Tumor Grade Improves the Prognostic Ability of American Joint Committee on Cancer Stage in Patients With Penile Carcinoma

Transcription

1 Tumor Grade Improves the Prognostic Ability of American Joint Committee on Cancer Stage in Patients With Penile Carcinoma Rodolphe Thuret,* Maxine Sun,* Firas Abdollah, Lars Budaus, Giovanni Lughezzani, Daniel Liberman, Monica Morgan, Rupinder Johal, Claudio Jeldres, Mathieu Latour, Shahrokh F. Shariat, François Iborra, Jacques Guiter, Jean-Jacques Patard, Paul Perrotte and Pierre I. Karakiewicz From the Cancer Prognostics and Health Outcomes Unit (RT, MS, FA, LB, GL, DL, MM, RJ, CJ, SFS, PIK) and Departments of Urology (DL, MM, RJ, CJ, PP, PIK) and Pathology (ML), University of Montreal, Montreal, Quebec, Canada, Departments of Urology, Centre Hospitalier Universitaire de Montpellier (RT, FI, JG), Montpellier and Rennes University (JJP), Rennes, France, and Vita-Salute San Raffaele University (FA, GL), Milan, Italy, and Martini-Clinic, Prostate Cancer Center Hamburg-Eppendorf (LB), Hamburg, Germany Purpose: Penile cancer is rare. Thus, predicting cancer specific mortality may be difficult. We devised an accurate and yet easily applicable predictive rule that compares favorably with 2 previous models (73.8% and 74.7% accuracy, respectively). Materials and Methods: We identified patients treated with primary tumor excision for all stages of penile squamous cell carcinoma between 1998 and Disease stage definitions using Surveillance, Epidemiology and End Results stage, American Joint Committee on Cancer stage and TNM classification, and tumor grade were used to predict cancer specific mortality. Predictive accuracy estimates were compared using the DeLong method for related AUCs. Results: Surveillance, Epidemiology and End Results stage alone (1 predictor variable) was least accurate (74.5%). American Joint Committee on Cancer stage with tumor grade (2 predictor variables) was the most simple and most accurate (80.9%, p 0.001). A benefit similar to that of American Joint Committee on Cancer stage with tumor grade was seen for TNM classification and TG (80.7%, p 0.8). However, this rule (4 predictor variables) was more complex than American Joint Committee on Cancer stage and tumor grade. Conclusions: American Joint Committee on Cancer stage combined with tumor grade is the simplest, most accurate cancer specific mortality prediction rule after primary tumor excision for penile squamous cell carcinoma. This method is also more accurate than 2 previous cancer specific mortality prediction rules. Key Words: penis; carcinoma, squamous cell; mortality; nomograms; neoplasm staging PENILE cancer is an uncommon malignancy in the Western world with an incidence of 0.1 to 0.9 new cases per 100,000 males yearly. 1 This rate is lower than in developing countries, such as in some African and South American countries, where penile cancer represents up to 10% of malignant diseases. 2 Of all penile cancers 95% are squamous cell variants. 3 Due to the rarity of SCCP in 2006 Kattan et al devised a nomogram for the standardized prediction of CSM after treatment for primary SCCP in Abbreviations and Acronyms AJCC American Joint Committee on Cancer CSM cancer specific mortality EAU European Association Urology ILND inguinal lymph node dissection PTE primary tumor excision SCCP penile squamous cell carcinoma SEER Surveillance, Epidemiology and End Results TG tumor grade Submitted for publication June 22, Supported by the University of Montreal Health Center Urology Specialists, Fonds de la Recherche en Santé du Quebec, University of Montreal Department of Surgery, University of Montreal Health Center Foundation (PIK) and Association Française d Urologie (RT). Supplementary material for this article can be obtained at staging.html. * Equal study contribution. Correspondence: Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Center, 1058, rue St-Denis, Montréal, Québec, Canada, H2X 3J4 (telephone: ; FAX: ; pierre. karakiewicz@umontreal.ca) /11/ /0 Vol. 185, , February 2011 THE JOURNAL OF UROLOGY Printed in U.S.A by AMERICAN UROLOGICAL ASSOCIATION EDUCATION AND RESEARCH, INC. DOI: /j.juro

