Many men receive external beam radiotherapy (EBRT) with curative intent for clinically localized prostate cancer. In an effort to give meaningful guidance to these men, several models have been developed to guide therapy and predict outcome. Most of these models have been based on standard pretreatment prognostic factors, including Gleason score, tumor stage (T classification), and, when available, prostate-specific antigen (PSA). In addition, treatment-related variables, such as radiation dose and androgen-deprivation therapy (ADT), have been added to predictive models. These pretreatment and treatment-related factors have been well studied and are the focus of this review. In addition to these standard predictors of outcome, several biomarkers have been studied in combination with radiotherapy.1-3 A comprehensive review of this topic is beyond the scope of this review, but the text below summarizes work published (at the time of this writing) by the Radiation Therapy Oncology Group (RTOG) focusing on several potential prognostic factors. The latter part of this review is focused on the “state of the art” of predictive models that have been based on pretreatment factors widely available outside of the research setting.
Predictive models are being used increasingly in effort to allow physician and patient expectations to be aligned with outcomes that are based on available data. Most predictive models for men who receive external beam radiotherapy for clinically localized prostate cancer are based on Gleason score, clinical tumor classification, and prostate-specific antigen (PSA) levels. More sophisticated models also have been developed that incorporate treatment-related variables, such as the dose of radiation and the use of androgen-deprivation therapy. Most of the predictive models applied to prostate cancer were derived using PSA recurrence rates as the major endpoint, but clinical endpoints have been incorporated increasingly into predictive models. Biomarkers also are increasingly being added to predictive models in an effort to strengthen them. The Radiation Therapy Oncology Group (RTOG) has completed studies on a wide range of markers using tissue from 2 phase 3 trials (RTOG 8610 and 9202). To date, preliminary assessments of p53; DNA ploidy; p16/retinoblastoma 1 protein; Ki-67; mouse double-minute p53 binding protein homolog; Bcl-2/Bcl-2–associated X protein; cytosine, adenine, and guanine repeats; cyclooxygenase-2; signal transducer and activator of transcription 3; cytochrome P450 3A4; and protein kinase A have been completed. Although they are not ready for widespread, routine use, there are reasons to believe that future models will combine these markers with traditional pretreatment and treatment-related variables and will improve our ability to predict outcome and select the optimal treatment. Cancer 2009;115(13 suppl):3112–20. © 2009 American Cancer Society.
Biomarkers in Radiotherapy for Prostate Cancer: Contributions From the RTOG Trials
The RTOG can be credited with having performed the most extensive studies of biomarkers in men with clinically localized prostate cancer who received EBRT on phase 3 randomized trials. A partial list of those studies is provided in Table 1. The source of the tissues used for the studies came from 2 phase 3 trials: RTOG 8610 and RTOG 9202. A total of 11 markers were studied in 15 articles between 1997 and 2008.
|Marker||Reference(s)||RTOG 8610||RTOG 9202|
|Distant Metastases||Cause-specific Survival||Overall Survival||Distant Metastases||Cause-specific Survival||Overall Survival|
|p53*||Grignon 1997,4 Che 20075||+||+||+*||+||+||+ (Among patients who received STADT)|
|DNA ploidy||Pollack 20033||−||Not reported||+||NYA||NYA||NYA|
|P16/pRB†||Chakravarti 2003,6 Chakravarti 20077||+||+||+/− (Borderline significance; P = .07)||+||+||−‡|
|Ki-67||Li 2004,8 Pollack 20049||+||+||− (70% vs 55%; P = .17)||+||+||+‡|
|MDM2||Khor 200510||+/− (Borderline significance; P = .06)||−||−||Work in progress||Work in progress||Work in progress|
|Bcl-2/Bax||Khor 2006,11 Khor 200612||−||−||−||−||−||−|
|Androgen receptor CAG repeats||Abdel-Wahab 200713||−||−||−||NYA||NYA||NYA|
|Stat3||Torres-Roca 200715||+||− (Inadequate sample size?)||− (Inadequate sample size?)||NYA||NYA||NYA|
|PKA||Khor 200817||+||+/− (Borderline significance; P = .08)||−||NYA||NYA||NYA|
The first biomarker evaluated by the RTOG was p53. The study population consisted of a subset of men entered on RTOG 8610 who received EBRT with or without combined ADT.4 This subset consisted of 129 of the 471 patients (27%) who entered the trial for whom there was sufficient tumor material for analysis. Abnormal p53 protein expression was detected in 23 tumors (18%). Statistically significant associations were noted between abnormal p53 protein expression and an increased risk of distant metastases (P = .04), a decreased probability of progression-free survival (P = .03), and a reduction in overall survival (P = .02). Patients who received EBRT with ADT who had tumors that exhibited abnormal p53 protein expression had a reduced time to the development of metastases (P = .001); however, this difference was not observed in the patients who received EBRT alone. On the basis of that preliminary analysis, the RTOG carefully evaluated this endpoint for patients who were treated on RTOG 9202.
