Perspective of the Analysis

For whom is a CE analysis conducted? The answer to this question determines the perspective of the analysis. Perspective describes the point of view from which the study is conducted-for example, it might be the patient, the provider, the payer, or the developer of a medical technology. The perspective of the analysis determines which costs are included. An analysis conducted from the perspective of an individual patient might incorporate only the costs that the patient bears directly, such as a 20 percent copayment. An analysis conducted from the perspective of a government program like Medicare would include Medicare payments but ignore deductibles and copayments and other expenses patients bear directly (out-of-pocket payments), as well as payments made on behalf of patients by family members or private supplemental insurance. Neither of these perspectives incorporates all of the resources used to treat the patient. The most widely used and best accepted perspective for conducting CE analysis is the societal perspective, which includes all costs. ²

Specifying the Alternatives

Every CE analysis implicitly or explicitly compares at least two alternatives. The first alternative, and usually the one that motivates the analysis, is the intervention under study. Because cost-effectiveness is relative, the cost-effectiveness of any intervention depends on the alternative to which it is compared. One possible alternative is to "do nothing"; another is standard therapy or an older, but well-studied therapy. "Do nothing" is only an appropriate alternative if it is used commonly or is otherwise justifiable. In the current study we have not included a "do nothing" alternative because androgen suppression is universally regarded as beneficial for patients with advanced prostate cancer. We have included older therapies (such as DES) that have fallen out of favor but are well studied.

After the intervention and the alternative(s) have been selected, the numerator of the cost-effectiveness ratio is calculated by subtracting the cost of the alternative from the cost of the intervention; thus the numerator is the difference in costs between the two interventions, not simply the cost of the intervention under primary consideration. Similarly, the denominator of the cost-effectiveness ratio is the difference between the health effects that result from using the intervention and the health effects with the alternative. In short, a cost-effectiveness ratio is always incremental: the numerator consists of the amount by which health care costs of the patient receiving the intervention exceed the costs of a patient receiving the alternative (incremental costs), and the denominator consists of the amount by which the health outcome expected if a patient is treated with the intervention exceeds the health outcome expected under the alternative (incremental effectiveness).

Measuring Health Effects

To be useful, the measure of health effects or outcomes in a CE analysis must accurately quantify the impact of the interventions on all aspects of health that patients value. Occasionally CE analyses use specialized, highly circumscribed measures of health effects. For example, the cost-effectiveness ratio might be reported as the cost per positive test or the cost per unit change in blood pressure. Such special-purpose effectiveness measures are of limited value unless they are directly linked to "final outcomes" like life expectancy. The effectiveness measure should be sufficiently global to make it possible to compare different kinds of health interventions and sufficiently sensitive to capture changes in health that lead to an appreciable improvement in a patient's sense of well-being.

The most widely used global health outcome measure has been life expectancy, which is readily measured, easily interpreted, and immediately comparable across a wide range of interventions. As a measure of health outcome, though, life expectancy has a significant weakness: it is not designed to detect changes in the quality of life. Thus it fails to quantify the benefits of interventions whose major purpose is pain relief or improvement in functional status. Potential improvement of quality of life is an important reason to treat advanced prostate cancer, which can be highly symptomatic.

The outcome measure that is usually considered most desirable, even though it has not been applied as often as life expectancy, is quality-adjusted life years (QALYs) or, as it is sometimes called, quality-adjusted life expectancy. As its name suggests, this measure is analogous to life expectancy. In calculating life expectancy, each year that an individual lives longer contributes an additional year to life expectancy. In calculating QALYs, each year of additional life does not necessarily add an equal amount. Life years marred by functional limitations, pain, and other burdens associated with illness receive less weight than years in good health. In calculating life expectancy, each additional year receives a weight of 1, but in calculating QALYs, each additional year receives a preference weight of between 0 (for years when health is so bad that it is considered no better than death) and 1 (usually defined as best health imaginable).

Interventions can raise QALYs either by lengthening life or by improving its quality. Similarly, an intervention that lengthens life produces more QALYs if it maintains or improves quality of life than if it adds years of life that are impaired by significant morbidity.

Technical Note

Life expectancy for the general population, adjusted for age, sex, and race, is usually calculated from life tables. Equivalent information for clinically defined groups, such as patients with heart disease, can often be obtained from observational studies or randomized clinical trials. Life expectancy for a 50-year-old woman is the sum of the probabilities that she is alive at age 51, age 52, age 53, and so on. In general, the formula for life expectancy is the sum of the probabilities that an individual will be alive at each age (denoted by i) in the future, up to the maximum life span, or

where F_i is the probability that the person who is now at the "current age" will still be alive at age i.

Quality-adjusted life years are calculated in the same way as life expectancy, with the exceptions that QALYs give less weight to years of life that will occur in the distant future than to years of life that will occur soon (time discounting), and they give less credit to years of life with substantial morbidity than to years of perfect health. To calculate QALYs, it is necessary first to measure "utility" or "preference weights." The preference weight for the health characterizing age i, denoted by qi, can range in value from 0 (for the worst state of health, usually assumed to be death or its equivalent) to 1 (corresponding to "perfect" health). Each such term is the expected value of quality adjustments for all possible states of health at age i. To illustrate the calculation, imagine that individuals alive at age 60 could be in one of only two possible states of health: perfect health, (qh = 1), occurring with probability 0.5, or suffering from heart disease (qd = 0.8), also occurring with probability 0.5. Then q ₆₀, the expected value of the preference weight corresponding to being alive at age 60, is (0.5 X 1) + (0.5 X 0.8) = 0.9. After estimating the value of q_i for each age i, it is possible to calculate the expected number of QALYs according to the formula:

where is a time discount factor whose value is between 0 and 1. As in the formula for life expectancy, F_i is the probability that the person is still alive at age i. If = 1, two years of life in which q_i = 0.33 contribute the same number of QALYs as one year in which q_i = 0.66. If there is no time discounting ( = 1) and if each year of life has perfect health, or quality adjustment is ignored (q_i = 1 for every value of i), this formula reduces to the formula for life expectancy.

Measuring Costs

Most CE analyses use direct costs, or the costs of preventing, diagnosing, and treating health conditions, as the principal measure of resources consumed as a result of choosing a particular health intervention. Indirect costs, or the value of lost wages due to morbidity or premature death, are ordinarily excluded (Gold, Seigel, Russell et al., 1996). Costs include the immediate costs of an intervention (and its alternative) as well as future costs generated by the treatment along with savings due to disease prevented. ³ Thus a full accounting of the costs of a hypertension screening program, for use in a CE analysis, would include the costs of administering the screening test(s), costs of additional diagnostic procedures and treatment administered to individuals found to have hypertension, costs of treatment complications; subtracted from these costs would be the savings resulting from preventing strokes, myocardial infarction, and other complications of hypertension.

In most circumstances, the specific measure of costs that is appropriate to use in CE analysis is the difference between the marginal cost of applying the intervention and the marginal cost of the alternative. Marginal cost refers to the cost of producing an added (small) quantity of a good or service. Usually marginal costs cannot be obtained directly from available accounting data, or from data on payments, reimbursements, or charges for health care. Consequently, CE studies use a wide variety of cost figures. ⁴ For example, available cost measures are often designed to capture average costs, which may not be appropriate because they include both marginal costs and a fraction of the fixed costs of producing health services. There is no consensus about the best practically achievable measure of costs to use. In markets that exhibit competitive behavior, such as competing health plans bidding for hospital or physician services in a large metropolitan market, the price paid for services may approximate marginal costs.

