Randomized trials have demonstrated that escalated-dose external-beam radiotherapy (EDRT) is better than standard-dose radiotherapy (SDRT) for patients with prostate cancer and that adding androgen-deprivation therapy (ADT) to SDRT is better than SDRT alone; however, no trials have compared EDRT versus SDRT plus ADT or EDRT versus EDRT plus ADT. The authors designed a model to estimate the results of various doses of radiotherapy (RT) combined with various durations of ADT.
From 1989 to 2007, 3215 men consecutively received definitive EDRT with or without ADT. In total, 2012 patients had complete records available for creating the nomogram. The duration of ADT varied for patients who received no RT (n = 1562), ≤6 months of RT (n = 145), from >6 months to <24 months of RT (n = 140), and ≥24 months of RT (n = 165) with a median follow-up of 65.7 months, 66.2 months, 60.1 months, and 63 months, respectively. The model included the following covariates: palpation T-category, biopsy Gleason score, the percentage of tumor cells with a Gleason pattern of 4 or 5, the percentage of tumor tissue, initial pretreatment prostate-specific antigen (PSA) level, ADT duration, and RT dose. Two nomograms, for outcomes with and without ADT, were created from a single competing-risks model. Biochemical failure was defined as a rise in serum PSA of 2 ng/mL over the post-treatment PSA nadir.
According to the results from analyzing representative intermediate-risk to high-risk patient parameters, the gains from increasing the RT dose from 70 Gray (Gy) to 80 Gy were far less than the gains from adding ≥3 months of ADT.
External-beam radiotherapy (RT) is commonly used to treat nonmetastatic prostate cancer with curative intent. Several randomized studies have demonstrated that escalated-dose RT (EDRT) without androgen-deprivation therapy (ADT) is more effective than standard-dose RT (SDRT) in providing long-term biochemical control.1,2 Thus, EDRT has become standard at most institutions.
Other randomized trials have indicated that the addition of ADT to SDRT improves both biochemical control and survival in patients with intermediate-risk to high-risk prostate cancer.3,4 In high-risk patients who received SDRT, extending the length of androgen deprivation from 4 to 6 months to 2.5 to 3.0 years also was effective.5,6 However, to our knowledge, no reported randomized trials have compared EDRT versus SDRT plus ADT or EDRT versus EDRT plus ADT. Consequently, there is considerable variability in the management of prostate cancer with external-beam RT. Although nomograms have been developed to estimate the outcome of patients who receive RT,7 none have estimated the impact of the length of ADT on patients who receive escalated RT doses. The objective of the current study was to develop a prediction tool that incorporates contemporary clinical and pathologic parameters to assist clinicians and patients in estimating the potential gains in biochemical control from adding various lengths of ADT to EDRT or SDRT.
MATERIALS AND METHODS
From April 1989 to August 2007, 3215 consecutive patients with clinically localized prostate cancer received definitive external-beam RT with or without ADT in the Department of Radiation Oncology at Fox Chase Cancer Center. We excluded 920 patients for whom information was lacking on biopsy core region involvement, Gleason score, RT dose, prostate-specific antigen (PSA), or T-category. We also excluded 10 patients who had received <60 Gray (Gy).
Radiotherapy and Androgen-Deprivation Therapy
Three-dimensional conformal radiotherapy
Patients who had a clinical T-category of T1/T2a-T2b and a Gleason score of 2 to 6 typically received treatment to the prostate with or without the proximal seminal vesicles. Patients who had more advanced prostate cancer (T2c/T3/T4 or Gleason score 7–10) received 46 to 50 Gy to a small pelvic field followed by a conformal boost to the prostate and seminal vesicles.8 The mean dose to the patient treatment volume (PTV) was typically between −5% and +7% of the prescribed dose, and 99% of the PTV received at least 95% of the prescribed dose.
The primary clinical target volume (CTV1) included the prostate, any extraprostatic extension, and the proximal seminal vesicles, defined as the first approximately 9 mm of the proximal seminal vesicles. In high-risk patients, additional volumes were defined. CTV2 comprised the distal seminal vesicles, and CTV3 comprised the periprostatic, periseminal vesicle, external iliac, obturator, and internal iliac lymph nodes. PTV margins were 8 mm in all dimensions except posterior (5 mm). The dose was prescribed to the PTV volumes, such that the PTV1 was planned to receive ≥95% of the prescription dose, and the PTV2 and PTV3 were planned to receive ≥95% of 56 Gy over the full treatment course.9
ADT was received by 450 men in the study group and was prescribed at the discretion of the treating physician. This consisted of a luteinizing hormone-releasing hormone (LHRH) agonist and/or an antiandrogen. The LHRH agonist was received by 434 men and consisted of leuprolide or goserelin depot injections. This was received in combination with an antiandrogen (eg, bicalutamide, flutamide, or nilutamide) at the outset by 229 patients. An additional 16 patients received an antiandrogen alone. The duration of ADT was used as a continuous covariate in the model. For purposes of illustration, 1562 patients received no ADT (none), 145 received ADT for ≤6 months (short term), 140 patients received ADT for >6 months to <24 months (intermediate term), and 165 patients received ADT for ≥24 months (long term); the median ADT duration in each group was 0 months, 3 months, 12 months, and 29 months, respectively.
