ABSTRACT
Objective
Incidental prostate cancer (PCa) patients are currently stratified to either cT1a vs. cT1b stage, depending on tumor volume (< 5% vs. > 5%) in the resected tissue. We tested for other models that could improve cancer‐specific survival (CSS) predictions.
Methods
Incidental (cT1a/cT1b) PCa patients were retrospectively identified within the Surveillance, Epidemiology, and End Results (SEER) database (2004–2015). Kaplan–Meier plots illustrated CSS at 5 years of follow‐up. Combination of variables resulted in three distinct stratification models based on cT1a vs. cT1b and Gleason score sum (GS). Multivariable Cox regression models that predicted CSS were used for area under the curve (AUC) quantification after 20‐fold cross validation.
Results
We identified a total of 5155 incidental PCa patients. CSS at 5 years was 98% for cT1a vs. 90% for cT1b (p < 0.0001). CSS at 5 years for combination of cT1a and GS 6 patients was 99% vs. 92% for patients with either cT1b or GS ≥ 7 (p < 0.0001). Finally, for patients with GS < 8, CSS at 5 years was 98% vs. 72% for patients with GS ≥ 8 (p < 0.0001). The multivariable adjusted AUC was 0.83 vs. 0.82 vs. 0.88, for each model, respectively.
Conclusion
Of all tested stratification models, the consideration of GS stratified between < 8 and ≥ 8 resulted in the best survival discrimination, as well as highest accuracy after cross‐validation. In consequence, the use of this model (GS < 8 vs. ≥ 8) appears better suited than the standard cT1a vs. cT1b substaging, when CSS represents the endpoint of interest.
Keywords: cT1a, cT1b, incidental, prostate cancer, staging
1. Introduction
Incidental prostate cancer refers to the unforeseen diagnosis of PCa after surgical treatment for benign prostate hyperplasia (BPH). According to the current American Joint Committee on Cancer (AJCC) TNM‐classification, this entity of PCa is currently sub staged into cT1a vs. cT1b, based on the percentage of PCa in the resected tissue (cT1a = < 5%, cT1b = > 5%) [1]. Since its introduction in 1992, this subclassification has remained unchanged despite numerous staging modifications, as well as advances in diagnosis and treatment of PCa [2, 3]. In this regard, it has been questioned whether the current tumor volume‐based stratification (cT1a vs. cT1b) is still appropriate or whether an alternative should be proposed for stratification and prognostication in the context of incidental PCa diagnosis [2, 4, 5]. Indeed, Mottet et al. recently discussed a modification of the standard stratification of incidental PCa between cT1a vs. cT1b, whereby ISUP grade is considered in addition to the established volume‐based stratification (cT1a vs. cT1b) [4]. Based on this proposal, cT1a patients should only include those individuals in whom the pathological diagnosis consists of ISUP grade 1. Conversely, patients with more than 5% PCa in the resected tissue or ISUP ≥ 2 are considered cT1b. Despite the novelty and logic of this novel proposed stratification of incidental PCa patients, it is unknown what extend of improvement this modification offers relative to the established cT1a vs. cT1b stratification. Moreover, it is unknown whether a different stratification, for example one on Gleason score sum (GS) alone may offer a similar advantage within a simpler framework. We addressed these unknowns, relying on the Surveillance, Epidemiology, and End Results (SEER) database. Specifically, we compared cancer‐specific survival (CSS) rates, as well as the cross‐validated accuracy for the above candidate predictor models of CSS.
2. Methods
Patients with histologically confirmed non‐metastatic adenocarcinoma of the prostate (International Classification of Disease for Oncology [ICD‐O‐3] code 8140/3 site code C61.9) who were ≥ 18 years old, were identified within the SEER database 2004–2015. Incidental prostate cancer (PCa) patients were defined as patients who underwent transurethral resection of the prostate (TURP) and were then subsequently diagnosed with cT1a/b PCa. Exclusion criteria consisted of metastatic patients (M1), patients who underwent TURP for suspected or known cancer (palliative TURP), patients with unknown subsequent treatment information or Gleason score, cases that were identified only at autopsy or death certificate [3], no available preoperative PSA value, and patients with no information regarding death (Supporting Information Figure).
2.1. Variables of Interest
Descriptive characteristics at baseline consisted of age at diagnosis, preoperative PSA (ng/mL; continuously coded), Gleason score (GS; 6 vs. 7 vs. ≥ 8). The subsequent treatment following TURP was categorized as either radical prostatectomy (RP), radiotherapy (RT), or no local treatment (NLT), which encompassed patients that neither underwent RP nor RT. Cancer‐specific survival (CSS) at 5 years was the endpoints in survival analyses, based on the median follow‐up of our study cohort [6].
