Conditional survival predictions after surgery for patients with penile carcinoma


  • Rodolphe Thuret MD,

    1. Cancer Prognostics and Health Outcomes Unit, University of Montreal, Montreal, Canada
    2. Department of Urology, Montpellier University, Montpellier, France
    Search for more papers by this author
    • The first 2 authors contributed equally to this work.

  • Maxine Sun BSc,

    1. Cancer Prognostics and Health Outcomes Unit, University of Montreal, Montreal, Canada
    Search for more papers by this author
    • The first 2 authors contributed equally to this work.

  • Firas Abdollah MD,

    1. Cancer Prognostics and Health Outcomes Unit, University of Montreal, Montreal, Canada
    2. Department of Urology, Vita-Salute San Raffaele University, Milan, Italy
    Search for more papers by this author
  • Jan Schmitges MD,

    1. Cancer Prognostics and Health Outcomes Unit, University of Montreal, Montreal, Canada
    2. Martini-Clinic, Prostate Cancer Center Hamburg-Eppendorf, Hamburg, Germany
    Search for more papers by this author
  • Shahrokh F. Shariat MD,

    1. Cancer Prognostics and Health Outcomes Unit, University of Montreal, Montreal, Canada
    2. Department of Urology, Weill Cornell Medical Center, New York, New York
    Search for more papers by this author
  • François Iborra MD,

    1. Department of Urology, Montpellier University, Montpellier, France
    Search for more papers by this author
  • Jacques Guiter MD,

    1. Department of Urology, Montpellier University, Montpellier, France
    Search for more papers by this author
  • Jean-Jacques Patard MD,

    1. Department of Urology, Rennes University, Rennes, France
    Search for more papers by this author
  • Paul Perrotte MD,

    1. Department of Urology, University of Montreal, Montreal, Canada
    Search for more papers by this author
  • Pierre I. Karakiewicz MD

    Corresponding author
    1. Cancer Prognostics and Health Outcomes Unit, University of Montreal, Montreal, Canada
    2. Department of Urology, University of Montreal, Montreal, Canada
    • Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Center (CHUM), 1058, rue St-Denis, Montréal, Québec, Canada, H2X 3J4
    Search for more papers by this author
    • Fax: (514) 227-5103



Conditional survival (CS) implies that, on average, long-term cancer survivors have a better prognosis than newly diagnosed individuals. The objective of the current study was to devise an accurate predictive tool that accounts for CS in men diagnosed with penile cancer.


Overall, 1245 patients treated with primary tumor excision (PTE) for pT1-3M0 squamous cell carcinoma of the penis (SCCP) between 1998 and 2006 were identified. Cox regression models were fitted for prediction of cancer-specific mortality (CSM). Nomogram development for prediction of CSM using CS methodology at 2 and 5 years was performed on 670 patients. External validation and calibration of the conditional nomogram was performed in 575 patients.


The 5-year CSM-free survival of patients at surgery was 84.3% and increased to 95.0% and 97.8% after 2 and 5 years of disease-free survival (DFS), respectively. The predicted probabilities varied by as much as 49% (57% vs 85%) when, for example, predictions of CSM-free survival at 5 years were made after PTE versus after 2 years of DFS. Within the external validation cohort, the accuracy of the conditional nomogram was 75.3% and 78.1% at 2 and 5 years after PTE.


The authors developed and externally validated the first conditional nomogram for predicting SCCP CSM-free survival that allows consideration of the length of survivorship. Cancer 2011;. © 2011 American Cancer Society.

Survival estimates are customarily reported as the probability of being free from cancer-specific mortality (CSM) from the time of surgery. In general, prognosis improves with disease-free survival (DFS). Patients with longer DFS have better prognoses than their counterparts who just received therapy. This favorable effect of survivorship, otherwise known as conditional survival (CS), has been previously reported in prostate, kidney, colon, breast, ovarian, head and neck squamous cell carcinoma, gastric, and rectal cancers.1-8 CS rates are important to patients. CS estimates may allow patients to quantify the improvements in prognosis, as time goes by. To clinicians, CS estimates are equally important as they may allow adjusting for the frequency and type of follow-up, as time goes by and as the risk of disease recurrence decreases.

