- Email: [email protected]
Development and Evaluation of a Nomogram to Predict Inguinal Lymph Node Metastasis in Patients With Penile Cancer and Clinically Negative Lymph Nodes
Development and Evaluation of a Nomogram to Predict Inguinal Lymph Node Metastasis in Patients With Penile Cancer and Clinically Negative Lymph Nodes Yao Zhu, Hai-Liang Zhang, Xu-Dong Yao, Shi-Lin Zhang, Bo Dai, Yi-Jun Shen and Ding-Wei Ye* From the Department of Urology, Fudan University Shanghai Cancer Center and Department of Oncology, Shanghai Medical College, Fudan University, Shanghai, People’s Republic of China
Purpose: Optimal management for penile cancer in patients with clinically negative lymph nodes is still under debate. We developed and evaluated a nomogram to stratify patients who are suitable candidates for further treatment. Materials and Methods: This study included 110 men with penile cancer and clinically negative lymph nodes from 1990 to 2008. All patients underwent primary tumor resection and regional lymphadenectomy. We retrospectively reviewed medical records and tumor slides. Statistical analysis was done in R with library rms. Results: The lymph node metastasis rate in the entire cohort was 23.6%. The final model, presented as a nomogram, included T stage, grade, lymphovascular invasion and p53 expression. Only lymphovascular invasion showed independent prognostic value on multivariate analysis (p ⫽ 0.024). The model also showed good calibration (bootstrap corrected concordance index 0.79). To determine the clinical usefulness of the nomogram we compared it with the European Association of Urology risk classification using decision curve analysis. At a 10% probability threshold our nomogram led to 1 positive result per 100 patients without an increase in the number of false-positive results. At this probability threshold the model also decreased 13 unnecessary interventions per 100 patients without missing more metastatic disease. Conclusions: We generated a nomogram in patients with clinically node negative penile cancer based on readily available pathological factors. The clinical usefulness of the nomogram was evidenced by decision curve analysis.
Abbreviations and Acronyms c-index ⫽ concordance index DSNB ⫽ dynamic sentinel LN biopsy EAU ⫽ European Association of Urology FNAB ⫽ fine needle aspiration biopsy LN ⫽ lymph node LNM ⫽ LN metastasis pt ⫽ threshold probability Submitted for publication December 18, 2009. * Correspondence: Department of Urology, Fudan University Shanghai Cancer Center, No. 270 Dong’an Rd., Shanghai, 200032, People’s Republic of China (telephone: 86-21-64175590; FAX: 86-2164434556; e-mail: [email protected]).
Key Words: penis; carcinoma, squamous cell; lymph nodes; neoplasm metastasis; nomograms OPTIMAL management for penile cancer in men with clinically negative LNs is still under debate. Today there are 2 main procedures in this setting, that is prophylactic lymphadenectomy and DSNB. Lymphadenectomy provides accurate staging information and can cure minor LN disease.1 DSNB has a low complication rate with acceptable false-negative results
at dedicated centers.2 However, the 2 interventions are invasive and have more than an 80% over treatment rate in the targeted population.1,2 Thus, predefined criteria for patients at risk are suboptimal and need further improvement. Since Ficarra et al first introduced a nomogram to predict inguinal LNM in patients with penile cancer,3 sev-
0022-5347/10/1842-0539/0 THE JOURNAL OF UROLOGY® © 2010 by AMERICAN UROLOGICAL ASSOCIATION EDUCATION
Vol. 184, 539-545, August 2010 Printed in U.S.A. DOI:10.1016/j.juro.2010.03.145
AND
RESEARCH, INC.
www.jurology.com
539
540
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
eral studies have been done to develop useful prediction models.4,5 If announced, these studies often included patients with clinically palpable LNs and showed that clinical palpability has a significant impact on LNM risk.3,5 According to our experience these cases may probably be classified as LN positive disease by FNAB. Using the current DSNB technique preoperative ultrasound and intraoperative palpation are also done to screen for LNM.6 Thus, it is doubtful whether a nomogram should be used instead of low cost detection techniques in LN palpable cases. In addition to the inappropriate patient population, another important problem is whether statistical performance translates into clinical usefulness. Penile cancer is curable when surgery is done at an early stage but the outcome after LN recurrence is often dismal.7 Thus, the probability threshold to stratify high and low risk cases may be 5% to 15%. In this situation a prediction model with good calibration in patients at more than 50% risk may have limited clinical significance. As introduced by Vickers and Elkin,8 decision curve analysis can identify the range of threshold probabilities in which a model is of value, the magnitude of benefit and which of several models is optimal. To overcome these limitations we developed a nomogram based on clinicopathological features in 110 patients with penile cancer and clinically negative LNs. We evaluated the clinical usefulness of the nomogram by decision curve analysis at different probability thresholds.
tected by immunohistochemical staining.11 The cutoff to define strong p53 expression was 20%, according to previous reports.11,12 The EAU risk classification of penile cancer defines risk as low—Tis, TaG1–2 or T1G1, intermediate—T1G2 and high—T2 or greater, or G3.13 Standard inguinal LN dissection was done in these patients. The borders included 2 cm above the inguinal ligament superiorly, the apex of the femoral triangle inferiorly, the adductor longus medially and the sartorius muscles laterally. All superficial and deep inguinal LNs in the field were removed. We used a standard method of processing LN specimens. Multivariate logistic regression analysis was used to test the association of clinicopathological features with inguinal LN status. This model was then used to predict the individual probability of positive inguinal LNs. Restricted cubic splines modeling was done using continuous variables when it could accommodate potential nonlinear effects and optimize Akaike’s information criterion. For model validation we assessed discrimination and calibration. Discrimination was measured using the c-index, which is similar to the ROC curve AUC.14 Bootstrap corrected c-indexes were used to better gauge expected future predictive accuracy. Calibration was assessed by visual inspection of the plots of predicted probability of metastasis vs actual outcome. We further used decision curve analysis to explore the clinical effects of our models.8 This method estimates a net benefit for prediction models by summing the benefits (positives) and subtracting the harms (false positives).
