Economic Evaluation of Drug Treatment for Psychiatric Disorders:

Gary A. Zarkin, Ph.D., Josephine A. Mauskopf, Ph.D., Henry Grabowski, Ph.D., Heather Luckey, M.P.A. and Richard Weisler, M.D.

During the 1980s and continuing into the 1990s, health care cost containment has received increasing attention as health care costs have risen more rapidly than inflation. At the same time, many new drugs have been introduced at such high prices that they have become a target for third-party payer and government efforts for cost containment.

In response to the public outcry over drug prices, many researchers have designed and performed studies to evaluate the costs and outcomes of new drug therapies. These cost and outcome studies, which we refer to as drug valuation studies, are important to drug companies, clinicians, patients, and public policymakers. Valuation studies can provide marketing information and enhance pharmaceutical companies' competitive advantage (23). For patients and their physicians, valuation studies can determine whether new therapies justify their potentially greater expense. In addition, valuation studies can help patients assess whether future health benefits justify a possible reduction in the quality of life caused by a current therapy. For policymakers, who must make resource allocation decisions, a valuation study that reflects a more complete picture of societal benefits can help differentiate between therapies with marginal differences in clinical efficacy.

The Food and Drug Administration (FDA) currently requires clinical trials to demonstrate the safety and clinical efficacy of new drugs before they can be sold in the US Although safety and clinical efficacy data clearly are of primary importance, they provide policymakers with insufficient information about the quality-of-life and economic implications of a new drug. Pharmaceutical companies have not had the incentive to collect the data required for drug valuation studies, at least for drug approval in the US market. However, in several countries (e.g., Australia), pricing and reimbursement decisions for new drugs are based on economic valuation studies (5, 10, 16). In the United States, insurance providers and managed care organizations are indicating a desire for economic valuation studies of drugs (48), and health care reform spawned by managed care may bring an increased demand for quality-of-life and economic evaluations of clinical trial data. These trends have induced US pharmaceutical companies to include quality-of-life and economic measures as integral components of clinical trial protocols.

These new protocols were first introduced for studies of treatment of nonpsychiatric conditions. However, published economic studies of psychotherapeutic medications are now becoming more common. To understand the role economics plays in evaluating new drug treatments for psychiatric disorders, we reviewed quality-of-life and economic evaluations of drug treatment for schizophrenia, major depression, and anxiety disorders. Although psychiatric disorders are prevalent and impose substantial individual and societal costs, we found relatively few published quality-of-life or economic evaluations of psychotherapeutic drugs (3, 17, 25, 31, 35, 37, 39, 41-43, 45, 46, 58).

This chapter describes the type of quality-of-life and economic data (which we also refer to as outcomes data) that are being collected in trials of drug treatment for psychiatric disorders and discusses the use of these data in drug valuation studies. Possible outcome measures that might be included in clinical trials are changes in patient quality of life and overall well-being, work days gained for patients and caregivers, and changes in medical resource use and associated medical care costs. We also describe how these enhanced clinical trials data can be used to develop disease outcome models. These models are useful because they capture the dynamics of disease/treatment patterns over the entire course of the disease, providing a basis for evaluating the effects of alternative therapeutic interventions over a patient's lifetime. Many of the initial parameters required for these models can be estimated using clinical trials outcome data. After the drug has been introduced to the market, however, the initial model parameters can be validated and improved with data from actual clinical practice and post-marketing studies.

This chapter briefly reviews drug valuation methods, including quality of life, cost-effectiveness, benefit-cost, and cost-utility analyses. We will also discuss the quality-of-life outcomes now being collected in many clinical trial protocols for depression, schizophrenia, and anxiety disorders; present the results of such studies; and discuss their limitations. We highlight recent additions of resource use data to clinical trial protocols and their use in valuation studies. A description of how clinical trials data can be used to develop disease/outcome models is also included. Finally, we summarize our discussion and review our suggestions for using clinical trials data in quality-of-life and economic analyses of drug therapy for psychiatric disorders.