Discounting

Most CE analyses place less weight on future costs and health effects than on immediate effects. The rationale for placing less weight on dollar costs that will be incurred in the future is familiar to anyone who has ever borrowed money or received interest on savings; a dollar today is worth more than a dollar tomorrow. If you borrow $5,000 dollars from a bank today, promising to repay the $5,000 dollars (the principal) and the accumulated interest in 7 years, by the end of the loan period, you might owe twice as much as you borrowed. Similarly, if you know that you will have a large expense in the future (retirement or a child's college costs), you can add to your savings today and let the money accumulate over time. When the time comes to spend the money, you will have the amount you saved as well as accumulated interest on the savings; thus future dollars are appropriately discounted relative to current dollars.

Future health effects are similarly discounted, although the precise rate of discount to use is controversial. In this analysis, we follow the recommendations of the Panel on Cost-Effectiveness in Health and Medicine that health effects be discounted at the same rate as costs (Gold, Seigel, Russell et al., 1996). When resources are being spent to increase health, failing to discount health effects that occur in the future or using a different rate of time discount for health effects than for monetary costs leads to inconsistent and paradoxical policy implications. Note here that we are focusing on societal rather than individual discount rates.

What discount rate should be used? For the base case analysis, we followed the recommendations of the Panel on Cost-Effectiveness in Health and Medicine (Gold, Seigel, Russell et al., 1996) and applied a discount rate of 3 percent to both costs and health effects. This value was estimated as the most appropriate for costs when data on real economic growth and the real consumption rate of interest was taken into account. As outlined above, we applied the same rate to health effects.

Sensitivity Analysis

Aspects of nearly every CE analysis are subject to uncertainty. The magnitude of a treatment effect may not be known with precision, the cost of a test or a drug may be unknown or may vary substantially, and the frequency of an uncommon but severe long-term adverse effect may be highly uncertain. Cost-effectiveness analyses usually include a sensitivity analysis to learn whether the results of the study are sensitive to (i.e., change substantially with) variation in the values of any of several uncertain numbers that are included in the analysis. For example, the frequency of an adverse effect of treatment may have little impact on the cost-effectiveness ratio of a screening test, particularly if the screening test seldom leads to use of the treatment; hence, even if the rate at which adverse effects occur may not be known with certainty, within a plausible range of values adverse effects have little influence on the principal conclusions of the analysis. This use of the term sensitivity is distinct from the sensitivity of a diagnostic test, or the likelihood that the test result will be positive in a patient with the disease. Sensitivity analyses can be used to gauge the likely validity of a CE analysis and to point to areas in which additional data that are more precise are needed.

Objectives of the Cost-Effectiveness Analysis

The CE analysis seeks to address the following questions:

• What is the most cost-effective drug within each class of antiandrogens? What is the most cost-effective androgen suppression therapy, when outcomes are measured in terms of life expectancy? What is the most cost-effective androgen suppression therapy, when outcomes are measured in quality-adjusted life years? How do the cost-effectiveness estimates change when assumptions about the efficacy of antiandrogen strategies change? Are the cost-effectiveness estimates sensitive to variation in the values of other uncertain numbers that are included in the analysis? For patients presenting with regional metastatic prostate cancer, what is the cost-effectiveness of early (when prostate cancer is diagnosed) compared to late (when disease progression occurs or symptomatic distant metastases develop) antiandrogen therapy? How might the cost-effectiveness estimates change if clinical strategies incorporating biochemical markers are included?

Markov Modeling

To model the time sequence of disease states and survival, along with associated costs, we constructed a semi-Markov model of advanced prostate cancer (Beck and Pauker, 1983) (Figure 13). Semi-Markov models are decision trees that include Markov nodes, branching points in the tree that lead into a Markov process, a type of state-transition model. In a Markov process, both specific health states and the possible ways to change health states (known as transitions) are defined. The prognosis of the patient (or cohort) in the analysis is thus described by the health states, the permissible transitions between states, and the rates of transition. In this model, we use a generalization of Markov processes in which some transition rates between states are not fixed but rather change with time ⁵ (Sonnenberg and Beck, 1993). We next describe the typical patient in the model (the base case), the health states, and the transition rates. A summary of model assumptions is presented in Table 20.

Figure

Figure 13. A Schematic of the Markov Model. At baseline, patients have local progression of disease. Death from other causes can occur in any cycle. Patients receiving androgen suppression therapy can have adverse effects that are fatal, (more...)

Table

Table 20. Summary of Model Assumptions.

The Base Case

The model follows a hypothetical cohort of patients with advanced prostate cancer over the course of their remaining lifetimes. The base case is a 65-year-old man with localized recurrence of prostate cancer but no distant metastases who is treated with antiandrogen therapy. We do not model the clinical course of a patient with prostate cancer untreated with antiandrogen therapy, a clinically unacceptable scenario. ⁶ The model's time horizon is 20 years, at which point virtually all patients have died of prostate cancer or other causes. In this model, decisions to initiate or change therapy are based on clinical events only. Decisions based on biochemical progression (such as a rising prostate specific antigen) and optimal timing of antiandrogen therapy are analyzed separately.

The Health States and Transitions

At any time, patients in the model are classified into one of several discrete health states, each with an associated cost and quality of life. The model starts with five basic health conditions associated with advanced prostate cancer (Roman numerals refer to the states as depicted in Figure 14): local recurrence indicating disease progression (I), asymptomatic distant metastases (II), symptomatic distant metastases (III), death from prostate cancer (IV), or death from other causes (V). Patients with prostate cancer may remain stable (Transitions 2, 4, and 6 in Figure 14) or experience disease progression. Progression occurs in a fixed sequence from having local recurrence to asymptomatic distant metastases (Transition 1) to symptomatic distant metastases (Transition 3) to death from prostate cancer (Transition 5). ⁷ At any time, patients may also die of causes other than prostate cancer (Transition 7). The cycle length for the Markov model (the interval at which transitions between health states occur) is 1 month. Further details of the health states and transitions are outlined below (see "Details of Health States in the Model" and "Details of Transitions Between Health States").

Figure

Figure 14. State-cycle Diagram of the Markov Cycle. Shown are the basic states in the model. Patients start in the state "Local Recurrence" (State I). If they experience disease progression, they first develop "Asymptomatic Distant Metastases" (State (more...)

Events Prior to Entry into the Model

The base case pertains to a patient previously diagnosed with localized (clinical stage A or B) and moderately differentiated (Gleason stage 5 or 6) prostate cancer. The patient previously received definitive local treatment with either radiotherapy or radical prostatectomy. Although the choice of initial therapy may affect the chances of developing recurrence, the model builds from an assumption that the choice does not affect the course of disease after recurrence, particularly the response to hormonal therapy. We do not include patients who opted for watchful waiting when initially diagnosed. We further assume that any short-term use of neoadjuvant hormonal therapy during the initial treatment for prostate cancer does not affect responses to antiandrogen therapy for recurrent disease.

Strategies in the Model

We evaluate four antiandrogen strategies: bilateral orchiectomy, an LHRH agonist (leuprolide, goserelin, and buserelin), diethylstilbestrol (DES), and nonsteroidal antiandrogens (NSAA, flutamide, bicalutamide, and nilutamide). We first present an analysis of the choice of LHRH agonist and of nonsteroidal antiandrogen. We then present an analysis of choice between agents, including combined androgen blockade. Not included in this model is a strategy of intermittent androgen suppression.

Further Assumptions of the Natural History Model

We assume that patients would initially have hormone-sensitive disease, manifest by a response to androgen suppression. We based this assumption on the observation that mutations in the androgen receptor gene that confer "hormone resistance" are rare in prostate specimens from patients who have not received prior androgen therapy (Bubley and Balk, 1996). However, patients eventually develop hormone-resistant disease (Garnick, 1997). We assume that new metastatic disease defines the development of hormone resistance. Further, all deaths attributable to prostate cancer occur in patients whose disease is no longer amenable to hormone manipulation.

