|Home | About | Journals | Submit | Contact Us | Français|
Clinical practice guidelines for localized prostate cancer endorse active surveillance, radical prostatectomy, external-beam radiation therapy (EBRT), and brachytherapy (BT) as alternatives which should be offered to men with clinically-localized disease [1, 2]. However, few high-quality comparative effectiveness studies exist to guide decisions among these alternatives. Recently, studies from large observational cohorts have identified differences in long-term oncologic outcomes across treatment modalities [3, 4], but randomized trials comparing treatments have not been completed. Absent consensus regarding optimal treatment, prostate cancer treatment is both preference- and supply-sensitive, and tremendous variation exists in primary management strategies . Differences across treatments in definitions of recurrence , HRQOL domains affected , and other considerations complicate efforts to compare surgical with radiation-based treatments. Substantial differences in cost also have been documented .
Decision and cost-effectiveness analyses have examined specific topics such as the utility of active surveillance  and proton-beam therapy , but no such analysis has yet addressed the larger question of relative cost-effectiveness, at various strata of disease risk, of the most commonly employed treatments—surgery vs. radiation therapy. We aimed to determine costs and quality-adjusted outcomes between surgery and radiation, including the various modalities within these two broad categories.
A four-phase literature search was conducted. In Phase 1, the published literature on local prostate cancer treatments was searched via Pubmed and yielded 7008 candidate articles. Limiting to English articles reporting on human subjects since 2002 reduced the pool to 3583, and further restricting to clinical trials (randomized or not), meta-analyses, and other explicit comparative studies yielded 988 articles. Titles and abstracts were then manually reviewed and studies were selected that reported a sample size of at least 20 men with clinically localized disease and did not combine results from different treatment modalities (e.g., BT and BT+EBRT). Meta-analyses were excluded at this stage, as were papers which were superseded by subsequent reports from the same cohort. A final set of 374 articles was thus identified at the end of Phase 1 (eTable 1).
Phases 2 and 3 of the literature search were performed concurrently. Systematic application of inclusion/ exclusion criteria specific to each clinical parameter was conducted for all articles from Phase 1 and 60 selected hand-picked manuscripts. For 3DCRT, in order to reflect contemporary practice only papers reporting results from dose-escalated series were included in the base-case analysis [11, 12]. Twenty-two cost and utility information sources were hand-selected. When duplicates were eliminated at the end of Phase 3, a total of 202 publications remained. In Phase 4, thirty additional articles were used in manuscript preparation, yielding a final set of 232 unique publications provided sources for all study data (eTable 2). The final list of references is presented in eTable 3. Probabilities for all outcomes were derived from the literature review and validated by the expert panel for the following outcomes.
A decision-analytic Markov model was developed to evaluate the clinical outcomes, quality-adjusted life years (QALYs), and lifetime costs for a hypothetical cohort of men with clinically localized (clinical stage ≤T3aN0M0) prostate cancer. Following each treatment analyzed (open radical prostatectomy [ORP], laparoscopic-assisted radical prostatectomy [LRP], robot-assisted radical prostatectomy [RARP], 3D conformal radiation therapy [3DCRT], intensity-modulated radiation therapy [IMRT], BT, and EBRT+BT), possible post-treatment health states for each one-month Markov cycle were remission, biochemical recurrence, metastasis, death from prostate cancer, and death from other causes. With each cycle, patients incurred costs, and those experiencing complications or adverse effects of treatment accrued disutilities. eFigure 1 presents the full decision tree. The analysis was stratified by clinical risk at diagnosis according to the 3-level classification endorsed by the clinical practice guideline; however, because this schema is frequently modified or adapted in various studies in the published literature, strict adherence to the risk criteria was not required for study inclusion.
Men undergoing ORP, LRP, or RARP were assigned probabilities of erectile dysfunction (ED) and incontinence at each Markov cycle (Table 1). 76% of surgery patients in biochemical recurrence were assumed to receive salvage treatment. Low-, intermediate-, and high-risk patients were 75%, 50%, and 25% likely, respectively, to receive salvage radiation and the remainder of patients received androgen deprivation therapy (ADT) alone. Salvage radiation was assumed to IMRT given with 6 months of ADT [4, 13]. Salvage radiation yielded a possibility of returning to the remission state, whereas salvage ADT alone did not. Both salvage modalities entailed costs and potential adverse effects.
