- Top of page
- PATIENTS AND METHODS
- CONFLICT OF INTEREST
Locally recurrent prostate cancer has been reported in 25–94% of prostatic biopsies after radiotherapy (RT) [1,2]. An increase in the serum PSA level after primary RT often represents the first clinical suspicion of treatment failure that might require additional therapy in the form of either salvage or hormonal ablative therapy [3,4]. There have been numerous definitions of biochemical failure which have been proposed over the years, including the American Society for Therapeutic Radiology and Oncology definition of three consecutive increases after the nadir PSA level, which was most frequently used, but the Phoenix definition of the ‘PSA nadir after RT + 2 ng/mL’ has gained acceptance among many radiotherapists and oncologists [5,6]. Patients with biopsy-confirmed locally recurrent prostate cancer might be suitable candidates for salvage prostatectomy or salvage cryotherapy (SC), provided this is in the clinical context of the absence of metastatic disease. A major hindrance to the widespread application of salvage prostatectomy has been the perceived significant morbidity (e.g. rectal injury, moderate/severe urinary incontinence) associated with this treatment, although recent reports would suggest a significant decline in its associated morbidity when performed at tertiary-care referral centres by experienced urological surgeons . By contrast, the major limitation to the more frequent use of SC has been the few patients treated at any one institution. This lack of data makes it difficult to give patients realistic expectations about SC. Although previous studies have established the prognostic importance of several pretreatment variables, including a serum PSA level of >10 ng/mL, a PSA doubling time of <16 months, clinical stage >T3, and a biopsy Gleason score of the local recurrence of >9 [8,9], there remains controversy over the treatment-specific outcomes of SC, and it remains unclear which patients might be optimal candidates for this treatment.
To address these issues, we gathered and pooled the retrospective databases of six tertiary-care referral centres that use SC for locally recurrent prostate cancer. We re-addressed the clinical predictors of treatment outcome in this large cohort of patients. Importantly, we developed a pretreatment nomogram that allows a prediction of the probability of biochemical failure after SC, based on pretreatment clinical variables.
PATIENTS AND METHODS
- Top of page
- PATIENTS AND METHODS
- CONFLICT OF INTEREST
Six tertiary-care referral centres participated in this study, i.e. Columbia University, University of Western Ontario (London), Triangle Urological Group, The University of Texas M.D. Anderson Cancer Center, Prostate Institute of America, and University of California in San Francisco. Each of the participating institutions obtained approval from their respective Institutional Review Board to conduct a retrospective chart review of their treated patients. Between September 1990 and October 2005, 797 patients with biopsy-confirmed locally recurrent prostate cancer were treated with SC at one of these centres. Databases from all participating institutions were gathered and merged to form a pooled database. Before data analysis, all data points were inspected to verify data consistency. All patient identifiers were removed, in compliance with the Health Insurance Portability and Accountability Act. Tumours from all patients were staged according to the 2002 TNM staging system. All patients had undergone a complete metastatic evaluation, including a complete history, physical examination, TRUS-guided prostatic biopsies documenting a local recurrence, serum PSA level, and radiographic imaging (chest X-ray, abdominal/pelvic CT, bone scan) before SC. All patients with radiographic or tissue-proven evidence of metastatic disease were excluded from the study.
The distribution of patients by participating centre was: Columbia University, 213 (26.7%); University of Western Ontario, 187 (23.5%); Triangle Urological Group, 153 (19.2%); M.D. Anderson Cancer Center, 110 (13.8%); Prostate Institute of America, 100 (12.5%); and University of California in San Francisco, 34 (4.3%). Neoadjuvant hormonal ablative therapy in the form of an antiandrogen with or without a LHRH agonist was used in a subset of patients for 3–6 months before SC; however, none of the patients received adjuvant hormonal ablative therapy. After SC patients were typically evaluated at 3–6-month intervals with a history, physical examination and serum PSA measurements. Patients suspected of having biochemical failure (defined as a serum PSA level of >0.5 ng/mL after SC ) were offered repeat TRUS-guided prostatic biopsies in an attempt to document persistent locally recurrent disease. Patients with an increasing PSA level after SC were also considered for repeat metastatic evaluation, including radiographic imaging (abdominal/pelvic CT and bone scan).