Details of Health States in the Model

Recall the five basic health conditions in the model (see Figure 14): local recurrence indicating disease progression (I), asymptomatic distant metastases (II), symptomatic distant metastases (III), death from prostate cancer (IV), or death from other causes (V). We assume that patients with local recurrence (stage I) are treated with hormone suppression and that when asymptomatic distant metastases (stage II) develop, the prostate cancer has developed hormone resistance. Alternatively, if hormone therapy is initiated when patients have asymptomatic distant metastases (stage II), hormone resistance develops when patients progress to symptomatic distant metastases (stage III). Similarly, if hormone therapy is initiated when patients have symptomatic distant metastases, they will have an initial period during which their disease is hormone sensitive before hormone resistance develops (not shown in Figure 14).

Health states are further defined according to the antiandrogen therapy used (first-line, second-line, or none) and adverse effects from treatment (present or absent). Note that not all combinations exist; for example, patients stop active therapy when they develop asymptomatic distant metastases. A complete listing of the health states in the base case model follows:

• Local recurrence, active therapy
• Local recurrence, active therapy with minor adverse effects
• Local recurrence, second-line therapy
• Local recurrence, no therapy
• Asymptomatic distant metastases, second-line therapy
• Asymptomatic distant metastases, no therapy
• Symptomatic distant metastases, second-line therapy
• Symptomatic distant metastases, no therapy
• Dead from prostate cancer
• Dead from other causes

In addition, the following states are included when combined androgen blockade strategies are considered. We have included states defined by the use of an alternative active therapy and by drug withdrawal (see "Drug Sequence"):

• Local recurrence, alternative active therapy
• Local recurrence, alternative active therapy with minor adverse effects
• Local recurrence, drug withdrawal
• Asymptomatic distant metastases, alternative active therapy
• Asymptomatic distant metastases, alternative active therapy with minor adverse effects
• Asymptomatic distant metastases, drug withdrawal
• Symptomatic distant metastases, alternative active therapy
• Symptomatic distant metastases, alternative active therapy with minor adverse effects
• Symptomatic distant metastases, drug withdrawal

Note that two clinical events, the occurrence of severe adverse effects and the development of local obstructive symptoms, are not modeled as separate states. Rather, the costs and quality of life effects associated with these events are captured in the transitions between states.

Details of Transitions Between Health States

Transitions between health states could occur at the end of each cycle and depend on the biological behavior of the cancer and the response to treatment (see Figures 13 and 14). Using the declining exponential approximation of life expectancy and assuming a constant transition rate from one state to the next, the transition probability is calculated by the formula p = 1-exp (-rate*t), where t represents the cycle length (Miller and Homan, 1994). Note that this assumption assumes that no patients are cured of prostate cancer, consistent with clinical data that the treatment of advanced prostate cancer is palliative (Garnick, 1997). The transition rate can be approximated by the reciprocal of the mean occupancy time in each state. However, when a substantial risk of competing illness exists, such approximations may be inaccurate and alternative modeling techniques are required (Black, Nease, and Welch,1996). To refine the transition probability estimates, we developed a simple transition model that incorporated mortality related to other causes as well as prostate cancer mortality.

We next describe the possible transitions between health states. We first focus on the transitions between the health conditions associated with prostate cancer (Table 21), then we detail transitions between health states as defined by the occurrence of local obstruction and adverse effects.

Table

Table 21. Summary of Transition Rates Under Different Scenarios.

Events Occurring During Transitions

We modeled two clinical events that do not define separate states but whose impact is captured in the transition between states (Figure 15). First, if a patient develops severe adverse effects necessitating discontinuation of medical antiandrogen therapy, he incurred a cost and a transient loss in quality of life. Further details about adverse effects modeling are presented below. Second, if a patient develops local obstructive symptoms, he also incurred a one-time added cost and a transient decrease in quality of life.

Figure

Figure 15. Details of Transitions Between States. Five illustrative states are shown: local recurrence, active therapy (corresponding to State I in Figure 14); asymptomatic distant metastases, second-line therapy (State II in Figure (more...)

To clarify these possible transitions, we first present an example. Figure 15 details the possible transitions from the state "local recurrence, active therapy" for a patient treated with monotherapy. If patients experience disease progression, they transition into the state "asymptomatic distant metastases, second-line therapy" (transition A in Figure 15). Note that patients may experience local obstruction, a severe adverse effect, or a minor adverse effect in the same cycle as they experience disease progression. Note also that patients start second-line therapy after disease progression (see "Changes in Antiandrogen Therapy").

Patients may also experience local obstruction without developing distant metastases (transition B), possibly with an adverse effect. We assumed that patients would change medications after developing local obstructive symptoms. Similarly, if patients develop severe adverse effects from therapy, they start second-line therapy (transition C). Patients who experience only mild adverse effects continue with their medications; hence they move to the state "local recurrence, adverse effect" (transition D). If patients experience neither obstructive symptoms nor any adverse effects, they remain in their initial state (transition E). Finally, patients may die of causes other than prostate cancer (transition F). To understand the model fully, it may be useful to directly compare Figure 15 to Figure 14. Transition A corresponds to transition 1 in Figure 14. Transitions B, C, D, and E correspond to transition 2 in Figure 14 (indicating that several events may still occur even though patients remain in the same health condition), and transition F corresponds to transition 7 in Figure 14.

Note that we assumed that these were independent events. For example, a patient may move from the state "local recurrence, active therapy" to the state "asymptomatic distant metastases, no therapy" because two events have occurred-he has developed disease progression and he has experienced a severe adverse effect. To continue with the example, let us suppose that this patient also had local obstruction, as in transition A in Figure 15. The probability of this transition is the product of the probability of developing asymptomatic distant metastases, the probability of experiencing a severe adverse effect, and the probability of having an episode of local obstruction.

Changes in Antiandrogen Therapy

We assumed that patients would change therapy for either of two reasons: they experienced a severe, dose-limiting adverse effect, or clinical signs or symptoms indicated that their prostate cancer was no longer sensitive to hormonal intervention. In the model, patients switched therapies if they developed distant metastases, if asymptomatic distant metastases became symptomatic, or if they developed local obstructive symptoms.

Adverse Effects

The frequency, type, and severity of associated toxicities differ among treatment strategies. Our analysis distinguishes among fatal adverse effects (such as fulminant hepatic failure from nonsteroidal antiandrogens), severe adverse effects that require discontinuation of treatment (such as angina from diethylstilbestrol), and bothersome but tolerable minor adverse effects (such as hot flushes with all treatments). We assume that fatal adverse effects can occur at any time. We estimate that the rate of fatal hepatic failure with NSAAs is 0.0003/100 patient-year (Wysowski and Fourcroy, 1996). The rate of excess cardiac death from 5 mg DES is 0.03/patient-year and from 1 mg DES is 0.01/patient year (Byar, 1973; Byar and Corle, 1988).

We assume that all the drugs can cause severe adverse effects, but they occur only in the first month of treatment. We assume that severe adverse effects are associated with a one time incremental cost and decrease in quality of life. Minor adverse effects, however, are associated with an incremental cost and loss of quality of life that last for the duration of drug use. Since DES is associated with thromboembolic and cardiovascular complications, we conduct a sensitivity analysis to test the assumption that severe adverse effects with DES are associated with higher costs and morbidity than other agents. In the base case, we assume that the risk of developing adverse effects with bicalutamide is independent of the risk of developing adverse effects to nilutamide. To estimate adverse effect rates, we pool the incidence of adverse effects across studies included in the meta-analysis that reported toxicities. The pooled estimates of drug withdrawal for each agent are shown in Table 22 .

Table

Table 22. Frequency of Stopping Antiandrogen Medication Due to Adverse Effects.

We assume that minor adverse effects were associated with all agents, consisting primarily of effects of androgen deprivation such as hot flushes, and occurring with equal frequency (55 percent) in all strategies.