Table 1. Clinicopathological characteristics in 110 patients penile cancer Variables
MATERIALS AND METHODS The study group included 110 men with penile squamous cell carcinoma and clinically negative LNs from January 1990 to December 2008. Of the patients 13 underwent surgery elsewhere and were included in analysis after a review of pathological slides and medical records. The remainder were treated at our cancer center. Since we routinely suggest prophylactic inguinal lymphadenectomy in patients with penile cancer (except Ta/Tis disease), pathological staging information on inguinal LNs was readily available. Physical examination of inguinal LNs was done 1 to 4 weeks after primary disease resection. The definition of clinically negative LNs was 1) impalpable inguinal LNs or 2) palpable inguinal LNs but negative FNAB results. Patient age and time from presentation to primary disease treatment were retrieved from medical records. Pathological review was done using primary tumor slides. Tumor stage was assigned using the 2002 American Joint Committee on Cancer TNM system.9 T2 stage was further divided into 2 subgroups based on invasion depth, including T2a— corpus spongiosum infiltration and T2b— corpora cavernosa involvement. Histological grade was classified by the Broders system.10 Lymphovascular invasion was defined as tumor emboli within endothelium lined spaces bounded by a thin wall. p53 expression was de-
Presentation–primary disease treatment (mos): Less than 3 3–6 months Greater than 6 T stage: T1 T2a T2b T3 Grade: G1 G2 G3 Lymphovascular invasion: Absent Present p53 Expression: Weak Strong EAU risk classification: Low Intermediate High N stage: N0 N1 N2 N3
No. Pts (%) 64 (58.2) 42 (38.2) 4 (3.6) 60 (54.5) 33 (30) 9 (8.2) 8 (7.3) 60 (54.5) 42 (38.2) 8 (7.3) 91 (82.7) 19 (17.3) 78 (70.9) 32 (29.1) 34 (30.9) 20 (18.2) 56 (50.9) 84 (76.4) 17 (15.5) 8 (7.3) 1 (0.9)
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
Table 2. Univariate and multivariate analysis of clinicopathological factors to predict inguinal LNM in 110 patients Univariate Variables
% LNM
T stage: T1 T2a T2b T3 Grade: G1 G2 G3 Lymphovascular invasion: Absent Present p53 Expression: Weak Strong
p Value
Multivariate* OR (95% CI)
0.0022 11.7 27.3 77.8 37.5
— 2.61 (0.68–10.1) 7.32 (0.66–81.52) 3.78 (0.44–32.66) 0.0025
10 40.5 37.5
— 2.77 (0.72–10.72) 6.89 (0.77–61.88) ⬍0.0001
13.2 73.7
— 6.75 (1.28–35.73) 0.032
17.9 37.5
— 3.22 (0.96–10.86)
p Value 0.30 — 0.17 0.10 0.22 0.16 — 0.14 0.09 0.024 — 0.058 —
* Based on comparison to first value in each set of categorized variables.
Since the value of a positive finding, such as identifying metastatic disease early, may be different from the disadvantages of a false-positive finding, such as unnecessary lymphadenectomy, the net benefit calculation weights true and false positives differently. Weighting is derived from the threshold probability of a disease at which a patient would elect intervention. A false-positive finding is weighted as pt/(1 ⫺ pt) compared with a positive finding. Because this threshold probability can vary among patients, net benefit is calculated across a range of probabilities. We chose 5% to 25% as our range. We would be surprised if any man would choose invasive intervention when told that the probability of LNM was less than 5% and few clinicians would do 20 or more invasive proce-
541
dures to find metastatic disease. Similarly it would be rare for a man to demand at least a 1/4 chance of metastasis before accepting treatment. Bootstrap resampling was used to correct model overfit on decision curve analysis. Interpretation of a decision curve is simple. The model with the highest net benefit at a particular threshold probability should be chosen and we estimated the net benefit at each threshold probability. All analysis was done with R 2.10.0 with the rms library and other functions.8,14 –16
RESULTS Table 1 lists clinicopathological characteristics in 110 men with a mean age of 52.9 years (median 54, range 20 to 75) with penile cancer. A median of 23 inguinal LNs (range 11 to 37) was removed per case with a mean of 1.6 positive inguinal LNs (range 1 to 5) per LN positive case. Four patients had bilateral inguinal LNM. On univariate analysis patient age and time from presentation to primary disease treatment had weak associations with LN status (p ⬎0.5). They were excluded from the final model, which consisted of T stage, grade, lymphovascular invasion and p53 expression. T2b disease was more often associated with LNM than T2a and T3 disease. Multivariate analysis revealed only lymphovascular invasion as a statistically significant predictor of LNM (table 2). The prediction model is presented as a nomogram that graphically shows the multivariate effects of prognostic factors (fig. 1). We further evaluated nomogram discrimination and calibration. The bootstrap corrected c-index of the model was 0.79. Figure 2 shows good calibration between predicted risk and
Figure 1. Nomogram predicting LNM risk in patients with clinically node negative penile cancer
542
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
observed incidence. In this patient population the EAU risk classification achieved a c-index of 0.71, which was inferior to that of the nomogram (p ⬍0.001). We then investigated the clinical usefulness of the nomogram and used EAU risk classification as a comparison. Figure 3 shows the distribution of nomogram predictions by EAU risk classifications. The risk of LNM in the EAU high risk category varied highly and some cases were overrated. Thus, better risk prediction is strongly needed to avoid over treatment in these patients. Figure 4, A shows the results of decision curve analysis. At most threshold probabilities the nomogram had a higher net benefit over the EAU risk classification. According to the predefined range of pt we constructed a plot to determine the net decrease in interventions by the nomogram and the EAU risk classification (fig. 4, B). Table 3 shows the net benefit and decrease at pt from 5% to 25% at a 5% interval. In our clinical practice we choose 10% as pt to treat clinically LN negative cases of penile cancer. In this circumstance our nomogram found 1 positive result per 100 cases without an increase in the number of false-positive results. Furthermore, it also decreased 13 interventions per 100 cases without missing more positive results.
