A preoperative clinical prognostic model for non-metastatic renal cell carcinoma

Authors


Luca Cindolo, Department of Urology, Medical School of University ‘Federico II’, via S. Pansini 5, 80131 Naples, Italy.
e-mail: lucacindolo@virgilio.it

Abstract

Authors from Naples, Paris and Rennes describe their efforts to develop a model for the preoperative prediction of outcome for non-metastatic renal cancer. It is valuable to both urologist and patient to develop such a model, particularly so on preoperative criteria. The results of their study are interesting, leading to possibly helpful findings.

In another article, authors from Iceland estimated the risk of developing prostate and other cancer among relatives of men from that country diagnosed with prostate cancer. They found that a family history is a risk factor for prostate cancer, with the risk potentially higher for relatives of patients who died from the disease.

From the relative dearth of papers on quality of life after radical prostatectomy there are now several, and the authors from Bristol report on the effects of erectile dysfunction on quality of life after this type of treatment. They had a very high response rate (91%) to their questionnaire, and found that erectile dysfunction has a profound effect on quality of life. This finding is not a surprise, but do we need to examine our management of prostate cancer in the light of it?

OBJECTIVE

To develop a model to predict the outcome before surgery for non-metastatic renal cell carcinoma (RCC).

PATIENTS AND METHODS

The records of 660 patients with non-metastatic RCC, operated at three European medical institutes, were reviewed. Univariate and multivariate analyses were used to assess the clinical and pathological variables affecting disease-free survival.

RESULTS

The median (range) follow-up was 42  (2–180) months; the disease recurred in 110 patients (16%). The 2- and 5-year overall survival was 87% and 54%, respectively. Five variables were significant in the univariate analysis, i.e. clinical presentation, clinical and pathological size, tumour grade and stage (P < 0.05). The preoperative variables, e.g. clinical presentation and clinical tumour size, were retained from the multivariate model. A recurrence risk formula (RRF) was constructed from this model, as (1.28 × presentation (asymptomatic = 0; symptomatic = 1) + (0.13 × clinical size)). Using this equation, the 2- and 5-year disease-free survival was 96% and 93% for an RRF of ≤ 1.2 and 83% and 68% for an RRF of > 1.2.

CONCLUSION

A formula was developed which, independent of stage, can be used to predict the rate of treatment failure in patients who undergo nephrectomy for non-metastatic RCC. The RRF might be useful for more accurate sub-grouping of good-prognosis patients, and for counselling patients before surgery, their personalized follow-up or adjuvant treatment once available.

Abbreviations
RRF

recurrence risk formula

HR

hazard ratio.

INTRODUCTION

Because of the variable biological behaviour of RCC an accurate assessment of prognostic features at presentation is mandatory for defining the disease outcome and prognosis. Recent studies identified clinical, serological and biomolecular markers as prognostic tools influencing survival [1]. To date, tumour stage and grade remain the best indicators for predicting survival after radical treatment [2–4] and none of the new prognostic tools have been validated for prognostic relevance in patients with RCC [5]. However, there is general agreement that tumour stage and grade are not accurate enough to predict the large spectrum of clinical outcomes that occur in RCC.

Recently, some attempts have been made to develop mathematical models using independent prognostic variables to predict the recurrence risk or probability of survival in RCC [6–9]. This attractive approach allows the prediction of each patient's risk. The published series, integrating data before and after surgery, and clinical or pathological variables, have always been based on experience from one centre.

The objective of the present study, through the development of a European multicentre RCC database, was to establish a recurrence risk formula (RRF) that allows the prediction of individual recurrence risks after radical treatment for non-metastatic RCC.

PATIENTS AND METHODS

In a multicentre retrospective study we reviewed the records of 883 patients operated for a kidney malignancy since 1987 in the urology departments of the authors’ institutions. Relevant demographic, clinical and pathological data, and management and survival data, were retrieved and entered into a spreadsheet. Patients with bilateral synchronous or metachronous disease at the time of surgery, von Hippel-Lindau disease, tumour extension beyond Gerota's fascia (cT4 or pT4), clinical or pathological lymphadenopathy (cN+ or pN+), and metastatic dissemination (M+), were excluded from the study.