Until recently, the prognosis of patients with squamous cell carcinoma of the penis (SCCP) could not be accurately predicted. Only Kaplan-Meier–derived survival estimates were available, and these could not be adjusted to the characteristics of individual patients. Kattan et al were the first to develop a model for individualized CSM predictions in Italian patients.9 Subsequently, Zini et al derived a highly generalizable model using the Surveillance, Epidemiology, and End Results (SEER) population of individuals with SCCP.10 Despite the qualities of these models, neither was externally validated. Even more importantly, neither can provide CS estimates. CS estimates may drastically differ from baseline CSM predictions, especially after several years of disease-free follow-up.

In the current study, we devised a model for prediction of CSM. Moreover, we added a feature capable of providing CS predictions. Finally, we performed an external validation of this model within the most contemporary SEER cohort.


Study population

Patients were identified by using the SEER database of the National Cancer Institute (NCI) program. This cohort covers approximately 26% of the US population and is considered representative of the United States regarding demographic composition, as well as of cancer incidence and mortality.4 The reliability of the SEER database is high.5, 11 The 17 SEER Registries include Alaska, Metropolitan Atlanta, Greater California, Los Angeles, San Francisco-Oakland, San Jose-Monterey, as well as Connecticut, New Jersey, Detroit (Metropolitan), Iowa, Kentucky, Utah, Louisiana, New Mexico, Rural Georgia, Seattle (Puget Sound, and Hawaii.

Within the 17 SEER tumor registries, we identified all men treated for primary SCCP between 1988 and 2006 and who underwent PTE (excisional biopsy, partial, or total penectomy) according to 2 diagnostic codes: the 10th revision of the International Classification of Disease for Oncology second edition (ICD-O-2; C60.0-60.9) and the ICD-O-3 codes for histological subtype (squamous cell carcinoma type; ICD-O-3: 8070-8076). Only patients with squamous cell histology were included. Tumor grade was stratified according to the SEER database among grades 1, 2, and 3. We relied on the 2002 tumor, node, metastasis (TNM) classification system to tabulate disease stage.12 The 2002 version of the TNM staging manual (American Joint Committee on Cancer [AJCC] and International Union Against Cancer [UICC]) was used instead of the 2009 version because of lack of information on converting the SEER data to the 2009 version.13 For patients treated with inguinal lymph node dissection (ILND), pathological N substages were used versus clinical N substages for patients who did not undergo an ILND. All stage assignments were obtained from the SEER database. Exclusions consisted of T4 stage and presence of distant metastases. Patients in SEER registries who were treated in Alaska (n = 5) and Rural Georgia (n = 2) were also excluded because of exceedingly low numbers of observations in these 2 registries.

PTE was considered as the start time of observation. For the purpose of this analysis, deaths from penile cancer were coded as cancer-specific events according to the SEER-specific cause of death recode (28030). All other deaths were considered as other-cause mortality (OCM).

Statistical Analyses

The SEER registries were used to randomly divide the entire cohort into 2 groups, namely the development and the external validation cohorts. The independent sample t and the chi-square tests were used to assess differences of respectively means and proportions between these 2 groups.

Within the development cohort, univariate and multivariate Cox regression models were fitted for prediction of CSM. Main predictors were T category (T1 vs T2 vs T3), N category (cN0 vs cN+ vs pN0 vs pN1 vs pN2 vs pN3) and tumor grade (1 vs 2 vs 3). Additional variables included age, race (white vs black vs other), and type of surgery (excisional biopsy vs partial penectomy vs total penectomy). Subsequently, a backward selection process, which used the Akaike Information Criterion as the stopping rule, was applied. This resulted in the identification of the most informative variables that were included in the final model. Thereafter, Cox regression coefficients were used to devise a nomogram. Within the external validation cohort, the nomogram was applied to each individual patient, and the nomogram predictions were compared with the actuarial survival. As a means of quantifying the nomogram's ability to discriminate among patients, we relied on the receiver-operating characteristics (ROC) calculation.14 In the context of time-to-event analysis, the ROC is slightly modified with the Harrell concordance index, which substitutes the ROC-derived area under the curve (AUC).15 However, its interpretation remains similar. The AUC under the ROC curve is the probability that given 2 randomly drawn patients, the patient who dies first has a higher probability of mortality. However, when both patients experience the event of interest, or when the 1 who did not experience the event has a shorter follow-up, the probability does not apply to the 2 patients.