Among the participating centres, complete pretreatment clinical data were available in 450 patients. The mean (range) is reported for age at diagnosis, serum PSA level at diagnosis, total dose of primary RT, nadir PSA level after RT and before SC, use of neoadjuvant and adjuvant hormonal therapy, rate of biochemical failure, and duration of follow-up. Also, frequency distributions were reported for initial biopsy Gleason score, initial clinical stage and local recurrence biopsy Gleason score.
Using logistic regression, we assessed the statistical significance of several potential prognostic factors in predicting biochemical failure. A univariate model was first used to estimate the odds ratios for each potential prognostic factor, including the 95% CI. The final model contained potential prognostic factors that were both statistically relevant, with P < 0.25 on univariate analysis, and clinically important. Regression diagnostics included: (i) identifying the correction functional form of the continuous covariate initial serum PSA level which lead to the log-log transformation; (ii) plots of the standardized Pearson residuals and the standardized deviance residuals for potential outliers; and (iii) the model fit was assessed using the Hosmer-Lemeshow goodness-of-fit test.
A pretreatment nomogram was developed based on the prognostic factors identified, to assess a patient’s estimated probability of biochemical failure, which was defined as a serum PSA level of >0.5 ng/mL after SC. The variables selected to create the nomogram were the initial serum PSA value, biopsy Gleason score at the time of diagnosis, and initial clinical T stage. Before developing the nomogram we explored the association between biochemical failure and clinical and pathological data using logistic regression methods. Our pretreatment nomogram represents a graphical representation of the relationship between an outcome and independent predictor variables. As a measure of predictive accuracy, we calculated the concordance index, which is equivalent to the area under the receiver operator characteristic (ROC) curve, and provides a measure of the model’s discriminative ability. Once we established what we believed to be a clinically useful pretreatment nomogram, we evaluated the predictive performance of the nomogram using 500 bootstrap samples, which provided validation indices such as the bootstrap biased-corrected concordance index and a plot used to assess calibration accuracy. SPLUS code was used in conjunction with the Harrell libraries to generate the nomogram, validation indices, and calibration plots.
The methods of Kaplan and Meier  were used to estimate the probability of biochemical failure-free survival in 277 patients for whom the serum PSA level at various times after SC were available. The duration of follow-up for the study population was calculated from the date of SC to the date of the last follow-up, death, or last contact. The median duration of follow-up for the entire study population was 3.4 years. All statistical analyses were performed using either STATA (Release 10; StataCorp LP, College Station, TX) or SPLUS (version 7; Insightful Corporation, Seattle, WA) statistical software with the Harrell libraries.
- Top of page
- PATIENTS AND METHODS
- CONFLICT OF INTEREST
Medical records were available for 450 of the 797 eligible patients; these 450 comprised the study population, and their clinical characteristics are shown in Table 1. The mean (range) patient age at the time of diagnosis with prostate cancer was 64.1 (50.7–78.2) years, and the mean serum PSA level was 17.8 (1.3–157.1) ng/mL. Most patients had localized disease at time of diagnosis, with clinical stage T1 in 77 (19.0%) and stage T2 in 195 (48.0%). The remaining patients had clinical stage T3 in 115 (28.3%) or T4 in 19 (4.7%). The biopsy Gleason score at time of diagnosis was frequently ≤6 (181, 52.3%), with Gleason score 7 in 105 (30.4%) and ≥8 in 60 (17.3%) patients. The mean (range) cumulative dose of primary RT in patients was 65.6 (31–78) Gy, with the mean nadir PSA level afterward being 2.3 (0.1–12.5) ng/mL. Before SC the mean serum PSA level was 7.8 (0.5–64.2) ng/mL and neoadjuvant hormonal therapy was used in 96 patients (37.9%). Before SC all patients had prostatic biopsies taken to document any local recurrence after RT, with a significant proportion of patients having high-grade disease, i.e. Gleason score ≥8 in 117 (45.2%) patients. None of the patients received adjuvant hormonal therapy after SC except when there was evidence of disease progression (i.e. biochemical failure or metastatic progression) which constitutes salvage hormonal ablative therapy. At a median follow-up of 3.4 (2.7–4.0) years after SC, the rate of biochemical failure was 66.0%. A Kaplan-Meier analysis of the probability of biochemical failure in 277 patients for whom serum PSA levels were available at various times after SC is shown in Fig. 1.
