The prognostic value of tumor size in penile cancer: A Surveillance,Epidemiology, and End Results database study

Background: The objective of this study was to investigate the prognostic value of tumor size on cancer-specic mortality (CSM) and lymph node metastasis for patients with penile squamous cell carcinoma (PSCC). Method: The patients diagnosed with PSCC between 2004 and 2015 were selected from the Surveillance, Epidemiology, and End Results (SEER) database. Restricted cubic spline functions were calculated to characterize the association between tumor size and the risk of CSM. The competing-risks model was used to evaluate the impact of tumor size on the cumulative incidence of CSM. The logistic regression analysis was performed to examine the association between tumor size and lymph node metastasis. Results: Totally, 1365 PSCC patients were analyzed, with 52.3% having tumors ≤ 30 mm, and 47.7% >30 mm. The restricted cubic splines showed that the risks of CSM increased as tumors enlarged. Following adjustment of competing events, the PSCC patients with tumors >30 mm were more likely to succumb to CSM in comparison with those with tumors ≤ 30 mm (hazard ratio [HR]=1.57, 95% condence interval [CI]: 1.23-2.01, P<0.001). In subgroup analyses, tumor size >30 mm was signicantly associated with an increased risk of CSM relative to tumor size ≤ 30 mm among patients with T1 (HR=1.56, 95%CI: 1.03-2.37, P=0.036) and T3 (HR=2.51, 95%CI: 1.41-4.45, P=0.002) classications. On logistic regression analysis, tumors >30 mm were signicantly associated with lymph node metastasis (odds ratio [OR]=1.46, 95% CI: 1.03-2.07, P=0.034). Conclusion: Larger tumors (>30 mm) were signicantly associated with higher risks of CSM and increased likelihood of lymph node metastasis for PSCC patients, which could be integrated into the development of a staging system for penile cancer.


Introduction
Penile cancer is an uncommon urogenital malignancy, which accounts for 0.2% of new cancer cases and 0.2% of cancer deaths worldwide [1]. The annual incidence of penile cancer is about 1.0 per 100000 males in developed countries, including Europe and the US [2,3]. Whereas, in the developing countries of Africa, Asia, and South America, it reaches to 2.0-4.0 per 100000 males [4]. PSCC, the most frequent histological subtype, composes 95% of all penile neoplasms [5].
For males, the cumulative probability of suffering penile cancer is only 0.09% from birth to 74-year-old [1]. The American Cancer Society estimates that in the US, the new cases and deaths of penile cancer in 2020 are 2200 and 440, respectively [6]. Due to the rare incidence, the investigations regarding the prognosis of this cancer are relatively less in comparison with other types of cancers. The presence and extent of regional lymph node involvement is the most signi cant predictor of prognosis among patients with PSCC. Other clinical and pathological variables, including tumor grade [7], tumor thickness [8], histological subtype [9], and lymphovascular invasion [10], are also associated with regional lymph node metastasis and patient survival. Tumor size, an independent prognostic factor in several types of cancer, is frequently utilized to evaluate tumor stage in the American Joint Committee on Cancer (AJCC) staging system [11]. However, the most recent AJCC staging system does not include tumor size as an important criterion for the classi cation of penile cancer.
To date, limited studies have reported that tumor size is a prognostic predictor of overall survival [12], disease-free survival [13], and cancer recurrence [14] for PSCC. Whereas the association of tumor size with CSM is still unclear. In addition, the association between tumor size and lymph node metastasis has been controversial, and there is a paucity of study focused on this issue. Therefore, the current study was based on the hypothesis that tumor size was signi cantly associated with CSM as well as lymph node metastasis for PSCC. The population-based SEER database was queried for the eligible patients, and the impact of tumor size on the prognosis of PSCC was evaluated in detail.

Data source
The SEER database of the National Cancer Institute program (http://www.seer.cancer.gov), covering approximately 28% of the US population [15], was used to select patients in this study. The cancer-related data were gathered in the SEER database from 1973, which could be freely provided to registered researchers. We used SEER*Stat (version 8.3.6) to extract the data, and our user name to access the database was 11910-Nov2019. All the collected data were de-identi ed, and hence, our study was exempted from institutional review board approval.

Study Population
The SEER database was queried for all cases who were diagnosed with PSCC between 2004 and 2015. The cases were identi ed using the primary site codes C60.0-60.9 for penile cancer and morphology codes 8051/3, 8052/3, 8070/3-8076/3, 8083/3, and 8560/3 for squamous cell carcinoma. PSCC was the rst and only primary diagnosis and con rmed by histological examination. To guarantee the accuracy of tumor size, only patients with cancer-directed surgery performed and tumor size ≤ 100 mm were included. The exclusion criteria were composed of age < 18 years at diagnosis, the presence of distant metastasis, and autopsy or death certi cate cases. The patients with unknown information about follow-up data, tumor size, tumor grade, and tumor stage were also excluded. The ow diagram of patient selection was shown in Fig. 1.