This nomogram was then complemented with a tool that predicts CS probabilities according to the duration of disease-free survival (DFS). The conditional survival estimates after PTE for SCCP were calculated by using the approach described by Skuladottir and Olsen,17 which was previously applied by other investigators.2, 18 This method is based on the multiplicative law of probability. When the probability of an event A and event B (P[A and B]), and the probability of event A (P[A]) are known, one is able to calculate the conditional probability of event B occurring given that event A has occurred (P[B|A] = P[A and B]/P[A]). In the current context, if probabilities of survival at 2 and 5 years that are computed immediately after PTE are respectively 90% and 70%, the CSM-free survival rate at 5 years after a 2-year disease-free interval is 77.8% (0.7/0.9).

The third part of the analyses relied on the validation cohort. Here the nomogram-predicted CSM rates were compared with actuarial CSM rates at 2 and 5 years. The accuracy of time-specific predictions was quantified using the val.surv method that relies on the Harrell concordance index. The latter substitutes ROC-derived AUC in time-to-event analyses.15, 16 A value of 1.0 represents ideal prediction versus 0.5, which is synonymous to a flip of a coin. The correlation between predicted and observed was graphically displayed by using the val.surv method.

All statistical analyses were performed by using the Statistical Package for Social Science, version 15.0 (SPSS, Chicago, Illinois) statistical software and R statistical package (R foundation for Statistical Computing, Vienna, Austria). All tests were 2-sided, with a significance level set at <.05.


Between 1988 and 2006, we identified 1245 patients with SCCP within 15 SEER registries (Table 1). Of those, 670 (53.8%) patients from 7 registries were grouped within the development cohort and the remaining 575 (46.2%) patients were used for external validation. In the development and validation cohorts, respectively, the average age was 65.6 years (median, 67 years) and 66.9 years (median, 68 years). More black patients were identified within the development cohort (10.6%) than in the validation cohort (7.5%; P = .002). The majority of patients were treated with partial penectomy in both cohorts (65.5% and 63.8%, respectively). Most patients were category T1 (59.9% and 61.9%, respectively) and cN0 (72.5% and 77.4%, respectively). Tumor grade 2 (46.0% and 45.7%, respectively) predominated in both cohorts.

Table 1. Descriptive Data for 670 Patients Used for Nomogram Development Cohort and 575 Patients Used for External Validation Cohort Treated for Squamous Cell Carcinoma of the Penis
 Development CohortValidation Cohort 
VariableNo. of Patients%No. of Patients%χ2 Test P
  1. SCCP indicates squamous cell carcinoma of the penis.

SEER registries     
 Atlanta (Metropolitan)375.5   
 Greater California21932.7   
 Los Angeles18026.9   
 San Francisco-Oakland6710.0   
 Connecticut  10818.8 
 Detroit (Metropolitan)  9516.5 
 Iowa  8314.4 
 Kentucky  8615.0 
 New Jersey  10718.6 
 New Mexico  579.9 
 San Jose-Monterey  254.3 
 Utah  142.4 
Age, y    .1
 Mean (median)65.6 (67.0) 66.9 (68.0)  
 Range22-102 25-96  
Race    .002
Surgery    .1
 Excisional biopsy14521.611119.3 
 Partial penectomy43965.536763.8 
 Total penectomy8612.89716.9 
T category    .6
N category    .4
Tumor grade    .4
SCCP-specific mortality7811.67613.2.6
Overall mortality26840.024843.1.7
Follow-up, mo    .1
 Mean (median)40.7 (27.0) 44.3 (30.0)  
 Range0-221 0-210  

Univariate analyses for prediction of CSM revealed that T category, N category, tumor grade, and surgery type achieved statistical significance (all P < .01; Table 2). In multivariate analyses, T3 patients showed a 2.6-fold higher chance of CSM than T1 patients (P = .002). Similarly, cN+ (P = .01) and pN3 (P = .02) patients had respectively 3.1- and 2.8-fold higher chance of CSM relative to their cN0 counterparts. Patients with grade 2 (G2) disease had a 2.8-fold higher chance of CSM than G1 patients (P = .004). The effect of age, race, and surgery type failed to reach statistical significance.

Table 2. Univariate and Multivariate Cox Regression Analyses for Prediction of Cancer-Specific Mortality in the Nomogram Development Cohort (n=670)
 Univariate AnalysesMultivariate Analyses
  1. HR indicates hazard ratio; CI, confidence intervals.