The decision tree for men undergoing 3DCRT, IMRT, BT, or EBRT+BT was similar, but incontinence was replaced with grade ≥2 gastrointestinal and/or genitourinary toxicity per RTOG criteria (Table 1) . Patients receiving a treatment including EBRT were assumed not to receive concurrent ADT if they were low-risk, 50% likely to receive 6 months of treatment if they were intermediate-risk, and 75% likely to receive 18 months of treatment if they were high-risk . 25% of brachytherapy patients were assumed to receive a short course of neoadjuvant ADT for prostate downsizing . Radiation patients in recurrence likewise had the possibility of salvage and return to the remission state with surgery, or of secondary treatment with ADT alone. 44% of radiation patients were assumed to receive salvage therapy, 4% with prostatectomy and 96% with ADT only [4, 13].
Short-term outcomes (surgical complications and acute radiation toxicity) could only accrue once. ED, incontinence, and delayed radiation toxicity could persist for multiple cycles, with a probability of resolution. Perioperative mortality was assumed to be 0.2% for RRP and 0.1% for LRP and RARP . Parameter estimates for other complications and adverse events are listed in Table 1.
Over 150 different definitions of biochemical recurrence have been proposed . We included studies reporting the most common: for surgery patients, PSA ≥0.2 ng/ml with or without verification, PSA >0.3 ng/ml, or PSA ≥0.4 ng/ml, also allowing for secondary treatment to define failure. For radiation patients, we included studies reporting outcomes using the ASTRO or Phoenix definitions, two PSA rises above the nadir to at least 1.0 ng/ml, or a PSA ≥0.4 ng/ml after nadir . The parameter estimates used for biochemical recurrence derived from the literature are listed in Table 2. For both surgical and radiation, success rates for salvage local therapy in returning the patient to the remission state were 70%, 60%, and 50% for low-, intermediate-, and high-risk disease, respectively [17–21].
Biochemical recurrence is itself an important endpoint to the extent that it leads to additional testing and treatment, and causes anxiety. However, definitions of recurrence following surgery and radiation are not comparable—by the nature of their calculation, reflecting the different biological effects of radiation and surgery, the radiation definitions shift the survival curves substantially to the right, and thus may introduce bias in favour of radiation [22, 23]. Moreover, recurrence by no means uniformly predicts progression to metastasis and prostate cancer-specific mortality (CSM) .
Therefore, estimates for time to metastasis from recurrence for prostatectomy [24, 25] and radiation patients  were determined based on the literature to account for these variances. The median times used in the model for surgery and radiation patients were 10 and 6 years, respectively for low-risk patients, 8 and 4 years for intermediate-risk patients, and 6 and 2 years for high-risk patients. These times were further varied in sensitivity analyses. Time to CSM following first onset of metastasis was assumed to be 3.5 years for all patients . Mortality from non-prostate cancer causes was based on National Center for Health Statistics actuarial data . Use of ADT was assumed to increase risk of non-prostate cancer mortality by 1% annually. Radiation therapy was assumed to be associated with an annual probability of bladder or rectal cancer of 0.16% starting 5 years after treatment ; mortality from these secondary pelvic malignancies was assumed to be 12.9% annually .
Each of the health states was assigned a utility weight, determined from the literature and the Cost-Effectiveness Analysis Registry (www.cearegistry.org). These utilities, listed in Table 3, were validated by the expert panel and extensively tested in sensitivity analyses. Disutility values for short- and long-term complications of surgery or radiation were subtracted from the health state utilities. Use of ADT was also assigned a fixed disutility value. For each cycle, the final utility score was multiplied by one month and discounted by 3% annually. These quality-adjusted life-months were summed over the lifetime to determine the QALYs.