DISCUSSION Early detection of inguinal LNM is of paramount importance for penile cancer management. Lont et al compared outcomes in clinically LN negative penile cancer cases managed by surveillance or further diagnosed by DSNB with subsequent LNM resection.7 Although the groups were similar in clinicopathological features, the disease specific survival rate was significantly better in the early treatment group. Of 20 patients with LNM during surveillance 13 died of penile cancer. Patients with recurrence
Figure 2. Nomogram calibration plot. C, c-index.
Figure 3. Nomogram predictions by EAU risk classification
were more likely to have bilateral LNM, extracapsular growth and an increased number of positive LNs, which are adverse predictors of survival. Thus, omitting surgery in patients with occult LN disease may probably eliminate the opportunity to cure the disease. However, complications of inguinal lymphadenectomy are hard to accept, especially since almost 80% of patients with impalpable LNs are over treated.1 DSNB in a dedicated group can achieve a 7% false-negative rate with a 4.7% complication rate.2 However, biopsy results were negative in 86% of explored groins when including all cases staged as T1G2 or greater. Since the estimated cost per case is up to €11,000, the benefit of avoiding unnecessary lymphadenectomy is not as obvious.17 Thus, better patient selection is needed to improve the efficacy of further invasive interventions.
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
543
Figure 4. Nomogram and EAU risk classification. Vertical line indicates 10% probability threshold cutoff. FUSCC, Fudan University Shanghai Cancer Center. A, benefit. B, decrease.
Several groups investigated potential LNM prognostic factors in penile cancer cases.18 Tumor stage and grade are commonly used predictors in clinical practice.19 EAU guidelines recommended a risk classification to select patients for early lymphadenectomy.13 Hegarty et al reported 100 cases of penile cancer prospectively managed according to the guideline.1 None of 17 patients at low risk had LN disease during followup. However, predictive accuracy was not promising in those at intermediate and high risk. Only 18% of men with unfavorable primary disease and impalpable nodes had positive LN disease at prophylactic lymphadenectomy. Our results were similar (fig. 3). Predicted probability varied in men at high risk, of whom almost half were at less than 30% risk for LNM.
To increase prognostic accuracy tumor stage and grade modification or the introduction other pathological variables has been investigated. In the current T staging system the same group (T2) includes tumor involvement of the corpus spongiosum or corpora cavernosa, which is considered to have a better outcome than urethral invasion (T3). In our study men with corpora cavernosa invasion were at higher risk for LNM than those with corpus spongiosum or urethral infiltration. These observations were confirmed in another large cohort.20 Among other prognostic factors lymphovascular invasion and p53 expression have gained the most attention. In a report of 175 patients lymphovascular invasion was identified as an independent predictor of LNM.21 Prognostic significance was retained in 115 patients with
544
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
Table 3. Nomogram net benefit and decrease vs EAU risk classification % Probability Threshold
Net Benefit/100 Cases
Net Decrease/100 Cases
5 10 15 20 25
0 1 2 3 3
5 13 14 11 8
clinically negative LNs (OR 5.381). Lopes et al noted that p53 expression was associated with LNM on multivariate analysis including lymphovascular invasion.12 Their results were confirmed by another report in which p53 expression had independent prognostic value after including other molecular variables, such as Ki-67, epithelial cadherin and matrix metalloproteinase 9.11 We observed that lymphovascular invasion was the only independent LNM prognostic factor in clinically LN negative cases of penile cancer. However, p53 expression failed to achieve statistical significance (p ⫽ 0.058). To facilitate the clinical use of prediction models several groups have developed penile cancer nomograms. In patients at low risk for LNM a nomogram could be added as a tool in the decision making process to provide a numerical estimate to help the clinician and patient weigh the pros and cons of this decision.22,23 In the nomogram of Ficarra et al lymphovascular invasion and clinically palpable inguinal LNs were 2 powerful predictors of LNM (bootstrap corrected c-index 0.876).3 In the same patient population Kattan et al constructed another useful nomogram to predict 5-year cancer specific survival in patients with penile cancer.24 Velazquez et al developed a nomogram to predict LNM in tumors invading 5 to 10 mm.4 Perineural invasion and grade were significant prognostic factors in this subgroup. However, they did not report the statistical performance of the nomogram. Recently Bhagat et al analyzed 106 patients with penile cancer to construct a nomogram.5 Grade, lymphovascular inva-
sion and clinical palpability were the strongest predictors in that model. The nomogram c-index was 0.74. Our nomogram has 4 readily available pathological variables and achieved a c-index of 0.79, superior to that of the EAU risk classification in this cohort. The populations in the 4 studies were different since we only included patients with clinically negative LNs. We performed FNAB in patients with palpable LNs because 1) many patients had palpable but inflammable LNs that could not be defined as exactly nonpalpable and 2) FNAB is an easy, minimally invasive technique to detect LNM in these men at high risk. Our nomogram has good statistical performance but its clinical importance may be limited. For penile cancer with clinically negative LNs common management is surveillance or surgery (lymphadenectomy or DSNB). Thus, various predicted risks may be eventually transformed into 2 groups. On decision curve analysis we evaluated the benefit of the nomogram using different cutoffs. We used the EAU risk classification for comparison. When a 10% cutoff was applied, our nomogram had a net benefit of 1/100 cases and a net decrease of 13/100. Our nomogram correctly classified a proportion of patients at high risk. Our study has limitations. Sample size was relatively small for statistical analysis and data were collected retrospectively. Given the rarity of penile cancer, our study is still a large cohort with inguinal LN pathological staging. Furthermore, the predictive accuracy of the nomogram should be tested in an external cohort, which represents the gold standard to confirm the validity of clinical prediction models.