Cost-outcome analysis aids policymakers in deciding whether the costs of a new drug treatment are justified by the benefits it generates. One method of analyzing the benefits of a new drug therapy is to list the relevant clinical endpoints and compare the differences in clinical endpoints between an old drug therapy and a new drug therapy. For example, analysts might compare the differences in symptoms and side effects of two alternative drug therapies. If the new drug therapy results in more symptomatic improvement and fewer side effects, compared with an older drug, the new drug would be more valuable from a purely clinical viewpoint. However, the clinical viewpoint neglects changes in the patient's overall health-related quality of life (HRQL) and differences in resource use, including differences in the cost of the treatment regimens. An expanded value analysis might compare the differences between an old drug therapy and a new drug therapy in terms of the relevant clinical, quality-of-life, and resource-use endpoints.

For many purposes, comparing all the relevant outcomes may be sufficient to demonstrate that one drug is more valuable than another. For example, if a new drug has better clinical outcomes, uses fewer resources (including expenditures for the drug itself), and is associated with a better quality of life than an older drug, the new drug is more valuable and should be used in place of the older drug. More often, a new drug might improve clinical outcomes and the quality of life, but cost more than a competing drug. In these cases, researchers must perform "cost-outcome" studies to account for changes in both costs and outcomes attributable to drug therapy and to provide a rational basis for assessing whether a drug's extra expense justifies the improvements in clinical and quality-of-life outcomes. Three cost-outcome valuation methods can be applied to drug valuation studies: cost-effectiveness analysis, benefit-cost analysis, and cost-utility analysis.

Cost-effectiveness analysis compares the differences in cost and outcome across alternative therapies. The outcome generally refers to a clinical outcome and is measured in its natural units. The results are usually expressed as the incremental cost (relative to an alternative treatment) per unit of incremental outcome change, yielding ratios such as cost per averted sick day or cost per life-year(s) gained.

To perform a cost-effectiveness analysis, a researcher should have one, unambiguous objective of the intervention yielding a single outcome measure of effectiveness (8). If there are many outcomes of interest, cost-effectiveness measures are often computed for each of the alternative outcomes (e.g., life years gained, days of work loss averted, days of caregiver time saved) (8); but if the therapy under study is not clearly superior for all possible outcomes, decision makers are left in a quandary as to the desirability of the therapy. Under these circumstances, a benefit-cost analysis might be performed. By translating all the costs and benefits (including the health-outcome and quality-of-life improvements) into dollars, benefit-cost analysis allows researchers to assess directly whether the benefits of treatment justify the treatment costs.

Benefit-cost analysis potentially provides the broadest estimate of the total value to society attributable to a drug therapy. In practice, however, measuring and quantifying all the costs and benefits of a drug therapy—especially the dollar value of quality-of-life changes—are difficult and often controversial. For example, some analysts have raised concerns about assigning dollar values to improvements in labor market productivity (9, 18). Furthermore, analysts are often uncomfortable assigning dollar values to changes in people's well-being (14).

Because of these concerns, many analysts turn to cost-utility analysis. Cost-utility analysis is similar to cost-effectiveness analysis in that it compares the incremental cost and outcome attributable to a particular therapy, but cost-utility analysis also accounts for changes in the quality of life caused by drug treatment (8). Thus, cost-utility analysis incorporates changes in the quality of life resulting from the clinical effect in addition to changes in the length of life. In cost-utility analysis, the entire array of health improvements is converted to a single common unit, most commonly quality-adjusted life-years (QALYs) gained, which makes comparing alternative treatments easier.

Recently, researchers have designed disease/outcome models to simulate the effect of therapeutic interventions on outcomes such as incidence, mortality, and resource use. The transition patterns between severity levels are estimated using state-transition (Markov) or decision-tree models that capture the dynamics of treatment patterns over the entire course of the disorder. State-transition or decision-tree models are especially useful for chronic diseases because clinical trials are of relatively short duration. For example, Weinstein et al. (57) used a state-transition model to simulate future trends in incidence, prevalence, mortality, and resource cost under alternative assumptions about preventive and therapeutic interventions for coronary heart disease (CHD). Their model allows for simulation of the initial outcomes attributable to the CHD event, as well as subsequent events (such as recurrence) in persons suffering from CHD.