To determine costs, medical resource utilization (office visits, procedures, hospitalizations, medications, imaging and laboratory tests, etc.) was assigned to each treatment, and subsequently to each health state, reflecting complications of treatment where relevant. All services and products were described using coding taxonomies applicable to the Medicare fee-for-service payment system and validated by a certified coding expert. Costs associated with the resources were derived from the Fiscal Year 2009 National Medicare Fee Schedules and, in the case of medications, the 2009 Drug Topics Redbook. Costs were validated by clinical experts. In the case of BT+EBRT, two-thirds of the EBRT treatments was assumed to be IMRT and one-third was assumed to be 3DCRT. In either case, the cost of salvage EBRT was assumed to be two-thirds the cost of EBRT given as primary monotherapy. Costs were determined from the payer perspective; thus capital and maintenance costs for equipment were not separately included, as these are purported to be reflected in aggregate payment to providers.. However, time spent by patients in treatment and recovery was estimated by the expert panel, and indirect costs were assessed by associating these times with wage losses based on 2008 Bureau of Labor Statistics hourly rates weighted by employment status and age cohort size and inflated by 2%.
QALY outcomes and cost differences among treatments were assessed using ANOVA; adjustment for multiple comparisons across treatments was made using the Tukey test. The study employed a cost-utility analysis, in which the marginal cost for a treatment with improved outcomes is determined in terms of cost per QALY gained. In the event that one treatment was found to be dominant—that is, more efficacious and less costly—then cost-minimization analysis was utilized in lieu of cost-utility analysis.
Probabilistic Monte Carlo simulation was employed to follow hypothetical prostate cancer patients undergoing the treatment alternatives. For critical variables, parameter distributions were used rather than fixed point estimates. A normal distribution centred at age 65 was assumed for age at first treatment, triangular distributions for treatment costs, and beta distributions for utilities and biochemical failure probabilities. These are illustrated in eFigure 2. The probability distributions were sampled 250 times and 250 first-order simulations were performed with each parameter set.
An extensive set of one-way and multi-way sensitivity analyses were performed to determine the effects of varying the parameter estimates for various cost and outcome variables. Where modality- and risk-specific comparisons allowed, validation of the model-based predictions of prostate cancer death with outcomes published from two large cancer centres  were conducted. The value ranges included for sensitivity analyses are included in Tables 1–4. The analyses were performed using TreeAge Pro 2009 (TreeAge Software, Williamstown, MA).
The results from the base case analysis are presented in Table 5. The likelihood of disease recurrence, progression, and mortality increased with increasing baseline disease risk, as did associated lifetime costs. QALYs for each of the modalities studied were relatively similar within a given risk stratum, and fell with increasing levels of risk. The differences across modalities were modest but statistically significant; among low-risk patients, 3DCRT was the least effective radiation modality (10.3 QALYs), and for intermediate- and high-risk patients EBRT+BT was the most effective radiation modality (10.1 and 9.1 QALYs, respectively, p<0.001). There were no significant differences among the surgical modalities in terms of QALYs (11.3, 10.3–10.4, and 9.2–9.3 QALYs, respectively, for low-, intermediate-, and high-risk), and, in all comparisons except EBRT+BT vs. ORP for high-risk patients, the surgical alternatives were statistically significantly more effective than the radiation modalities in terms of QALYs.
As a validation test of the oncologic outcomes resulting from our model, we compared rates of CSM derived from the model for IMRT and ORP patients to those published in a large, multi-centre academic series reported by Zelefsky et al. Assuming a starting age of 60 for ORP patients and 69 for IMRT patients, as was reported in the Zelefsky et al series, CSM rates at 8 years in our model were 0.9%, 3.2%, and 8.8% for low-, intermediate-, and high-risk IMRT patients, respectively, and 0.3%, 2.0%, and 5.0% for ORP patients. These results matched closely to the published rates of 0%, 4.5%, and 9.5% for IMRT and 0%, 1.9%, and 3.8% for ORP (Figure 1) (.
As summarized in Table 5, given similar biochemical outcomes and payer and patient costs across the surgical modalities, lifetime costs were statistically and clinically similar within risk strata across the surgical modalities (approximately $20,000, $28,500, and $35,500, respectively, for low-, intermediate, and high-risk patients). Lifetime costs for radiation, conversely, varied substantially across modalities within risk strata. For low- and intermediate-risk patients, BT was less expensive than the other modalities ($25,067 and $32,553 for low- and intermediate-risk); for high-risk patients, BT ($43,952) and 3DCRT ($42,397) were both less expensive than BT+EBRT ($50,376) or IMRT ($53,539, p<0.001). Regardless of risk, the radiation modalities consistently entailed higher costs than the surgical modalities in each risk stratum (p=0.008).