CONCLUSIONS We generated a nomogram in men with clinically node negative penile cancer based on readily available pathological factors. Decision curve analysis evidenced the clinical usefulness of the nomogram. Using a 10% probability threshold our nomogram was helpful for decreasing unnecessary intervention.
REFERENCES 1. Hegarty PK, Kayes O, Freeman A et al: A prospective study of 100 cases of penile cancer managed according to European Association of Urology guidelines. BJU Int 2006; 98: 526. 2. Leijte JA, Hughes B, Graafland NM et al: Twocenter evaluation of dynamic sentinel node biopsy for squamous cell carcinoma of the penis. J Clin Oncol 2009; 27: 3325. 3. Ficarra V, Zattoni F, Artibani W et al: Nomogram predictive of pathological inguinal lymph node involvement in patients with squamous
cell carcinoma of the penis. J Urol 2006; 175: 1700. 4. Velazquez EF, Ayala G, Liu H et al: Histologic grade and perineural invasion are more important than tumor thickness as predictor of nodal metastasis in penile squamous cell carcinoma invading 5 to 10 mm. Am J Surg Pathol 2008; 32: 974. 5. Bhagat SK, Gopalakrishnan G, Kekre NS et al: Factors predicting inguinal node metastasis in
squamous cell cancer of penis. World J Urol 2009. 6. Leijte JA, Kroon BK, Valdes Olmos RA et al: Reliability and safety of current dynamic sentinel node biopsy for penile carcinoma. Eur Urol 2007; 52: 170. 7. Lont AP, Horenblas S, Tanis PJ et al: Management of clinically node negative penile carcinoma: improved survival after the introduction of dynamic sentinel node biopsy. J Urol 2003; 170: 783.
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
8. Vickers AJ and Elkin EB: Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making 2006; 26: 565. 9. Penis. In: AJCC Cancer Staging Manual, 6th ed. American Joint Committee on Cancer. New York: Springer 2002; pp 303–308. 10. Broders AC: Squamous-cell epithelioma of the skin. Ann Surg 1921; 73: 141. 11. Zhu Y, Zhou XY, Yao XD et al: The prognostic significance of p53, Ki-67, epithelial cadherin and matrix metalloproteinase-9 in penile squamous cell carcinoma treated with surgery BJU Int 2007; 100: 204. 12. Lopes A, Bezerra AL, Pinto CA et al: p53 as a new prognostic factor for lymph node metastasis in penile carcinoma: analysis of 82 patients treated with amputation and bilateral lymphadenectomy. J Urol 2002; 168: 81. 13. Solsona E, Algaba F, Horenblas S et al: EAU guidelines on penile cancer. Eur Urol 2004; 46: 1.
14. Harrell FEJ: The rms Package for R: Regression Modeling Strategies, R package version 2009; 2.1-0. Available at http://biostat.mc.vanderbilt. edu/wiki/Main/Rrms. Accessed November 12, 2009. 15. Steyerberg EW: Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. New York: Springer 2008. 16. R: A Language and Environment for Statistical Computing. R Development Core Team. Vienna, Austria: R Foundation for Statistical Computing 2009. 17. Hegarty PK, Rees RW, Borley NC et al: Contemporary management of penile cancer. BJU Int 2008; 102: 928. 18. Ficarra V, Novara G, Boscolo-Berto R et al: How accurate are present risk group assignment tools in penile cancer? World J Urol 2009; 27: 155. 19. Pettaway CA, Lynch DF and Davis JW. Tumors of the penis. In: Campbell-Walsh Urology, 9th ed. Edited by AJ Wein, LR Kavoussi, AC Novick et al. Philadelphia: Saunders Elsevier 2007; chapt 31.
545
20. Leijte JA, Gallee M, Antonini N et al: Evaluation of current TNM classification of penile carcinoma. J Urol 2008; 180: 933. 21. Ficarra V, Zattoni F, Cunico SC et al: Lymphatic and vascular embolizations are independent predictive variables of inguinal lymph node involvement in patients with squamous cell carcinoma of the penis: Gruppo Uro-Oncologico del Nord Est (Northeast Uro-Oncological Group) Penile Cancer data base data. Cancer 2005; 103: 2507. 22. Ross PL, Gerigk C, Gonen M et al: Comparisons of nomograms and urologists’ predictions in prostate cancer. Semin Urol Oncol 2002; 20: 82. 23. Specht MC, Kattan MW, Gonen M et al: Predicting nonsentinel node status after positive sentinel lymph biopsy for breast cancer: clinicians versus nomogram. Ann Surg Oncol 2005; 12: 654. 24. 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: 2103.