Given that psychiatric disorders tend to be chronic and recurring, we recommend using state-transition or decision-tree models in drug valuation analyses of psychotherapeutic drugs. These disease/outcome models allow researchers to consider treatment outcomes in a dynamic context over the lifetime of the individual. The models incorporate the potential resource savings from reducing the intensity and length of acute episodes, as well as the gains that accrue from preventing future acute episodes. An example of a decision-tree model applied to the acute and maintenance phases of major depression will be presented later.

Two recent reviews of the literature on quality-of-life measurement and analysis in psychiatric disorders (43, 58) found that valid and reliable HRQL measures have been developed for use in clinical trials of patients with psychiatric disorders (including schizophrenia, anxiety, and major depression), but that very few published analyses compare the impact of new pharmaceutical products on quality of life using data from random controlled trials. Most of the published studies used data from uncontrolled or cross-sectional studies. The small number of quality-of-life evaluations of psychotherapeutic drugs may be attributable to the difficulty in measuring psychological states with reliability and validity. Disease states and concomitant quality-of-life outcomes may be easier to measure for nonpsychiatric illness than for psychiatric disorders. Traditionally, clinical trials for pharmacotherapy of depression, schizophrenia, and anxiety disorders focused on safety and clinical efficacy. For acute treatment, efficacy was determined by general or disease-specific measures indicating the presence, frequency, and intensity of symptoms, behaviors, or feelings as rated by the physician (40). Common clinical outcome measures of general psychopathology include the Brief Psychiatric Rating Scale, Hopkins Symptom Checklist, Global Assessment Scale, and Clinical Global Impressions scale. Disease-specific scales, such as the Hamilton Rating Scales for Depression and Anxiety and the Schedule for Affective Disorders and Schizophrenia, are more popular among clinical researchers (11).

Similarly, trials of maintenance drugs have not collected the specific quality-of-life outcomes required for valuation studies. Maintenance therapies generally measure outcomes such as relapse rates, time between episodes (survival time), number and severity of subsequent episodes after treatment, and duration and severity of symptoms.

Measurement of HRQL for patients with psychiatric illness usually includes measures of mental, physical, social, and role functioning and overall well-being (43). Two types of HRQL questionnaires can be used: generic questionnaires, which are designed to measure level of functioning for all aspects of health, or disease-specific questionnaires, which are designed to measure the level of impairment for those aspects of health associated with a specific disease and its treatment. HRQL questionnaires should be completed by the patient whenever possible. Several studies have shown that patients with severe mental illness such as schizophrenia can successfully complete these questionnaires despite their psychological impairments (4, 24, 56).

HRQL has been measured more frequently in recent trials (43, 58). The instruments used include measures of social functioning (e.g., Social Adjustment Scale and the Work and Disability Scale) and quality of life (e.g., Short Form 36 [SF36], Quality of Life in Depression Scale [QLDS], Quality of Life Scale [QLS], Sertraline Quality of Life Battery [SQOLB], SmithKline Beecham Quality of Life Scale [SBQOL]; Depression Quality of Life Battery [DQOLB]—see Revicki and Murray {43}). These instruments allow investigators to measure emotional and social functioning, well-being, disability, and overall health status attributable to mental illness and its treatment (20, 43).

Although quality-of-life measures are being used more extensively to evaluate treatment outcomes in people with schizophrenia and major depression, most published studies present results from observational studies or uncontrolled clinical trials.

There are two exceptions for schizophrenia. First a study by Carpenter et al. (6) compared continuous versus intermittent dosing with fluphenazine. They presented HRQL results using the QLS (22). On this scale, no difference between treatment groups was observed over a two-year time period, although in the second year they did see some differences between the treatments in other disease outcome measures. Similarly, Lurie et al. (30) found no difference in measures of health status and physical, role, and social functioning between fee-for-service and pre-paid health care programs. Uncontrolled studies of HRQL in schizophrenia (e.g., Meltzer et al. {34}, Simpson et al. {47}) have shown that improvements in HRQL scale scores over time were associated with improvements in clinical symptoms.