2 502 TUMOR GRADE IMPROVES PROGNOSTIC ABILITY OF STAGE IN PENILE CANCER a cohort of 175 Italian men. 4 This model relied on 8 predictors and was between 72.8% and 74.7% accurate depending on whether clinical or pathological N substage was used. Several predictors represented detailed pathological variables, ie tumor thickness, growth pattern or venous and/or lymphatic embolization. In 2009 Zini et al devised a simpler model in a North American cohort of 856 men. 5 Only 2 variables were used, including SEER stage and TG, and 73.8% accuracy was recorded. Despite the merits of this model it may have oversimplified the determinants of the SCCP CSM rate. Thus, we revisited the subject of SCCP CSM prediction and tested the ability of 2002 AJCC stage, the 2002 TNM classification, and the combination of AJCC stage or the TNM classification with TG to predict CSM. Our hypothesis was that more detailed and yet universally available information would result in better ability to predict CSM. MATERIALS AND METHODS Study Population In 17 SEER registries we identified all men treated for primary SCCP between 1988 and PTE consisted of excisional biopsy, or partial or total penectomy. The 2 diagnostic coding systems used were ICD-O-2 (C ) and ICD-O-3 codes for histological subtype (squamous cell carcinoma type, ICD-O-3, ). The SEER database includes approximately 26% of the American population and is considered representative of the United States in regard to demographic composition, and cancer incidence and mortality. 6 Only patients with squamous cell histology were included in analysis. TG was stratified according to the SEER database as grades I to III. Cases of confirmed distant metastasis were included and considered M1 even when tumor or nodal stage was unknown. Five patients treated in Alaskan and 2 treated in rural Georgian SEER registries were excluded from study due to exceedingly few observations. Disease stage was tabulated in 3 ways. 1) We relied on SEER stage (localized vs regional vs distant). 2) We used AJCC stage. 3) We relied on the 2002 TNM classification. 7 In patients treated with ILND pathological N substage was used in analysis vs clinical N substage in patients not treated with ILND. All stage assignment was obtained from the SEER database. PTE was considered the start of observation. The cause of death was defined according to the SEER specific cause of death code (28030). For the purpose of analysis death from another cause was considered a censored event or as other cause mortality on competing risks analysis. Statistical Analysis We relied on Kaplan-Meier plots to graphically illustrate CSM-free survival rates in the overall population, and after stratifying by SEER and AJCC stage as well as according to the TNM classification and TG. Three Cox regression models were used to test the accuracy of SEER stage vs AJCC stage vs TNM classification to predict CSM. An additional 3 Cox regression models were fitted using the same disease stages with the addition of TG. Multivariate Cox regression coefficients of the 3 models that relied on disease stage alone as well as of the 3 additional models that relied on disease stage and on TG were then used to generate nomograms. The prognostic ability of nomograms was quantified with ROC derived AUC estimates. 