Two evidence-based therapies exist for the treatment of high-risk prostate cancer (PCA): external-beam radiotherapy (RT) with hormone therapy (H) (RT + H) and radical prostatectomy (S) with adjuvant radiotherapy (S + RT). Each of these strategies is associated with different rates of local control, distant metastasis (DM), and toxicity. By using decision analysis, the authors of this report compared the quality-adjusted life expectancy (QALE) between men with high-risk PCA who received RT + H versus S + RT versus a hypothetical trimodality therapy (S + RT + H).
The authors developed a Markov model to describe lifetime health states after treatment for high-risk PCA. Probabilities and utilities were extrapolated from the literature. Toxicities after radiotherapy were based on intensity-modulated radiotherapy series, and patients were exposed to risks of diabetes, cardiovascular disease, and fracture for 5 years after completing H. Deterministic and probabilistic sensitivity analyses were performed to model uncertainty in outcome rates, toxicities, and utilities.
RT + H resulted in a higher QALE compared with S + RT over a wide range of assumptions, nearly always resulting in an increase of >1 quality-adjusted life year with outcomes highly sensitive to the risk of increased all-cause mortality from H. S + RT + H typically was superior to RT + H, albeit by small margins (<0.5 quality-adjusted life year), with results sensitive to assumptions about toxicity and radiotherapy efficacy.
Prostate cancer is diagnosed in >200,000 men annually, and it is the most common malignancy among men in the United States other than non-melanoma skin cancer.1 Although the rise of prostate-specific antigen (PSA) screening has decreased the prevalence of locally advanced disease, approximately 5% to 15% of men still present with extension through the prostatic capsule, which is categorized as clinical stage T3 (cT3) disease.1, 2 Furthermore, a recent study demonstrated that almost 25% of men present with classic “high-risk” disease, which is defined as tumors with Gleason scores of 8 to 10, clinical stage T2c or greater disease, or a PSA level >20 ng/dL.3
There are 2 evidence-based treatment paradigms for managing high-risk prostate cancer: external-beam radiation therapy (RT) with adjuvant androgen-deprivation therapy (RT + H) and radical prostatectomy with adjuvant radiotherapy (S + RT), and the latter approach is reserved for men whose prostatectomy specimens reveal pathologic T3 disease or positive margins.4 These 2 therapies differ dramatically in terms of both treatment efficacy and toxicity profiles. Whereas RT + H is associated with a lower risk of distant metastases, S + RT leads to improved local control rates without the benefit of systemic therapy. In addition, quality-of-life for men who receive RT + H is dominated by the effects of hormone therapy and proctopathy, whereas the health states after prostatectomy and adjuvant radiotherapy are primarily a function of sexual and urinary toxicities. To our knowledge, no randomized trial has ever compared these 2 approaches, nor has any prospective trial described the outcome of adjuvant S + RT with concurrent, combined androgen blockade, which may maximize both local and distant disease control.
It is unlikely that any clinical trial will compare these strategies, all of which have significantly different benefits and risks and may be indicated in certain patients over others. Given the meaningful prevalence of high-risk prostate cancer, it is important to determine the optimal treatment paradigm for this population of men. Therefore, we have developed a decision model to analyze treatment outcomes, defined as quality-adjusted life expectancy (QALE), after treatment of high-risk prostate cancer with RT + H, S + RT, and S + RT + H. Our results were tested against a wide range of disease, treatment, and utility assumptions to determine the optimal treatment approaches for men with different disease and clinical characteristics.
MATERIALS AND METHODS
We designed a Markov model to simulate the clinical history of 3 hypothetical cohorts of men aged 65 years with nonmetastatic, high-risk prostate cancer.5 Cohort 1 consisted of men who received RT + H, Cohort 2 consisted of men who received S + RT, and Cohort 3 consisted of men who received trimodality therapy (S + RT + H). Patients in the trimodality arm were treated with combined androgen blockage, including a gonadotropin-releasing hormone agonist.
We assumed that this overall cohort roughly approximated men who present with PSA levels between 6.1 ng/dL and 10 ng/dL, a biopsy Gleason score between 8 and 10, and clinical stage T2b/c disease. Because it is distinctly possible that patients who present with clinical high-risk prostate cancer may have more favorable disease characteristics at prostatectomy, we assumed that 20% of patients who underwent surgery (based on the Partin tables with this assumption6) avoided all adjuvant therapy, because of favorable, pathologic T2 disease with negative margins. This situation was associated with more favorable local recurrence and toxicity outcomes. The percentage of prostatectomy-alone patients varied widely in sensitivity analyses.