T category         
 T1Ref.  Ref.  Ref.  
N category         
 cN0Ref.  Ref.  Ref.  
Tumor grade         
 G1Ref.  Ref.  Ref.  
Age      ---
 Continuously coded1.00.9-   
 WhiteRef.  Ref.  ---
Surgery type         
 Excisional biopsyRef.  Ref.  ---
 Partial penectomy1.30.7-   
 Total penectomy2.81.3-5.9.0071.70.7-3.9.2   

After backward variable selection, only T and N categories as well as tumor grade remained in the model. These variables were then included in a nomogram predicting CSM-free survival (Fig. 1). Within this nomogram, the N category represented the most powerful predictor of CSM. Specifically, presence of clinically positive nodes (cN+) exerted virtually the same effect on CSM as the presence of pathologically confirmed N2 or N3 category. Clinical category N0 conferred a higher probability of CSM than pN0 category. The effect of tumor grade was the least influential.

Figure 1.

Nomogram predicts cancer-specific mortality (CSM)-free rate. Nomogram instructions: locate patient value for pathological T category. Draw line straight up to point axis to determine how many points toward probability of freedom from penile CSM patient receives for value of pathological stage. Repeat process for each additional variable. Sum points for each predictor.

Figure 2 shows the effect of CS within the development cohort, according to Kaplan-Meier and life-table methodologies. The duration of DFS had an important effect on CSM-free rates. For example, the 5-year CSM-free survival of patients immediately after PTE was 84.3% (95% confidence interval [CI], 80.4-87.4%). This rate increased to 95.0% (95% CI, 91.6-97.1%) after 2 years of DFS and to 97.8% (95% CI, 93.3-99.3%) after 5 years of DFS.

Figure 2.

Kaplan-Meier plots show survival rates in development cohort according to survival after primary tumor excision.

The nomogram-predicted CS probabilities of CSM-free survival at 2 and 5 years after PTE are shown in Figure 3. For example, in a patient with T2 (17 points), G2 (13 points), and cN0 (45 points) SCCP, the nomogram-calculated probability of being CSM-free at 2 and 5 years is respectively 67% and 57% immediately after the surgery. In comparison, the probability of being CSM-free at 2 years after a disease-free interval (DFI) of 1 year since surgery is 85% (improvement of 18%; Fig. 3A). Similarly, the immediate probability of CSM-free survival at 5 years (Fig. 3B) after a 1-year and a 2-year DFI after PTE increased from 57% to 73% (+16%) and 85% (+28%), respectively.

Figure 3.

To calculate conditional prediction of freedom from penile cancer-specific survival (CSM) at (A) 2 years and (B) 5 years after primary tumor excision (PTE), locate the value corresponding to sum on total point axis obtained from Figure 1. Within panels, y-axis indicates survival time due to PTE, and x-axis indicates nomogram-predicted risk points. Draw a straight line up from that point on the total point axis. Then draw a horizontal line on the y-axis that corresponds to months of disease-free survival (DFS) between PTE and today. Use the intersection of those 2 lines to identify a slanted line that crosses it or passes right next to it. Follow that slanted line to the x-axis to read the probability of remaining free of penile CSM at prespecified number of years of DFS after PTE.

The application of the Harrell concordance index to the newly devised nomogram resulted in predictive accuracy of 75.3% and 78.1% at 2 and 5 years, respectively. The val.surv calibration plot illustrating the nomogram predicted and the actuarial probability of CSM-free survival showed minimal departure from ideal prediction (Fig. 4).

Figure 4.

Nomogram calibration plot demonstrates good agreement between predicted and observed mortality rates.


Penile cancer represents an uncommon malignancy in the Western world, with an incidence of 0.1 to 0.9 new cases per 100,000 males per year.19 To provide more individualized and more accurate predictions of CSM, 2 groups of investigators devised nomograms for predicting CSM-free probabilities.9, 10 Despite multiple merits of both models, neither is capable of providing adjusted CSM-free survival that is adjusted according to DFS. This limitation is important because after 1 or several years of DFS, the CSM predictions made at baseline no longer apply. This observation is particularly pertinent in malignancies with a relatively slow natural history. In the majority of cases, SCCP qualifies for this attribute. In consequence, CS should be considered after a period of DFS. If such consideration is not made, then an overly pessimistic prognosis will be given to the patient.20 Similarly, excessively frequent follow-up may be recommended. Both situations should be avoided. To avoid such scenarios, CS prediction should be generated, as described by Skuladottir and Olsen.17 As demonstrated in the original study of these 2 investigators, survival estimates based on DFI after treatment are better than predictions obtained immediately after surgery, when a DFI has been recorded. In the current article, we describe a model capable of dynamically predicting individualized CSM-free survival after adjustment for DFI.