Efficacy of Antiandrogen Therapy

We assume that all interventions are as efficacious as orchiectomy in delaying disease progression when patients are on active therapy. We distinguish between efficacy and effectiveness. ⁸ We use efficacy in a narrow sense to indicate only how the various strategies influence disease transition rates (the impact of the therapies on the natural history of prostate cancer). In contrast, we use effectiveness as a broad term that incorporates not only effects of drugs on disease progression, but also effects on mortality from other causes (such as with DES), differing toxicities, and differing effects on quality-of-life rates (the impact of the therapies on the net health benefits).

The estimates of the efficacy of combined androgen blockade deserve special attention for several reasons. For example, these estimates are based on clinical trials in which this strategy may have been used much as monotherapies are typically used. However, combined androgen blockade including a nonsteroidal antiandrogen has two potential advantages over monotherapies that may not be captured in the meta-analysis: first, patients who are intolerant of one agent can start another, and second, a brief window exists after drug withdrawal during which the risk of disease progression is decreased. We repeated our analysis with and without incorporating these advantages.

We used several different methods to estimate the efficacy of combined androgen blockade. For the base case, we assumed that combined androgen blockade is not more effective than monotherapies because the current meta-analysis of 2-year survival (grouped by specific strategy) suggested that there was no significant difference between strategies. Nevertheless, the point estimates from the meta-analysis did suggest a trend toward decreased disease progression with combined androgen blockade. Therefore, we repeated our analysis using these efficacy estimates.

Cost Estimates

Because we use the societal perspective to estimate costs, all costs associated with the treatment strategy, regardless of who bears them, are considered. As in other contexts, the costs may vary across settings. For example, the price a pharmacy pays to acquire is one "cost," whereas the amount it charges an individual patient is another, and the price paid by a health plan or pharmacy benefits manager to a manufacturer is another. There may be geographic variation in each of these measures of drug costs. Although there is some ambiguity about which of these costs is the appropriate measure to use, and all may be at best rough proxies for the conceptually preferred measure of long-run marginal cost (Gold, Seigel, Russell et al., 1996), many CE analyses use the final prices charged for such a product or service. Nevertheless, variation in pricing means that the cost figures used in a CE analysis do not apply to every purchaser of the technology under study. Thus, it is particularly important to assess the consequences of cost variation for estimated cost-effectiveness.

In this analysis, we attempt to approximate the price paid by the health care system, payer, or patient as the cost of the new technologies. Although this might appear to be a "health care system" perspective, it is actually a societal perspective that includes costs borne by patients and others in addition to those of the payer or health system.

Drug costs are derived from the manufacturer's wholesale drug price (Table 23) (Drug Topics Red Book, 1997). We assumed that when an LHRH agonist was used, nilutamide was added for the first 2 weeks to avoid worsening of androgen-dependent symptoms of prostate cancer. Orchiectomy costs are estimated from a previous published CE analysis (Hillner, McLeod, Crawford et al., 1995). All costs are updated to 1998 dollars using the gross domestic product (GDP) deflator with an Internet-based inflation calculator (Parametric Cost Estimating Reference Manual, 1998). Krahn and coworkers estimated costs for treating prostate cancer patients at the New England Medical Center (Krahn, Mahoney, Eckman et al., 1994). Direct inpatient costs were obtained from the Clinical Cost Manager at the Center; outpatient costs and professional services were estimated by adjusting the charges for services by the proportion covered for each cost center. They estimated the yearly cost following local recurrence to be $290. Updated to 1998 dollars, we estimate this cost to be $320 (see Table 23).

Table

Table 23. Model Inputs.

Taplin et al., estimated that the 3-month cost of "continuing care" for prostate cancer patients with "local" disease was $1,277 and for "regional" disease was $1,356 (Taplin, Barlow, Urban et al., 1995). Costs were estimated at the Group Health Cooperative of Puget Sound and included materials, disposable goods, physician and other salaries, administrative overhead, inpatient costs, and hospice and visiting nurse services. However, these costs include treatment with antiandrogens, which we modeled as a separate cost. Assuming that the 3-month cost of antiandrogen therapy is approximately $900 to $1,250 (the costs of nilutamide and goserelin, respectively), the estimates of these treatment costs are similar to those reported by Krahn et al. (1994).

We assumed that the cost of treating asymptomatic distant disease was the same as the cost of local recurrence, based on Taplin and colleagues' finding that disease costs for regional disease were only marginally higher than those for local disease ($1,356 versus $1,277, including antiandrogens; difference not statistically significant) (Table 23).

To estimate the cost of treating symptomatic distant disease, we again used estimates from Krahn et al. (1994) and Taplin et al. (1995). Krahn et al. estimated that the annual costs of treating a patient with metastatic disease, including antiandrogen therapy, was $2,225. Taplin et al. estimated this cost to be $1,807. We assume that the majority of these costs is due to the use of antiandrogen therapy and that the other costs of metastatic disease total $375. Updated to 1998 dollars, this cost estimate is $410 (see Table 23).

We assumed that minor adverse effects were associated with the cost of one additional office visit per year, with a small incremental cost associated with increased physician visits and laboratory tests. We estimated the cost of one office visit to be $30 ( Coley, Barry, Fleming et al., 1997b ) (see Table 23 ).

Krahn et al. (1994) estimated that the yearly cost of terminal care was $37,290. Taplin et al. (1995) estimated that the 6-month cost was $15,581. In our model, this cost was incurred during the transition to death as a one-time cost. Hence, we based our estimate on Taplin and coworkers' estimate, updated to 1998 dollars and rounded off to $17,000 (see Table 23).

Our model included two other costs incurred during the transition between states. The cost of a severe adverse effect is incurred when a patient develops an adverse effect severe enough to necessitate discontinuing antiandrogen therapy. Frequent severe adverse effects include diarrhea and hepatotoxicity. Krahn estimated that the cost incurred at the first evidence of metastatic disease was $190; updated to 1998 dollars this estimate is $200. We assumed that severe adverse effects would be associated with lower costs and estimated this amount to be $150 (see Table 23).

The other cost included during the transition between states was the cost of a local obstructive event. We estimated the cost of treating an obstruction from Coley and coworkers' estimate of the cost of transurethral prostatectomy for local obstruction, which was $4,830 (Coley, Barry, Fleming et al., 1997b) (see Table 23).

Quality-of-Life Adjustments

As noted previously, QALYs can be calculated by multiplying the time spent in different states by the utility or value assigned to those states and adding the products together. Although in theory measures like QALYs make it possible to compare treatments that have disparate effects (for example, one primarily affects survival, and the other affects morbidity), in practice the quality-of-life measures used to construct QALYs often fall short. Frequently quality-of-life measures are disease-specific and emphasize the sensitivity of the measurement instrument over the ability to generalize from its results to societal preferences.

Among the common symptoms of metastatic prostate cancer and its treatment are impotence, loss of sexual desire, appetite and weight loss, pain, fatigue, diminished social function, impaired mental health, and less frequently, nausea and vomiting, hot flashes, hair loss, headaches, diarrhea, and breast tenderness (Denis 1995; Garnick 1997; Lucas, Strijdom, Berk et al., 1995; Presant, Soloway, Klioze et al., 1987). A small number of studies have estimated how patients value different health states associated with advanced prostate cancer and with hormonal treatment for prostate cancer. Although the quality of life of patients with asymptomatic disease is similar to that of the general population, quality of life decreases with development of progressive disease. Causes of diminished quality of life include both the disease and its treatment. Although several studies have reported changes in disease-specific measures of quality of life, none of them estimated the decrease in quality of life associated with undergoing surgical or medical castration or the quality of life associated with palliative measures such as the administration of prednisone.