We conducted an extensive set of sensitivity analyses, varying the key parameters in the model to determine which exert the most influence on the model outcomes (Table 6). Many of these analyses had no or minimal impact on the model. For example, varying the probabilities, duration, disutility penalties, and costs for functional outcomes—including incontinence, erectile dysfunction, and late radiation toxicity—had no substantial impact on the relative costs and benefits for any of the treatment modalities. Incorporating patient time costs into the model increased total costs for all modalities by roughly $4,000 to $7,000, but again had little effect on the relative costs among modalities; the same was true of varying probabilities of salvage therapy use, secondary malignancy rates and costs, other costs such as those assigned to salvage therapy, and the discount rate (including a non-discounted analysis).
In the base case, the median age for all patients was 65; if surgical patients were assumed to be younger and radiation patients older, reflecting actual practice , the differences in costs and CSM between surgical and radiation patients were reduced, but corresponding differences in QALYs and overall survival increased. Varying the costs of salvage therapy and management of biochemical failure and metastasis, as well as the probability of mortality attributable to ADT use, impacted the model for intermediate- and high-risk patients only. While varying the estimates for these parameters resulted in changes of approximately 5%–15% from the base case results, the changes were not sufficiently different by treatment modality to alter conclusions related to the relative costs and benefits of the modalities. Including reported literature on non-dose-escalated 3DCRT resulted in substantially worse survival and QALY outcomes for this modality.
Varying the time from biochemical recurrence to metastasis had the greatest impact on clinical and economic outcomes. Varying the assumption of a 4-year differential in terms of time between recurrence and metastasis had a strong effect on CSM estimates for men at intermediate- and high-risk. The CSM rates cross at 0 years differential for intermediate-risk tumours and at 1 year for high-risk tumours  (Figure 1). These changes resulted in differences in QALYs and costs as well.
Absent consensus defining optimal management of localized prostate cancer, patterns of management vary tremendously [15, 32, 33]. To date, clinical trials of intervention vs. conservative management have been completed [34, 35]. but those comparing surgery to radiation have no . One such trial has now accrued, but results will not be available for several years . In the interim, cost-effectiveness analyses may shed important light on the question of which modality or modalities offer the best value relative to cost. These analyses are notably scarce in prostate cancer, however; a recent systematic review identified only 22 studies published through 2007, compared to 86, for example, in breast cancer .
Our model found, in the context of the US reimbursement system, statistically significant but relatively modest differences among treatment modalities in terms of QALYs (Table 5). In general, surgery was preferred over radiation for lower-risk men, whereas combination EBRT+BT compared favourably for high-risk men. However, across the risk spectrum, radiation was consistently more expensive. Some treatment strategies are thus considered dominated: IMRT for low- and intermediate-risk men, for example, is no more effective than surgery or brachytherapy, and is substantially more expensive. These findings generally were robust to a wide range of sensitivity analyses. The assumption which led to the greatest change in outcome in sensitivity analysis was the differential in time from recurrence to metastasis between surgical and radiation patients. These effects were most dramatic in intermediate and high risk patients and could lead to changes in conclusions related to the relative costs and benefits of radiation and surgery for prostate cancer. Future research related to correction for different recurrence definitions is warranted.
Our findings also are consistent with other recent studies based on carefully risk-adjusted retrospective studies of prospectively collected cohorts, which have found consistent evidence for improved distal clinical outcomes following surgery compared to EBRT. A study from the community-based CaPSURE registry found a roughly 2-fold increase in CSM among men treated with a variety of radiation therapy approaches compared to surgery . The Zelefsky series likewise found a 3-fold difference in CSM comparing RRP patients to those receiving high-dose IMRT. Of note, both studies found the greatest differences among men at relatively high levels of risk, and neither included men treated with BT. Another multicenter academic series, reached similar conclusions; this study did include BT patients, whose outcomes were better in some analyses than those of EBRT patients .