Table 1. The clinical characteristics of the 450 patients
|Variable||Mean (range) or n (%)|
|Age at diagnosis, years|| 64.1 (50.7–78.2)|
|Serum PSA at diagnosis, ng/mL|| 17.8 (1.3–157.1)|
|Clinical stage at diagnosis*|
| T1|| 77 (19.0)|
| T2||195 (48.0)|
| T3|| 115 (28.3)|
| T4|| 19 (4.7)|
|Initial biopsy Gleason score†|
| ≤7||285 (82.7)|
| ≥8|| 60 (17.3)|
|Primary RT dose, Gy|| 65.6 (31–120)|
|Nadir PSA level after RT, ng/mL|| 2.2 (0.1–12.5)|
|Serum PSA level before SC, ng/mL|| 7.8 (0.5–64.2)|
|Local recurrence biopsy Gleason score‡|
| ≤7||141 (54.6)|
| ≥8|| 117 (45.4)|
|Neoadjuvant hormonal therapy (before SC))|| 96 (38.1|
|Adjuvant hormonal therapy (after SC)|| 0|
|Rate of biochemical failure, % (PSA level >0.5 ng/mL)|| 66.0|
|Follow-up, years, median (95% CI)|| 3.4 (2.7–4.0)|
Potential predictors of biochemical failure were used to construct a logistic regression model, with the univariate and multivariate models shown in Table 2. In the multivariate analysis, serum PSA level at the time of diagnosis was identified as a statistically significant predictor of probability of biochemical failure after SC, with 1 unit increases in the log-log of the serum PSA associated with an odds ratio of 3.8 (P < 0.001). The Gleason score at diagnosis was also a statistically significant predictor of biochemical failure on multivariate analysis, with patients presenting with an initial Gleason score of ≥8 having odds ratio of 2.9, vs patients with scores of ≤7 (P = 0.015). Despite clinical T stage at diagnosis not being a statistically significant predictor of biochemical failure on multivariate analysis, nevertheless there were increases in the probability of biochemical failure after SC between patients with clinical T stages of T1, T2, T3 and T4, with odds ratios of 1.2, 1.7 and 3.1, respectively, with stage T1 serving as the reference category. The Hosmer-Lemeshow goodness-of-fit statistic for the multivariate model was 8.97, with a corresponding P = 0.345, suggesting that the null hypothesis that the model fits the data should not be rejected.
Table 2. Results of univariate logistic regression analysis with biochemical failure as the dependent variable (277 patients)
|Variable||Odds ratio (95% CI)||P|
|Serum PSA at time of diagnosis|
|1 unit change in log-log of initial PSA level||4.6 (2.1–10.2)||<0.001|
|Biopsy Gleason score at diagnosis|
| >7||3.4 (1.5–7.7)||0.002|
|Clinical stage at diagnosis|
| T2||1.4 (0.8–2.8)||0.260|
| T3||2.4 (1.2–4.9)||0.019|
| T4||5.4 (1.1–26.3)||0.039|
On the basis of our multivariate analysis of predictors of biochemical failure, a pretreatment nomogram was developed for use in predicting the risk of biochemical failure after SC (Fig. 2). By applying this nomogram to an individual patient’s pretreatment variables (initial serum PSA level, biopsy Gleason score and clinical T stage at diagnosis), the number of points attributed to these individual three variables were cumulated to produce a total number of points for that patient. A vertical line is then drawn from the line indicating the total number of points to the line indicating the probability of biochemical failure, thereby predicting the patient’s likely outcome from SC. To assess the performance of our pretreatment nomogram in predicting the probability of biochemical failure with SC, an ROC curve was generated (Fig. 3). The concordance index (equivalent to the area under the ROC curve) was 0.70, representing acceptable discrimination of the model. Furthermore, the bootstrap bias-corrected estimate of the concordance index was 0.68. Finally, a plot of apparent and bias-corrected calibration was drawn to validate the model for calibration accuracy, using 500 bootstrap samples (Fig. 4).