The Markov simulation described in this report allows these men to transition between different health states in fixed increments of time.7 The model starts patients off in the well state (ie, with no evidence of disease [NED]), having received their definitive therapy. These states are collectively titled NED (Fig. 1). The other potential health states are local recurrence, distant metastasis, death from other causes, and death from prostate cancer. Patients arrived at the latter health state after having developed a distant metastasis.
The primary outcome of this analysis was QALE, which was established over a lifetime horizon. The cycle length was 1 month. The model was created and analyzed using Data TreeAge Pro (TreeAge Software, Inc., Williamstown, Mass).
Model Assumptions and Data
Patterns of failure and mortality
Assumptions and data sources for the decision model are provided in Table 1. All probabilities were extracted from the published literature. Outcome probabilities for RT + H were calibrated from Radiation Therapy Oncology Group (RTOG) trial 92-02, a phase 3 trial of radiotherapy with adjuvant long-term or short-term androgen deprivation.9, 23 Given the underlying disease heterogeneity in prospective studies in this high-risk population, in a secondary analysis, we also calibrated the model against European Organization for Research and Treatment (EORTC) trial 22863, in which the patients who received RT + H experienced improved local control but worsened distant control compared with the patients in RTOG 92-02.24 Outcome probabilities for S + RT were calibrated with Southwestern Oncology Group (SWOG) trial 8794, a phase 3 trial of radical prostatectomy with adjuvant radiotherapy.8
Table 1. Probabilities and Utilities Used in the Model
Abbreviations: H, androgen-deprivation therapy; NA, not applicable; NED, no evidence of disease; RT, external-beam irradiation; S, radical prostatectomy.
The local recurrence rate in patients with pathologic T2 prostate cancer who underwent prostatectomy alone was taken from a large retrospective study.10 Randomized trials have demonstrated the superiority of RT + H compared with radiotherapy alone and S + RT compared with prostatectomy alone, but the exact survival rates are highly variable. Therefore, sensitivity analyses tested a wide range of disease and treatment assumptions.
To our knowledge, outcome data for S + RT + H have not been published. To estimate the favorable effect of combined androgen blockade on local and distant failure rates after prostatectomy and adjuvant radiotherapy, the risk of local and distant recurrence for the trimodality cohort was determined by multiplying the baseline risk of recurrence (after S + RT) by the hazard ratios for hormone therapy derived from RTOG 92-02 and RTOG 86-10.9, 12
The risk of distant metastasis as the site of first failure after treatment was extracted from the results of RTOG 92-029 and was calibrated to result in survival estimates comparable to data in the reported literature. We assumed that the rate of metastasis as first failure was constant over the 10 years of follow-up. Before any therapy, we assumed that this underlying rate of distant metastases as first failure was equal in all 3 arms (ie, the identical patient population before any treatment approach), although it was reduced by hormone therapy.
We also assumed that the hazard ratio of local recurrence evolving into metastasis was constant across the time window and equal across all cohorts. On the basis of prior data, the hazard ratio of developing a metastasis after a local recurrence was 1.8913; that is, the baseline hazard rate for distant metastasis was increased by a factor of 1.89. This variable was tested in sensitivity analyses. We assumed that all deaths from prostate cancer were the result of metastasis. Patients could die from other causes after any health state based on 2006 life tables for males in the United States.25
Toxicities were defined as the likelihood of grade 2 or greater sexual, rectal, and urinary side effects, as classified by the toxicity criteria of the RTOG and EORTC.26 Patients were allowed to have any combination of these toxicities, with the assumption that the development of any toxicity was independent of the others. We assumed that all toxicities from surgery or radiotherapy occurred within 3 years of therapy; after 3 years, we assumed no onset of toxicity. We assumed that toxicities lasted throughout the remainder of the patient's life; and we assumed that baseline toxicity rates for the 2 surgery arms were equal, although the effects of hormone therapy were accounted for in the trimodality arm. Specific toxicity rates are detailed in Table 1. Of particular note, we assumed that the probability of sexual toxicity in the 2 hormone therapy arms (RT + H and S + RT + H) was 100% for the first 3 years after therapy because of the suppressive effects of androgen deprivation on libido and potency. Radiation-related toxicity rates were extracted from intensity-modulated radiotherapy-based studies.14, 15, 18
Androgen-deprivation therapy is associated with significant acute and long-term sequelae beyond changes in libido, including increased risks of diabetes, heart disease, and osteoporotic fracture.27 We modeled these complications in the hormone-containing arms by exposing patients to a risk of these 3 disease conditions for an additional 5 years after completing antiandrogen therapy.21 If patients developed 1 of these conditions, then they were considered to have complications of hormone therapy, with the decreased utility secondary to hormones, for life. Whether hormone therapy is associated with an increased risk of cardiovascular mortality and/or all-cause mortality is highly controversial28; nevertheless, we performed a sensitivity analysis on this question by assuming that hormone therapy is associated with an increase in all-cause mortality of 1.04 or 1.96 according to Nanda et al.29
Stewart et al elicited patient-perspective utilities for health states and toxicities associated with treatment using standard gambles with men aged ≥60 years who were treated for prostate cancer.22 We used their derived utilities for all health outcomes, treatment preferences, and toxicities, as detailed in Table 1. Prior data suggest that the utility function in prostate cancer is multiplicative.22 Therefore, a given patient's utility was a multiplicative function of: (baseline utility, dependent on likelihood of recurrence)*(utility because of presence of hormone therapy)*(utility based on presence of toxicities).