8,9 In Cox regression models the AUC was substituted for by Harrell s concordance index, 10,11 which was used in the current analysis. The prognostic ability of the 3 staging systems with and without TG was tested for predicting the 5-year CSM-free rate. We chose this time point based on the observation that virtually all cancer related events occurred within 5 years after PTE. We tested the statistical significance of the differences in accuracy estimates using the method of DeLong et al to compare related AUCs. 12 We then graphically explored the calibration of all models using the val.surv method. A separate plot was generated for each model without (data not shown) or with the inclusion of TG. Since a proportion of patients with SCCP die of other causes, we used univariate and multivariate competing risks regression models, as described by Fine and Gray, 13 Table 1. Characteristics of 1,324 patients diagnosed with SCCP between 1988 and 2006 in 15 SEER registries No. Pts (%) Race: White 1,134 (85.6) Black 126 (9.5) Other 64 (4.8) Treatment type: Excisional biopsy 286 (21.6) Partial penectomy 835 (63.1) Radical penectomy 203 (15.3) Tumor grade: I 410 (31.0) II 606 (45.8) III 308 (23.3) SEER stage : Localized 729 (55.1) Regional 515 (38.9) Metastatic 80 (6.0) AJCC stage: I 697 (52.6) II 301 (22.7) III 189 (14.3) IV 137 (10.3) 2002 TNM substage : T T1 763 (57.6) T2 334 (25.2) T3 163 (12.3) T4 28 (2.1) TX 36 (2.7) N cn0 948 (71.6) cn1 3/X 72 (5.4) pn0 127 (9.6) pn1 58 (4.4) pn2 62 (4.7) pn3 57 (4.3) M M0 1,273 (96.1) M1 51 (3.9)

3 TUMOR GRADE IMPROVES PROGNOSTIC ABILITY OF STAGE IN PENILE CANCER 503 to test the significance of the variables in predicting CSM. Competing risks regression models allow us to account for the effect of other cause mortality. All tests were 2 sided with significance considered at RESULTS Between 1988 and 2006 we identified a total of 1,324 patients with SCCP in 15 SEER registries (table 1). Average age was 66.3 years (median 68.0, range 22 to 102) and 85.6% of the men were white. Partial penectomy, excisional biopsy and radical penectomy were done in 63.1%, 21.6% and 15.3% of patients, respectively. SCCP was grades I to III in 410 (31.1%), 606 (45.8%) and 308 men (23.3%), respectively. According to SEER stage 729 patients (55.1%) had localized, 515 (38.9%) had regional and 80 (6.0%) had distant disease. AJCC stage was I to IV in 697 (52.6%), 301 (22.7%), 189 (14.3%) and 137 patients (10.3%), respectively. When the TNM classification was used, T stage was T1 to T4 in 763 (57.6%), 334 (25.2%), 163 (12.3%) and 28 patients (2.1%), respectively. Of all patients 304 (23.0%) underwent ILND. Pathological N substage was pn0 in 127 patients (9.6%), pn1 in 58 (4.4%), pn2 in 62 (4.7%) and pn3 in 57 (4.3%). Of the 1,020 patients who did not undergo ILND clinical N substage was cn0 in 948 (71.6%) and cn1-3/x in 72 (5.4%). Figure 1, A and B show cumulative CSM incidence-free rates according to Fine and Gray, 13 and CSM-free rates according to the Kaplan-Meier method, respectively. The 2 curves could be virtually superimposed. At 5 years the cumulative incidencefree CSM rate was 83.3% vs 81.1% by the Kaplan- Meier method. After stratifying by SEER stage the 5-year CSMfree survival rate was 91.1%, 75.8% and 18.1% for localized, regional and distant stage, respectively (fig. 1, C). After stratification by AJCC stage the 5-year CSM-free survival rate was 91.4%, 82.7%, 69.4% and 33.0% for stages I to IV, respectively (fig. 1, D). Figure 2 shows stratification by T, N and M stages, and TG. The 5-year CSM-free survival rate was 80.3%, 88.7%, 83.7% and 78.2% in patients with T2 any N any M, pn0 any T any M, M0 any T any N and TG II any T any N any M stage, respectively (fig. 2). In Cox regression models all examined variables achieved statistical significance (p 0.001, table 2). SEER stage, AJCC stage and TNM classification Figure 1. Cumulative incidence plots show CSM-free and other cause mortality-free rates in overall population of 1,324 men (A). Kaplan-Meier survival curve reveals CSM-free rate in overall cohort (B). Kaplan-Meier survival curve shows CSM-free rate by localized vs regional vs distant SEER stage (C). Kaplan-Meier survival curve demonstrates CSM-free rate by AJCC stages I vs II vs III vs IV (D).