A commonly used global quality-of-life measure is the SF-36 questionnaire (Patrick and Erickson, 1993). A study using this instrument reported that patients with metastatic prostate cancer in remission had a quality of life comparable to the general quality of life in the United States for men of the same age (Albertsen, Aaronson, Muller et al., 1997). The SF-36, like other so-called "statistically weighted" measures, assigns the same weight or value to all items in the questionnaire. In contrast, measurements of "utility" or "preferences" for health states, also called preference weighted measures, use observed preferences for different health states to weight items differently. Although statistically weighted measures are usually considered inappropriate for cost-effectiveness analysis (Gold, Seigel, Russell et al., 1996), results derived with statistically weighted measures and preference weighted measures tend to follow the same trends (Patrick and Erickson, 1993).

Not surprisingly, physician estimates of patient's likely tradeoffs between time without symptoms and stable prostate cancer without treatment-related toxicity produced a lower utility estimate of 0.92. This means that the physicians believed that patients would equally value 1 year with stable or asymptomatic metastatic prostate cancer and 0.92 years of disease-free life (Bennett, Matchar, McCrory et al., 1996). Although this represents an important decrease in quality of life, it is less than that for moderate angina or the need to use a walking stick for ambulation (Nord, 1992). It is also important to note that it is lower than the estimates used for a similar state by The Prostate Patient Outcomes Research Team (Fleming, Wasson, Albertsen et al., 1993). In the following analysis, we use the best available data to estimate quality-of-life weights, preferring patients' preferences to that of proxies and preference measures to statistically weight measures. Where the less desirable measures are the only ones available or estimates are unreliable, we include a wide range for the values in sensitivity analyses.

There is a consistent decrease in quality of life with increasingly severe disease. After separating SF-36 results for patients in remission into minimal and extensive disease based on the number and location of bone metastases, patients with minimal disease in remission had significantly better quality of life in two dimensions, social functioning and mental health, compared to patients with extensive disease in remission (Albertsen, Aaronson, Muller et al., 1997). Similarly, the development of gastrointestinal adverse effects due to flutamide was associated with a utility of 0.83 (Bennett, Matchar, McCrory et al., 1996), despite the caveat that these adverse effects were not severe enough to warrant discontinuation of treatment.

Utility scores decline as prostate cancer advances. Early progression of metastatic cancer was associated with a utility of 0.83, the same as that associated with stable metastatic disease with gastrointestinal adverse effects due to flutamide (Bennett, Matchar, McCrory et al., 1996). Late progression of disease was associated with a much greater decrease in utility, to 0.42 (Bennett, Matchar, McCrory et al., 1996). Patients with progressive metastatic prostate cancer, defined as continuing increase in PSA levels of 4 ng/ml, had a lower overall level of quality of life compared to a comparable population based on SF-36 results. Patients with progressive disease had significantly worse (p<0.05) scores on four of eight dimensions including bodily pain, vitality, social functioning, and mental health compared to all patients in remission. Patients with progressive disease also had significantly worse (p<0.05) scores on the bodily pain dimension compared to patients with extensive disease in remission (Albertsen, Aaronson, Muller et al., 1997). In this study, patients in remission and with progressive disease had received treatment with an LHRH agonist and flutamide, although the requirements for inclusion base on duration of treatment varied across the two groups. Unfortunately, it is unclear to what extent these decrements in quality of life result from adverse effects of different treatment strategies.

The decrease in quality of life accompanying disease progression is consistent with extrapolations from the Quality of Well-Being (QWB) Index, a preference weighted measure of health-related quality of life. For example, if disease in remission causes primarily impairments in social and physical activity, the QWB would assign a 6.1 percent reduction in quality of life compared to good health. More severe symptom groupings characteristic of symptomatic advanced disease, such as general tiredness or weight loss, urinary dysfunction, gastrointestinal upset, and hot spells, are associated with decrements of 24.4 percent to 29.2 percent in the QWB Index (Kaplan and Anderson, 1988; Patrick and Erickson, 1993.)

The quality-of-life values assigned to different stages of metastatic prostate cancer in the above studies are similar to the estimates of quality of life used in this analysis (see Table 23). Previous decision analyses have produced similar estimates (Hillner, McLeod, Crawford et al., 1995; Krahn, Mahoney, Eckman et al., 1994). There is a decline in quality of life with early prostatic cancer of roughly 10 percent, followed by a much larger decrement of at least 50 percent with advanced disease. Asymptomatic disease is not associated with a substantial decline in quality of life.

We make several assumptions regarding quality of life. First, we assume that orchiectomy is not associated with a lower quality of life than antiandrogen drug therapy. Because the utility attached to orchiectomy may vary greatly among patients with prostate cancer, we tested a wide range of alternative values for this value in sensitivity analyses. Second, we assume that all patients respond initially to antiandrogen therapy. Third, we assume that there is no decrement in quality of life for patients who use antiandrogens and do not experience adverse effects; thus patients who stop medications do not experience an improved quality of life. Fourth, we assume that the quality of life with symptomatic distant metastases was relatively high while receiving antiandrogen therapy (including second-line therapy and hormone withdrawal) but fell to a low level once all hormonal manipulations were exhausted.

Timing of Antiandrogen Therapy

The meta-analysis indicates that, for patients presenting with regional metastatic disease, equivalent survival benefits are obtained if antiandrogen therapy is initiated early (when prostate cancer is diagnosed) or late (when disease progression occurs or when symptomatic distant metastases develop). We modify the assumptions in our decision tree to evaluate the cost-effectiveness of these approaches (see Table 20). First, we redefined the starting state (state I in Figure 14) in our model, focusing on a typical patient who presents with stage C disease (see Table 21). Second, we estimated the transition rate from this state to asymptomatic metastatic disease as 0.066 per patient-year, as outlined above (see "Local Recurrence to Asymptomatic Distant Metastases"). We further assume that antiandrogen therapy can be started either when stage C disease is diagnosed, when asymptomatic distant metastases are detected (i.e., clinical progression occurs), or when symptomatic distant metastases develop.

If antiandrogen therapy is started early, we assume that the transition rates from asymptomatic to symptomatic distant metastases (transition 3 in Figure 14) and from symptomatic distant metastases to death were unchanged (transition 5 in Figure 14) and were similar to the base case model. When antiandrogen therapy is started upon the development of asymptomatic distant metastases, we modified the transition rates accordingly. As stated above, we estimated that the transition rate from asymptomatic to symptomatic distant metastases for patients with hormone sensitive disease using antiandrogen therapy was 0.399/patient-year (see "Asymptomatic to Symptomatic Distant Metastases"). We then modified the transition from the state "Stage C prostate cancer at diagnosis" to "asymptomatic distant metastases" such that the overall survival in the model was unchanged (see Table 21).

Last, we calculated transition rates corresponding to a practice of initiating antiandrogen therapy when patients develop symptomatic distant metastases. We assumed that the transition rates for patients not receiving antiandrogen treatment were the same as those for patients with hormone-resistant prostate cancer (see Table 21). Note that all patients still develop a stage when they have hormone-resistant symptomatic distant metastases with an associated poor quality of life; the effect of starting antiandrogens late as modeled is to delay the occurrence of this very poor health state.

Incorporating Biochemical Markers

Current practice incorporates information from biochemical markers into treatment decisions, although natural history data relating changes in markers to outcomes after treatment failure are scant. To simulate this situation, we assumed that asymptomatic distant metastases were identified earlier in the disease process (see Table 20). Consequently, we redefined states I and II in Figure 14 as the first and second occurrence of biochemical failure (for example, a rise in PSA). Thus, we increased the rate at which patients progress from state I to state II (0.14/patient-year) to approximate the time spent between first and second biochemical relapse (Zietman, Dallow, McManus et al., 1996) (see Table 21). We also estimated the rate at which patients progress from second PSA failure to symptomatic distant metastases as 0.05/patient-year; doing so yielded an undiscounted survival advantage of 2.2 years, which is within the range of estimates of increased observation with biochemical monitoring (whether this is due to lead time bias or better treatment) (Lillis and Thompson, 1996).