One published study demonstrated the impact of treatment on quality of life using data from randomized, controlled trials for major depression. This study by Lonnqvist et al. (29) compared moclobemide with fluoxetine using the SF20 and a symptom checklist. They showed a difference at two weeks in the physical functioning scale of the SF20, but no difference at six weeks. In addition, a recent review of the pharmacoeconomics of sertraline (7) cited several abstracts that demonstrated a quality-of-life benefit with sertraline, compared with other antidepressants. Souêtre et al. (49), using a cross-sectional, nonrandomized design, found a quality-of-life benefit for fluoxetine, compared with tricyclics. Furthermore, several studies indicated that improvements in HRQL scale scores occurred at the same time as improvements in the clinical symptoms of major depression. For example, Turner (55) demonstrated statistically significant improvements in all nine domains of the sertraline quality-of-life battery in six-week and eight-week open-label trials of sertraline. Patients also had significant improvements in the clinical measures used in the studies. Similar results have been shown for the QLDS (36, 19) and for the DQOLB (44). Improvements in patient functional status were also shown in an uncontrolled study of treatment of depressed patients with bupropion SR, using the clinician-rated Work and Social Disability Scale (33).

The main conclusion that can be drawn from the published studies of the impact of treatment on HRQL in psychiatric illness is that improvements in clinical symptoms are associated with improvements in patients' HRQL, but the ability of the scales to distinguish between the effects of different treatments is still in question.

HRQL measures provide meaningful information about the drug's effect on functioning and well-being that decision makers can use to assess drug value. However, these measures do not provide the direct, quantifiable, utility-based, quality-of-life measure that economists prefer (described below), nor are they substitutes for direct, quantifiable measures of productivity and resource use that are necessary for economic valuation studies. Apart from using an expert elicitation process to link scores on these scales to preference-weighted well-being measures, resources used, or productivity levels, we have no direct way to determine a drug's economic value based on these scales.

To perform an economic assessment of drug value, economists often measure quality-of-life changes as the difference in the patient's utility between perfect health and alternative impaired health states (8, 53, 54), where utility is a measure of overall patient well-being that lies between 0 and 1. Several techniques exist to elicit individuals' utility of alternative health states. One technique, category scaling, asks a person to place several alternative health states in the appropriate place on a line bounded by zero (death) and one (perfect health). Other quality-of-life valuation techniques include the standard gamble and time trade-off methods (8). Using the standard gamble method, people are asked to choose between a certain less-than-perfect health outcome and an outcome of either perfect health with probability 1-p or death with probability p. The probability of death at which the person is indifferent between the choices is equal to the utility of the certain less-than-perfect outcome (8). A typical standard gamble question might be: "Imagine that there is a new, free medication available that will either completely cure your mental illness or will kill you. Suppose that 50 percent of the people who take the new medication are cured of the disease, and 50 percent die. Would you risk taking the new medication?" If the individual answered "yes," the probability of dying would be successively increased until the individual answered "no." Similarly, if the individual initially answered "no," the probability of death would be successively reduced until their answer changed to "yes." In either case, the probability at which the answer to the question changes represents the utility of being mentally ill (51, 3, 41). In the time trade-off method, a person is asked to give the length of time in perfect health that is equivalent to a full lifetime in selected, less-than-perfect health states (54). These methods may be limited, however, because people do not necessarily give consistent or sensible answers to these elicitation methods (51). These problems may be particularly acute for patients with psychiatric disorders.

Although hundreds of clinical trials have been conducted to determine the efficacy of medications for major depression, schizophrenia, and anxiety disorders, there are very few published economic valuation studies of these drugs. A review of the literature on health-care cost-benefit and cost-effective analyses between 1979 and 1990 found only nine published studies on psychiatric medications, compared with almost 200 studies on medications for nonpsychiatric diseases (13).

Until recently, clinical trials for pharmacotherapy of psychiatric disorders have not provided the data economists need to conduct drug valuation studies. Because of these limitations, most published economic studies of the costs of psychiatric disorders have relied on secondary data for their estimates (e.g., Andrews et al. {2}, Stoudemire et al. {50}). Very few have been able to determine the economic value of alleviating a psychotic episode or preventing recurrence.

Recently, researchers have started to collect economic outcomes as part of their clinical trial protocols, especially for expensive drugs with improved efficacy relative to alternative treatments. For example, clozapine, a relatively new treatment for neuroleptic-resistant schizophrenia, has been the subject of several cost-effectiveness studies in the past several years (25, 35, 42).