Purpose: Optimal management for penile cancer in patients with clinically negative lymph nodes is still under debate. We developed and evaluated a nomogram to stratify patients who are suitable candidates for further treatment. Materials and Methods: This study included 110 men with penile cancer and clinically negative lymph nodes from 1990 to 2008. All patients underwent primary tumor resection and regional lymphadenectomy. We retrospectively reviewed medical records and tumor slides. Statistical analysis was done in R with library rms. Results: The lymph node metastasis rate in the entire cohort was 23.6%. The final model, presented as a nomogram, included T stage, grade, lymphovascular invasion and p53 expression. Only lymphovascular invasion showed independent prognostic value on multivariate analysis (p ⫽ 0.024). The model also showed good calibration (bootstrap corrected concordance index 0.79). To determine the clinical usefulness of the nomogram we compared it with the European Association of Urology risk classification using decision curve analysis. At a 10% probability threshold our nomogram led to 1 positive result per 100 patients without an increase in the number of false-positive results. At this probability threshold the model also decreased 13 unnecessary interventions per 100 patients without missing more metastatic disease. Conclusions: We generated a nomogram in patients with clinically node negative penile cancer based on readily available pathological factors. The clinical usefulness of the nomogram was evidenced by decision curve analysis.
Abbreviations and Acronyms c-index ⫽ concordance index DSNB ⫽ dynamic sentinel LN biopsy EAU ⫽ European Association of Urology FNAB ⫽ fine needle aspiration biopsy LN ⫽ lymph node LNM ⫽ LN metastasis pt ⫽ threshold probability Submitted for publication December 18, 2009. * Correspondence: Department of Urology, Fudan University Shanghai Cancer Center, No. 270 Dong’an Rd., Shanghai, 200032, People’s Republic of China (telephone: 86-21-64175590; FAX: 86-2164434556; e-mail: [email protected]).
Key Words: penis; carcinoma, squamous cell; lymph nodes; neoplasm metastasis; nomograms OPTIMAL management for penile cancer in men with clinically negative LNs is still under debate. Today there are 2 main procedures in this setting, that is prophylactic lymphadenectomy and DSNB. Lymphadenectomy provides accurate staging information and can cure minor LN disease.1 DSNB has a low complication rate with acceptable false-negative results
at dedicated centers.2 However, the 2 interventions are invasive and have more than an 80% over treatment rate in the targeted population.1,2 Thus, predefined criteria for patients at risk are suboptimal and need further improvement. Since Ficarra et al first introduced a nomogram to predict inguinal LNM in patients with penile cancer,3 sev-
0022-5347/10/1842-0539/0 THE JOURNAL OF UROLOGY® © 2010 by AMERICAN UROLOGICAL ASSOCIATION EDUCATION
Vol. 184, 539-545, August 2010 Printed in U.S.A. DOI:10.1016/j.juro.2010.03.145
AND
RESEARCH, INC.
www.jurology.com
539
540
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
eral studies have been done to develop useful prediction models.4,5 If announced, these studies often included patients with clinically palpable LNs and showed that clinical palpability has a significant impact on LNM risk.3,5 According to our experience these cases may probably be classified as LN positive disease by FNAB. Using the current DSNB technique preoperative ultrasound and intraoperative palpation are also done to screen for LNM.6 Thus, it is doubtful whether a nomogram should be used instead of low cost detection techniques in LN palpable cases. In addition to the inappropriate patient population, another important problem is whether statistical performance translates into clinical usefulness. Penile cancer is curable when surgery is done at an early stage but the outcome after LN recurrence is often dismal.7 Thus, the probability threshold to stratify high and low risk cases may be 5% to 15%. In this situation a prediction model with good calibration in patients at more than 50% risk may have limited clinical significance. As introduced by Vickers and Elkin,8 decision curve analysis can identify the range of threshold probabilities in which a model is of value, the magnitude of benefit and which of several models is optimal. To overcome these limitations we developed a nomogram based on clinicopathological features in 110 patients with penile cancer and clinically negative LNs. We evaluated the clinical usefulness of the nomogram by decision curve analysis at different probability thresholds.
tected by immunohistochemical staining.11 The cutoff to define strong p53 expression was 20%, according to previous reports.11,12 The EAU risk classification of penile cancer defines risk as low—Tis, TaG1–2 or T1G1, intermediate—T1G2 and high—T2 or greater, or G3.13 Standard inguinal LN dissection was done in these patients. The borders included 2 cm above the inguinal ligament superiorly, the apex of the femoral triangle inferiorly, the adductor longus medially and the sartorius muscles laterally. All superficial and deep inguinal LNs in the field were removed. We used a standard method of processing LN specimens. Multivariate logistic regression analysis was used to test the association of clinicopathological features with inguinal LN status. This model was then used to predict the individual probability of positive inguinal LNs. Restricted cubic splines modeling was done using continuous variables when it could accommodate potential nonlinear effects and optimize Akaike’s information criterion. For model validation we assessed discrimination and calibration. Discrimination was measured using the c-index, which is similar to the ROC curve AUC.14 Bootstrap corrected c-indexes were used to better gauge expected future predictive accuracy. Calibration was assessed by visual inspection of the plots of predicted probability of metastasis vs actual outcome. We further used decision curve analysis to explore the clinical effects of our models.8 This method estimates a net benefit for prediction models by summing the benefits (positives) and subtracting the harms (false positives).