Choice of Agent

We first identify the most economically attractive agent of the LHRH agonists and the nonsteroidal antiandrogens. The results of the meta-analysis suggest that the efficacy of the major antiandrogen treatment strategies is similar. Furthermore, the pooled adverse effect results suggest that all of the LHRH agonists have similar adverse effect profiles. In addition, the nonsteroidal antiandrogens have similar adverse effect profiles. We therefore used a cost-minimization analysis to identify the least expensive agent in each group.

We first consider the LHRH agonists. Not including administration costs, leuprolide is less expensive when administered every month rather than every 3 months ($5,064 versus $6,282 in annual drug costs). Even assuming a $25 charge per administration, the more frequent schedule is more advantageous ($5,364 versus $6,382). Buserelin is not yet commercially available in the United States. Compared to the least costly leuprolide regimen ($5,064), goserelin is marginally less expensive ($4,995). Because less frequent dosing results in lower costs for administration, the most economically advantageous LHRH agonist is goserelin administered every 3 months.

We next consider the nonsteroidal antiandrogens. Although monotherapy with a nonsteroidal antiandrogen is not considered the standard of care in the United States, this strategy has been extensively studied and we include it for completeness. Flutamide is marginally less expensive than bicalutamide ($3,694 versus $3,890), both of which are more expensive than nilutamide ($2,842). The cost of nilutamide in the first year is increased because nilutamide is given in higher doses during the first month of therapy (300 mg/d), but is still the least expensive NSAA ($3,079). Thus, the most economically advantageous nonsteroidal antiandrogen is nilutamide.

Comparison of Monotherapies

Of the monotherapies, DES was associated with the shortest average survival, 6.9 years, a consequence of the relatively high rate of cardiovascular mortality associated with its use. Thus, although the strategies are of equal efficacy, DES is the least effective. Using an NSAA was associated with a survival of 7.3 years, whereas orchiectomy was associated with a survival of 7.6 years. Using an LHRH agonist was associated with an average survival of 7.5 years. After an annual discount rate of 3 percent was applied, the relative rankings remained similar: DES is associated with an average survival of 5.9 discounted life years, NSAA with 6.3 discounted life years, and orchiectomy and LHRH agonists with 6.5 discounted life years each (Figure 16, top). The difference in quality-adjusted life years is less than the difference in life expectancy; DES is associated with a quality-adjusted survival of 4.6 years, NSAA with 4.9 discounted QALYs, and orchiectomy and LHRH agonists with 5.1 discounted QALYs each (Figure 16, bottom).

Figure

Figure 16. Lifetime Costs and Effectiveness of Each Strategy. Each dot represents the discounted lifetime costs and effectiveness of each therapy. Effectiveness is presented before and after adjusting for quality of life (top and bottom (more...)

Figure

Figure 16. Lifetime Costs and Effectiveness of Each Strategy. Each dot represents the discounted lifetime costs and effectiveness of each therapy. Effectiveness is presented before and after adjusting for quality of life (top and bottom (more...)

In contrast to the very similar survival estimates, the lifetime costs of the strategies differ markedly. The lowest lifetime cost, about $4,100, is associated with DES therapy. The cost of orchiectomy is $7,500, whereas the cost associated with monotherapy with an NSAA is $18,000 and for treatment with an LHRH agonist is $30,900. After a time discount rate of 3 percent is applied, these estimates are $3,600, $7,000, $15,700, and $27,000, respectively.

Cost-effectiveness ratios can be calculated by dividing the incremental cost of one therapy relative to another by the incremental health effect. The same information is presented graphically in Figure 16, where the slope of a line connecting any two points represents the inverse of the incremental cost-effectiveness ratio of those two therapies. Thus, a near vertical line would indicate that an intervention increases life expectancy at minimal cost.

DES is the least expensive option. Orchiectomy is both slightly more effective and more expensive. The incremental cost-effectiveness of orchiectomy relative to DES is $6,100 per life year gained and $7,500 per QALY. Monotherapy with LHRH agonists or NSAA is dominated by other strategies both before and after quality-of-life adjustments are applied. Thus, orchiectomy is the most cost-effective strategy.

Combined Androgen Blockade

We compared the effects with combined androgen blockade (CAB) to monotherapies. We compared two regimens for which we had cost and toxicity data-nilutamide plus goserelin and nilutamide plus orchiectomy. The model predicts that both regimens are associated with undiscounted crude survivals of 7.5 years, discounted survival of 6.5 years, and discounted quality-adjusted survivals of 5.0 quality-adjusted life years respectively. The undiscounted lifetime costs of each therapy are $46,200 and $23,200 and the discounted lifetime costs are $40,300 and $20,700, respectively. Of course, if CAB is no more efficacious than monotherapies, then CAB cannot be more cost-effective (they are dominated strategies). If CAB is slightly more effective, it is no longer dominated by other strategies but the cost-effectiveness ratios are very high (see "Sensitivity Analysis"). These estimates include potential advantages to using combined androgen blockade due to increased therapeutic choices and benefits from drug withdrawal.

Alternative Efficacy Assumptions

We repeated the analysis (including CAB) using the point estimates of relative hazards from the meta-analysis (see Table 20). The meta-analysis yielded estimates for the relative hazards for survival. We assumed that these results were valid estimates of drug efficacy in all stages of disease when patients had cancer that was hormone-sensitive. Under these assumptions, the least cost is again obtained with DES, followed by orchiectomy (Figure 17). Orchiectomy is associated with a higher quality-adjusted survival (5.10 QALYs) than DES (4.68 QALYs), LHRH agonists (4.89 QALYs), or NSAA monotherapy (4.61 QALYs). The cost-effectiveness ratio of orchiectomy relative to DES under these assumptions is $8,100 per QALY.

Figure

Figure 17. Costs and Outcomes With All Monotherapies and Combined Androgen Blockade.

Combined androgen blockade with nilutamide and orchiectomy has a discounted lifetime cost of $20,900 and a quality-adjusted survival of 5.08 QALYs. This strategy is eliminated by strict dominance by orchiectomy. Combined androgen blockade with nilutamide and goserelin is the most expensive therapy ($41,400 discounted lifetime costs) but is also associated with the highest quality-adjusted life expectancy (5.13 QALYs). Compared to the alternative CAB regimen, the cost-effectiveness of this strategy is $413,300 per QALY. Compared to orchiectomy, the cost-effectiveness of this CAB strategy is $1,110,000 per QALY.

We repeated the analysis assuming greater efficacy for both combined androgen blockade regimens. We first assumed that the relative risk for disease progression was 0.86, based on the meta-analysis results of 5-year survival when all CAB strategies were grouped together. Under these assumptions, the cost-effectiveness ratio of NSAA plus orchiectomy relative to orchiectomy alone was $74,800/QALY. The cost-effectiveness ratio of NSAA plus LHRH agonist relative to orchiectomy was $196,700/QALY. At a relative risk of 0.78 (the lower limit of the 95 percent confidence interval from the same meta-analysis result), the cost-effectiveness of NSAA plus orchiectomy relative to orchiectomy alone was $43,400/QALY and of NSAA plus LHRH agonist relative to orchiectomy was $110,900/QALY.

Sensitivity Analysis

We performed extensive sensitivity analysis to test the robustness of our findings and identify important areas of uncertainty. The relative rankings of strategies were not sensitive to assumptions about the natural history or cost of care of prostate cancer. Similarly, the cost-effectiveness of orchiectomy relative to DES remained less than $30,000 per QALY throughout these assumptions. Even with high cost estimates for orchiectomy, the cost-effectiveness ratio of orchiectomy relative to DES remained favorable.