Tumor tissue sufficient for analysis of p53 status was available from 777 patients on RTOG 9202.5 Abnormal p53 was noted in 22% of these patients and was associated with cause-specific survival (P = .014) and the risk of distant metastasis (P = .013). In the subgroup of patients who received short-term ADT, there was a correlation noted between p53 status and cause-specific survival (P = .004). When these patients were divided into subgroups according to p53 status, only the subgroup of patients with abnormal p53 were found to have a significant association between the assigned treatment and cause-specific survival (P < .01). Unfortunately, because all patients on RTOG 9202 received ADT, the question of a possible unfavorable interaction between abnormal p53 expression and the use of short-term ADT compared with EBRT alone (raised in the previous analysis based on RTOG 8610) could not be resolved.
DNA ploidy was evaluated using a cohort of 149 men (33% of the total number) who were treated on RTOG 8610.3 In that cohort, 50% of men received EBRT alone, and 50% received EBRT with short-term ADT. DNA ploidy was independent of other pretreatment variables. The 5-year overall survival rate was 70% for the men with diploid tumors and 42% for the men with nondiploid tumors. Although having a nondiploid tumor was associated with a reduced overall survival, there was no correlation with distant metastasis. The investigators postulated that patients with nondiploid tumors might be less responsive to salvage ADT and that the use of short-term ADT might not be advisable for patients with nondiploid tumors.
On the basis of a subset of patients treated on RTOG 8610, loss of p16 expression was associated with an increased risk of local failure, distant metastasis, and disease-specific survival (P < .01, P < .03, and P < .01, respectively), and there was a borderline association with overall survival (P = .07).6 In an analysis of 612 patients from RTOG 9202, reduced expression was associated with an increased rate of distant metastases (P < .04).7 Among patients with high immunostaining for p16, the use of long-term ADT was associated with an increase in cause-specific survival and a decreased incidence of distant metastasis compared with short-term ADT (P < .001 and P < .01, respectively). These results suggest that p16 expression may be associated with increased hormone sensitivity.
Ki-67 was the next marker evaluated by the RTOG using tissue from RTOG 8610. Diagnostic material from 108 patients was available for Ki-67 analysis from 60 patients who received EBRT alone and from 48 patients who received EBRT with short-term ADT.8 A Ki-67 staining index ≤3.5% and >3.5% was associated with a 5-year risk of distant metastasis of 13.5% and 50.8%, respectively (P = .0005); a 5-year risk of disease-specific survival of 97.3% and 67.7%, respectively (P = .0039); and a 5-year overall survival rate of 70% and 55%, respectively (P = .17). These trends were confirmed in multivariate analyses. An additional assessment of Ki-67 also was made using tissue from 537 patients who were treated on RTOG 9202.9 When it was analyzed as a continuous variable, the Ki-67 staining index was found to be associated with the risk of distant metastasis (P < .0001), disease-specific survival (P < .0001), and overall survival (P < .01), and it was the most significant predictor of the first 2 endpoints. On the basis of a subset analysis, the authors hypothesized that there may be a subgroup of patients that does not require long-term ADT. This observation suggests that Ki-67 may be useful in selecting patients for short-term ADT and in stratifying patients placed on future trials.