4 504 TUMOR GRADE IMPROVES PROGNOSTIC ABILITY OF STAGE IN PENILE CANCER Figure 2. Kaplan-Meier plots show CSM-free rate after stratification by T (A), N(B) and M(C) stages, and TG (D) had 74.5%, 77.2% and 78.0% predictive accuracy, respectively (fig. 3). TG as a single predictor of CSM had 66.1% accuracy. The accuracy gain related to the inclusion of TG with SEER stage, AJCC stage and TNM classification was 3.3%, 3.7% and 2.7%, respectively. All accuracy gains were statistically significant (each p 0.02) and resulted in an estimated predictive accuracy of 77.8%, 80.9% and Table 2. Univariate and multivariate Cox regression analysis to predict CSM in 1,324 patients Staging System Alone Staging System Grade* Predictors (stage) HR (95% CI) p Value % AUC HR (95% CI) p Value % AUC SEER: Localized vs regional 3.0 ( ) 2.6 ( ) Localized vs distant 19.5 ( ) 16.1 ( ) AJCC: II vs I 2.0 ( ) 1.8 ( ) III vs I 4.0 ( ) 3.2 ( ) IV vs I 13.4 ( ) 11.0 ( ) TNM T: T2 vs T1 1.9 ( ) 1.3 ( ) T3 vs T1 3.8 ( ) 2.1 ( ) T4 vs T1 3.0 ( ) 1.5 ( ) TX vs T ( ) 1.6 ( ) N: cn1-3/x vs cn0 9.3 ( ) 3.2 ( ) pn0 vs cn0 1.1 ( ) 0.9 ( ) pn1 vs cn0 4.0 ( ) 2.5 ( ) pn2 vs cn0 4.6 ( ) 2.7 ( ) pn3 vs cn0 8.4 ( ) 4.6 ( ) M M1 vs M ( ) 4.8 ( ) * Tumor grade variable not shown.

5 TUMOR GRADE IMPROVES PROGNOSTIC ABILITY OF STAGE IN PENILE CANCER 505 Figure 3. Accuracy by concordance index of 5-year CSM-free results of SEER and AJCC stages, and TNM classification without and with TG using method of DeLong et al 12 for related AUCs. Reported p values represent statistical significance of accuracy comparisons using method of DeLong et al. 80.7% for SEER stage, AJCC stage and TNM classification, respectively, combined with TG. Similarly the comparison of accuracy estimates among the 3 staging systems after including TG also achieved statistical significance (each p 0.01, fig. 3). Exceptions were the comparisons between the combination of AJCC stage and TG, and the combination of TNM classification and TG (80.9% and 80.7%, respectively, p 0.8). In competing risks regression models adjusted for other cause mortality SEER stage, AJCC stage and TNM classification achieved statistical significance (each p 0.001, table 3). The statistical significance of all disease staging methods was maintained after including TG (each p 0.01) except for select substage comparisons of the TNM classification. Figure 4, A, C and E shows the effect of each staging systems and of TG on 5-year CSM-free predictions using nomograms. In the nomograms based on SEER stage and TG, and on AJCC stage and TG distant metastasis was the most powerful CSM predictor while for AJCC stage IV SCCP was the most powerful predictor (fig. 4, A and E). In each nomogram high TG was a less influential variable. In the nomogram based on TNM classification and TG pathologically confirmed N3 substage had virtually the same effect as distant metastasis (fig. 4, C). TG II and III had an intermediate effect. T substage was an even less influential variable. Figure 4, B, D and F shows that all 3 models without or combined with TG showed good calibration, as evidenced by the virtually perfect (1:1) relationship between predicted and observed rates. DISCUSSION Despite the availability of 2 models for the standardized prediction of CSM limitations affect these prediction rules. 4,5 The model devised by Kattan et al relied on highly detailed pathological variables, ie tumor thickness and growth pattern, which cannot be obtained from routine pathological assessment of penectomy specimens. 4 Thus, the model of Kattan et al cannot be routinely used in clinical practice. Also, this model was devised in a relatively limited sample of 175 patients treated at a total of 11 Italian Table 3. Univariate and multivariate competing risks regression models to predict CSM in 1,324 patients Staging System Alone Staging System Grade* Predictors (stage) HR (95% CI) p Value HR (95% CI) p Value SEER: Regional vs localized 2.9 ( ) ( ) Distant vs localized 14.3 ( ) ( ) AJCC: II vs I 2.0 ( ) ( ) III vs I 3.7 ( ) ( ) IV vs I 10.8 ( ) ( ) T: T2 vs T1 1.8 ( ) ( ) 0.2 T3 vs T1 3.6 ( ) ( ) T4 vs T1 2.8 ( ) ( ) 0.4 TX vs T ( ) ( ) 0.6 N: cn1 3/X vs cn0 7.7 ( ) ( ) pn0 vs cn0 1.1 ( ) ( ) 0.8 pn1 vs cn0 3.8 ( ) ( ) pn2 vs cn0 4.1 ( ) ( ) pn3 vs cn0 7.1 ( ) ( ) M1 vs M ( ) ( ) * Tumor grade variable not shown.