On the basis of the importance of CS, we developed a novel model predicting CSM by using a large population-based North American cohort. The CS methodology was applied to the predictions of this model at 2 and 5 years after PTE for SCCP. A formal external validation was then performed.

A total of 1245 patients with SCCP were identified within the SEER database. Of those, 670 patients were randomly assigned to the development cohort, which was used to construct the model for prediction of CSM. After backward variable selection, only 3 variables qualified for inclusion in the final model: tumor category, node category, and tumor grade. In external validation, the combined accuracy of these variables in predicting CSM was 75.3% and 78.1% at 2 and 5 years, respectively, after PTE. These results are comparable to Zini et al's, who devised a nomogram relying on 2 variables with 73.8% accuracy.10 Compared with this nomogram, our model is more accurate and only requires 1 additional variable. Most importantly, our nomogram was externally validated, unlike that of Zini et al. Last but not least, our model is equally or more generalizable than that of Zini, based on a larger and more contemporary patient population. The nomogram of Kattan et al, like that of Zini et al, was not externally validated.9 In addition, weaknesses of that nomogram consist of it being specific to Italian men and requiring numerous variables,8 of which not all are routinely reported.

The main advantage of our nomogram consists of CSM predictions that are adjusted according to DFI. Indeed, the duration of DFS substantially influenced CSM-free rates. For example, the 5-year CSM-free survival of patients immediately after surgery was 84.3% and increased to 97.8% given a CS of 5 years. The consideration of DFI also substantially modified the predicted CSM-free survival. For example, for an individual with T2cN0G2 disease, the 2- and 5-year CSM-free estimates made immediately after PTE are 67.0% and 57.0%. Conversely, the 2-year CSM-free survival predicted estimates considering a 1-year DFI increases from 67.0% to 85.0% (+18%). Similarly, the 5-year CSM-free survival predicted estimates based on a 2-year DFI increases from 57.0% to 85.0% (+28%). These differences are neither trivial for patients nor are they unimportant to clinical decisions when the frequency and the type of follow-up are decided upon.

Our data indicate that mortality from SCCP is rare beyond 5 years of follow-up. This observation supports the recommendations of the European Association of Urology (EAU) guidelines.21 The latter suggests clinical follow-up for up to 5 years after PTE in patients treated for SCCP. Subsequently, only regular self-examination is recommended.

Our study is not devoid of limitations. First, the developed model may have benefitted from a higher accuracy than the one that was reported. Nonetheless, the current study represents the sole tool to date capable of providing CS probabilities in the context of patients treated for SCCP. In consequence, whereas the current nomogram is of only moderate accuracy, in the absence of other tools that account for CS in the same setting, its use may be considered valuable to clinical decisions. Second, lack of central pathology review might represent a weakness. Central pathology review could have contributed to higher accuracy of pathologically assessed variables.22 Third, the retrospective nature of the study represents a limitation that is shared with previous studies addressing SCCP outcomes.23-26 Fourth, other variables, such as perineural invasion, lymphovascular invasion, smoking history status, or surgical margin status may also have achieved independent predictor status.27 However, these variables were not available in the SEER database and could not be considered in our model. Fifth, the predicted probabilities do not account for adjuvant (radiotherapy) or salvage therapy (radiotherapy or chemotherapy). These treatments may have no or, at best, minimal impact on survival.28 However, their effects could not be directly tested in this study. In addition, the nomogram cannot be applied to patients treated with modalities other than PTE, such as surveillance, local tumor destruction with neodymium:yttrium-aluminum-garnet (Nd:YAG) laser therapy, or brachytherapy.29-31 The survival of these individuals may be different. Sixth, our model cannot be applied to patients with metastatic SCCP. We have excluded T4 and metastatic patients because of the exceedingly low numbers of such individuals. But >90% of patients with SCCP have T1-3M0 SCCP at presentation.26, 32 Finally, while the current model was externally validated, it may be primordial importance to externally validated on cohorts other than that of the SEER database.

In summary, an accurate nomogram was developed and externally validated. In addition, this represents the first tool capable of providing an individualized assessment of CSM-free survival with adjustment for DFI in patients treated with PTE for SCCP. These considerations are crucial for patients and clinicians who are making clinical decisions.


Pierre I. Karakiewicz is partly supported by the University of Montreal Health Center Urology Specialists, Fonds de la Recherche en Santé du Quebec, the University of Montreal Department of Surgery, and the University of Montreal Health Center (CHUM) Foundation. Rodolphe Thuret is partly supported by the Association Française d”Urologie (AFU).