The RTOG also evaluated the association between mouse double-minute p53 binding protein homolog (MDM2) expression (an oncoprotein that promotes p53 degradation) and outcomes using tissue from RTOG 8610.10 Adequate archival diagnostic tissue specimens from 108 patients with MDM2 overexpression had >5% nuclear staining in 47 specimens (44%). In multivariate analysis, there was trend toward a correlation with the risk of distant metastasis at 5 years (P = .06). A more comprehensive and promising study combining MDM2 with Ki-67 based on patients who were treated on RTOG 9202 is near completion (but the results are not yet available). The results indicate that MDM2 compliments Ki-67, that it is a much stronger predictor of outcome than p53, and that the combination of MDM2 and Ki-67 has promise in identifying men at a particularly high risk of distant metastases.
Bcl-2 and Bcl-2–associated X protein (Bax) expression levels also were evaluated using tissue from RTOG 8610.11 Suitable diagnostic tissue was available from 119 patients (26%) for the Bcl-2 analysis and from 104 patients (23%) for the Bax analysis. Bcl-2 overexpression was observed in 26% of patients (n = 30), and abnormal Bax expression was observed in 47% of patients (n = 49). On the basis of these analyses, neither protein was related to outcome. A follow-up investigation was performed using tissue from 502 patients for Bcl-2 and from 343 patients for Bax who were treated on RTOG 9202.12 Bcl-2 was positive in 45.6% specimens, and Bax expression was altered in 53.9% of specimens. The combination of negative Bcl-2 and normal Bax expression was related to reduced biochemical failure (P = .036), particularly among those who received short-term ADT, suggesting that long-term ADT may be advised when either Bcl-2 or Bax is expressed abnormally.
A quantitative assessment of cytosine, adenine, and guanine (CAG) base pair repeats on the androgen receptor gene was also performed on a subset of patients who were treated on RTOG 8610.13 CAG repeats were measured in 94 tumor specimens and did not significantly influence local control, the risk of metastasis, cause-specific survival, or overall survival; however, patients with short repeats who received short-term ADT did appear to have a higher local control rate.
Cyclooxygenase-2 (Cox-2) expression also was evaluated using tissue from 586 men who were treated on RTOG 9202 who had sufficient tissue for immunohistochemical staining and image analyses.14 The intensity of Cox-2 staining was predictive of the risk of distant metastasis (P = .0004) and the risk of biochemical failure using both the American Society of Therapeutic Radiology and Oncology (ASTRO) definitions and the Phoenix definitions (P = .008 and P = .014, respectively), particularly among those who received short-term ADT.
Signal transducer and activator of transcription 3 (Stat-3) expression was evaluated in a subset of 62 patients who had sufficient tissue from RTOG 8610.15 Activated STAT3 was correlated inversely with the development of distant metastasis (P = .04) but not with survival or local control; although, because of the small sample size, this conclusion has to be interpreted with caution.
Polymorphisms in the androgen receptor cytochrome P450 3A4 (CYP3A4) were evaluated for a subset of patients who were treated on RTOG 9202 to understand how variation polymorphisms in CYP3A4 correlated with outcomes and race.16 Tissue specimens from 56 men of African American origin and from 54 of European-American patients were studied. There was a strong association between race and CYP3A4 polymorphisms: Seventy-five percent of European-American men had the wild type compared with only 25% of African-American men (P < .0001), but there was no association between the CYP3A4 polymorphisms that we studied and outcomes.
Archival diagnostic tissue samples from 80 patients who were treated on RTOG 8610 were used to study the predictive value of protein kinase A RI-alpha (PKA) expression.17 On multivariate analyses, there was a correlation between overexpression and increased biochemical failure (P = .03), increased distant metastasis (P = .018), and a trend toward increased cause-specific mortality (P = .08). The analysis of PKA expression and patient outcome from RTOG 9202 is nearing completion and validates the findings from RTOG 8610.