2.2. Statistical Analysis
First, baseline characteristics of the cohort were tabulated. Kruskal–Wallis and Chi‐square tests were used to assess differences in medians and proportions. Second, Kaplan–Meier plots depicted CSS according to specific stratification models (cT1a vs. cT1b; cT1a and GS 6 vs. cT1b or GS ≥ 7; GS 6 vs. GS ≥ 7). Third, for each stratification model, the area under the curve (AUC) was derived after 20‐fold cross‐validation, based on multivariable Cox regression models according to Heagerty's method [7]. Here, adjustment covariables consisted of age (continuously coded), preoperative PSA (continuously coded), and subsequent treatment (RP vs. RT vs. NLT). In all statistical analyses, R software environment for statistical computing and graphics (R version 4.1.2; R Foundation for Statistical Computing, Vienna, Austria) was used [8]. All tests were two‐sided, with a level of significance set at p < 0.05. All statistical analyses, along with the accompanying figures and tables, adhere to the current guidelines for statistical reporting in urology [9, 10].
3. Results
3.1. Descriptive Characteristics
A total of 5155 incidental PCa patients were identified between 2004 and 2015. Of all patients, 3035 (59%) were cT1a vs. 2120 (41%) were cT1b. Compared to cT1a, cT1b patients were older (72 vs. 70; p < 0.001), had higher preoperative PSA values (5 vs. 4; p < 0.001), harbored higher proportions of Gleason ≥ 8 (23% vs. 3%; p < 0.001) and were more likely to undergo either RP (20% vs. 13%, p < 0.001) or RT (22% vs. 7%; p < 0.001; Table 1). Corresponding baseline, tumor, and treatment characteristics were tabulated according to the three stratification models in Table 1. The median follow‐up of the overall study population was 59 months.
Table 1.
Descriptive characteristics of 5155 incidental prostate cancer patients according to different prognostication models (current cT1a vs. cT1b; cT1a and GS 6 vs. cT1b or GS ≥ 7; GS < 8 vs. GS ≥ 8), identified within the Surveillance, Epidemiology, and End Results (SEER) database (2004–2015).
| Current cT1a vs. cT1b classification | cT1a and GS 6 vs. cT1b or GS ≥ 7 | GS < 8 vs. GS ≥ 8 | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Characteristic | N | cT1a N = 3035 (59%)a | cT1b N = 2120 (41%)a | p‐valueb | cT1a & GS 6 N = 2409 (47%)a | cT1b or GS ≥ 7 N = 2746 (53%)a | p‐valueb | GS < 8 N = 4575 (89%)a | GS ≥ 8 N = 580 (11%)a | p‐valueb |
| Age | 5155 | 70 (64, 76) | 72 (64, 79) | < 0.001 | 69 (63, 75) | 72 (65, 79) | < 0.001 | 70 (63, 76) | 78 (70, 83) | < 0.001 |
| Preoperative PSA | 5155 | 4 (2, 7) | 5 (3, 13) | < 0.001 | 3 (2, 6) | 5 (3, 12) | < 0.001 | 4 (2, 7) | 10 (4, 31) | < 0.001 |
| Gleason score | 5155 | < 0.001 | < 0.001 | < 0.001 | ||||||
| ≤ 6 | 2409 (79%) | 893 (42%) | 2409 (100%) | 893 (33%) | 3302 (72%) | — | ||||
| 7 | 530 (17%) | 743 (35%) | — | 1273 (46%) | 1273 (28%) | — | ||||
| ≥ 8 | 96 (3.2%) | 484 (23%) | — | 580 (21%) | — | 580 (100%) | ||||
| Treatment | 5155 | < 0.001 | < 0.001 | < 0.001 | ||||||
| NLT | 2429 (80%) | 1226 (58%) | 2031 (84%) | 1624 (59%) | 3310 (72%) | 345 (59%) | ||||
| RP | 382 (13%) | 422 (20%) | 239 (9.9%) | 565 (21%) | 741 (16%) | 63 (11%) | ||||
| RT | 224 (7.4%) | 472 (22%) | 139 (5.8%) | 557 (20%) | 524 (11%) | 172 (30%) | ||||
Median (IQR); n (%).
Wilcoxon rank‐sum test; Pearson's Chi‐square test.
3.2. Cancer‐Specific Survival in Incidental Prostate Cancer Patients According to Established cT1a vs. cT1b Stratification
Of all 5155 incidental PCa patients, 3035 (59%) were cT1a vs. 2120 (41%) were cT1b. CSS at 5 years was 98% for cT1a vs. 90% for cT1b (p < 0.0001, Figure 1). The AUC based on a multivariable Cox regression predicting CSM was 0.83 (Table 2).