Many clinical trial protocols now incorporate some or all of the following economic outcomes for psychotherapeutic drugs:

·mortality

·resource-use measures

·labor market and household productivity measures

Conducting a resource-use analysis for treating psychiatric disorders entails collecting data during the clinical trial on inpatient and outpatient resource use. Health-care resource costs can be estimated separately from the clinical trial using standard charge schedules and cost-to-charge ratios. Examples of economic outcome studies that have collected data on hospitalization costs and physician charges for psychiatric disorders include Meltzer et al. (35), Revicki et al. (42), Kamlet et al. (27), Simon et al. (46), Simon et al. (45), Revicki et al. (41), and Aberg-Wistedt et al. (1).

Simon et al. (46) provide an excellent example of the use of random control trials to study the consequences of alternative treatment decisions under usual clinical care conditions. The study compares the clinical, quality-of-life, and economic outcomes of initially prescribing fluoxetine (a serotonin reuptake inhibitor) versus imipramine or desipramine (tricyclic antidepressants). HMO patients were randomly assigned initially to one of the three drugs, and patients and physicians made subsequent treatment decisions following usual clinical practice patterns. Patients beginning therapy with fluoxetine reported fewer adverse events and were more likely to continue the original medication. Quality-of-life and clinical outcomes (after three months) were essentially identical for all three drugs. Total health care costs over six months were also approximately equal, with the higher cost of fluoxetine offset by lower outpatient and inpatient costs.

Other important endpoints that should be collected in clinical trials are labor market and household productivity effects. These measures include the days of work the patient gained, the level of function of those days gained, and the paid and non-paid caregiver time avoided as a result of treatment. Examples of questions that might be added to assess these effects include the following: "Since we saw you last, how many days of work did you miss because of panic attacks?" and "Since we saw you last, how many days did a non-paid caregiver miss work to take care of you while you felt depressed?"

A dollar value of lost work time can be calculated by multiplying the foregone days of work attributable to the disorder by a wage measure. However, this method has some limitations. First, if the patient is not employed, no wage measure exists. Thus, an estimated wage would have to be developed for those patients who are too ill to work, are unemployed, are retired, or are homemakers. Second, patients may be reluctant to provide income information during the clinical trial. Finally, this method of valuing productivity changes, known as the human capital approach, is controversial. Grabowski and Hansen (18) suggest that such a procedure "measures health and quality of life as though they are a unit of production, not something of intrinsic value." As a result, the human capital approach to valuing the productivity gains of a new depression drug treatment gives more weight to high wage earners than low wage earners. In spite of their limitations, productivity questions are now included in many clinical trial protocols.

Economists often employ an alternative method of valuing drugs —individuals' willingness to pay for a drug rather than go without it. Although this willingness-to-pay definition is considerably more encompassing than cost-outcome analyses, and it captures economists' fundamental definition of value, willingness to pay must be measured by a preference elicitation method such as questionnaire-based models or revealed preference methods (18). Because these elicitation methods are still relatively new, they are not yet widely accepted by the medical community (18).

The willingness-to-pay method determines the amounts of money people are willing to pay to avoid various possible health states (51). For example, an investigator might ask the patient: "Consider all the effects of your anxiety on your life. How much would you currently be willing to pay each week, realistically, to get rid of your anxiety and all the problems it brings?" (52) This method is potentially limited because respondents may not be able to give consistent or rational personal judgments on their willingness to pay (51). Furthermore, willingness-to-pay estimates are controversial because the magnitude of the estimates is likely to be affected by the respondents' income level, such that higher-income respondents might have a greater willingness to pay for good health (and hence a greater value for good health) than lower-income respondents.

We recognize that methods of valuing economic outcomes, particularly productivity and willingness-to-pay approaches, have limitations and are currently controversial. Although none of these techniques has been widely used or accepted as standard, conceptual work to advance their use in clinical trials for mentally ill persons is under way. The development of common definitions and standards of measurement will allow researchers to accumulate comparable data across studies and across populations (28). In turn, these data will help economists improve their economic valuation studies of drug treatment.

Clinical trials data can be used in conjunction with other data sources to develop disease/outcome models that highlight clinical decision points, alternative treatment choices, and the resulting possible outcomes. Using these models, researchers can illustrate the temporal and logical sequence of the disease/treatment dynamic, combine the results of acute and maintenance clinical trials, and control for natural recovery that may occur apart from drug treatment.