Sensitivity analyses allow the modeler to adjust the assumptions of the model. All parameters listed with ranges in Tables 2 and 3 were placed into 1-way sensitivity analyses. When 1 strategy was clinically more effective (ie, higher QALE) than the other, that strategy was described as “superior” to the other strategy.
Table 2. One-Way Sensitivity Analyses: Radical Prostatectomy Plus External-Beam Irradiation Versus External-Beam Irradiation Plus Androgen-Deprivation Therapy
Abbreviations: H, androgen-deprivation therapy; NED, no evidence of disease; RT, external-beam irradiation; S, radical prostatectomy.
Indicates a 0.5 to 1 QALY difference.
Indicates a <0.5 quality-adjusted life-year (QALY) difference.
Indicates a <1 quality-adjusted life-month difference.
Probabilistic sensitivity analysis is a technique in which unknown parameters are assigned a probability distribution according to prior data, and a series of Monte Carlo simulations are carried out in which the value of the unknown parameter(s) is drawn from those distributions. To determine the extent to which the model depended on health state utilities, we performed a series of probabilistic sensitivity analyses for the utility values representing health states with bowel, urinary, sexual, and hormone toxicities. The distribution for each parameter was based on a normal distribution with the mean and standard deviation values taken from Stewart et al.22
Before proceeding with the analyses, we first verified that the Markov model adequately described the disease process of high-risk prostate cancer. The model predicted 10-year local recurrence rates of 8% and 13% for S + RT and RT + H, respectively. These values are in line with the reported values of 8% and 12.3% from SWOG 8794 and RTOG 92-02, respectively.8, 9 Furthermore, the model predicted a 10-year distant metastasis rate of 17% after radiotherapy, similar to the value reported by RTOG 92-02. In fact, modeled prostate cancer-specific mortality in the RT + H arm was 16%, comparable to the 10-year disease-specific survival probability of 83.9% in the long-term follow-up of RTOG 92-02.
In the secondary analysis in which the model was calibrated with EORTC 22863, the predicted 10-year local recurrence rate after RT + H was 6.2%, and the predicted 10-year risk of distant metastasis was 48%, consistent with the trial results of 6% and 49% for these parameters.24 These outcome similarities suggested that the decision model provided a realistic emulation of the health states after treatment of high-risk prostate cancer with these modalities.
In the base case, the QALEs after treatment with RT + H, S + RT, and S + RT + H were 9.3, 8.0, and 9.5 quality-adjusted life years (QALYs). In the head-to-head comparison between primary radiotherapy and surgery plus adjuvant radiotherapy, radiotherapy with androgen-deprivation therapy was associated with an improvement of 1.3 QALYs. Conversely, when comparing RT + H with trimodality therapy, S + RT + H was only slightly superior, accruing 0.2 additional QALYs.
In the secondary validation of the model using data from EORTC 22863, the QALEs after treatment with RT + H, S + RT, and S + RT + H were 8.7, 6.4, and 8.4 QALYs, respectively. The dramatically higher risk of distant metastases in that trial emphasized the relative benefits of hormone therapy; in fact, RT + H was the optimal strategy because of the low risk of local recurrence in that arm in the EORTC study.
One-Way Sensitivity Analyses
Sensitivity analyses test the strength of a model's conclusions by varying model parameters and by comparing those results with original values. Table 2 displays the results of several sensitivity analyses when comparing RT + H with S + RT. In the majority of cases, RT + H resulted in a more than 1 QALY difference compared with S + RT. In the analysis of the lower bound of age, RT + H resulted in a >3 QALY difference over S + RT. It is important to note that the primary radiotherapy strategy was superior even when heavily penalizing patients on hormone therapy, either with significant but non-life threatening late complications or with a low utility. These results support the notion that RT + H is more effective than S + RT in nearly all clinical and treatment parameters.