Measurements Of Variables
The demographic and clinicopathological characteristics of the eligible patients were collected, including age at diagnosis, race ( [16]. Additionally, we used multivariable logistic regression models to examine the association between tumor size and lymph node status. All statistical analyses were performed using R software (version 3.5.3, http://www.r-project.org/). The R package "cmprsk" was used to build cumulative incidence function curves and perform competing risk analysis. The signi cance level was set at P < 0.05, two-sided.

Demographic and clinicopathological characteristics
Between 2004 and 2015, 1365 PSCC patients who met the inclusion criteria were identi ed from the SEER database. The demographic and clinicopathological characteristics of the entire cohort and the subgroups strati ed by tumor size were presented in Table 1. The distribution of tumor sizes was presented in Fig. 2. 52.3% of tumors in the study population were at most 30 mm, and 47.7% of tumors were larger than 30 mm. Figure 3 exhibited the boxplots of tumor sizes in subgroups based on T and N classi cation, and signi cant differences could be observed (all P < 0.001). The curves for the cumulative incidence functions of CSM strati ed by tumor size was displayed in Fig. 4. After controlling competitive risk events, a signi cant difference was observed for the cumulative incidence of CSM between two subgroups (P < 0.001).
The effects of tumor size as both a continuous and categorical variable on CSM were separately analyzed. As shown in Fig. 5, restricted cubic splines were used to exible model and visualize the association between tumor size and CSM. The plot depicted an apparent tendency that the risks of CSM increased as tumors enlarged. Multivariable competing-risks regression analyses of CSM were displayed in Table 2. When tumor size was incorporated into the competing-risks regression model as a categorical variable, the estimates indicated that the PSCC patients with tumors > 30 mm were more likely to succumb to CSM than those with tumors ≤ 30 mm (HR = 1.57, 95%CI: 1.23-2.01, P < 0.001). Race, Tumor grade, T classi cation, and N classi cation were independent predictors for CSM. However, no signi cant association with CSM was found with age at diagnosis, marital status, chemotherapy, and radiotherapy. Abbreviations: HR, hazard ratio; 95% CI, 95% con dence interval.

Sensitivity analyses for the association between tumor size and CSM
In the current study, subgroup analyses strati ed by T classi cation were performed to verify the association between tumor size and CSM. As shown in Fig. 4, there were statistical differences in cumulative incidence of CSM across the subgroups strati ed by T classi cation (all P < 0.05  Abbreviations: HR, hazard ratio; 95% CI, 95% con dence interval; CSM, cancer-speci c mortality.
All models were adjusted for age at diagnosis, race, marital status, tumor grade, N classi cation, chemotherapy, and radiotherapy.
Association of tumor size with lymph node status.
Tumor size as a categorical variable was analyzed by using multivariable logistic regression models. Compared with tumors ≤ 30 mm, the odds ratio (OR) for lymph node metastasis was 1.46 (95% CI: 1.03-2.07, P = 0.034) for tumors > 30 mm. Additionally, age at diagnosis, tumor grade, T classi cation, chemotherapy, and radiotherapy were all signi cantly associated with lymph node metastasis. However, no statistical signi cance was reached with respect to race and marital status. The results of multivariable logistic regression analysis were presented in Table 4.