- Top of page
- PATIENTS AND METHODS
- CONFLICT OF INTEREST
SC has become an increasingly popular and acceptable treatment choice for patients with locally recurrent prostate cancer after primary RT [1–4]. However, one of the major hindrances to the widespread application of SC in many tertiary-care referral centres is the limited available data on the treatment outcomes of SC. Furthermore, several authors have raised concerns about the oncological outcomes of SC compared to salvage surgery, as well as the potential treatment-related morbidities with this treatment [7,11]. In an attempt to gain a greater insight in terms of the treatment outcomes that can be expected with SC we gathered SC databases from many of the major centres using SC in North America. In a large sample of patients from these tertiary-care referral centres, we assessed the outcomes of SC for locally recurrent prostate cancer. In the present patients the biochemical failure rate was 66.0% at a median follow-up of 3.4 years after SC. Furthermore, on multivariate analysis of potential predictors of biochemical failure, we identified serum PSA level at diagnosis, the initial biopsy Gleason score, and the clinical stage at diagnosis as important predictors of treatment outcome. We then developed a pretreatment nomogram allowing a prediction of the likely outcome of SC for an individual patient. This has important clinical implications, as it might help to select patients best suited for SC. Similarly, use of this pretreatment nomogram might help to guide treatment discussions with a patient. That this pretreatment nomogram was developed from a multicentre cohort of patients adds further validity and enhances the general applicability of this nomogram, as it represents the realistic outcomes of SC when applied across several centres.
Previous clinical studies, including reports from several participating centres, identified the prognostic importance of serum PSA level, PSA doubling time, clinical stage before RT, and Gleason score before SC [1,2,9]. In the present study, we showed that predictors of biochemical failure after SC are determined at the time of initial diagnosis with prostate cancer. Although several explanations can be postulated as to why predictors of outcome for SC are determined by clinical variables at diagnosis with prostate cancer, we believe that these variables might reflect the inherent tumour biology (i.e. local aggressiveness and metastatic potential) of these tumours, as well as tumour volume.
Overall, we report a biochemical failure-free rate of 39.6% with SC at a median follow-up of 3.4 years. The primary objective of this study was not to compare the treatment-specific outcomes of SC with those of other forms of salvage therapy, but previous salvage prostatectomy series reported biochemical failure-free rates of 55–65% at 5 years using varying definitions of biochemical failure. However, we suggest caution in concluding from this that salvage prostatectomy is better than SC . In the present study, we used a very rigorous definition of biochemical failure, consisting of a serum PSA level of >0.5 ng/mL after SC. Also, there have been major technological advances in SC, with most centres currently using third-generation devices , whereas >90% of the present patients were treated using second-generation devices because the study extended over a 15-year period during which third-generation devices were not yet available. We also believe that many of the patients treated with SC earlier in most centres had a higher incidence of unfavourable clinical features (e.g. high Gleason grade, elevated serum PSA level, advanced clinical stage) and were at high risk of loco-regional and systemic progression. As such, we feel that an accurate comparison between salvage therapies could only be made prospectively using comparable treatment groups and with a common definition of biochemical failure between the treatment arms. Although we encourage and support such a study, we feel it might be difficult to accomplish because of the few patients currently being offered salvage therapy, although development of a multicentre consortium would facilitate this effort.
There are several limitations to the present study: First, complete medical records were available for 450 of the patients, reflecting differences in the way data were gathered across the participating centres, and clearly representing a significant limitation. Furthermore, the pooled study cohort was treated at several centres over an extensive study period, and this could have affected our results. Although this is an inherent limitation resulting from the study being retrospective, a prospective study with similar endpoints would not be feasible because too few patients are treated with SC at individual centres to make a similar prospective study feasible. Second, internal validation of our pretreatment nomogram using both an ROC curve and a calibration plot showed that this model had acceptable discrimination (concordance index 0.70) in terms of predicting the risk of biochemical failure with SC for an individual patient. Clearly, our pretreatment nomogram could be further optimized, and we hope subsequent studies will be conducted to refine this nomogram and thereby improve its predictive capabilities. Also, our pretreatment nomogram will need to be validated when applied to patients treated solely with third-generation cryotherapy devices.
In conclusion, we validated the clinical utility of SC as a treatment choice for patients with locally recurrent prostate cancer after primary RT. The most important predictors of treatment outcome with SC in our large, multicentre patient cohort included serum PSA level, biopsy Gleason score and clinical stage at diagnosis. Importantly, we developed a pretreatment nomogram that can be used to predict the potential outcome of SC for an individual patient. This nomogram might prove useful when determining which patients are best suited for this treatment, as well as providing patients with information on which to base realistic expectations of their treatment outcomes.