Statistical Analysis and Model Development
Analyses of variance and chi-square tests were used to compare variables between the ADT subgroups (Table 1). Two nomograms—1 with ADT and 1 without ADT—were created from a single Fine and Gray competing-risks model. These nomograms predicted the probability of biochemical failure (BF) at 8 years after RT. BF was defined using the PSA-based “nadir + 2 ng/mL” definition and was derived from the cumulative incidence function, conditional on a set of disease-related patient characteristics. Specifically, the percentage of tumor cells with a Gleason pattern of 4 or 5 (%GP4/5),10 the percentage of biopsy tumor tissue that was positive for adenocarcinoma (PPT), the initial pretreatment PSA (iPSA) value, ADT duration, and RT dose were used as continuous covariates; whereas palpation T-category was used as a categorical variable. Gleason score was used as a categorical variable in the no-ADT nomogram and as a continuous variable in the ADT nomogram.
Abbreviations: %GP4/5, percentage of Gleason pattern 4 or 5 tumor cells; ADT, androgen-deprivation therapy; Gy, Gray; iPSA, initial prostate-specific antigen; PPT, percentage of biopsy tumor tissue positive for adenocarcinoma; RT, radiotherapy; SD, standard deviation.
This table displays patient characteristics according to the duration of ADT. Values are reported as mean ± SD values for continuous variables and percentages for categorical variables. Tests for differences included analyses of variance for continuous variables and chi-square tests for categorical variables.
No. of patients
Follow-up: Mean ± SD, mo
70.38 ± 34.14
73.15 ± 36.31
62.63 ± 34.55
68.49 ± 28.28
ADT duration: Mean ± SD, mo
0 ± 0
3.01 ± 1.28
13.90 ± 6.08
31.60 ± 10.31
RT dose: Mean ± SD, Gy
78.37 ± 3.63
78.43 ± 4.11
79.98 ± 3.57
80.56 ± 3.18
Patients who received <76 Gy, %
iPSA: Mean ± SD, ng/mL
7.32 ± 5.14
10.11 ± 8.63
18.94 ± 22.65
15.50 ± 16.88
Gleason score, %
Tumor category, %
PPT: Mean ± SD, %
11.38 ± 12.94
17.36 ± 18.38
25.88 ± 21.90
30.53 ± 22.99
%GP4/5: Mean ± SD %
2.04 ± 5.66
4.85 ± 9.47
12.52 ± 6.01
19.28 ± 19.39
The PPT was calculated as the average percentage of the tumor core length that was positive from each region sampled; then, the average of all region percentages was calculated. For example, if sextant biopsies were obtained and more cores were taken from 1 sextant over another, then the sextant percentage still would have the same weighting as the other sextant percentages.
In developing the nomogram, we used estimated model coefficients to assign points to each characteristic, and predictions from the model were used to map cumulative point totals to the probability of having the nadir + 2 ng/mL outcome by 8 years. We used restricted cubic splines as described by Harrell11 with 3 knots to model PSA, PPT, and %GP4/5. Because these variables had a large effect on BF, we believed it was important to use splines to allow for nonlinear effects. For Gleason score in the no-ADT model, we dichotomized the Gleason score into ≥7 versus <7 to similarly account for nonlinearity. We included interactions between all prognostic variables, including spline transformation terms, and whether an individual received any ADT. We also included main effects terms for the duration of ADT. The model, hence, allowed us to create 2 separate nomograms but also to adjust predictions based on the duration of ADT (Fig. 1). The probability of failure was determined as follows: For each parameter, the location on the horizontal line for a specific patient's values is mapped to several points by referring down to the horizontal line labeled total points. The points are then summed for each parameter to obtain the total, which is mapped to a probability of BF at 8 years.