Our results are also generally consistent with other recently published studies on costs and outcomes of treatment. A recent Medicare study demonstrated statistically significant but relatively modest benefits for IMRT over conventional radiation therapy in some but not all quality of life domains . Another Medicare study found that while the marginal costs of robotic compared open prostatectomy were relatively modest and declined over time through the middle part of the last decade, the costs of IMRT compared to conventional radiation were very high, and relatively stable . Neither of these studies included brachytherapy patients. Our analysis found relatively minor differences between ORRP and RARP. Indeed, a recent meta-analysis found advantages for RARP in terms of short-term perioperative outcomes and margin rates, but no large study has yet demonstrated clear advantages for either approach in terms of longer-term oncologic or quality of life outcomes . In the context of a lifetime decision analysis, any impact of short-term outcomes will generally be limited.
As described above, the absolute rates of risk-stratified CSM in our model corresponded fairly closely to those reported by Zelefsky et al . However, the relative difference in mortality between radiation and surgery patients was lower in our model than in either the Zelefsky et al study or the CaPSURE study, suggesting that our analysis is relatively conservative in its estimation of the life-year and QALY differences between the surgical and radiation modalities. Our cost assumptions are generally consistent with those recently determined by another CaPSURE study .
Several limitations to this analysis should be considered. Primary ADT monotherapy for localized disease is commonly used in practice , but outcomes of this approach in the U.S. are sparsely reported, and it is not included as a standard option in the practice guideline . Active surveillance, conversely, is rapidly gaining acceptance—including endorsement in practice guidelines —as a viable option for men with low-risk disease [43, 44], and for carefully selected men with intermediate-risk disease . A recent cost-effectiveness analysis in fact found slightly greater QALYs for surveillance compared to immediate treatment for low-risk disease . This study did not include costs, but another did find cost savings for initial surveillance over treatment, depending in part on likelihood and timing of delayed treatment among patients initially surveilled . We agree entirely that for low-risk disease active surveillance may well be preferred to any of the modalities included in this analysis. However, neither long-term oncologic outcomes nor HRQOL outcomes have been reported to date. Therefore, to avoid adding additional layers of complexity, active surveillance was included in our model, but will certainly be the subject of future modelling efforts.
Multiple assumptions underlie the model. Utilities for various post-treatment health states, for example, are based on the best available in the literature, but these have not been extensively validated. Our literature review began in 2002; thus not all studies used to derive probabilities reflected the most recent improvements in treatment modalities. Other variables, such as increased mortality attributable to ADT or secondary malignancy, are the subject of significant ongoing controversy. Fortunately, none of these factors proved to be strong determinants of overall QALYs or costs, and were tested in sensitivity analyses with only minor impacts. It is important to stress that the economic analysis was performed from the U.S. payer perspective, with the additional incorporation in sensitivity analysis of indirect patient time costs. This approach does not account for hospital investments in capital equipment, disposables, and maintenance. These costs are theoretically reflected in insurance payments, but in fact in the U.S. government and private payers reimburse at substantially higher levels for IMRT, for example, compared with 3DCRT, but do not do so for RARP vs. ORP. Particularly germane to the question of the cost-effectiveness of RARP vs. ORP, then, is the fact that the costs associated with the robotic platform which are absorbed by hospitals are not reflected.
These assumptions clearly reflect the present situation in the U.S., and will vary substantially across other health care systems. Despite these caveats, we believe that through incorporation of both QALYs and costs, consistent risk-stratification, inclusion of multiple modalities within surgery and radiation, and use of a lifetime horizon, this analysis is the most comprehensive economic analysis yet undertaken for this disease. With the exception of the time to metastasis from recurrence, the findings are robust to sensitivity analyses, and may inform future policy discussions regarding strategies to improve efficiency and reduce variation in localized prostate cancer care.
Avalere Health LLC and Veritas Health Economics Consulting were commissioned by Intuitive Surgical (Sunnyvale, CA) to perform the cost effectiveness analysis. However, the study sponsor had no role whatsoever in the collection, analysis, or interpretation of the data; in writing or approving the manuscript; or in the decision to submit for publication. Dr. Ramakrishna was compensated as a consultant to Avalere. None of the other non-Avalere/Veritas authors received any direct financial or other remuneration for their work on this study. Dr. Cooperberg’s effort was supported by National Institutes of Health/National Cancer Institute (5RC1CA146596), and by the Agency for Healthcare Research and Quality (1U01CA88160).