Table 1. Clinicopathological characteristics in 110 patients penile cancer Variables
MATERIALS AND METHODS The study group included 110 men with penile squamous cell carcinoma and clinically negative LNs from January 1990 to December 2008. Of the patients 13 underwent surgery elsewhere and were included in analysis after a review of pathological slides and medical records. The remainder were treated at our cancer center. Since we routinely suggest prophylactic inguinal lymphadenectomy in patients with penile cancer (except Ta/Tis disease), pathological staging information on inguinal LNs was readily available. Physical examination of inguinal LNs was done 1 to 4 weeks after primary disease resection. The definition of clinically negative LNs was 1) impalpable inguinal LNs or 2) palpable inguinal LNs but negative FNAB results. Patient age and time from presentation to primary disease treatment were retrieved from medical records. Pathological review was done using primary tumor slides. Tumor stage was assigned using the 2002 American Joint Committee on Cancer TNM system.9 T2 stage was further divided into 2 subgroups based on invasion depth, including T2a— corpus spongiosum infiltration and T2b— corpora cavernosa involvement. Histological grade was classified by the Broders system.10 Lymphovascular invasion was defined as tumor emboli within endothelium lined spaces bounded by a thin wall. p53 expression was de-
Presentation–primary disease treatment (mos): Less than 3 3–6 months Greater than 6 T stage: T1 T2a T2b T3 Grade: G1 G2 G3 Lymphovascular invasion: Absent Present p53 Expression: Weak Strong EAU risk classification: Low Intermediate High N stage: N0 N1 N2 N3
No. Pts (%) 64 (58.2) 42 (38.2) 4 (3.6) 60 (54.5) 33 (30) 9 (8.2) 8 (7.3) 60 (54.5) 42 (38.2) 8 (7.3) 91 (82.7) 19 (17.3) 78 (70.9) 32 (29.1) 34 (30.9) 20 (18.2) 56 (50.9) 84 (76.4) 17 (15.5) 8 (7.3) 1 (0.9)
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
Table 2. Univariate and multivariate analysis of clinicopathological factors to predict inguinal LNM in 110 patients Univariate Variables
% LNM
T stage: T1 T2a T2b T3 Grade: G1 G2 G3 Lymphovascular invasion: Absent Present p53 Expression: Weak Strong
p Value
Multivariate* OR (95% CI)
0.0022 11.7 27.3 77.8 37.5
— 2.61 (0.68–10.1) 7.32 (0.66–81.52) 3.78 (0.44–32.66) 0.0025
10 40.5 37.5
— 2.77 (0.72–10.72) 6.89 (0.77–61.88) ⬍0.0001
13.2 73.7
— 6.75 (1.28–35.73) 0.032
17.9 37.5
— 3.22 (0.96–10.86)
p Value 0.30 — 0.17 0.10 0.22 0.16 — 0.14 0.09 0.024 — 0.058 —
* Based on comparison to first value in each set of categorized variables.
Since the value of a positive finding, such as identifying metastatic disease early, may be different from the disadvantages of a false-positive finding, such as unnecessary lymphadenectomy, the net benefit calculation weights true and false positives differently. Weighting is derived from the threshold probability of a disease at which a patient would elect intervention. A false-positive finding is weighted as pt/(1 ⫺ pt) compared with a positive finding. Because this threshold probability can vary among patients, net benefit is calculated across a range of probabilities. We chose 5% to 25% as our range. We would be surprised if any man would choose invasive intervention when told that the probability of LNM was less than 5% and few clinicians would do 20 or more invasive proce-
541
dures to find metastatic disease. Similarly it would be rare for a man to demand at least a 1/4 chance of metastasis before accepting treatment. Bootstrap resampling was used to correct model overfit on decision curve analysis. Interpretation of a decision curve is simple. The model with the highest net benefit at a particular threshold probability should be chosen and we estimated the net benefit at each threshold probability. All analysis was done with R 2.10.0 with the rms library and other functions.8,14 –16
RESULTS Table 1 lists clinicopathological characteristics in 110 men with a mean age of 52.9 years (median 54, range 20 to 75) with penile cancer. A median of 23 inguinal LNs (range 11 to 37) was removed per case with a mean of 1.6 positive inguinal LNs (range 1 to 5) per LN positive case. Four patients had bilateral inguinal LNM. On univariate analysis patient age and time from presentation to primary disease treatment had weak associations with LN status (p ⬎0.5). They were excluded from the final model, which consisted of T stage, grade, lymphovascular invasion and p53 expression. T2b disease was more often associated with LNM than T2a and T3 disease. Multivariate analysis revealed only lymphovascular invasion as a statistically significant predictor of LNM (table 2). The prediction model is presented as a nomogram that graphically shows the multivariate effects of prognostic factors (fig. 1). We further evaluated nomogram discrimination and calibration. The bootstrap corrected c-index of the model was 0.79. Figure 2 shows good calibration between predicted risk and
Figure 1. Nomogram predicting LNM risk in patients with clinically node negative penile cancer
542
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
observed incidence. In this patient population the EAU risk classification achieved a c-index of 0.71, which was inferior to that of the nomogram (p ⬍0.001). We then investigated the clinical usefulness of the nomogram and used EAU risk classification as a comparison. Figure 3 shows the distribution of nomogram predictions by EAU risk classifications. The risk of LNM in the EAU high risk category varied highly and some cases were overrated. Thus, better risk prediction is strongly needed to avoid over treatment in these patients. Figure 4, A shows the results of decision curve analysis. At most threshold probabilities the nomogram had a higher net benefit over the EAU risk classification. According to the predefined range of pt we constructed a plot to determine the net decrease in interventions by the nomogram and the EAU risk classification (fig. 4, B). Table 3 shows the net benefit and decrease at pt from 5% to 25% at a 5% interval. In our clinical practice we choose 10% as pt to treat clinically LN negative cases of penile cancer. In this circumstance our nomogram found 1 positive result per 100 cases without an increase in the number of false-positive results. Furthermore, it also decreased 13 interventions per 100 cases without missing more positive results.