The variables were not bounded in the regression model before fitting the models. However, for 3 variables in which there were local regions of points estimates that were nonmonotone or nearly flat, we selected bounding points at or near the local minimum/maximum, at which we believed there was no significant difference beyond the bounding point (Fig. 2).
The predictive accuracy of the model was assessed in 2 ways. First, the Harrell concordance statistic (C-statistic) for the Fine and Gray model11 was estimated using the imputation methods of Ruan and Gray.12 Second, the calibration method described by Kattan and colleagues was used.13 For each individual, we predicted the probability of each outcome at 8 years after fitting the competing-risk regression using data only from the other 2012 individuals in the data set. We then averaged the predicted probabilities from the model within quintiles defined by the magnitude of the predictions. Within each quintile of individuals, we estimated the marginal cumulative incidence of failure using the method described by Gray.14 We then plotted the marginal estimates versus the average of the model predictions.
In total, 2012 patients were analyzed. Figure 3 illustrates the unadjusted cumulative incidence BF curves divided according to the length of ADT. Of the 263 men who had BF, 56 developed distant metastases, and 237 died. Patient characteristics are detailed according to the length of ADT in Table 1. The median follow-up was 65.7 months, 66.2 months, 60.1 months, and 63 months for those who received no RT, short-term RT, intermediate-term RT, and long-term ADT, respectively (P = .0364). In addition, as ADT duration increased, T-category, Gleason score, %GP4/5, PPT and RT dose increased, whereas the percentage of patients who received <76 Gy decreased.
Figure 4 presents the results from model calibration. The model is well calibrated because the average of the predicted probabilities within each quintile is close to the estimated marginal cumulative incidence probability and falls on a 45-degree line. The Harrell C-statistic for the model was 0.76.
The nomograms allow for estimates of the results from various durations of ADT and RT dose by comparing the probabilities of failure. Table 2 provides example case parameters to illustrate the potential of the nomograms. Patients were entered into the model assuming an RT dose of 70 Gy or 80 Gy; however, other doses also could be entered. The table indicates that the probabilities are associated with a P value related to a comparison of the use of no ADT versus a specified duration of ADT. For illustrative purposes, a patient with a T1/T2 tumor, a Gleason score of 7, an iPSA of 10 ng/mL, a %GP4/5 of 5%, and a PPT of 10% (Table 2, Example Patient A) would be expected to experience a reduction in the 8-year risk of BF from 40% to 35% by increasing the dose from 70 Gy to 80 Gy. Adding 6 months of ADT reduced the BF rate to 20% and 18% for 70 Gy and 80 Gy, respectively. By comparison, a patient with the same T-category, Gleason score, and iPSA but with a %GP4/5 of 30% and a PPT of 50% (Table 2, Example Patient B) would be expected to experience a drop in the 8 year BF rate from 81% to 75% by increasing the dose from 70 Gy to 80 Gy. Adding 6 months of ADT reduced the 8-year BF rate by over half to 38% and 34% for 70 Gy and 80 Gy, respectively.
Table 2. The Estimated 8-Year Risk of Biochemical Failure for Patients With T1/T2 Tumors and Gleason Score 7 Disease Who Received 70 or 80 Gy and Various Lengths of Androgen-Deprivation Therapya
8-Year BF Rate After RT With or Without ADT, %
P: Pairwise Comparisons by Length of ADT
%GP 4/5, %
RT Dose, Gy
0 vs 3 mo
0 vs 6 mo
0 vs 12 mo
0 vs 24 mo
Abbreviations: %GP4/5, percentage of Gleason pattern 4 or 5 tumor cells; ADT, androgen-deprivation therapy; BF, biochemical failure; iPSA, initial prostate-specific antigen; PPT, percentage of biopsy tumor tissue positive for adenocarcinoma.
Various values were entered into the nomogram for each of the parameters to derive 8-year probabilities of BF for sample patients under conditions of 70 Gy versus 80 Gy and various durations of ADT. Pairwise comparisons were performed of the linear combination of coefficients that were used to generate the probabilities to estimate the statistical difference of these probabilities. The Gleason score (GS) is determined as the highest sum of GS1 and GS2.
Table 2 also provides pairwise BF comparisons for no ADT versus ADT for increasing lengths of time. In general, the P values were more significant with higher tumor burden and iPSA, as observed by the P values for Example Patients A, B, and C in Table 2. In addition, as the length of ADT was increased, the significance increased.
In Figure 5, the cumulative incidences of distant metastasis and prostate cancer death by quintile, predicted by nomograms probabilities, are presented. There are strong associations between the 8 year BF risk and distant metastasis (Gray test for cumulative incidence; P < .0001) as well as cause-specific death (P = .009).
