Validation of the cell cycle progression score to differentiate indolent from aggressive prostate cancer in men diagnosed through transurethral resection of the prostate biopsy

Abstract Background Validation of biomarker‐based prognostic models to improve risk stratification in men with localized prostate cancer (PrCa) remains a clinical need. It has previously been shown that the cell cycle progression (CCP) test provides significant, independent prognostic information for men who were incidentally diagnosed with PrCa after transurethral resection of the prostate (TURP) and were conservatively managed. Aim The results have been extended in a newly analyzed retrospective cohort of UK men diagnosed through TURP biopsy (TURP1B; N = 305). Methods and Results The CCP score was derived from TURP biopsy tissue and combined with a modified UCSF Cancer of the Prostate Risk Assessment score (CAPRA) to generate the clinical cell‐cycle risk score (CCR). The primary endpoint was PrCa‐specific mortality (PSM). Hazard ratios (HR) were calculated for a one‐unit change in score. Median follow‐up was 9.6 (IQR: 5.4, 14.1) years, and 67 (22%) men died from PrCa within 10 years of diagnosis. The median CCP score was 1.1 (IQR: 0.6, 1.7). In univariate analyses, CCR proved a significant prognosticator of PSM (HR per unit score change = 2.28; 95% CI: 1.89, 2.74; P = 1.0 × 10−19). In multivariate analyses, CCR remained a significant prognosticator of PSM after adjusting for CAPRA (HR per unit score change = 4.36; 95% CI: 2.65, 7.16; P = 1.3 × 10−8), indicating that its molecular component, CCP, provides significant, independent prognostic information. Conclusion These findings validate a combined clinicopathologic and molecular prognostic model for conservatively managed men who are diagnosed through TURP, supporting the use of CCR to inform clinical management.


| INTRODUCTION
Increasingly, clinical management of cancer patients is being tailored to the individual's risk of having or developing aggressive disease. This is particularly important for patients with newly diagnosed prostate cancer (PrCa), wherein treatment options range from conservative management with active surveillance or watchful waiting, to definitive treatment such as prostatectomy or primary radiation with or without concurrent androgen deprivation therapy (ADT).
For patients with low-risk disease, conservative management is becoming increasingly popular. A recent survey of clinical practice in the United States found that approximately 40% of men with low-risk clinical features are opting for active surveillance. 1 For these patients, deferred treatment can be more appropriate because the risk of disease progression is low, and it offers the opportunity to avoid unnecessary treatment-associated morbidities such as impotence and incontinence.
Conversely, when high risk can be established, definitive treatment is appropriate. Randomized clinical trial results 2,3 and clinical practice guidelines [4][5][6][7] both support the use of either surgery and adjuvant radiation, or primary radiation and ADT in appropriately selected patients.
However, there are potential risks associated with any clinical management decision. The opportunity to cure localized disease may be lost for patients who are inappropriately managed by AS, or, conversely, patients who are inaccurately identified as high risk may suffer needlessly from treatment-associated morbidities. Therefore, accurate risk estimates are essential. Various prognostic models have been developed and validated to help predict disease outcome. For example, the UCSF Cancer of the Prostate Risk Assessment (CAPRA) score combines Gleason score, prostate-specific antigen (PSA) level, clinical T Stage, percent positive cores, and age into a validated prognostic model for biochemical recurrence (BCR) after definitive treatment and for prostate cancer-specific mortality (PSM) after conservative management. 8,9 Another prognostic model, PREDICT, uses similar variables as CAPRA to predict distal oncologic outcomes but also estimates the absolute benefit of treatment. 10  settings and shown to add independent prognostic information that is not captured by standard clinicopathologic features. [12][13][14][15] The score has been combined with clinical variables into a combined clinical cell-cycle risk score (CCR) to calculate a 10-year predicted risk of PSM. 12 The objective of this study was to validate a predefined model (CCR) that combines molecular and clinicopathologic variables to predict risk of aggressive disease in a retrospective cohort of men who were incidentally diagnosed with PrCa by TURP biopsy and were conservatively managed. These patients were collected as part of a TURP cohort, part of which was used previously to evaluate the molecular CCP score, 16 but the patients used in the present analysis were not included in the previous analysis and the CCR score which combines CCP with CAPRA 12 was not evaluated in the previous analysis. We also aimed to expand the body of evidence showing that CCP provides significant prognostic information that is not captured by clinicopathologic features.

| Cohort
The TURP1B cohort was part of a large, previously described cohort of conservatively managed men, but this subgroup has not been analyzed previously. The full TURP cohort was a population-based, retrospective, watchful-waiting cohort identified from six cancer registries in Great Britain. Within each region, cases from collaborating hospitals were reviewed; full details of these cases have been reported. 16 Men were included in the study if they had clinically localized PrCa diagnosed by TURP from 1990 through 1996, were younger than 76 years at diagnosis, and had a baseline PSA measurement <100 ng/mL. Patients were excluded if they were treated with radical prostatectomy or radiation therapy, or if they died, or showed evidence of metastatic disease within 6 months of diagnosis. Men who had hormone therapy before diagnostic biopsy were also excluded due to the potential effects of hormone treatment on Gleason score interpretation. Original histological specimens for difficult cases were centrally reviewed by a panel of expert urological pathologists to confirm the TURP diagnosis and, when necessary, to reassign Gleason scores for all tumors using a contemporary interpretation of the Gleason scoring system. 17 Follow-up information was obtained through the cancer registries, with the final review conducted in January 2010. Deaths were divided into two categories, those from PrCa and those from other causes, according to the World Health Organization's standardized criteria. 18 A sub-cohort of a large cohort of conservatively managed men who were diagnosed by TURP biopsy has been previously reported. 16 The sub-cohort analyzed here, TURP1B, is an independent subset of the full cohort which has not been previously analyzed. This new cohort comprised 305 men who were eligible for the study, generated a valid CCP score, and had a complete set of clinicopathologic features to allow calculation of a CAPRA score ( Figure 1). A previous independent TURP sub-cohort (TURP1A) was collected at the same time as TURP1B, and has been reported previously 16 and was also used to help develop the CCR score. 12 TURP1A comprised 330 men who met the same eligibility criteria as for TURP1B.