DISCUSSION Early detection of inguinal LNM is of paramount importance for penile cancer management. Lont et al compared outcomes in clinically LN negative penile cancer cases managed by surveillance or further diagnosed by DSNB with subsequent LNM resection.7 Although the groups were similar in clinicopathological features, the disease specific survival rate was significantly better in the early treatment group. Of 20 patients with LNM during surveillance 13 died of penile cancer. Patients with recurrence
Figure 2. Nomogram calibration plot. C, c-index.
Figure 3. Nomogram predictions by EAU risk classification
were more likely to have bilateral LNM, extracapsular growth and an increased number of positive LNs, which are adverse predictors of survival. Thus, omitting surgery in patients with occult LN disease may probably eliminate the opportunity to cure the disease. However, complications of inguinal lymphadenectomy are hard to accept, especially since almost 80% of patients with impalpable LNs are over treated.1 DSNB in a dedicated group can achieve a 7% false-negative rate with a 4.7% complication rate.2 However, biopsy results were negative in 86% of explored groins when including all cases staged as T1G2 or greater. Since the estimated cost per case is up to €11,000, the benefit of avoiding unnecessary lymphadenectomy is not as obvious.17 Thus, better patient selection is needed to improve the efficacy of further invasive interventions.
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
543
Figure 4. Nomogram and EAU risk classification. Vertical line indicates 10% probability threshold cutoff. FUSCC, Fudan University Shanghai Cancer Center. A, benefit. B, decrease.
Several groups investigated potential LNM prognostic factors in penile cancer cases.18 Tumor stage and grade are commonly used predictors in clinical practice.19 EAU guidelines recommended a risk classification to select patients for early lymphadenectomy.13 Hegarty et al reported 100 cases of penile cancer prospectively managed according to the guideline.1 None of 17 patients at low risk had LN disease during followup. However, predictive accuracy was not promising in those at intermediate and high risk. Only 18% of men with unfavorable primary disease and impalpable nodes had positive LN disease at prophylactic lymphadenectomy. Our results were similar (fig. 3). Predicted probability varied in men at high risk, of whom almost half were at less than 30% risk for LNM.
To increase prognostic accuracy tumor stage and grade modification or the introduction other pathological variables has been investigated. In the current T staging system the same group (T2) includes tumor involvement of the corpus spongiosum or corpora cavernosa, which is considered to have a better outcome than urethral invasion (T3). In our study men with corpora cavernosa invasion were at higher risk for LNM than those with corpus spongiosum or urethral infiltration. These observations were confirmed in another large cohort.20 Among other prognostic factors lymphovascular invasion and p53 expression have gained the most attention. In a report of 175 patients lymphovascular invasion was identified as an independent predictor of LNM.21 Prognostic significance was retained in 115 patients with
544
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
Table 3. Nomogram net benefit and decrease vs EAU risk classification % Probability Threshold
Net Benefit/100 Cases
Net Decrease/100 Cases
5 10 15 20 25
0 1 2 3 3
5 13 14 11 8
clinically negative LNs (OR 5.381). Lopes et al noted that p53 expression was associated with LNM on multivariate analysis including lymphovascular invasion.12 Their results were confirmed by another report in which p53 expression had independent prognostic value after including other molecular variables, such as Ki-67, epithelial cadherin and matrix metalloproteinase 9.11 We observed that lymphovascular invasion was the only independent LNM prognostic factor in clinically LN negative cases of penile cancer. However, p53 expression failed to achieve statistical significance (p ⫽ 0.058). To facilitate the clinical use of prediction models several groups have developed penile cancer nomograms. In patients at low risk for LNM a nomogram could be added as a tool in the decision making process to provide a numerical estimate to help the clinician and patient weigh the pros and cons of this decision.22,23 In the nomogram of Ficarra et al lymphovascular invasion and clinically palpable inguinal LNs were 2 powerful predictors of LNM (bootstrap corrected c-index 0.876).3 In the same patient population Kattan et al constructed another useful nomogram to predict 5-year cancer specific survival in patients with penile cancer.24 Velazquez et al developed a nomogram to predict LNM in tumors invading 5 to 10 mm.4 Perineural invasion and grade were significant prognostic factors in this subgroup. However, they did not report the statistical performance of the nomogram. Recently Bhagat et al analyzed 106 patients with penile cancer to construct a nomogram.5 Grade, lymphovascular inva-
sion and clinical palpability were the strongest predictors in that model. The nomogram c-index was 0.74. Our nomogram has 4 readily available pathological variables and achieved a c-index of 0.79, superior to that of the EAU risk classification in this cohort. The populations in the 4 studies were different since we only included patients with clinically negative LNs. We performed FNAB in patients with palpable LNs because 1) many patients had palpable but inflammable LNs that could not be defined as exactly nonpalpable and 2) FNAB is an easy, minimally invasive technique to detect LNM in these men at high risk. Our nomogram has good statistical performance but its clinical importance may be limited. For penile cancer with clinically negative LNs common management is surveillance or surgery (lymphadenectomy or DSNB). Thus, various predicted risks may be eventually transformed into 2 groups. On decision curve analysis we evaluated the benefit of the nomogram using different cutoffs. We used the EAU risk classification for comparison. When a 10% cutoff was applied, our nomogram had a net benefit of 1/100 cases and a net decrease of 13/100. Our nomogram correctly classified a proportion of patients at high risk. Our study has limitations. Sample size was relatively small for statistical analysis and data were collected retrospectively. Given the rarity of penile cancer, our study is still a large cohort with inguinal LN pathological staging. Furthermore, the predictive accuracy of the nomogram should be tested in an external cohort, which represents the gold standard to confirm the validity of clinical prediction models.