The American Joint Commission on Cancer uses pretreatment PSA, Gleason score, and T-category to classify a patient's risk of relapse into 4 groups. Such binned risk-group associations are inherently heterogeneous. Moreover, it has been demonstrated that tumor burden parameters (eg, the percentage positive biopsies or the percentage positive tumor tissue) are significant determinants of prostate cancer outcome and are not included in the American Joint Commission on Cancer groupings. A patient who has a high Gleason score with low volume, a low T-category, and a low pretreatment PSA level may have a lower risk of BF than a patient who has an intermediate Gleason score with high volume and the same T-category and PSA. Nomograms allow for the consideration of such individual nuances. The nomogram we have devised takes this a step further by providing for estimate of the impact of treatment parameters, such as RT dose and length of ADT.
Several nomograms have been described to predict the risk of failure after prostatectomy or RT15; however, there are limitations and caveats.16 Only some of these have incorporated biopsy variables, such as the percentage positive biopsy cores (PPC) or the percentage biopsy tissue positive for adenocarcinoma (PPT). In a surgical series by Freedland et al, PPT predicted outcome to a greater degree than PSA or Gleason score in the patients who underwent radical prostatectomy.17 Although PPC has been predictive of outcome in RT series, as reported by D'Amico et al18 and others, the men in those earlier reports received low RT doses (median dose, 70.4 Gy; range, 69.3-70.4 Gy). Williams et al7 reported that PPC discriminated poorly in men who received higher doses. Our experience is that PPT is a better predictor of BF than PPC for men who receive doses >72 Gy. Therefore, PPT, and not PPC, was included as a covariate representing tumor burden in our nomogram.
Several surgical series have demonstrated that the biopsy tissue %GP4/5 is associated significantly with outcome.19 The %GP4/5 was related to biochemical and clinical outcome independent of RT dose, T-category, iPSA, Gleason score 3 + 4, or Gleason score 4 + 3 disease classification.10 On the basis of these data, %GP4/5 was included in the model. The %GP4/5 is available for most patients; the primary Gleason pattern, secondary Gleason pattern, and percentage core length involvement is all that is needed to calculate the %GP4/5.10
A strength of our analysis is that no variable selection was done; rather, all variables of interest were entered into the model and are reported as such. To our knowledge, this is the first nomogram to incorporate %GP4/5 and PPT, each of which has demonstrated independent predictive value, but neither of which has previously been entered into a nomogram.
The second important component of the nomogram is its design, which factors the impact of RT dose and the length of ADT into outcome estimates. Both RT dose and neoadjuvant/adjuvant ADT reduce long-term BF, but there are no published randomized trials incorporating both for the treatment of prostate cancer. Thus, the anticipated benefit is largely left up to the subjective impression of the treating physician. The nomogram would be a useful guide, and the Example Patients in Table 2 are illustrative. The data indicate that, in terms of reducing BF, the gains are far greater when ADT is added compared with going from 70 Gy to 80 Gy. The data from randomized trials support this relationship from the perspective that RT has rarely been shown to have a survival impact,20 whereas the addition of ADT to standard doses of RT has much more consistently been related to improvements in survival.5–9 Of note, we did not include pelvic lymph node treatment as 1 of the covariates in the model, because close to 90% of the high-risk patients received such treatment.
Another correlation that we observed by applying the nomogram to hypothetical patients was that for most, short (≤6 months) to intermediate (from >6 months to <2 years) term ADT would be favored over extending ADT to 2 years or more. There are no randomized trials comparing 1 year to 2 years of ADT directly; however, the results with 2 to 3 years of androgen deprivation have been much more consistently positive than those comparing 6 months to 1 year.
Using 8-year BF as the main endpoint allowed us to draw conclusions about long-term efficacy; however, does not consider the potential for effects of salvage ADT survival after ADT. We approached this question by examining the association between the 8-year BF risk and distant metastasis and cause-specific death (Fig. 5). Strong correlations were observed, supporting the supposition that the BF endpoint estimates translate into long-term clinical outcome consequences.
In summary, a novel dual-nomogram prediction tool was devised that provides individualized estimates of the relative effects of dose escalation and various durations of ADT. This prediction tool incorporated newer measures of tumor burden and Gleason grade to estimate the 8-year risk of BF. Such a tool would be a useful as an adjunct to physicians and patients in decision-making and also could be incorporated into the design of clinical protocol eligibility and/or stratification.
This study was supported in part by grants CA-006927 and CA-101984 from the National Cancer Institute and by grant 09-BW from the Bankhead Coley Cancer Research Program.