Presented at the 86th annual meeting of the American Radium Society, Napa, California, May 1–5, 2004.
In Radiation Therapy Oncology Group (RTOG) trial 92-02, after men received neoadjuvant hormone cytoreduction and radiotherapy for locally advanced prostate carcinoma, they were randomized to receive either 2 years of long-term androgen-deprivation (LTAD) or no further treatment (short-term androgen-deprivation [STAD]). The specific objective of the current study was to determine whether LTAD was a cost-effective treatment for patients with locally advanced prostate carcinoma.
The cost-effectiveness of LTAD was tested using a Markov model that was designed using proprietary software. The analysis took a payor's perspective. Unit costs were obtained by estimation using a global Medicare fee schedule. Costs and outcomes were discounted by 3%. Distributions were sampled at random from the treatment utilities, transition probabilities, and costs using a second-order Monte Carlo simulation technique.
The expected mean cost was $32,564 for LTAD compared with $33,039 for STAD after accounting for the additional cost of salvage treatment for men who were treated with STAD. The mean number of quality-adjusted life years (QALYs) for men who received LTAD was 4.13 QALYs compared with a mean of 3.68 QALYs for men who received STAD. The cost-effectiveness acceptability curve analysis showed a 91% probability that LTAD was cost-effective compared with STAD. Although overall survival was similar in the LTAD and STAD groups, the patients who received LTAD experienced gains in QALYs and had lower costs, because LTAD prevented biochemical failure and the necessitating salvage hormone therapy.
The advantage of long-term hormone therapy in selected high-risk patients with adenocarcinoma of the prostate has been documented in randomized trials by the Radiation Therapy Oncology Group (RTOG) and the European Organization for the Research and Treatment of Cancer (EORTC).1–5 One group reported that patients who had clinical T3 disease or evidence of regional lymph node involvement and who were treated with indefinite hormone therapy had decreased local failures and distant metastases, improved biochemical control (defined as a prostate-specific antigen [PSA] level < 1.5 ng/mL), and improved absolute survival compared with men who were given goserelin at the time they developed recurrent disease.6 This improvement, however, comes with increased clinical and economic cost. Patients who received long-term androgen-deprivation (LTAD) reportedly had increased toxicity. Bolla et al. reported a significantly higher rate of Grade 1–3 incontinence (19% vs. 16%; P = 0.002) with the addition of LTAD compared with radiation therapy (RT) alone.4 Patients who received LTAD on RTOG 9202 also reported greater late gastrointestinal (GI) toxicity compared with patients who received short-term androgen-deprivation (STAD).3 Patients who received both hormone suppression and RT also reportedly had lower utilities compared with patients who received RT alone.7 Some of the reasons for the lower utilities in men who received androgen suppression include the above-mentioned factors as well as sexual dysfunction and hot flashes. The specific objective of the current study was to evaluate, by means of a Markov model, the cost-effectiveness of LTAD in the treatment of men with locally advanced prostate carcinoma.
MATERIAL AND METHODS
Randomized Trial 9202
RTOG 9202 was a prospective, randomized trial that compared STAD plus RT with LTAD plus RT in the treatment of men with histologically confirmed adenocarcinoma of the prostate (clinical T2C–T4 tumors according to the 1992 American Joint Committee on Cancer tumor, lymph node, metastasis [TNM] system). The men in the study had a Karnofsky performance status (KPS) of ≥ 70 and pretreatment PSA levels < 150 ng/mL. The RT was similar in both arms: Patients were treated using a 4-field technique with megavoltage (MV) energy (> 6 MV). The dose was prescribed to the center of the prostate with the prostate receiving 65.0–70.0 Gray (Gy) for patients with T2C tumors and 67.5–70.0 Gy for patients with T3–T4 tumors at 1.8–2.0 Gy per day. Regional lymph nodes received 44.0–46.0 Gy in fractions of 1.8–2.0 Gy.