CONCLUSIONS We generated a nomogram in men with clinically node negative penile cancer based on readily available pathological factors. Decision curve analysis evidenced the clinical usefulness of the nomogram. Using a 10% probability threshold our nomogram was helpful for decreasing unnecessary intervention.
REFERENCES 1. Hegarty PK, Kayes O, Freeman A et al: A prospective study of 100 cases of penile cancer managed according to European Association of Urology guidelines. BJU Int 2006; 98: 526. 2. Leijte JA, Hughes B, Graafland NM et al: Twocenter evaluation of dynamic sentinel node biopsy for squamous cell carcinoma of the penis. J Clin Oncol 2009; 27: 3325. 3. Ficarra V, Zattoni F, Artibani W et al: Nomogram predictive of pathological inguinal lymph node involvement in patients with squamous
cell carcinoma of the penis. J Urol 2006; 175: 1700. 4. Velazquez EF, Ayala G, Liu H et al: Histologic grade and perineural invasion are more important than tumor thickness as predictor of nodal metastasis in penile squamous cell carcinoma invading 5 to 10 mm. Am J Surg Pathol 2008; 32: 974. 5. Bhagat SK, Gopalakrishnan G, Kekre NS et al: Factors predicting inguinal node metastasis in
squamous cell cancer of penis. World J Urol 2009. 6. Leijte JA, Kroon BK, Valdes Olmos RA et al: Reliability and safety of current dynamic sentinel node biopsy for penile carcinoma. Eur Urol 2007; 52: 170. 7. Lont AP, Horenblas S, Tanis PJ et al: Management of clinically node negative penile carcinoma: improved survival after the introduction of dynamic sentinel node biopsy. J Urol 2003; 170: 783.
NOMOGRAM TO PREDICT INGUINAL LYMPH NODE METASTASIS IN PENILE CANCER
8. Vickers AJ and Elkin EB: Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making 2006; 26: 565. 9. Penis. In: AJCC Cancer Staging Manual, 6th ed. American Joint Committee on Cancer. New York: Springer 2002; pp 303–308. 10. Broders AC: Squamous-cell epithelioma of the skin. Ann Surg 1921; 73: 141. 11. Zhu Y, Zhou XY, Yao XD et al: The prognostic significance of p53, Ki-67, epithelial cadherin and matrix metalloproteinase-9 in penile squamous cell carcinoma treated with surgery BJU Int 2007; 100: 204. 12. Lopes A, Bezerra AL, Pinto CA et al: p53 as a new prognostic factor for lymph node metastasis in penile carcinoma: analysis of 82 patients treated with amputation and bilateral lymphadenectomy. J Urol 2002; 168: 81. 13. Solsona E, Algaba F, Horenblas S et al: EAU guidelines on penile cancer. Eur Urol 2004; 46: 1.
14. Harrell FEJ: The rms Package for R: Regression Modeling Strategies, R package version 2009; 2.1-0. Available at http://biostat.mc.vanderbilt. edu/wiki/Main/Rrms. Accessed November 12, 2009. 15. Steyerberg EW: Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. New York: Springer 2008. 16. R: A Language and Environment for Statistical Computing. R Development Core Team. Vienna, Austria: R Foundation for Statistical Computing 2009. 17. Hegarty PK, Rees RW, Borley NC et al: Contemporary management of penile cancer. BJU Int 2008; 102: 928. 18. Ficarra V, Novara G, Boscolo-Berto R et al: How accurate are present risk group assignment tools in penile cancer? World J Urol 2009; 27: 155. 19. Pettaway CA, Lynch DF and Davis JW. Tumors of the penis. In: Campbell-Walsh Urology, 9th ed. Edited by AJ Wein, LR Kavoussi, AC Novick et al. Philadelphia: Saunders Elsevier 2007; chapt 31.
545
20. Leijte JA, Gallee M, Antonini N et al: Evaluation of current TNM classification of penile carcinoma. J Urol 2008; 180: 933. 21. Ficarra V, Zattoni F, Cunico SC et al: Lymphatic and vascular embolizations are independent predictive variables of inguinal lymph node involvement in patients with squamous cell carcinoma of the penis: Gruppo Uro-Oncologico del Nord Est (Northeast Uro-Oncological Group) Penile Cancer data base data. Cancer 2005; 103: 2507. 22. Ross PL, Gerigk C, Gonen M et al: Comparisons of nomograms and urologists’ predictions in prostate cancer. Semin Urol Oncol 2002; 20: 82. 23. Specht MC, Kattan MW, Gonen M et al: Predicting nonsentinel node status after positive sentinel lymph biopsy for breast cancer: clinicians versus nomogram. Ann Surg Oncol 2005; 12: 654. 24. 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: 2103.