Cost utility analysis of maintenance treatment for recurrent depression

Cost Utility Analysis of Maintenance Treatment for Recurrent Depression Mark S. Kamlet, PhD, Nancy Paul, MS, Joel Greenhouse, PhD, David Kupfer, MD, Ellen Frank, PhD, and Martcia Wade, PhD Carnegie Mellon University and University of Pittsburgh (M.S.K., J.G.), Carnegie Mellon University (NJ’.), University of Pittsburgh (D.K., E.F.), and Urban lnstitute and Carnegie Mellon

University

(M. IV.)

ABSTRACT: This paper presents a cost-utility analysis of three maintenance treatments for recurrent depression: interpersonal therapy (VT-Ml, imipramine drug therapy (Drug), and a combination of the two. We base our analysis on the results of the University of Pittsburgh’s Controlled Clinical Trial of Maintenance Therapies for Recurrent Depression. We construct a Markovian state-transition model to incorporate clinical effectiveness into cost and quality-of-life impacts; we assign empirical values to the parameters of this model; and we then use Monte Carlo analysis to compare the relative cost effectiveness of the different maintenance treatments. For the patients who met the eligibility standards for the study, Drug maintenance treatment is cost-effective in the strongest sense of the term compared to either a placebo group or IPT-M: it both improves expected lifetime health (measured in quality-adjusted life years, or QALYs) and reduces direct medical costs. This is true even when relatively severe side effects of the drug are considered. Compared to the placebo group, VT-M and the combination of IPT-M and Drug each improve expected lifetime health, although in neither case are expected direct medical costs reduced. Still, the cost of the resulting health improvements, under $5000/QALY, are very reasonable. A similar conclusion holds comparing Drug and IPT-M to IPT-M alone. All of the above conclusions are quite robust to sensitivity analyses.

KEY WORDS: model

Recurrent depression, cost-utility analysis, Monte Carlo simulation, mixture survival

INTRODUCTION Clinical depression affects l-3% of the U.S. population during any 6-month period, according to conservative estimates 111. The lifetime incidence of depression is over 15%. Depression is a debilitating illness that, as described by Stoudemire et al., Address reprint requests to: Nancy Paul, MS, Carnegie Mel/on University, Department Baker Hall 232, Pittsburgh, Pennsylvania 15213. Received June 23, 1993; revised May 12,1994. Controlled Clinical Trials 16~17-40 0 Elsevier Science Inc. 1995 655 Avenue of the Americas, New York, New York 10010

of Statistics,

0197.2456/95/$9.50 SSDI 0197-2456(94)00020-4

18

M.S. Kamlet et al. usually leads to withdrawal from social, work, and family activities, decreased motivation, feelings of hopelessness, low self-esteem, pessimism, and self-blame. Withdrawal from family relationships can result in separation, divorce, and parent-child problems. Irritability, anger, and withdrawal can cause depressed patients to become isolated in their misery as others are driven away by their despair. Depression may lead to poor work performance, absenteeism, and unemployment. The cognitive impairments (difficulty in concentrating and memory loss) can resemble a form of dementia. . Depressed individuals have a high rate of physical complaints, especially pain, headaches, insomnia, and digestive problems [Z].

is linked not only to morbidity but to mortality. Monkoff et al. 131 suggest that 60% or more of people who commit suicide have clinically significant depression as the primary psychiatric disorder, accounting for about 16,000 deaths per year ]21. For many people, depression is a recurrent phenomenon. It has been estimated that 50% of those suffering one depressive episode will suffer another within the ensuing 10 years and that those who have experienced two depressive episodes have almost a 90% chance of experiencing a third. Once the disorder has become recurrent, rates of relapse are high. Keller et al. [4,5] indicate that individuals with three or more lifetime episodes of depression may have relapse rates as high as 40% within 15 weeks after recovery from a given episode. Klerman [61 indicates that 65% of recurrent depressives have some degree of relapse within the first year if untreated. Given the debilitating nature of depression and the high rate of relapse, attention has recently focused on ways to treat patients between depressive episodes. These “maintenance treatmentCare intended to lower the probability and/or duration of future episodes. Early research on maintenance treatments involved the use of lithium carbonate [71, although this generally proved less effective than hoped [Bl. Subsequent research has indicated that imipramine is more effective than lithium carbonate in protecting against depressive recurrences, particularly for individuals with severe index episodes 191. There has been substantial interest not only in drug maintenance therapies for recurrent depression but in psychotherapy as well. Klerman et al. [lo] suggest that while medication affects neurovegetative symptoms of depression, such as sleep and appetite, psychotherapy affects aspects of depression such as mood, self-esteem, and social functioning. Based on the premise that depression develops in a psychosocial context, recent research has examined the effects of interpersonal therapy (IPT) as a maintenance treatment. IPT helps the patient understand and renegotiate difficult interpersonal relations. It is thought to be instrumental in recovery from depressive episodes and may also help prevent future episodes. This approach helps the patient develop more effective strategies for dealing with social and interpersonal problems associated with the onset of depression. IP’I has been found effective in relieving depressive symptoms and in delaying or preventing recurrences in several studies, including those of DiMascio et al. 1111, Weismann [12,13], Prusoff et al. [14], and Klerman et al. [15]. Note, however, that all of these studies involved follow-ups of 1 year or less.