All patients received flutamide (Eulexin; Schering Plough, Kenilworth, NJ) at a dose of 250 mg orally 3 times daily with monthly goserelin acetate (Zoladex; AstraZeneca Pharmaceuticals, Wilmington, DE) at a dose of 3.6 mg subcutaneously beginning 2 months prior to RT and continuing until the RT was completed. Patients who were randomized to the STAD arm received no further treatment, whereas patients who were randomized to the LTAD arm received monthly subcutaneous injections of goserelin 3.6 mg for an additional 2 years after the completion of RT.
The Markov model was set up with the following states: no disease progression, hormone-responsive disease progression, hormone-unresponsive disease progression, and death. This model is similar to the model used by Hummel et al. and Fleming et al. to evaluate different treatments for prostate carcinoma.8, 9 The allowable state transitions were from no disease progression, to hormone-responsive disease progression, to hormone-unresponsive disease progression, to death. The transitions from states of no disease progression to states of hormone-responsive and hormone-nonresponsive disease progression were biochemical determinations. Patients were allowed only forward transitions. Patients spent 1 year in each state before transitioning to another state or staying in the same state. At the completion of therapy, the base case was set up for all patients to enter the no disease progression state. Patients could then either enter a state of hormone-responsive disease progression or stay in the state of no disease progression. From the state of hormone-responsive disease progression, patients could stay in the state of hormone-responsive disease progression, or they could transition to the state of hormone-unresponsive disease progressive, in which they remained for 1 year before dying.
Yearly transition probability estimates for state transitions, assuming constant rates, were calculated from rates obtained from the results of RTOG trial 92-02 in the literature by using the following equation: annual rate = [− ln(1 − P)/n], in which P is the probability of the occurrence of interest, and n is the number of years the rate is measured.3 The Markov termination limit was 10 years. It was found that patients in the LTAD arm who had a PSA failure had a higher death rate from prostate carcinoma compared with patients in the STAD arm. The death rates for patients who failed after androgen deprivation, in either the LTAD arm or the STAD arm, were used to calculate transition probabilities from the state of hormone-responsive disease progression to hormone-unresponsive disease progression.10 The annual probability of the event was calculated by using the following formula: annual probability = 1 − exp (− annual rate). DataPro decision-analysis software (TreeAge Software, Inc. Williamstown, MA) was used to analyze the Markov model. Patients who received LTAD had a greater chance of having bowel toxicity, which could decrease the quality of life and utilities. A sensitivity analysis that included the probability of RTOG Grade 3 bowel toxicity was included in the analysis. Patients who experienced a GI toxicity were modeled to have a decrease in utility of 0.05 for the period of time spent with GI toxicity.
Probabilistic sensitivity analysis was performed using a Monte-Carlo simulation. Patient utilities and costs were sampled from the distributions described below. The distributions were sampled once per patient, and each simulation had 1000 trials. One-way and two-way sensitivity analyses were performed on the cost and utility values. Benefits were discounted by 3% per year.
Cost and Economic Assumptions
Table 1 lists the economic and utility assumptions that were used to inform the model. We assumed a modified payor's perspective (i.e., Medicare). RT costs were based on global billing to account for the technical and professional components and were based on modeling of the data. RTOG has used modeling in the past to estimate cost for treatment and has found that modeling can approximate the true cost of treatment for complex treatments.11 The clinical trial was reviewed, and resource-utilization procedures (e.g., treatment simulations and treatments) were identified and quantified. The procedures were matched to the relative Current Procedural Terminology codes and to the appropriate Relative Value Units (RVU) to obtain unit costs. The relative Resource-Based RVU (RBRVU) or unit costs were totaled and multiplied by a conversion factor, $34.592/RVU, to arrive at a dollar amount for each arm of the clinical trial. The two treatment arms differed only in their use of total androgen suppression (TAS); therefore, the cost of RT was the same in both treatment arms. The costs of hormones, goserelin, and flutamide, were calculated based on average wholesale prices obtained from the Drug Red Book. We assumed a $100 administration fee for the goserelin. Costs were discounted at 3% per year in keeping with the Panel on Cost-Effectiveness in Health and Medicine.12 The mean cost of all therapies, including chemotherapy, for the last year of life for a patient with metastatic prostate carcinoma was obtained from the literature and was estimated at $24,000.13 We assumed that all other costs of care were equal across both arms of the study.