In contrast, Table 3 displays the results of sensitivity analyses when comparing RT + H with trimodality therapy. In almost all cases, treatment with S + RT + H led to a higher QALE. However, the differences were small, typically <0.5 QALYs and often <1 quality-adjusted life month. In addition, RT + H was superior when its efficacy was improved. The model was particularly sensitive to toxicity, because RT + H was optimal when the risks of rectal or urinary toxicity after surgery were raised, when the impotence utility was lowered, or when the local recurrence utility was raised. Indeed, the local recurrence utility threshold at which RT + H became superior was 0.65.
Results were not particularly affected by metastasis risk and risk of death from metastasis, because hormone therapy was delivered in both treatments. Younger men particularly benefited from primary surgery, because they had more disease-free years to live after the more aggressive treatment. In total, these results suggest that trimodality therapy may be a superior therapy to RT + H, but the relative difference is small enough to be quite sensitive to toxicity rates and patient utilities.
Finally, if we assumed that hormone therapy increased all-cause mortality by 4%, then the difference between S + RT and RT + H was reduced from 1.3 QALYs to 1.2 QALYs, and the difference between RT + H and S + RT + H was stable at 0.2 QALYs. However, if we assumed a significant increase in all-cause mortality from hormone therapy (ie, because of pre-existing cardiac disease; hazard ratio, 1.96), then S + RT was the optimal strategy and was associated with a 1.2-QALE benefit over RT + H and a 0.5-QALE benefit over trimodality therapy. In fact, S + RT crossed the threshold of superiority versus the other 2 strategies when the relative increase in all-cause mortality associated with hormone therapy was 1.6 times the baseline mortality rate.
Probabilistic Sensitivity Analyses
Probabilistic sensitivity analysis allows the modeler to determine the optimal treatment strategy over a distribution of potential parameter values. It is noteworthy that the results were not changed appreciably from the baseline outcomes. The QALE comparison between RT + H and S + RT favored RT + H for each variable, with a benefit of 1.3 QALYs for all utility parameters. Similarly, the comparisons between RT + H and trimodality therapy favored S + RT + H each time, with a difference in QALE of 0.2 QALYs in each analysis.
Decision analysis provides an evidence-based method of comparing clinical outcomes after treatment with RT + H, S + RT, and S + RT + H, and sensitivity analyses enable the modeler to test these results against a wide range of assumptions, supporting or constricting the conclusions of the model. The results of our model suggest that RT + H is a superior treatment to S + RT across a wide range of outcome, toxicity, and utility assumptions but that it is distinctly sensitive to the life-threatening complications of hormone therapy. In nearly all 1-way sensitivity analyses, RT + H resulted in a QALE of >1 QALY compared with S + RT. In fact, primary radiotherapy was optimal even when assuming a significant rate of toxicity or high disutility from androgen-deprivation therapy. However, when we assumed that hormone therapy significantly increased all-cause mortality, S + RT was clearly the superior paradigm. That said, this latter scenario is more likely when the patient has significant cardiovascular comorbidities, such that it is questionable whether the individual could even tolerate a radical prostatectomy. Finally, it is notable that the probabilistic sensitivity analysis also supported the superiority of primary radiotherapy in this setting, because the average benefit of radiotherapy with androgen deprivation was 1.3 QALYs.
These results prompted us to analyze the comparative effectiveness of RT + H with a hypothetical trimodality therapy consisting of radical prostatectomy with adjuvant radiotherapy and hormone therapy (S + RT + H). In principle, this treatment approach should lead to an optimal oncologic outcome with superior local control (from S + RT) and distant control (from H). Indeed, this trimodality therapy was the superior strategy in almost all cases, most notably in younger men. Yet the absolute difference in QALE was small, typically <0.5 QALYs and often much smaller. One-way sensitivity analyses indicated that the efficacy of RT + H, the probabilities of toxicity, and the patient utilities from those toxicities drove the superior strategy in this comparison. For example, primary radiotherapy became superior when the local recurrence risk of RT + H was dropped to 6%. Whether this assumption is realistic is debatable, because there is a fair amount of uncertainty regarding the local recurrence rate after high-dose radiotherapy with androgen deprivation. Although patients on EORTC 22863 did experience this local recurrence risk (perhaps it was so low because of the high competing risk of distant metastasis in that trial), in general, it is unlikely that non-surgical strategies will lead to such a low local relapse rate: consider that, in a classic dose-escalation trial for low-risk and intermediate-risk disease, without hormone therapy, the estimated local recurrence risk was 33% in the high-dose arm.30 Furthermore, when the local recurrence utility was raised beyond 0.65, RT + H was the optimal treatment. Because, in actuality, many local recurrences are not symptomatic (although they may be associated with anxiety, which may reduce the evaluation of the health state), this particular sensitivity analysis argues that RT + H may be superior for many men. Nevertheless, the probabilistic sensitivity analyses suggested a consistent, albeit very small (0.2-QALY) benefit to using S + RT + H after adjusting for the potential spectrum of utility values.