The clinical presentation was categorized as symptomatic (local or systemic) and incidental. Tumours that caused acute flank pain, lumbar pain, palpable mass or haematuria were identified as local symptomatic tumours. Tumours that caused weight loss, and paraneoplastic syndromes (anaemia, fever, asthenia, hypercalcaemia) were identified as systemic symptomatic tumours. All patients had preoperative CT or MRI and the findings used to assign clinical stage according to the 1997 TNM system [2]. The largest diameter of the tumour, measured on CT or MRI, was recorded as the clinical size. Pathological staging and grading were assigned according to the 1997 TNM classification. Patients were followed up periodically by a physical examination, routine laboratory evaluation, chest X-ray, and CT at 6–12 months.

Recurrence was defined by every new occurrence of kidney cancer after nephrectomy, and local, metastatic recurrence or death from kidney cancer. The disease-free survival time was calculated from the date of the surgery.

Differences in survival between prognostic groups were evaluated in a univariate analysis by the log-rank test, and the respective influence on survival of the different variables, significant at P < 0.05, calculated in a forward stepwise fashion according to the Cox regression method. The RRF was calculated using coefficients of the Cox model (βi), the hazard ratio (HR) being estimated as exp(βi). This was used to subdivide the patients into two subsets (good and poor prognosis). Survival curves were plotted and median and 5-year survival rates calculated from life tables, together with the respective 95% CI.

RESULTS

Of a total of 883 patients, 182 had advanced disease (T4 and/or any TN+ and/or any T and any NM+), 12 had bilateral synchronous or metachronous disease, 12 primary tumours not fitting Heidelberg criteria (sarcoma, Wilm's tumour, epidermoid carcinoma), nine Von Hippel-Lindau disease, eight metastatic neoplasm affecting the kidney (four lymphomas, two lung cancer and two ovarian cancer), and were excluded from study. Thus, 660 patients with sporadic unilateral non-metastatic RCC were available for analysis. The series comprised 404 (61%) men and 256 (39%) women (1.6 : 1) with a mean (sd, range) age of 60.6 (12.1, 21–89) years; 315 (48%) of the patients were incidentally diagnosed after having abdominal ultrasonography or CT for unrelated conditions, 265 (40%) presented with local symptoms related to a renal mass, and 80 (12%) complained of systemic symptoms related to renal neoplasms (Table 1).

Table 1.  Clinical and pathological features of the 660 patients
FeatureN (%)
Presentation
Incidental315 (48)
Local symptomatic265 (40)
Systemic symptomatic 80 (12)
Nephrectomy type
Radical594 (90)
Partial (nephron-sparing) 66 (10)
Surgical approach
Open611 (93)
Laparoscopy 49 (7.5)
Clinical T stage
T1418 (63)
T2 63 (9)
T3a 89 (13)
T3b/c 90 (14)
Pathological T stage
T1338 (51)
T2127 (19)
T3a102 (15)
T3b/c 93 (14)
Tumour grade
1139 (21)
2265 (40)
3230 (35)
undetermined 26 (4)
Histology
Papillary 21 (3)
Chromophobe 24 (4)
Conventional615 (93)

The mean (sd, range) tumour size was 6.8 (3.3, 1–21) cm; there were no differences between clinical (cTNM) and pathological (pTNM) size, at 6.8 (3.26) and 6.9 (3.29) cm (Student's t-test, P = 0.83). The clinical and pathological stages are also shown in Table 1, with the histology.

The median (range) follow-up after surgery was 42 (2–196) months; the overall median survival from histological diagnosis was 47 (0–196) months and the mean disease-free survival 53.8 months. The 2 and 5-year overall survival was 87% and 54%, respectively.

Disease recurred in 110 of the 660 patients (16%), with a mean time to recurrence of 27 months, comprising 20 (18%) with stage T1, 24 (21%) with T2 and 66 (61%) with T3. None of the 139 patients with a T1 tumour of < 4 cm had tumour recurrence. There were five recurrences after 66 nephron-sparing interventions (7.5%; two T1, one T2, two T3a), compared to 105 events (17.6%) among patients after radical nephrectomy (P = 0.036). The disease-free survival curves for patients with pT1 or with pT2–3 tumours are shown in Fig. 1a. The 2- and 5-year disease-free survival was 97% and 93% for pT1, and 79% and 62% for pT2–3 tumours.