Table 1. Economic and Utility Assumptions Used to Inform the Model
AD: androgen deprivation; RT: radiotherapy.
Initial cost of AD and RT
Hormone treatment after biochemical failure
Cost of treatment in the last yr of life
RT and AD
Costs were sampled at random using a second-order Monte-Carlo simulation. The modeled costs were sampled by using a normal distribution with the modeled calculated cost used as the mean and a standard deviation that was 60% of the modeled calculated cost.
No utility values were available from trial participants. Utilities for the hormone treatment, and RT hormone levels were obtained directly from the current health of 95 patients with prostate carcinoma who had received these treatment modalities at some point in the past.7 Note that, although these patients had received the treatments at some point, not all were receiving them at the time that utilities were elicited. Utility decrements for dysfunction were small (0.08–0.14 [sexual], 0.06–0.13 [urinary], and 0.01–0.13 [bowel]), as reported by Krahn et al.,7 who reported more sexual, urinary, and bowel dysfunction associated with a lower utility score and reflecting a poorer quality of life. The utility for the last year of life while patients were receiving chemotherapy and other supportive therapy was obtained from the literature and was estimated at 0.42.14
Distributions were sampled for each utility at random from the treatment utilities by using a second-order Monte-Carlo simulation technique. Utilities were modeled using a β distribution. This produced a distribution for incremental cost-effectiveness that was used to construct 95% uncertainty intervals and cost-effectiveness acceptability curves.
Distributions were sampled for each state transition probability at random from the treatment probabilities by using a second-order Monte-Carlo simulation technique for overall survival and disease progression and were obtained from the results of the previously published RTOG trial 92-02.3, 10 State transition probabilities were modeled by using a β distribution and were calculated from the formulas discussed above. Patients who received STAD only had a 28.1% 5-year disease-free survival rate compared with 46.4% for patients who received LTAD. No difference was noted in overall survival between the STAD group and the LTAD group. Patients who exhibited a biochemical failure after LTAD had a worse progression-free survival compared with patients who initially received STAD.10 Table 2 lists the transition probabilities between health states for the 2 groups of patients. The dying transition probabilities reflect dying of all causes. Although these transition probabilities differ, they are not different statistically from one another.
Table 2. Transition Probabilities Used to Inform the Model
A separate analysis was performed on the subgroup of patients with Gleason scores ≥ 8. The values of the variables used to inform the model are listed in Table 3. Patients who were treated with LTAD in this subgroup had improved overall survival compared with patients who were treated with STAD.3
Table 3. Transition Probabilities in Patients with Gleason Score ≥ 8 Used to Inform the Model
The expected mean cost for patients who were treated with LTAD was $32,564, compared with $33,059 for patients who were treated with STAD. The mean number of quality-adjusted life years (QALYs) for patients who were treated with LTAD was 4.13 QALYs compared with 3.68 QALYs in the STAD group. In this analysis, patients who were treated with LTAD dominated patients who were treated with STAD, because the comparator, LTAD, was less costly and provided greater QALYs than the standard treatment, STAD. The cost-effective acceptability curve showed that LTAD had a 91% probability of being cost-effective at the $50,000 per QALY, willingness-to-pay level. Although it was costlier up-front, LTAD prevented subsequent failures and the need for salvage hormone therapy. Figure 1 shows an incremental cost-effectiveness scatterplot that compares STAD with LTAD. Nearly all of the data points reside in Quadrant IV, with LTAD more effective and less costly compared with STAD.
