereas no significant differences were found between pT1b vs pT2a (P = 0.365), pT2a vs pT2b (P = 0.650), pT2b vs pT3a (P = 0.151), pT3b vs pT3c (P = 0.084) and pT3c vs pT4 (P = 0.776 [Fig. 4]). Harrell's c-index to predict OS for the 2002 and 2010 versions of the TNM classification system was 0.643 and 0.633, respectively. After multivariate adjustment for regional lymph node invasion and presence of distant metastases, the c-index for the prediction of OS increased for the 2002 and 2010 versions of the TNM classification system to 0.673 and 0.667, respectively. Regarding CSS, Harrell's c-index to predict cancer-related death for the 2002 and 2010 versions of the TNM classification system was 0.756 and 0.739, respectively. After multivariate adjustment for regional lymph node invasion and presence of distant metastases, the c-index for the prediction of CSS increased to 0.856 and 0.845, respectively. For calculation of the predictive ability for MFS by Harrell's c-index and DCA, we excluded 158 patients with M1 disease. Considering the 2581 RCC cases with M0 disease at diagnosis, Harrell's c-index to predict MFS for the 2002 and 2010 versions of the TNM classification system was 0.744 and 0.726, respectively. Finally, DCA showed that, within this cohort, the use of the 2010 version of the TNM classification system was not associated with a net benefit in predicting OS, CSS or MFS compared with the 2002 version (Figs 5 and 6).

Table 1. Patients with RCC classified by the 2002 and 2010 TNM system (n = 2739)
2002 TNM classification2010 TNM classification
Pathological T stage n %Pathological T stage n %
pT1a97935.7pT1a97935.7
pT1b52619.2pT1b52619.2
pT21686.2pT2a1244.5
pT2b441.6
pT3a60522.1pT3a99736.5
pT3b43515.9pT3b431.6
pT3c70.3pT3c70.3
pT4190.7pT4190.7
Lymph node classification n %Lymph node classification n %
N0 + Nx257794.1N0 + Nx257794.1
N1722.6N11625.9
N2903.3   
M stage n %M stage n %
M0258194.2M0258194.2
M11585.8M11585.8
Figure 1.

Kaplan–Meier curves regarding MFS for the 2002 version of the TNM classification system for RCC.

Figure 2.

Kaplan–Meier curves regarding OS for the 2002 version of the TNM classification system for RCC.

Figure 3.

Kaplan–Meier curves regarding MFS for the 2010 version of the TNM classification system for RCC.

Figure 4.

Kaplan–Meier curves regarding OS for the 2010 version of the TNM classification system for RCC.

Figure 5.

Decision curve analysis showing that there was no gain in net benefit associated with the use of the 2010 version of the TNM classification system for prediction of A, MFS and B, OS. Model 1 refers to the 2002 version, model 2 to the 2010 version of the TNM classification system.

Figure 6.

Decision curve analysis showing that there was no gain in net benefit associated with the use of the 2010 version of the TNM classification system for prediction of CSS. Model 1 refers to the 2002 version, model 2 to the 2010 version of the TNM classification system.

DISCUSSION

With the introduction of the latest version of the TNM classification system for RCC in 2010, several studies in different tumour entities have addressed the clinical utility and predictive ability of these prognostic staging systems [14,15]. In the present study, we assessed the predictive ability of the novel 2010 version of the TNM classification system for RCC with regard to three widely used clinical endpoints, namely MFS, CSS and OS. However, using different statistical methods, the predictive ability was slightly inferior to the 2002 version of the TNM classification system for all endpoints tested. As the present study is a single-centre retrospective analysis, the results should be discussed in the context of other previously published reports.

Until now, published studies regarding the predictive value of the TNM classification system for RCC have only used CSS as a validated endpoint. In line with the present data, Lee et al. [16] reported results from an Asian RCC cohort, including 1691 patients, and found that the predictive ability of the 2010 version of the TNM classification system was not superior to the 2002 version. In contrast to the present data and the study by Lee et al., a recently published smaller series from the north-east of England showed a better correlation of pT3 tumours of the 2010 TNM classification system in predicting poor clinical outcome [17]. In another large US study comprising 3996 patients with RCC, it was demonstrated that the 2010 version provides a modest improvement in predictive ability with regard to CSS [18], and in a large multicentre cohort of 5300 patients with RCC, Novara et al. [8] found a comparable CSS rate between patients with pT2b and pT3a tumours, as well as between pT3c and pT4 tumours, whereas pT3a and pT3b tumour stages included patients with heterogenous outcomes. The recent 2010 version of the TNM classification system suggests a sub-classification of pT2 stages (i.e. organ-confined) according to the size threshold of 10 cm. In the present study, we were not able to demonstrate a significant difference between pT2a and pT2b tumours with regard to clinical outcome. Interestingly, in a large study by Waalkes et al. [19] the authors suggest that the new sub-classification of pT2 RCC into these two different stages does not yield additional prognostic information. It is of note, that although stratified staging for higher pT stages is important for the risk assessment of disease recurrence, the potential metastatic risk for localized pT1a tumours should also result in a consequent treatment and management of this small pT category [20].

As with all retrospective analyses, the present study has limitations inherent to its design. Many different pathologists and surgeons were involved over a long period of time and no central pathological review was performed. Moreover, the 2002 and 2010 versions of the TNM classification system do not represent a prospectively established variable.

In conclusion, we were able to validate the 2010 version of the TNM classification system for RCC and confirm its predictive ability for the three tested clinical endpoints MFS, CSS and OS. Comparing the predictive ability of the 2002 and 2010 versions of the TNM classification system, we were not able to find a significant superiority of the more recent version with regard to the evaluated endpoints. Our data seem to support a critical re-assessment and further improvement of the 2010 TNM classification system for the benefit of patients with RCC.

CONFLICT OF INTEREST

None of the contributing authors have any conflicts of interest, specific financial interests or relationships and affiliations relevant to the subject matter or materials discussed in the manuscript.

Ancillary