Figure 1.
Kaplan–Meier curves depicting cancer‐specific survival of incidental prostate cancer patients, according to (a) cT1a vs. cT1b (b) cT1a & GS 6 vs. cT1b or GS ≥ 7 (c) GS < 8 vs. GS ≥ 8. [Color figure can be viewed at wileyonlinelibrary.com]
Table 2.
Cancer‐specific survival (CSS) rates, as well as area under the curve (AUC) of the tested incidental prostate cancer stratification models.
| cT1a vs. cT1b | cT1a & GS 6 vs. cT1b or GS ≥ 7 | GS < 8 vs. GS ≥ 8 | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Stratification | N (%) | CSSa | AUCb | Stratification | N (%) | CSS a | AUC b | Stratification | N (%) | CSS a | AUC b |
| cT1a | 3055 (59%) | 98% | 0.83 | cT1a & GS 6 | 2409 (47%) | 99% | 0.82 | GS < 8 | 4575 (89%) | 98% | 0.88 |
| cT1b | 2120 (41%) | 90% | cT1b or GS ≥ 7 | 2746 (53%) | 92% | GS ≥ 8 | 580 (11%) | 72% | |||
At 5 years.
Based on a multivariable Cox regression model, after cross‐validation.
3.3. Cancer‐Specific Survival in Incidental Prostate Cancer Patients According to the Modified Stratification (cT1a & GS 6 vs. cT1b or GS ≥ 7) Discussed by Mottet et al.
Of all 5155 incidental PCa patients, 2409 (47%) were cT1a with GS 6, and 2746 (53%) were cT1b or harbored GS ≥ 7. CSS at 5 years was 99% for cT1a GS 6 vs. 92% for cT1b/GS ≥ 7 (p < 0.0001, Figure 1). The AUC based on a multivariable Cox regression predicting CSM was 0.82 (Table 2).
3.4. Cancer‐Specific Survival in Incidental Prostate Cancer Patients According to Gleason Score Sum (GS < 8 vs. GS ≥ 8)
Of all 5155 incidental PCa patients, 4575 (89%) had a GS < 8 and 580 (11%) had a GS ≥ 8.
CSS at 5 years was 98% for patients with GS < 8 and 72% for patients with GS ≥ 8 (p < 0.0001, Figure 1).
The AUC based on a multivariable Cox regression predicting CSM was 0.88 (Table 2).
4. Discussion
Current guidelines recommend stratifying incidental prostate cancer based on tumor volume (< 5% vs. > 5%), distinguishing between cT1a and cT1b stages [1]. We tested, whether a recently discussed modification (Mottet et al.) of this approach might be more accurate or if an equally simplistic stratification that is not based on tumor volume, but on Gleason score sum (GS) might be equally or possible even more accurate. Our testing revealed several noteworthy observations.
First, we identified a total of 5155 incidental PCa patients within the Surveillance, Epidemiology, and End Results (SEER) database (2004–2015). Although an incidental PCa diagnosis in not uncommon after surgery for benign prostate hyperplasia (BPH), it only accounts for 1.5% of all PCa diagnosis in the United States [11, 12]. As a result, incidental PCa research is often limited by small sample sizes which represent a rate‐limiting factor in meaningful analyses [13, 14, 15]. In consequence, limited evidence regarding incidental PCa is available and relatively few studies may be used for guideline recommendations [2, 4]. For example, a contemporary meta‐analysis, relied on 55 combined studies and yielded 2327 incidental PCa patient observations [11]. Conversely, in the current study that relied on the SEER database, more than twice as many individuals could be identified. In consequence, large‐scale databases such as the SEER or the National Cancer Database (NCDB) offer a unique perspective when analyses on rare entities such as incidental PCa are required.