Several conclusions can be reached based on these hypothesis-generating studies. Despite these interesting observations, taken as a whole, the major conclusion remains. Biomarkers as major predictors of outcome are not yet ready for routine use in clinical practice. However, the relations observed are promising. A model is being developed that incorporates the key markers identified in patients from RTOG 9202; preliminary evidence indicates that such a model will add significantly to classic clinical, pathologic, and treatment-related covariates in predicting distant metastasis at 10 years. The preliminary findings of these studies support the need for validation studies (which are in progress). The most robust observations are related to the clinical relevance of p16, Ki-67, MDM2, Cox-2, and PKA, as observed in Table 1, and most of these markers appear to be useful for identifying subsets of patients who do not appear to benefit from long-term ADT.
Predictive Models Using Standard Pretreatment Clinical Features
Predictive models for men with clinically localized prostate cancer have evolved over the last 10 years. Ross et al reported >40 models that could be used to predict various outcomes for patients with prostate cancer.18 Those authors included 6 models involving pretreatment variables that may be useful for predicting PSA recurrences after EBRT.18 The models were based on data from retrospective and prospective studies. These algorithms generally have been based on mathematical constructs19, 20 or simple additive models.21, 22 Some models have helped to provide a rationale for determining what volumes should be selected for irradiation.23, 24 In addition to the variety of data sources and the methods used to create predictive models, these models also have varied in their endpoints. Most predictive models are designed to predict PSA failure (typically at 5 years), whereas a few have been used to predict survival.19-22, 25 More recently, surrogate endpoints, such as PSA doubling time and the time to distant metastasis, have been considered in an attempt to shorten the time required to reach clinical endpoints that have a closer association with survival.26, 27
The shortfalls of each of these models have become more apparent as we have accumulated empirical data and learned more about prostate cancer. For example, it is now known that early nomograms that were used to determine the risk of lymph node involvement most likely were inaccurate. In these early series, the type of lymph node dissection was not extensive enough to identify all of the lymph nodes that were involved.23, 28-30 Early predictive models also tended to assume that the impact of a Gleason score of 7 was the same as a PSA of, for example, 10 ng/mL and 20 ng/mL and ignored the impact of having multiple adverse prognostic features and the impact of the percentage of positive biopsies.21, 22 Subsequent updates of these models addressed some but not all of these issues.31
Kattan et al have contributed mightily to the area of predictive models with many predictive nomograms that encompass a variety of treatment options, including 3-dimensional conformal EBRT.19 The nomograms have been based on >1000 patients and have used a combination of statistical methods, including 8 prediction techniques and Cox proportional hazard regression methods, an artificial neural network, and recursive partitioning, and were validated later. Despite the eloquence of this nomogram and its wide-scale adoption, it is clear now that there are severe limitations to its use. Among the shortcomings of the first Kattan EBRT nomogram is that it excludes consideration of several factors that have demonstrated of prognostic significance. For example, this nomogram ignores the potential impact of whole-pelvic radiotherapy and assumes that the benefits of neoadjuvant ADT are uniform across prognostic subgroups.32 This early nomogram by Kattan et al also is problematic because of the relatively short follow-up, which compromises the accuracy of predictions when based on the ASTRO consensus definition of ADT.19, 33, 34 It also has been established that the use of the ASTRO definition is inappropriate for men who receive ADT.33, 34
An additional concern is that the nomogram appears to be most accurate for the “average patient.” When applied to patients at extremes, it seems to render predictions that do not jive with clinical experience. For example, Figure 1 provides 2 such examples. In the first example (Fig. 1a), a patient has a PSA of 4 ng/mL, clinical T3c disease, a Gleason score of 10, and receives 72 grays (Gy) along with short-term ADT. The nomogram predicts a 5-year control rate of 70% (Fig. 1a). In fact, such patients usually fare very poorly and are expected to have a very high recurrence rate (and mortality rate). In the second example (Fig. 1b), a patient with a low T-classification and a low Gleason score who receives low-dose radiation but no hormone therapy but has a high PSA (25 ng/mL) is predicted to have a worse outcome. In fact, however, his survival would be expected to be longer than that of the patient in the first example. These conclusions are highlighted by the stark contrast between the weighting for this nomogram and a more recent nomogram that was correlated with the time to metastasis and death (see below)27, 35 The more recent nomogram is predictive of the time to metastatic disease after EBRT, as illustrated in Figure 2. This more clinically relevant nomogram has been validated and also is predictive of survival (Fig. 2).27, 35
Note, however, that the 2 nomograms are weighted very differently, although they were created by the same investigators. They include similar variables and presumably similar (or the same) patients as the first nomogram. This means that the true significance of a high PSA (eg, 25 ng/mL) and a high Gleason score (eg, 10) discussed above are reversed when metastasis and survival are the primary endpoints (as noted earlier) (Fig. 3).