Discussion
In the current study, we evaluated the association of tumor size with CSM as well as lymph node metastasis based on the SEER database. The PSCC patients with tumors > 30 mm were more likely to succumb to CSM than their counterparts with tumors ≤ 30 mm. Additionally, tumor size > 30 mm was signi cantly associated with an increased risk of lymph node metastasis.
Due to the rarity of PSCC, only scarce retrospective studies were published to investigate the prognostic value of tumor size. The existing limited literature revealed that tumor size was a prognostic predictor for overall survival, disease-free survival, and cancer recurrence [12][13][14]. Escande et al identi ed 201 patients with invasive penile carcinoma to examine the association between brachytherapy and long-term clinical outcomes. Patients with tumors greater than 4 cm had worse overall survival (HR = 2.2, 95% CI: 1.1-4.4, P = 0.028) and disease-free survival (HR = 2.4, 95% CI: 1.2-4.8, P = 0.01) than those with tumors at most 4 cm [12]. Mao and colleagues reported that tumors larger than 30 mm predicted worse overall survival relative to those less than 30 mm for node-positive penile cancer (HR = 1.71, 95% CI: 1.28-2.28, P < 0.001) [13]. However, to date, the association of tumor size and cancer-speci c survival has not yet been fully explored. In a recent study, Cox regression models were used to identify the risk factors of cancer-speci c survival for PSCC patients with node-positive disease. On multivariable analysis, tumor size ≥ 30 mm was signi cantly associated with worse cancer-speci c survival [13]. Of note, lymph node-negative patients were not involved in this study, and thus, we did not know the effect of tumor size on CSM for lymph node-negative patients. Additionally, the Cox regression model used was not suitable for analysis of competing events, which might overestimate the cumulative incidence of each event and lead to competitive risk bias [17]. The primary interest of our study was CSM. The other causes leading to death, such as cardiovascular disease, suicide, and accident, were regarded as competing events, which could hinder the occurrence of the primary interest. In this case, the competing-risks model was more appropriate to deal with multiple end events. Subgroup analyses were also carried out to verify the robustness of the ndings. The patients with tumors ≤ 30 mm had survival bene ts than those with tumors > 30 mm among T1 and T3 subgroups. However, the statistical signi cance did not reach in T2 subgroup, which might be attributed to the relatively small sample size that reduced statistical power to detect small effects.
As we knew, the involvement of regional lymph nodes was a valuable pathological factor for predicting the prognosis of PSCC patients [15,18]. Therefore, it was consequential to accurately recognize the risk factors associated with regional lymph node metastasis. Previous studies showed that tumor size was associated with an increased likelihood of regional lymph node metastasis for breast cancer [19], thyroid cancer [20], and lung cancer [21]. However, this topic did not reach a consensus on penile cancer. A study from Italy investigated the predictors of lymph node metastasis using 175 PSCC patients. The authors rstly used the Chi-square test to screen the clinical variables. Because of no statistical signi cance in univariable analysis, tumor size was not incorporated and analyzed in the multivariable analysis [22].
Another study based on the Ontario Cancer Registry was performed to evaluate the predictive variables of lymph node metastasis. The cohort was composed of 380 PSCC patients, of whom 63 patients had pathologically con rmed lymph node metastasis. In univariable analysis, tumor size (> 3 cm vs. ≤3 cm) was associated with an increased risk of lymph node metastasis (P = 0.040). However, it was not statistically signi cant at the conventional level (5%) in multivariable analysis [23]. Commonly, the reliable estimation of predictor effects needed at least 100 events. The requirement of sample size was at least 10 events per variable (EPV), and preferably 20 [24]. In the two studies, the number of events was only 71 and 63, and the values of EPV were respectively 15 and 9, less than 20. The relatively small sample size and low values of EPV might decrease the statistical power and the reliability of these results to some extent. In our study, 271 of 1365 patients were con rmed with lymph node-positive disease, and the value of EPV in multivariable analysis exceeded 30. Therefore, our study had su cient sample size and statistical power to obtain stable and reliable estimations of predictor effects. Our study showed tumors > 30 mm was signi cantly associated with an increased likelihood of lymph node metastasis compared to tumors ≤ 30 mm (OR = 1.46, 95%CI: 1.03-2.07, P = 0.034). Consistent with our ndings, Chalya and colleagues identi ed 236 penile cancer patients from a medical center in Tanzania, of whom 154 patients had lymph node metastasis at diagnosis. The multivariable logistic regression analysis showed that tumor size was an independent predictor for lymph node metastasis (≥ 20 mm vs. <20 mm, OR = 2.9, 95% CI: 1.1-6.4, P = 0.011) [14]. Kearns and collaborators reported that increasing tumor size was signi cantly associated with lymph node metastasis after adjusting T classi cation, tumor grade, and lymphovascular invasion (≥ 40 mm vs. <15 mm, OR = 2.9, 95% CI: 1.31-6.41, P = 0.009) as well [7].
The current study had its own advantages. The population-based SEER database was employed in this study, providing us a relatively large sample size, which was bene cial for the investigations on cancers with low incidence. In addition, this study systemically investigated the prognostic value of tumor size for PSCC by using multiple statistical methods, including restricted cubic splines, competing-risks models, and logistic regression models. These reliable results exhibited the potential prognostic value of tumor size in the development of a staging system for penile cancer. However, some limitations should be noted. First of all, the selection bias was inevitable due to the retrospective nature of this study. Second, although we had adjusted for many known patient factors, the estimation would be more accurate if more valuable covariables which were not recorded in the SEER database could be incorporated. Third, since the eligible patients were diagnosis in 2004-2015, the relatively short follow-up might in uence the estimation of cumulative incidence of CSM.

Conclusions
To our knowledge, the prognostic value of tumor size was rstly investigated for PSCC in this study.
Compared to tumors ≤ 30 mm, tumors > 30 mm were signi cantly associated with worse cancer-speci c survival and increased likelihood of lymph node metastasis. These ndings could help clinicians and investigators to perform patient counseling, design clinical trials, and improve cancer staging systems. Further investigations are needed to con rm the prognostic value of tumor size for penile cancer.

Declarations
Ethics approval and consent to participate Our study was exempted from institutional review board approval because of using the de-identi ed data in the SEER database.

Consent for publication
Not applicable.
Availability of data and materials.
The datasets generated and/or analyzed during the current study are available in the SEER repository (http://www.seer.cancer.gov).

Competing interests
The authors declare that they have no con ict of interest.

Funding
Not applicable.
Authors' contributions Wenwen Zheng and Zhiyu Zhang participated in study design. Xilei Xie, Weiwei Zhu, Kangqi Li, Qingna Wu, and Jiaojiao Chen were responsible for data collection and analysis. Wenwen Zheng was involved in drafting the manuscript. Congxiao Lu, Quan Zhao, Shengqiang Yu, and Chenyu Guo made appropriate revisions. All authors have read and approved the nal manuscript.