Second, current guidelines recommend stratifying incidental prostate cancer according to stage (cT1a vs. cT1b), which is based on the tumor volume (< 5% vs. > 5%) of the resected tissue at BPH surgery. Recently, Mottet et al. discussed a modified stratification model, that concomitantly considers tumor volume (< 5% vs. > 5%) and GS (< 7 vs. ≥ 7) [4]. Within a novel stratification scheme (GS < 8 vs. GS ≥ 8), we postulated, that possibly the use of GS alone may represent an equally powerful stratification tool. Tabulation of patient characteristics according to the three above‐described schemes revealed important differences in patient counts, as well as in their age and tumor characteristics. For example, the guideline recommended stratification (cT1a vs. cT1b) resulted in two tiers of 59% (cT1a) and 41% (cT1b). The modified model that combines stage (cT1a/b) and GS divided the population into virtually equally sized halves (47% vs. 53%). Conversely, the stratification based on GS < 8 vs. GS ≥ 8 resulted in an 89% vs. 11% division of the cohort. In consequence, the different stratification schemes result in potentially, drastically different patient populations that are determined by the sheer number of observations. Expectedly, the stratification according to GS < 8 vs. GS ≥ 8, where only 11% of the incidental PCa population harbored GS ≥ 8, resulted in an older patient population with substantially higher preoperative PSA values. Conversely, the established, guideline recommended stratification, where the majority of incidental PCa patients were defined as cT1a, resulted in a relatively young patient population with low PSA values. Interestingly, the modified stratification scheme discussed by Mottet et al. resulted in a very similar age and PSA distribution as the established guideline recommended stratification. These observations suggest few, if any differences between the established and the modified tumor‐volume based stratification schemes.
Third, in the final part of the study we tested the ability of the established, guideline recommended stratification (cT1a vs. cT1b) to predict CSS after cross‐validation. Subsequently, we contrasted 5‐year CSS predictions, as well as AUC values of the established, guideline recommend stratification relative to the proposed, modified scheme where stage and GS were combined (cT1a & GS 6 vs. cT1b or GS ≥ 7) [4]. Last but not least, we quantified the same metrics when an alternative stratification solely based on GS (GS < 8 vs. GS ≥ 8) was applied. Based on the similarity of the subgroups generated by the established, guideline recommended stratification and the modified stratification discussed by Mottet et al., the resulting CSS and AUC values were also similar. Specifically, 5‐year CSS rates were 98% for cT1a vs. 90% for cT1b and 99% for cT1a and GS 6 vs. 92% for cT1b or GS ≥ 7. The resulting, cross‐validated AUC of the established, guideline recommended stratification model was 0.83 vs. 0.82 for the modified scheme discussed by Mottet et al. Conversely, the novel stratification scheme solely based on GS (GS < 8 vs. GS ≥ 8) stratified the patient population into two distinct arms, where 5‐year CSS was 98% for low‐risk patients (GS < 8) and 72% for high‐risk patients (GS ≥ 8). The resulting AUC after cross‐validation was 0.88 and exceeded the recoded AUC values of the two stratification alternatives.
Taken together, the stratification model solely based on GS (GS < 8 vs. GS ≥ 8) appears most beneficial in clinical decision making. Its application results in a relatively small population of truly high‐risk patients, in whom CSS is drastically inferior (72%) to the bulk of low‐risk patients in whom CSS is virtually perfect (98%). Not surprisingly this stratification also results in the highest AUC after cross‐validation (0.88) that substantially exceeds the recorded AUC of the alternative models (0.83; 0.82). Although we relied on a cross‐validation methodology, this stratification of incidental PCa patients should undergo formal, external validation and possibly measured against more detailed alternatives that integrate additional risk factors that were not available in the SEER database, such as, for example, postoperative PSA values and prostate volume.
Despite the novelty of our study, several limitations should be acknowledged. First the SEER database lacks confirmatory biopsy information after BPH surgery. Additionally, it does not provide information regarding preoperative prostate volume nor about postoperative PSA values. These variables are undoubtedly very important and should ideally be taken into consideration when designing novel subclassification models for incidental PCa. Despite this limitation, we believe that our analyses offer valuable insights into this specific entity of PCa, where data is scarce. Second, the SEER database does not allow to differentiate between active surveillance and watchful waiting. In consequence, the treatment variable NLT cannot be further stratified. However, it can be postulated that the majority of patients who underwent NLT were included in either one of these treatment modalities. Third, it is important to highlight that certain CSS rates we reported, especially for patients with higher GS, are lower than expected for this entity of PCa, where excellent survival outcomes are generally expected. We suspect that this observation may, to some extent, be explained by the older age of patients with elevated GS and the significant number of patients who received NLT despite higher GS. Nevertheless, our multivariable models adjust for both age and subsequent treatment modality to account for this.
5. Conclusion
Of all tested stratification models, the consideration of GS stratified between < 8 and ≥ 8 resulted in the best survival discrimination, as well as highest accuracy after internal cross‐validation. In consequence, the use of this model (GS < 8 vs. ≥ 8) appears better suited than the standard cT1a vs. cT1b substaging, when CSS represents the endpoint of interest.
Funding
The authors have nothing to report.