| Statistical analysis
For the primary endpoint, time to PrCa specific mortality, patients with events or last follow-up after 10 years were censored at 10 years. Hazard ratios (HR) with 95% profile likelihood-based confidence intervals and two-sided P-values from partial likelihood ratio tests were reported for all Cox proportional hazards models. Confidence intervals for risk estimates were based on the log-log transformation. After examining the Schoenfeld residuals, time dependence of predictors in Cox proportional hazards models was accounted for using untransformed time. The distribution of continuous variables was compared using the Wilcoxon rank-sum test, and the distribution of ordered categorical variables was compared using the Goodman-Kruskal test. Calibration was tested using the Greenwood-Nam-D'Agostino test. P-values were considered significant at the two-sided 0.05 level. All analyses were carried out using R software, version 3.5.0 or higher (R Core Team, Vienna, Austria).

| RESULTS
Median follow-up for TURP1B was 9.6 years (IQR 5.4, 14.1), and 67 (22%) men died from PrCa within 10 years of diagnosis. The median CCP score was 1.1 (IQR: 0.6, 1.7). Compared with other clinicopathologic features, the CCP score correlated most strongly with Gleason score (Pearson r = 0.55). Median follow-up for TURP1A was 9.8 years (IQR: 4.9, 11.6), and 67 (20%) men died from PrCa within 10-years of disease diagnosis. The demographic features of TURP1A and TURP1B were similar except for a shift to higher clinical stage in TURP1B (Table 1).

F I G U R E 1 Patient flow for TURP1B
In univariate analyses, the CCP score was a significant prognostic variable for PSM in the TURP1B cohort, with an HR per unit score change of 3.23 (95% CI: 2.50, 4.17, P = 1.9 Â 10 À16 ; Table 2). The CCR score was also a significant univariate prognosticator of PSM in the TURP1B cohort, with an HR per unit score change of 2.28 (95%CI: 1.89, 2.74, P = 1.0 Â 10 À19 ; Table 2). There was no evidence for time dependence within this model indicating that the prognostic value of CCR was similar across all time points (test using Schoenfeld residuals P = 0.13). The 10-year Kaplan-Meier estimate of the probability of progression to PSM separated by CCR quartile groups in the TURP1B cohort is illustrated in Figure 2. The CCR score was used to generate a risk curve for progression to PSM within 10 years ( Figure 3) and compared with the CCR-based risk curve generated from TURP1A (Figure 3(A)). Although the curves were similar, there was evidence of overfitting in TURP1A, which had been used in part to train the CCR score. 12 Risk predicted from the TURP1A model was also not significantly different from the observed risk in TURP1B, indicating good calibration (P = 0.28, Supplementary Figure).
We also compared the CCR risk curve from TURP1B to that generated and validated for a separate cohort of patients diagnosed with needle biopsy (Figure 3(B)). 9 Interestingly, men diagnosed with TURP had a higher risk of harboring aggressive disease compared with that among men who were diagnosed with needle biopsy, but had otherwise similar clinicopathologic and molecular prognostic features.
In multivariate analyses (Table 2), the CCR score remained a significant prognostic factor of PSM after adjusting for CAPRA (HR per unit score = 4.36 (95% CI: 2.65, 7.16, P = 1.3 Â 10 À8 ), indicating that the molecular component of CCR provides significant and independent prognostic information. CAPRA was also significant in this analysis (P = 0.0063), but the HR was less than 1. This suggests that CAPRA is over-weighted within the CCR score when evaluated by the TURP1B cohort. The amount of new prognostic information provided by the CCR score can be illustrated by comparing the difference in predicted risk between CCR-only and CAPRA-only prognostic models (Figure 4). In addition, the c-index for the CAPRA-only model was 0.76 and improved to 0.80 for CCR.

| DISCUSSION
With increasing frequency, men with low-risk PrCa are choosing conservative clinical management regimens such as active surveillance. Validated biomarkers that add independent risk discrimination are useful adjuncts for identifying suitable candidates for such management. They can also be used to help guide the intensity of invention if immediate treatment is deemed necessary. Here, we have provided additional evidence that the F I G U R E 3 Clinical cell-cycle risk score-based risk curve for TURP1B compared with (A) TURP1A and (B) previously published risk data for men diagnosed using needle biopsy F I G U R E 2 Kaplan-Meier risk curves for prostate cancer-specific mortality, by clinical cell-cycle risk score quartiles CCP score provides a significant and substantial amount of independent prognostic information in men who were incidentally diagnosed with PrCa after TURP and conservatively managed. We have also validated a predefined prognostic model (the CCR score) that combines molecular and clinicopathologic data to predict the 10-year risk of PSM.
The CCR score was trained in more than 1000 men, 12 including a previous cohort of men diagnosed with TURP biopsy. 16