Figure 1.

Kaplan-Meier survival curves for a, patients with stage pT1 (green line) and pT2–3 (red dashed line; Wilcoxon test, P < 0.001) and b, stratified for patients according the RRF ≤1.2 (green line) or >1.2 (red dashed line; Wilcoxon test, P < 0.001).

Univariate and multivariate analyses of the patients are shown in Table 2. Five variables emerged as statistically significant from the univariate analysis, i.e. clinical presentation, clinical and pathological size, tumour grade and tumour stage. Incidental tumours had a better prognosis than symptomatic (P < 0.001). In the multivariate analysis clinical presentation, clinical size, T stage and tumour grade remained significant. The clinical presentation (symptomatic vs incidental tumours) and clinical size were associated with an HR (95% CI) of 3.69  (2.11–6.08) and 1.14 (1.08–1.21), respectively, in the multivariate model. The two variables that were available before surgery were then selected from the multivariate model and only clinical presentation (asymptomatic = 0; symptomatic = 1) and clinical size incorporated into the RRF as (1.28 × clinical presentation) + (0.13 × clinical size). The RRF was calculated for each patient and used for stratification. A threshold of 1.2 was selected that generated both a good (≤1.2) and a poor prognosis group (>1.2), with survival rates equivalent to risk groups generated in accordance with pathological TNM stage (pT1 vs pT2–3). The distribution of events according to pTNM staging system and the RRF are shown in Table 3. It was possible to identify 82% and 85% of the events using the TNM staging and RRF (threshold 1.2), respectively. If the patients were stratified into two sets according to TNM (pT1 and pT2–3) there were 20 and 90 events, respectively, but when using the selection criteria based on RRF there were 16 and 94 events.

Table 2.  Prognostic factors for RCC analysed in a univariate and a forward, stepwise multivariate Cox proportional hazards model
VariableHR (sem, 95% CI)P
Univariate analysis
Sex1.07 (0.21, 0.73–1.58)0.715
Age1.01 (0.01, 0.99–1.03)0.179
Clinical presentation4.66 (1.15, 2.87–7.57)< 0.001
pTNM5.54 (1.37, 3.41–9.00)< 0.001
Clinical size1.2 (0.03, 1.14–1.26)< 0.001
Pathological size1.2 (0.03, 1.14–1.26)< 0.001
Cellular type0.56 (0.23, 0.24–1.29)0.175
Cellular grade2.99 (0.15, 2.20–4.08)< 0.001
Multivariate analysis
Clinical presentation3.69 (0.97, 2.11–6.08)< 0.001
Clinical size1.14 (0.03, 1.08–1.21)< 0.001
pTNM3.87 (0.27, 1.61–5.71)< 0.001
Cellular grade3.79 (0.37, 1.8–7.9)< 0.001
Table 3.  Distribution of recurrence according to pTNM stage and RRF
VariableNo. of
patients
N (%) of
recurrences
5-year disease-
free survival (%)
pT133820 (6)93
pT2–332290 (28)62
RRF ≤ 1.228816 (5)93
RRF > 1.237294 (25)68

The disease-free survival curves based on the RRF (good vs poor prognosis) are shown in Fig. 1b. The 2- and 5-year disease-free survival was 96% and 93% for an RRF of ≤ 1.2 and 83% and 68% for an RRF of > 1.2.

DISCUSSION

Tumour stage [2] and grade [3] are the best indicators of survival after radical treatment for RCC [10]. Within tumour stage, tumour size is also a recognized prognostic variable separating T1 and T2 tumours. However, the optimum prognostic threshold is still a subject of debate [11,12].

The profile of diagnosed renal tumours has developed considerably [13]; in the 1970s, ≈ 10% of renal tumours were discovered incidentally [14] and indeed > 60% of treated renal tumours are totally asymptomatic [15]. These asymptomatic tumours are of lower stage and grade than symptomatic lesions [16,17]. Also, it seems that the mode of presentation of renal tumours could be a clinical prognostic variable independent of stage and grade [17–19]. Three clinical variables seem to have prognostic relevance; the absence of symptoms, symptoms like haematuria, flank pain or palpable mass, and a change in general health status [20]. Thus it appeared logical and potentially useful to rely on these new clinical prognostic variables in a mathematical model to predict disease recurrence in non-metastatic renal tumours.