We suggest that clinicians develop disease/outcome models, either using a decision-tree or Markov approach, concurrently with clinical trial protocols to help guide choices of outcome measures and to ensure that the trials collect data on the appropriate economic endpoints. Well-designed clinical trial protocols can estimate the transition probabilities between alternative disease states that are needed for the disease/outcome model. Decision-tree models provide a useful framework for organizing cost-effectiveness, cost-utility, and benefit-cost analyses.

Figure 1 and Figure 2 illustrate treatment choices, disease/treatment dynamics, and outcomes using a simplified example of a decision-tree model. Our focus in these figures is on modeling the treatment of major depression, but this is meant to be illustrative of modeling psychiatric disorders generally. The basic structure of this decision-tree model would be essentially the same for anxiety disorders or schizophrenia.

Figure 1 represents the disorder/treatment dynamics for an acute depressive episode; Figure 2 represents a phase in which the patient is stable and under maintenance therapy. We make several simplifying assumptions in this model: 1) depression occurs at four basic severity levels (i.e., mild, moderate, severe without psychotic symptoms, and severe with psychotic symptoms); 2) the patient is only affected by unipolar depression; 3) four treatment choices (i.e., drugs only, psychotherapy only, drugs and psychotherapy combined, and ECT) are available; and 4) treatment switching decisions are made only at eight weeks and six months after the start of treatment.

As illustrated in Figure 1, the patient enters treatment in one of the four possible severity levels. The physician selects a treatment based on the patient's initial severity level. After the initial treatment visit, the patient is reexamined periodically; in subsequent visits, the patient's health status will have improved, worsened, or remained the same. Alternative treatments may affect the length of time spent in each severity level and, hence, the functional loss of the patient. Each severity level is associated with health-care resource use, days of lost work or reduced productivity, and quality-of-life-changes that can be measured at various time intervals.

Depending on the success of the acute treatment and the patient's history of depression, several treatment paths may be taken at the eight-week and six-month time points. Those patients who are asymptomatic after eight weeks or six months and have no previous history of depression can end treatment at either point. Those patients who still have mild depression or worse at the eight-week or six-month time points are still in the acute phase; their treatment will be evaluated and possibly changed. Asymptomatic or mildly depressed patients with a history of three or more prior depressive episodes may be treated with continuation therapy to avoid relapse at the eight-week point, or these patients may enter maintenance therapy to avoid recurrence at the six-month point. Most patients with a history of recurrent depressive episodes will likely proceed to maintenance therapy.

Figure 2 summarizes the disorder/treatment dynamics for the maintenance phase of depression treatment. Patients enter maintenance therapy with either mild residual symptoms or no symptoms. After the initial maintenance treatment visit, the patient periodically visits the physician, who assesses the efficacy of the treatment. At the six-month assessment point, the patient's mental status will have either improved, worsened, or stayed the same.

This example of depression/treatment dynamics illustrates the connection between changes in disease stage, resource use, and quality-of-life endpoints. Measuring the movements between severity levels provides estimates of transition probabilities for decision-tree models. These estimates can then be used in conjunction with estimates of resource use to simulate the costs and benefits of new drugs in the context of the complete drug/treatment dynamics.

Kamlet et al. (27) recently operationalized this type of decision model in their cost-utility analysis of maintenance therapy for recurrent depression. The study compared three maintenance treatments for recurrent depression: interpersonal therapy, imipramine therapy, and a combination of the two. They drew on results from the Pittsburgh Recurrent Depression Project (15), in which the clinical effectiveness of the alternative treatments in delaying recurrence of depression was evaluated. Although the trial lasted only three years, Kamlet et al. developed a lifetime model of acute episodes followed by stable periods under maintenance treatment. To implement their cost-utility model, they required estimates of the following variables: time until recurrence, time spent in a depressive episode, probability of suicide, quality of life during a depressed episode and during maintenance treatment, and direct costs per episode of acute and maintenance treatment. Because their clinical trial data were limited to the maintenance phase, the analysts had to base the other parameter values on secondary data supplemented with assumptions or consensus of expert opinion. They combined all these clinical, quality-of-life, and economic elements into a Markovian state-transition model. They found that imipramine therapy results in improved QALYs and lower medical costs, compared with interpersonal therapy. Furthermore, the combination of interpersonal therapy and imipramine therapy compared to interpersonal therapy alone results in a cost per QALY estimate below $5,000.