6 506 TUMOR GRADE IMPROVES PROGNOSTIC ABILITY OF STAGE IN PENILE CANCER Figure 4. Nomograms predicting CSM-free rate 5 years after primary tumor excision using SEER stage (A), TNM classification (C) and AJCC stage (E) combined with TG. Calibration between predicted (x axis) and observed (y axis) 5-year CSMfree rate for SEER stage (B), TNM classification (D) and AJCC stage (F) models. tertiary care centers. These considerations may limit the generalizability of this model. Moreover, its accuracy was also intermediate at 74.7%. Zini et al attempted to circumvent the limitations related to the complexity of the nomogram of Kattan et al 4 and the resulting limited clinical applicability. 5 They devised a simple tool in a cohort of 856 North American patients. Despite its similar accuracy and lesser complexity relative to the nomogram of Kattan et al 4 this tool may represent an excessively simplified prediction rule since it relies only on SEER stage and grade. 5 Based on these considerations we compared the ability of SEER stage, AJCC stage and the TNM classification with or without TG to predict CSM-free survival after PTE for SCCP. Our results confirmed that TNM classification yielded the most accurate prediction of 5-year CSM-free survival (78.0%) when considering disease stage alone (fig. 3). When disease stage was combined with TG, AJCC stage and TNM classification yielded the highest and the second highest predictive accuracy (80.9% and 80.7%, respectively, fig. 3). Relative to the nomogram of Kattan et al 4 and to the SEER stage based nomogram 5 the AJCC nomogram had 6.2% and 7.1% better predictions, respectively. This suggests that if predictions are generated in 1,000 consecutive patients, between 62 and 71 may be ranked incorrectly if 1 of the 2 models were chosen over AJCC stage combined with TG. This is not negligible, especially in patients who are potentially misclassified. Combining SEER stage with TG resulted in the least accurate CSM prediction relative to AJCC stage combined with TG (77.8% vs 80.9%). This finding validates our hypothesis that more detailed assessment of tumor stage (AJCC vs SEER stage) yields better accuracy. Our findings are more generalizable than those in previous reports since we relied on 15 SEER registries vs 9 in the analysis by Zini et al. 5 Moreover, our sample size was larger than previously published patient samples, that is 1,324 vs 856 patients described by Zini et al 5 and 175 described by Kattan et al. 4 Our findings are predominantly applicable to North American patients. The natural history of treated SCCP may be significantly different in South America or Asia. Thus, our tool should be validated in cohorts outside Western countries before its use in clinical practice. Application of the current nomogram may affect followup type and frequency. EAU provides guidelines in that regard. 14 However, important heterogeneity may exist between patients in the same EAU risk groupings. For example, 2 men without evidence of lymph node metastasis may have a substantially different 5-year CSM-free probability. Patient 1 may harbor a T2G1 primary tumor (AJCC stage II) without lymph node metastasis. His 5-year CSM-free rate would be 91% according to the nomogram based on AJCC stage and TG. Patient 2 may harbor a T3G2 primary tumor (AJCC stage III) without lymph node metastasis. His 5-year CSMfree rate would be 68% according to the same nomogram. The 23% difference in the 5-year CSM-free rate between the 2 patients may be used as a strategy for more frequent followup in patient 2 relative to patient 1. However, according to the EAU guidelines the same type of followup is recommended, namely regular physician or self-examination with ultrasound investigation of the groin every 6 months for 2 years. This ability to better discriminate between the risk of CSM in these 2 men may be interpreted as an added value of the current nomogram relative to the EAU guidelines. Our prognostic rule is based on Cox regression modeling. However, since a number of patients with SCCP may die of another cause, we confirmed the validity of our predictions in a competing risks regression model. 13 The latter avoids overestimating