Patients in the high-risk group (based on Gleason scores ≥ 8) who were treated with LTAD had an expected mean cost of $31,820 with mean of 4.16 QALYs compared with $28,688 and a mean of 3.48 QALYs. The cost-effective ratio comparing LTAD with STAD was $4605 per QALY. The cost-effective acceptability curve showed that LTAD would have a 95.6% probability of being cost-effective at the $50,000 per QALY, willingness-to-pay level. The proportion of patients who died in the STAD group was greater than the proportion in the LTAD group, resulting in greater mean QALYs for the LTAD group, which may be responsible for the higher costs associated with LTAD treatment.
A sensitivity analysis was performed adjusting several variables in the model. We varied the length of the analysis to determine whether the finding of cost-effectiveness was dependent on the time frame of the analysis. LTAD still was considered cost-effective, even when the Markov termination condition was only 5 years. The mean expected cost for LTAD was $19,634 with a mean of 2.57 QALYs compared with $18,229 and 2.53 QALYs for STAD. This resulted in an incremental cost-effective ratio of $35,125 per QALY. Even with shorter follow-up, LTAD is considered cost-effective compared with STAD, although it did not dominate STAD when the Markov termination condition was set at 10 years, and the ratio is very close to the $50,000 per QALY limit, above which it would not be considered cost-effective. LTAD dominated STAD (i.e., was less costly and resulted in increased QALYs compared with STAD) when the Markov termination condition was set at 15 years. LTAD was less costly compared with STAD in the long run, because it prevented biochemical failure, which would necessitate salvage hormone therapy.
A sensitivity analysis also was performed to address the assumption that costs were distributed normally. The real distribution of the cost of care was not known, because costs were modeled. Cost also may be distributed in a γ distribution. LTAD still was cost-effective compared with STAD when a γ distribution was used for cost.
Another sensitivity analysis was performed to assess the impact of GI toxicity in patients who received LTAD. It was found that patients in the LTAD arm had increased GI toxicity. A Markov transition state was created with the utility of patients experiencing GI toxicity decreased by 0.05–0.15 in patients who transitioned into this state. At a Markov termination condition of 5 years, the mean number of QALYs was decreased by 0.01 year in patients who received LTAD when a decrease in utility was added for GI toxicity. This decrease in utility for patients who received LTAD increased the cost-effectiveness ratio to $47,333 per QALY from $35,125 per QALY, but the ratio remained below the cost-effectiveness ceiling of $50,000 per QALY. The results remained unchanged at Markov termination conditions of 10 years and 15 years, regardless of the decrease in utility caused by GI toxicity. The median utility of patients undergoing RT was 0.74.7 We believe we were justified in decreasing the utility of patients experiencing GI toxicity by only 0.1, because Saigal et al. previously reported that patients who experienced radiation proctitis had a median and mean utility of 0.75 and 0.6, respectively.15
A threshold analysis also was performed varying all of the probabilities and utilities within clinically relevant ranges. LTAD still was the favored treatment, except when the probability of disease progression in the observed patients receiving LTAD was increased above 0.22.
If a clinical trial was performed with 2000 patients distributed equally between LTAD and STAD, LTAD would prevent a hormone failure in 258 patients. Four hundred forty-nine patients receiving LTAD would not have had a hormone failure compared with only 191 patients receiving STAD.
LTAD still was considered cost-effective when the costs of hormones and chemotherapy were changed in the sensitivity analysis. Varying the utilities of the hormones also did not change the overall finding that LTAD was cost-effective in the sensitivity analysis.