Two assumptions were important in determining the cost-effectiveness of orchiectomy relative to DES. First, if the excess cardiovascular mortality from DES is less than we assumed, the associated quality-adjusted survival improved although it was always less than that associated with orchiectomy. If there was no excess mortality associated with DES, then the incremental cost-effectiveness of orchiectomy is $13,100/QALY.

The model was also sensitive to assumptions about the quality of life associated with orchiectomy (Figure 18). In the base case, we assumed that the quality of life associated with orchiectomy was similar to that of medical antiandrogen therapy. If the quality of life associated with orchiectomy is 0.88, only slightly less than the quality of life associated with medical antiandrogen therapy (0.92), the incremental cost-effectiveness ratio of substituting LHRH agonists for orchiectomy is less than $100,000/QALY. If the quality of life associated with orchiectomy is less than 0.83, the incremental cost-effectiveness ratio is less than $50,000/QALY. Under such circumstances, orchiectomy is strictly dominated by the other strategies. Comparing NSAAs to DES, the incremental cost-effectiveness ratio is $43,200/QALY. The model was relatively insensitive to other quality-of-life weights.

Figure

Figure 18. Sensitivity to Quality of Life of Orchiectomy. The cost-effectiveness of orchiectomy relative to LHRH agonists is plotted against the quality-of-life adjustment associated with orchiectomy. If the quality of life associated (more...)

Another key uncertainty is the efficacy of combined androgen blockade. The current meta-analysis indicates that the relative mortality hazard with combination of a NSAA and orchiectomy compared to orchiectomy alone is 0.94 (95 percent confidence interval 0.78 to 1.14). We conducted sensitivity analyses on the efficacy of combined androgen blockade (Figure 19, panel A). Even at the low end of the 95 percent confidence intervals, the cost-effectiveness of combined androgen blockade with an NSAA and an LHRH agonist exceeds $100,000/QALY. Combined androgen blockade with an NSAA and orchiectomy is less expensive, however. For this combination to have a cost-effectiveness ratio (relative to orchiectomy alone) of less than $50,000/QALY, the relative hazard associated with the combination therapy needs to be 0.80 or less (Figure 19, panel B).

Figure

Figure 19. Sensitivity Analysis on the Efficacy of Combined Androgen Blockade. Panel A

Figure

Figure 19. Sensitivity Analysis on the Efficacy of Combined Androgen Blockade. Panel B
The cost effectiveness of combined androgen blockade (CAB) relative to orchiectomy at different assumptions about the efficacy of (more...)

We also conducted a sensitivity analysis on the quality of life associated with a adverse effect of treatment that was not dose-limiting because patients may vary considerably in how they rate the associated adverse effects. In our base case, we assigned a quality-of-life weight of 0.55 to this state. If adverse effects are judged as severely bothersome (assigned a much lower quality-of-life weight), orchiectomy becomes less favorable because it is irreversible. The incremental cost-effectiveness of orchiectomy relative to DES is $35,200 per QALY and of flutamide relative to orchiectomy is $361,100 per QALY. Conversely, if adverse effects are judged to be very minor (assigned a much higher quality-of-life weight), orchiectomy becomes considerably more favorable with an incremental cost-effectiveness relative to DES of $6,500 per QALY, with all other strategies eliminated by strict dominance.

The cost-effectiveness estimates were less sensitive to several other values assigned to baseline variables. The model was also insensitive to the cost of orchiectomy-even if orchiectomy cost as much as $7,000 (Krahn, Mahoney, Eckman et al., 1994) the cost-effectiveness relative to DES would be $15,400 per QALY and all other strategies would be dominated. As well, the model was insensitive to the cost of a local obstruction. If all obstructions were associated with costs of $300, the cost-effectiveness of orchiectomy relative to DES was $7,400 per QALY. The model was also relatively insensitive to assumptions about the age of the man with prostate cancer at the start of the analysis. For 50-year-old men, the cost-effectiveness of orchiectomy relative to DES is $5,800 per QALY, whereas for 75-year-old men the corresponding estimate is $11,000 per QALY. All other strategies are dominated in all age groups.

The relative cost-effectiveness of antiandrogen strategies for patients who have symptomatic distant metastases at presentation changes only slightly. DES is still the least expensive option with the worst quality-adjusted life expectancy. Orchiectomy is associated with an incremental cost-effectiveness of $31,500 /QALY. All other strategies are dominated.

Timing of Antiandrogen Therapy

We compared the costs and effects of initiating antiandrogen therapy early (at the time of diagnosis) or late (when asymptomatic distant metastases develop or when symptomatic distant metastases develop). Details about the assumptions are given in the section "Timing of Antiandrogen Therapy" and Table 20 . Focusing first on orchiectomy, the greatest quality-of-life benefit is obtained by performing surgery late (7.0 QALYs), slightly fewer benefits result when antiandrogen therapy is started at the occurrence of asymptomatic distant metastases (6.8 QALYs), and the least benefits accrue when antiandrogens are started at the time of diagnosis (6.2 QALYs) (Figure 20, top). Greater benefits result when orchiectomy is performed late in the disease because this is when patients' quality of life is at its nadir. The least cost is also obtained by performing orchiectomy late ($5,200 with symptomatic distant disease, $5,600 with asymptomatic distant disease) rather than early ($7,400), primarily because fewer patients will require therapy (as some will have died of other causes). These results were only mildly sensitive to assumptions about the discount rate. Thus, waiting to initiate therapy until the patient has developed symptomatic metastases is both the least expensive and most beneficial therapeutic strategy. Similar results are obtained when we analyze the timing decision with respect to combined androgen blockade with a nonsteroidal antiandrogen and an LHRH agonist (Figure 20, bottom). The least benefit is obtained when medications are started early in the disease (5.9 QALYs) and the highest benefit is obtained from starting late in the disease (6.8 QALYs if started when patients are asymptomatic and 7.0 QALYs if started when patients are symptomatic). The highest costs are again obtained when medication is started early in the disease. The medication costs are slightly higher if medications are started when patients are symptomatic ($14,700) rather than at the first identification of an asymptomatic distant metastasis ($12,200). This increase in total costs occurs because antiandrogens delay the progression from the state in which they are started until the next. In other words, they prolong the time spent in any given state. Starting antian drogens when patients have symptomatic disease (a relatively expensive state) results in higher costs than starting when patients have asymptomatic disease (a relatively inexpensive state). Thus, starting combined androgen blockade when patients develop symptomatic distant metastases is more expensive but more beneficial than starting when patients develop the first sign of disease progression (asymptomatic distant metastases), with a cost-effectiveness ratio of $14,300 per QALY. Similar results were seen for all evaluated drug therapies ( Table 24 ).

Figure

Figure 20. Optimal Timing of Antiandrogen Therapy.

Figure

Figure 20. Optimal Timing of Antiandrogen Therapy. Costs and quality-adjusted survival when orchiectomy is performed at the time of diagnosis with local spread, when asymptomatic distant metastases develop, or when symptomatic distant (more...)

Table

Table 24. Costs and Effectiveness of Strategies According to Time of Initiation.

Incorporating Biochemical Markers

We modified the states and transition rate to estimate the consequences of incorporating biochemical markers. Details about the assumptions are given in Table 20. The relative rankings of the strategies remain unchanged (Figure 21). The incremental cost-effectiveness of orchiectomy relative to DES under these assumptions is $7,000 per QALY, and all other strategies are eliminated by strict dominance.

Figure

Figure 21. Incorporating Biochemical Markers. Costs and effectiveness when decisions to change therapy are based on biochemical markers (PSA) as well as clinical outcomes. See text for details of assumptions.