Our finding that trimodality therapy may offer a viable alternative to the 2 evidence-based, standard-of-care therapies for advanced prostate cancer stands in some contrast to prospective randomized trials. Several studies have demonstrated that neoadjuvant hormone therapy before prostatectomy results in decreases in margin positivity and the risk of progression but produce no difference in survival.31-33 However, in our current analysis, the hypothetical patient has a significantly higher risk of both local and distant recurrence than in those studies, such that the synergy of hormone therapy with adjuvant radiotherapy may increase local control and may improve metastasis-free survival. Of course, a major drawback to such aggressive treatment is increased toxicity, because the patient is exposed to potential complications from all 3 modalities.28, 29 However, our model did account for this increased toxicity.
The nature of our model-based analysis introduces several limitations. First, our assumptions were based entirely on published data, although we strove to include only widely accepted values from a variety of trials. Nevertheless, a decision analysis cannot replace a properly performed randomized controlled trial that actually treats patients with the study disease and can account for unknown confounders. That said, sensitivity analyses were used to investigate the validity of model assumptions, and our model proved to be robust. For example, although the survival and patterns-of-failure data were modeled on well established but slightly older cooperative group studies, we varied these probabilities over a wide range of potential values—accounting for significantly improved or reduced outcomes over time—and the results were essentially identical.
Moreover, we intentionally biased the model against RT + H in some cases, such as assuming 100% impotence in the 3 years after radiotherapy. Furthermore, the impotence utility, which was set at 0.89 in the baseline model, is quite possibly an overestimate, which would heavily favor the surgical arms.34-36 Similarly, by assuming a very low utility for the local recurrence health state (more frequent after RT + H treatment), we further biased the results against primary radiotherapy. It must also be noted that the recurrence rates for the trimodality therapy, which was superior to primary radiotherapy in most cases, were derived empirically by assuming that the efficacy of androgen-deprivation therapy in the adjuvant radiotherapy setting is equivalent to that in the definitive scenario. These values, although logically sound, are not based on level 1 evidence; hence, toxicity and recurrence rates may be higher or lower than assumed. Finally, we did not explicitly account for the anxiety prompted by a rise in PSA without clinical symptoms; however, we have attempted to account for this limitation by assuming a utility decrement for the NED health state in the first 10 years after treatment based on the report by Stewart et al.22 This health state includes the underlying anxiety of living with the uncertainty of a treated cancer that may recur.
In summary, we have demonstrated that primary radiotherapy with concurrent androgen deprivation may be superior to surgery and adjuvant radiotherapy for high-risk prostate cancer. However, this result is tightly related to the complications of hormone therapy; men whose mortality may be affected by it may benefit from surgery and adjuvant radiotherapy alone. Furthermore, trimodality therapy, consisting of radical prostatectomy followed by adjuvant radiation treatment and hormone therapy, actually may lead to superior outcomes in some patient populations, particularly among younger men. Indeed, although there are limits to the extent to which decision analysis can support or reject a clinical treatment paradigm, the current analysis argues against the use of radical prostatectomy and adjuvant radiotherapy alone for otherwise healthy men. Conversely, there does appear to be validity in further testing a strategy involving all 3 treatment modalities, which may optimize both local and distant control.
Whether or not this latter, more aggressive approach is successful for a given patient will depend on several factors that are patient-dependent, focusing primarily on the risk of toxicity and the patient's utility if he experiences that toxicity. Furthermore, a formal evaluation of the likelihood of serious complications from hormone therapy may be indicated before making the decision to deliver any therapy, because such complications may define the best treatment approach for the patient. To maximize quality-adjusted survival for a given man with high-risk prostate cancer, such tiny absolute differences in QALE between treatment approaches mandate careful elicitation of patient preferences on the potential health states after therapy.