The RTOG and EORTC have found that the addition of LTAD to RT improves disease-free survival in patients with locally advanced prostate carcinoma.1, 3, 4 The addition of hormone therapy is associated with increased economic and quality-of-life costs. Although it is relatively minimal, the addition of LTAD can increase the cost of treatment by approximately $6800 per year. Neymark et al. published the results of the economic analysis of EORTC 22863. LTAD increased the cost per year by 14,395 in 1998 French francs in EORTC trial 22863.16 Increased side effects also were reported in addition to the increased cost, which would decrease quality of life and quality-adjusted survival. Approximately 6% of patients withdrew before the completion of the planned 3 years of hormone therapy were given in the EORTC trial because of toxic side effects or patient's wishes.5 Similarly, in RTOG trial 92-02, a statistically significantly higher rate of side effects was noted in patients who received LTAD compared with patients who received STAD.3 Increased late toxicity in patients who received androgen deprivation for the treatment of prostate carcinoma also was reported by others.17, 18
The increases in toxicity and cost must be counterbalanced by an increase in incremental quality-adjusted survival. An incremental increase in quality-adjusted survival does occur to justify the increase in cost and toxicity experienced by patients who receive LTAD. The cost-effectiveness acceptability curve shows a very high probability that LTAD will be cost-effective at all willingness-to-pay ceilings. In the current sensitivity analysis, the benefit was more pronounced as the time frame of the analysis grew longer. This most likely represents the prevention of progressive disease, which would require salvage hormone therapy. LTAD was cost-effective even at a relatively short horizon of 5 years.
Sandler et al. recently reported a lower hormone salvage rate among patients who received LTAD on RTOG trial 92-02 compared with patients who received STAD.10 From the results of the current study, we hypothesize that patients who initially received hormone therapy may have had disease that became resistant to subsequent hormone therapy administered at the time of a biochemical failure. However, LTAD still was cost-effective, even when this higher progression rate was entered into the model.
The sensitivity analysis failed to reveal any variables that, when altered, resulted in LTAD not being cost-effective. The use of orchiectomy is less expensive compared with monthly hormone injections, but the effects are permanent.19 The decision to choose either orchiectomy or medical castration for the treatment of advanced prostate carcinoma can be difficult. Clark et al. surveyed 201 men who underwent orchiectomy or medical castration for the treatment of metastatic prostate carcinoma. The majority of those men were happy with their decision; however, of the 23% of men who expressed regret about their decision, the majority had undergone orchiectomy.20 The reason for that regret was not a result of the treatment but of the actual process of deciding on a treatment and the treatment itself. Men with regret were much less likely to state that they were satisfied with the role they played in the treatment decision process, and they believed that orchiectomy was the treatment for a “massive problem,” whereas the injections were less definitive.20 Men who choose orchiectomy eliminate the need for monthly injections, whereas others prefer to “just get castrated, and get it over with.”20
GI complications have been associated with the use of LTAD and RT in the treatment of locally advanced prostate carcinoma. To our knowledge, the exact etiology for the increased GI toxicity has not been identified. One hypothesis is that the use of androgen deprivation causes the prostate to shrink, resulting in a larger volume of rectum being in the RT treatment field; however, in RTOG trial 92-02, both sets of patients had shrinkage of the prostate. Others have hypothesized that the adjuvant hormones slow the reparative process of the irradiated rectum, thereby increasing the susceptibility to develop a late rectal injury.18 We reduced the utility of patients experiencing rectal toxicity in our model, and LTAD still was found to be cost-effective, even when the utility of the hormone therapy was decreased to 0.6. We did not include the potential for increased cost for the treatment or prevention of osteoporosis in patients who received LTAD. LTAD may increase the risk of osteoporosis, decreasing the QALYs and increasing the cost of medical care if the patient sustains a fracture.21–23
The results of the current study compare favorably and confirm the work of Neymark et al., who published the economic analysis of EORTC trial 22863, which evaluated LTAD in a similar group of patients.16 Unlike our analysis, those authors obtained actual cost data from 90 patients who were treated on the trial at Centre Hospitalier Universitaire (Grenoble, France). We used actual clinical data to build our model but modeled the cost involved in the treatment.
Despite its increased cost and toxicity, the current results indicate that LTAD was cost-effective in the patient population studied, confirming the work of the EORTC. The increases in cost and the added toxicity are justified, because QALYs are increased by the prevention of future biochemical failures.