In the present study of 660 sporadic unilateral non-metastatic renal tumours the prognostic factors identified in the univariate analysis were consistent with those in other studies, and widely accepted as clinical prognostic variables. Specifically, we confirmed that clinical presentation, clinical or pathological tumour size and TNM stage were variables to consider when a prognostic judgement is required. Also, stepwise multivariate analysis identified clinical presentation and tumour size as independent prognostic factors with tumour stage and grade. The multivariate analysis allowed the development of a formula to predict the risk of recurrence or progression in localized RCC. From this equation the population was divided in two groups whose survival was adjusted to the TNM distribution. The model the predicted a 5-year disease-free survival of 93% and 62%, for an RRF of ≤1.2 and >1.2, respectively.

Others have developed mathematical models using clinical or pathological variables to predict the outcome of RCC after radical treatment. Kattan et al.[7], from a series of 601 patients with localized disease, combined symptoms (incidental, local or systemic), histology (chromophobe, conventional or papillary), tumour size and pathological stage. This gave a nomogram predicting the 5-year recurrence probability with an accuracy as an area of 0.74 under the ROC curve. Zisman et al.[8], in a series of 661 patients (metastatic or not) constructed the UCLA Integrated Staging System, which included variables like the TNM stage, tumour grade and Eastern Cooperative Oncology Group performance status score. The clinical variables (performance) matched with pathological data provided five powerfully predictive survival-stratification groups. The same group, in a series of 292 localized tumours, developed a survival formula based on two independent variables obtained in a multivariate analysis. This formula, predicting the patient's probability of survival was:

100 × (Sb)e((1.09 × performance status) + (0.57 × (grade − 1))

where Sb was ‘baseline survivorship estimate’. Yaycioglu et al.[6], in a series of 296 patients, developed a clinical model that included the mode of presentation of the tumours (symptomatic or not) and its clinical size determined by CT. Their results are similar to those in the present study, as these two variables emerged as independent prognostic factors in a multivariate analysis; they also obtained a RRF (1.55 × presentation + 0.19 × clinical size) separating high- and low-risk patients for recurrence. These series have some limitations; they reflect the experience of only one institution with sometimes heterogeneous groups of patients, and the use of mixed variables before and after surgery.

On this point, the present study used only preoperative clinical variables in a well-defined cohort of patients. Also, to our knowledge, it is the first algorithm constructed from a multicentre experience and from a European series. Although the RRF threshold has been calibrated on the pTNM distribution it does not mean that the formula gives information similar to that classification. The information provided by the formula is not only beneficial to patients in the interim period between diagnosis and nephrectomy, but is also important in providing the patient with pertinent information as soon as they are diagnosed. It might also be interesting to pre-select for adjuvant clinical trials in high-risk patients. The identification of new biological targets is a rationale for such a strategy [21,22]. Also, the early detection of high-risk patients might help to obtain good compliance with a close follow-up. The important point is that tumour size and symptoms are independent variables, and thus the present model is potentially useful for refining the TNM stage classification. T1 or G2 tumours might be reclassified as being of high risk according to the RRF, while some T3 or G3 tumours might be considered as being of low risk. For instance, a symptomatic T1 tumour of 6 cm should be considered a high risk for progression (RRF 2.12). Finally, the increasing number of small tumours diagnosed leads to the development of conservative treatment options which unfortunately do not provide histological specimens (radiofrequency, high-intensity focused ultrasound, cryoablation). In this context, as noted by Yaycioglu et al.[6], mathematical models using purely preoperative clinical variables could be useful for refining inclusion and exclusion criteria. However, for such purposes the different formulae that have been published must be compared and validated in large multicentre studies.

In conclusion, from preoperative variables and using the Cox proportional hazard regression analysis, a simple mathematical model (the RRF) was developed that can accurately predict the risk of recurrence in patients after surgery for non-metastatic RCC. Once validated with external data this kind of representation will be a useful tool in preoperative counselling, stratification into treatment options, planning follow-up strategies and adjuvant therapy in patients with newly diagnosed RCC.

ACKNOWLEDGEMENTS

We acknowledge Dr Giovanna Messina for the statistical analysis support.

Ancillary