The first predictive model for overall and disease-specific survival after EBRT for localized prostate cancer was published by the RTOG in 2000, as illustrated in Figure 4.25 This model was based on data from what to our knowledge are the largest prospective phase 3 trials evaluated to date and also was useful for defining who should be prescribed ADT and whether it should be prescribed short term or long term.36 This model subsequently was validated using multi-institutional data and is useful in first elucidating the impact of pretreatment PSA on overall and cause-specific survival.37
With additional follow-up, it has been demonstrated that other models, including the popular risk-stratification schemes proposed by D'Amico et al, can predict survival.31 Because of these and other similar data, a modification of the current American Joint Commission for Cancer staging system has been proposed.38 This revised staging system is shown in Table 2 and appears to be a relatively simple way for the American Joint Commission for Cancer system to be improved by incorporating PSA, Gleason score, and T classification in a way that reflects current practice habits.
|Tx||Tumor cannot be assessed|
|T0||No evidence of tumor|
|Tis||Carcinoma in situ (PIN)|
|New stage I||T1-T2, and GS ≤6, and PSA <10 ng/mL|
|New stage II||(T1-T2, and GS ≤6, and PSA <10-20 ng/mL) or (T1-T2, and GS 7, and PSA <20 ng/mL)|
|Stage IIA||T1-T2, and GS ≤6, and PSA from 10 ng/mL to <20 ng/mL|
|Stage IIB||T1-T2, and GS 7, and PSA <20 ng/mL|
|New stage III||(T1-T2, and GS ≤6, and PSA ≥20 ng/mL), or (T1-T2, and GS 7, and PSA ≥20 ng/mL), or (T1-T2 and GS 8-10 or clinical T3 disease)|
|Stage IIIA||(T1-T2, and GS ≤6, and PSA ≥20 ng/mL) or (T1-T2, and GS 8-10, and PSA <20 ng/mL)|
|Stage IIIB||T1-T2, and GS ≥7, and PSA ≥20 ng/mL|
|Stage IIIC||Clinical T3, seminal vesicle or bladder neck invasion|
|Lymph node status|
|Nx||Lymph nodes cannot be assessed|
|N0||No regional lymph node involved|
|N1||Metastases, regional lymph node(s)|
By using the nomogram for predicting the probability of metastatic disease after EBRT developed by Kattan et al, several investigators have reported the ability to predict overall and cause-specific survival that improved upon the RTOG risk groups, as illustrated in Figure 4. The major value of this model is that it can be used not only to predict the time to metastatic disease but also as a surrogate for disease-specific and, ultimately, overall survival, thus potentially shortening the time-critical endpoints in the setting of clinical trials.27, 35
Despite the major progress that has been made over the past 10 years or more, to our knowledge none of the models proposed to date completely address the complex array of factors that contribute to the outcome of men who are treated with EBRT. For example, in addition to the biomarkers discussed earlier (as well others not covered here), it has been demonstrated that socioeconomic factors, such as Karnofsky performance status and age, also have an impact on patient outcome.39 Thus, with time, it is likely that far more comprehensive and accurate models will be made available and will allow us to improve our ability to predict outcomes for men who receive EBRT for clinically localized prostate cancer.
Conflict of Interest Disclosures
Sponsored by ASTRA Zeneca and the European School of Oncology.
Supported by RTOG U10 CA21661, Community Clinical Oncology Program U10 CA37422, and Stat U10 CA32115 grants from the National Cancer Institute.