Ethics Statement
All analyses and their reporting followed the SEER reporting guidelines. Due to the anonymously coded design of the SEER database, study‐specific Institutional Review Board ethics approval was not required.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Supporting File
Acknowledgments
The authors have nothing to report.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
References
- 1. Mottet N., van den Bergh R. C. N., Briers E., et al., “EAU‐EANM‐ESTRO‐ESUR‐SIOG Guidelines on Prostate Cancer‐2020 Update. Part 1: Screening, Diagnosis, and Local Treatment With Curative Intent,” European Urology 79 (2021): 243–262, 10.1016/J.EURURO.2020.09.042. [DOI] [PubMed] [Google Scholar]
- 2. Capitanio U., Autorino R., Bandini M., et al., “Incidental Prostate Cancer (cT1a–cT1b) Is a Relevant Clinical and Research Entity and Should Be Fully Discussed in the International Prostate Cancer Guidelines,” European Urology Oncology 5 (2022): 256–258, 10.1016/j.euo.2021.03.005. [DOI] [PubMed] [Google Scholar]
- 3. Capitanio U., “Contemporary Management of Patients With T1a and T1b Prostate Cancer,” Current Opinion in Urology 21 (2011): 252–256, 10.1097/MOU.0B013E328344E4AD. [DOI] [PubMed] [Google Scholar]
- 4. Mottet N., Rouviere O., and van der Kwast T. H., “Incidental Prostate Cancer: A Real Need for Expansion in Guidelines?,” European Urology Oncology 5 (2022): 259–260, 10.1016/j.euo.2021.03.005. [DOI] [PubMed] [Google Scholar]
- 5. Capitanio U., Cheng L., Lopez‐Beltran A., et al., “The Importance of Interaction Between Urologists and Pathologists in Incidental Prostate Cancer Management,” European Urology 60 (2011): 75–77, 10.1016/J.EURURO.2011.03.012. [DOI] [PubMed] [Google Scholar]
- 6.SEER Cause of Death Recode, Accessed June 12, 2023. https://seer.cancer.gov/codrecode/.
- 7. Heagerty P. J., Lumley T., and Pepe M. S., “Time‐Dependent ROC Curves for Censored Survival Data and a Diagnostic Marker,” Biometrics 56 (2000): 337–344, 10.1111/J.0006-341X.2000.00337.X. [DOI] [PubMed] [Google Scholar]
- 8. R: The R Project for Statistical Computing , Accessed June 4, 2023, https://www.r-project.org/.
- 9. Assel M., Sjoberg D., Elders A., et al., “Guidelines for Reporting of Statistics for Clinical Research in Urology,” Outcomes/Epidemiology/Socioeconomics 201 (2019): 595–604, 10.1097/JU.0000000000000001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Vickers A. J., Assel M. J., Sjoberg D. D., et al., “Guidelines for Reporting of Figures and Tables for Clinical Research in Urology,” BJU International 126 (2020): 14–25, 10.1111/BJU.15102. [DOI] [PubMed] [Google Scholar]
- 11. Cheng B. K. C., Castellani D., Chan I. S. H., et al., “Incidence, Predictive Factors and Oncological Outcomes of Incidental Prostate Cancer After Endoscopic Enucleation of the Prostate: A Systematic Review and Meta‐Analysis,” World Journal of Urology 40 (2022): 87–101, 10.1007/S00345-021-03756-9/TABLES/3. [DOI] [PubMed] [Google Scholar]
- 12. Scheipner L., Incesu R.‐B., Morra S., et al., “Characteristics of Incidental Prostate Cancer in the United States,” Prostate Cancer and Prostatic Diseases 27 (2023): 1–6, 10.1038/s41391-023-00742-7. [DOI] [PubMed] [Google Scholar]
- 13. Hutchison D., Peabody H., Kuperus J. M., et al., “Management of Prostate Cancer After Holmium Laser Enucleation of the Prostate,” Urologic Oncology: Seminars and Original Investigations 39 (2021): 297.e1–297.e8, 10.1016/j.urolonc.2020.11.003. [DOI] [PubMed] [Google Scholar]
- 14. Han J. H., Chung D. H., Cho M. C., et al., “Natural History of Incidentally Diagnosed Prostate Cancer After Holmium Laser Enucleation of the Prostate,” PLoS One 18 (2023): e0278931, 10.1371/journal.pone.0278931. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Capogrosso P., Capitanio U., Vertosick E. A., et al., “Temporal Trend in Incidental Prostate Cancer Detection at Surgery for Benign Prostatic Hyperplasia,” Urology 122 (2018): 152–157, 10.1016/j.urology.2018.07.028. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting File
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.