Discussion

Our evaluation of the cost-effectiveness of androgen suppression therapy for men with advanced prostate cancer incorporated several therapeutic strategies, considering not only a wide range of treatments but also issues such as the optimum time to initiate therapy. It considered how alternative treatments affect longevity, quality of life, and costs and examined how recent advances in clinical care might alter the principal findings. The results suggest that neither the most expensive nor the least expensive treatment options are good values. DES, the least costly therapy, is associated with the shortest life expectancy and the fewest quality-adjusted life years. Since the true excess mortality associated with low doses is unknown, DES may be considered as an appropriate agent for future studies. Failing this, alternative approaches will be preferred.

Orchiectomy represents a highly cost-effective alternative to DES. Under most of the assumptions we explored, its cost-effectiveness ratio is less than $30,000/QALY when compared with DES, a range that is usually considered to represent a very good value. Monotherapy with a nonsteroidal antiandrogen or an LHRH agonist costs significantly more and produces minimal additional benefit. Combined androgen blockade may yield slight additional benefit, but at an even greater increase in cost. DES and orchiectomy are used infrequently in the United States, as regimens that are more expensive have grown in popularity. The meta-analysis suggests, and the cost-effectiveness analysis confirms, that the extra benefit from these new therapies is small, even after accounting for differential toxicities. However, this conclusion is based on the assumption that orchiectomy is not associated with a worse quality of life than other antiandrogen therapies. Yet preferences toward orchiectomy may vary greatly among patients. Unlike medications, it is irreversible, and some men appear to believe that the treatment is unacceptable. For such individuals, orchiectomy would not produce the gain in QALYs that we estimate occurs on average. For other men, the psychological drawbacks of orchiectomy are minimal, and the treatment will be far more attractive. Medical therapy itself has significant drawbacks, such as increased office visits, pain with injectable medications, and associated psychological consequences (such as the constant reminder that patients have cancer). We did not include these factors in our analysis, but an increase in costs or a decrease in quality of life would strengthen our base case results. Nevertheless, the objective evidence supporting a decreased quality of life associated with orchiectomy is scant. Although many patients may be clear about how they value this health state, we present these results in detail to emphasize the uncertainty about the average patient's values and their importance to the f inal cost-effectiveness estimates.

The most expensive strategy is combined androgen blockade. Its efficacy is controversial; the meta-analysis suggests that it is not significantly more efficacious than orchiectomy. Note that we refer here only to the average benefit; a minority of patients may still experience a large benefit. Focusing on this average benefit and setting the cost-effectiveness threshold at $100,000/QALY, combined androgen blockade with a nonsteroidal antiandrogen and an LHRH agonist must decrease the risk of disease progression by 20 percent compared to orchiectomy before this strategy is considered cost-effective, a value within the range of estimates from the meta-analysis. Alternative combined androgen blockade strategies (such as orchiectomy and a NSAA) are less expensive and hence require less efficacy to be cost-effective. For these therapies, cost-effectiveness ratios are less than $100,000/QALY if the relative hazard of disease progression is less than 0.9, also a value within the range of estimates from the meta-analysis.

Our analysis of LHRH agonists and NSAAs used a cost-minimization approach that builds on the assumption that the efficacy and toxicities of drugs within a certain classification are identical. Although this assumption appears reasonable, the toxicities of NSAAs have not been rigorously evaluated. Flutamide, nilutamide, and bicalutamide may actually have dissimilar long-term adverse effects. Further toxicity data would clarify this choice and are a subject for future research.

Our model suggests that for patients diagnosed with locally progressive cancer at the time of diagnosis, initiating antiandrogen therapy early, when patients enjoy a good quality of life, will result in higher costs and no additional benefits and possible harm compared to deferring therapy. In the absence of a survival benefit, starting antiandrogen therapy when patients have advanced disease (whether orchiectomy or medications) is of greatest benefit to the patient. Indeed, our analysis implies that quality of life improves more when antiandrogen therapy is initiated at a later stage. A small empirical study found that patients with asymptomatic prostate cancer who did not receive hormonal therapy had similar or better quality of life than patients who received hormonal therapy (Herr, Kornblith, and Ofman, 1993). If this result is generally valid, the cost advantages of delaying prostate cancer would be strongly backed by clinical benefit. Furthermore, because we modeled the effects of late therapy as a delay in the occurrence of very poor health (rather than as an increase in quality of life), we may have underestimated the benefits of late therapy. If such a bias exists, starting antiandrogens when distant metastases develop would be even more economically attractive.

Our analysis of the optimal timing of antiandrogen therapy focuses only on patients who present with metastatic cancer at the time of diagnosis, but not patients who use antiandrogens as adjunctive therapy at the time of definitive treatment for localized prostate cancer. The meta-analysis indicates that for the latter patients, a survival benefit from early initiation of antiandrogens exists. Whether this approach is cost-effective is a topic for future study.

A previous model of antiandrogens has been published and updated (Bennett, Matchar, McCrory et al., 1996; Hillner, McLeod, Crawford et al., 1995). This analysis examined only the incremental cost-effectiveness of combined androgen blockade with orchiectomy plus flutamide compared to orchiectomy alone. The cost-effectiveness estimates were roughly of the same order when similar assumptions about this combined androgen blockade strategy were compared. If the efficacy of this strategy was 25 percent as assumed in the original previous cost-effectiveness analysis (relative hazard 0.75), that model estimated a cost-effectiveness ratio of $25,300 per QALY and corresponding estimate from our study is $37,600 per QALY. If the efficacy of this strategy was 9 percent, the previous cost-effectiveness estimate was $60,900 per QALY and the current estimate is $138,200 per QALY. Differences between estimates exist because the previous model used a different definition of health states (classifying asymptomatic distant metastases and symptomatic distant metastases as a single health state), did not include all features of the current model (for example, local obstruction), and used some different variables in the base case analysis (for example, a discount rate of 5 percent).

Our model has several limitations, including the need to simplify a number of assumptions. Nevertheless, its findings change little when uncertain values entered in the model are varied over wide ranges in one-way sensitivity analyses. Our model also did not include therapies unapproved for use in the United States. Lastly, our model does not indicate how individual patients, clinicians, or health plans might value the benefits or costs in the model differently.

Improvements in monitoring the stage and extent of prostate cancer can lead to earlier detection of metastatic disease (Smith and Pienta, 1997; Zietman, Dallow, McManus et al., 1996). Such advances might lead to the earlier initiation of antiandrogen therapy. Current practice often uses the PSA as an indicator of advancing disease, a strategy that we were unable to evaluate fully because no published evidence establishes the efficacy of this practice (we also did not evaluate the use of PSA as a prognostic indicator). However, redefining the states and transitions in our model to approximate the effects of such an "early detection" strategy incorporating PSA monitoring led to virtually no incremental improvement in outcomes. Consequently, our results are likely to continue to apply to evolving monitoring strategies. Without further advances in treatment, improved monitoring is likely to have little effect on the cost-effectiveness of treatment strategies for advanced prostate cancer.

Cost-effectiveness analysis can inform health policy decisions by indicating the most efficient use of scarce resources but does not substitute for the full decisionmaking process, which must account for factors beyond economics. Furthermore, cost-effectiveness analyses can identify key factors that may be important for decisionmakers. For example, the current analysis indicates that if there is substantial heterogeneity in patient preferences toward the effects of orchiectomy and alternative therapies, it will be important to individualize therapy. Other individual preferences such as attitude to antiandrogen adverse effects should also be considered. Appropriate application of cost-effectiveness results would include a careful assessment of each individual's values with respect to surgical and medical castration. For patients whose quality of life would diminish substantially if they underwent orchiectomy, the use of LHRH agonists or NSAA may represent reasonable alternatives, even though the average patient's quality of life may not fall substantially with orchiectomy.