Two recent articles, Revicki et al. (41) and Anton and Revicki (3), used decision modeling techniques and a Markovian state-transition model to compare antidepressive treatment in women with nefazodone, a new antidepressant with a dual mechanism of action on the serotonergic system (41), to treatment with imipramine or fluoxetine. As in Kamlet et al. (27), the data for the model were derived from a variety of sources: the medical literature, clinical trials data, and physician judgment. The base case results show that nefazodone treatment costs $Can1,447 less per patient than imipramine and increases QALYs by 0.72 (41). Similarly, nefazodone treatment costs slightly less than fluoxetine treatment ($Can14) and increases QALYs by 0.11. The authors conclude that "nefazodone may be a cost-effective treatment for major depression compared with imipramine and fluoxetine." Another recent article (21) also developed a Markov state-transition model to compare maintenance therapy with a serotonin reuptake inhibitor (sertraline) to episodic treatment with a tricyclic antidepressant (dothiepin). Jönsson and Bebbington (26) used a decision-tree model to measure the overall direct costs of depression in the UK and to illustrate issues in the evaluation of alternative antidepressant therapies.

Although disease/outcome models are a useful tool for economic evaluation of new drugs, analysts must use caution when attempting to generalize the results of clinical trials beyond the trial setting. First, the results of individual clinical trials—which often evaluate short-term effects and are directed at the acute phase of a disorder—may provide a misleading picture of the entire disorder/treatment dynamics of that drug. This is especially likely to be a problem for drugs used for treating diseases with a prolonged duration or chronic diseases that recur over a person's lifetime (as with many psychiatric disorders). Another limitation of clinical trials data is that patients in a clinical trial may not be representative of the diseased population (45). Because trials are typically conducted in tertiary care settings, patients participating in the trial are a highly selected subset of all those who have the disorder being studied. Furthermore, the clinical trial takes place in an idealized setting, where patient compliance with dosage is monitored more closely than with typical patients in an actual clinical setting. Thus, the drug's efficacy is likely to be overestimated when applied to patients outside the clinical trial setting. Finally, clinical trials usually compare the new drug with placebo rather than with an existing drug or the suggested medical therapy, leaving policymakers with little information about comparative outcomes (16, 38 ).

Because of these limitations, we suggest performing a sensitivity analysis of model parameters and validating parameter estimates from clinical trials against other data. Sensitivity analysis involves substituting a range of estimates for the probabilities and resource-use estimates to see whether they alter the model's conclusions. Eisenberg et al. (12) suggests comparing clinical trial data with data from other trials, expert opinion, and information from large databases such as Medicare claims files. Finally, researchers can validate decision-tree models with data from actual clinical practice or from post-marketing economics studies. For example, randomized trials can be developed with the primary purpose of estimating differential resource use across alternative drugs. As more information becomes available, researchers will be able to build better models of the mental disease/treatment dynamics.

Health outcomes measurement and economic valuation studies will help patients, physicians, pharmaceutical companies, and policymakers decide whether new therapies justify their additional expense. Given the advent of expensive new drugs and limited health care resources, comprehensive health outcomes assessment and economic valuation studies based on clinical trials data are critical to sound decision making.

In the past, clinical trials have not provided adequate data for a comprehensive valuation of psychotherapeutic drugs. However, researchers are increasingly recognizing the importance of incorporating quality-of-life and economic outcomes into clinical trial protocols. Trials are now collecting data on resource use, lost labor market productivity, and quality of life that are needed for sound economic analysis.

These enhanced data can be used to develop disease/outcome models to guide a comparison of the costs and outcomes of alternative drug therapies. Disease/outcome models can be developed that incorporate the results of separate acute and maintenance trials into a single model. These models can be validated and improved with data from other trials, expert opinion, post-marketing studies, and large databases such as Medicare claims files. As clinical trial protocols continue evolving to collect more economic information, researchers will be able to build more sophisticated economic models to evaluate the merits of new psychotherapeutic drugs in the context of the mental disorder/treatment dynamics.

published 2000