7 TUMOR GRADE IMPROVES PROGNOSTIC ABILITY OF STAGE IN PENILE CANCER 507 the effect of CSM rates by adjusting for other cause mortality. CSM Kaplan-Meier based estimations were virtually the same as cumulative incidence estimates derived from the Fine and Gray methodology. 13 For example, at 5 years the CSM-free rate was 83.3% for the Fine and Gray technique vs 81.1% for the Kaplan-Meier method. Based on the striking similarity of the results obtained with the 2 modeling techniques we relied on Cox regression models to develop our prediction rules, as done previously. 5,15 Our study is not devoid of limitations. 1) The lack of central pathology review may have contributed to a higher accuracy of pathologically assessed variables. However, no previously reported nomogram relied on a central pathology review. 4,5 2) Other variables may predict CSM in penile cancer. For example, lymphovascular invasion is an independent prognostic factor for lymph node metastasis and for CSM Phimosis, smoking status or a self-reported history of condyloma are also well recognized prognostic factors. 19,20 Unfortunately these variables could not be obtained from the SEER database and could not be included in our models. CONCLUSIONS Despite the mentioned limitations to our knowledge our prognostic rule represents the simplest and yet most accurate method to predict CSM. It relies on AJCC stage and TG, and provides 81% accurate predictions. It is highly generalizable in men diagnosed and treated for SCCP in the United States. Ideally external validation should be performed. REFERENCES 1. Jemal A, Siegel R, Ward E et al: Cancer statistics, CA Cancer J Clin 2007; 57: Misra S, Chaturvedi A and Misra NC: Penile carcinoma: a challenge for the developing world. Lancet Oncol 2004; 5: Kroon BK, Horemblas S and Niewec OE: Contemporary management of penile squamous cell carcinoma. J Surg Oncol 2005; 89: Kattan MW, Ficarra V, Artibani W et al: Nomogram predictive of cancer specific survival in patients undergoing partial or total amputation for squamous cell carcinoma of the penis. J Urol 2006; 175: Zini L, Cloutier V, Isbarn H et al: A simple and accurate model for prediction of cancer specific mortality in patients treated with surgery for primary penile squamous cell carcinoma. Clin Cancer Res 2009; 15: Ries LA: SEER Cancer Statistics Review. Available at /. Accessed May Sobin LH and Wittekind C: TNM Classification of Malignant Tumours. London: Wiley-Liss Harrell FE Jr, Lee KL and Mark DB: Multivariate prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15: Harrell FE, Califf RM, Pryor DB et al: Evaluating the yield of medical tests. JAMA 1982; 247: Atkinson AC: A note on the generalized information criterion for choice of a model. Biometrika 1980; 67: Grambsch PM and Therneau TM: Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994; 81: DeLong ER, DeLong DM and Clarke-Pearson DL: Comparing the areas under two or more correlated receiver operating characteristics curves: a nonparametric approach. Biometrics 1988; 44: Fine JP and Gray RJ: A proportional hazards model for the substitution of a comparing risk. JASA 1999; 94: European Association Urology Guidelines. Available at professional-resources/guidelines/online/. Accessed May Karakiewicz PI, Briganti A, Chun FKH et al: Multiinstitutional validation of a new renal cancerspecific survival nomogram. J Clin Oncol 2007; 25: Ornellas AA, Nobrega BL, Wei Kin Chin E et al: Prognostic factors in invasive squamous cell carcinoma of the penis: analysis of 196 patients treated at the Brazilian National Cancer Institute. J Urol 2008; 180: Bhagat SK, Gopalakrishnan G, Kekre NS et al: Factors predicting inguinal node metastasis in squamous cell cancer of penis. World J Urol 2010; 28: Lopes A, Hidalgo GS, Kowalski LP et al: Prognostic factors in carcinoma of the penis: multivariate analysis of 145 patients treated with amputation and lymphadenectomy. J Urol 1996; 156: Daling JR, Madeleine MM, Johnson LG et al: Penile cancer: importance of circumcision, human papillomavirus and smoking in in situ and invasive disease. Int J Cancer 2005; 116: Nordenvall C, Chang ET, Adami HO et al: Cancer risk among patients with condylomata acuminata. Int J Cancer 2006; 119: 888.