Cost Utility Analysis of Maintenance

Treatment for Recurrent Depression

19

In view of the promise of maintenance drug and therapy treatment, some have recommended maintenance treatment as generally desirable for recurrent depression. Angst 1161, for instance, suggests that long-term medication should be initiated in unipolar depression after the third episode, especially when the time interval between the latest two episodes is relatively short. However, there are trade-offs involved in any maintenance treatment. Some of these trade-offs involve economic considerations. Drug maintenance treatment involves the cost of medication, periodic blood tests, office visits, and so forth. Psychotherapy similarly involves the time of health care professionals (as well as the patient’s time). In addition, there are nonmonetary trade-offs to consider. In the case of imipramine maintenance treatment, for instance, some patients experience side effects such as blood pressure and ECG changes, interactions with other medications, sexual dysfunction, unwanted weight gain, constipation, lethargy, and dry mouth [17]. This paper presents a cost-utility analysis (CUA) of three maintenance treatments for recurrent depression-drug maintenance treatment, psychotherapy maintenance treatment, and the combination of the two. CUA evaluates a health intervention by comparing the incremental societal costs of a health intervention and the incremental health benefits that result from it. Typically, the outcome of a CUA is expressed as a ratio, with the units being dollars per quality-adjusted life years ($/QALY’s) 118-201. We will refer to this ratio as the cost-effectiveness ratio. We base our analysis on the results of the University of Pittsburgh’s Randomized Controlled Clinical Trial of Maintenance Therapies for Recurrent Depression 1211. This study was conducted at the Western Psychiatric Institute and Clinic, University of Pittsburgh, under principal investigators Dr. Ellen Frank and Dr. David Kupfer, and funded by the National Institute of Mental Health. In this study, individuals with recurrent depression were randomly assigned to one of five maintenance treatment protocols: maintenance IPT (IPT-M) alone, IPT-M with imipramine drug therapy, IPT-M with a placebo drug, imipramine drug therapy alone, and placebo alone. The next section provides a brief description of the trial (see also 1211). Following that is an analysis of the clinical effectiveness of the different maintenance treatments in preventing or delaying the recurrence of depression. The fourth section provides a state-transition model for incorporating this information into a cost-utility analysis. The fifth section discusses the assignment of empirical values to the parameters of the model. The sixth section presents the empirical results. The last section provides a brief conclusion. CLINICAL EFFECTIVENESS OF THE MAINTENANCE FOR RECURRENT DEPRESSION

TREATMENTS

The design and results of the randomized controlled clinical trial on recurrent depression are reported in Frank et al. [211. Briefly, to enter the trial, subjects between the ages of 21 and 65 years were required to be seen initially in their third or greater episode of unipolar depression, with the immediately preceding episode being no more than 2.5 years before the onset of the

MS. Kamlet et al.

20

present episode. In addition, patients were required to report a minimum lo-week remission according to Research Diagnostic Criteria (no more than two symptoms present to no more than a mild degree with absence of functional impairment) [221 between the current or index episode and the immediately preceding episode. A minimum Hamilton Rating Scale for Depression (HRSD) [23] score of 15 and a minimum Raskin Severity of Depression [24] score of 7 were also required. Patients were excluded if they met criteria for any other axis 1 diagnosis, except generalized anxiety disorder or panic disorder, or if they met criteria for antisocial or borderline personality disorder. Patients with some borderline or antisocial features were, however, included. Medical exclusions were based on any condition considered to be incompatible with imipramine therapy. Following a preliminary evaluation, all eligible subjects then received the same short-term treatment regimen consisting of a combination of imipramine hydrochloride (150-300 mg) and IPT [lo]. Treatment sessions were scheduled weekly for 12 weeks, then biweekly for 8 weeks, and then monthly. Patients continued to receive combined treatment for an additional 17 weeks, during which both HRSD and Raskin scores and imipramine dosages were required to remain stable. Patients who completed this initial stabilization phase began the experimental phase of the study and were randomly assigned to one of the five maintenance treatments for a period of 3 years or until they experienced a recurrence of illness. The five maintenance treatments were (1) a maintenance form of IPT (IPT-M) offered alone; (2) IPT-M with active imipramine therapy continued at the acute treatment dosage; (3) IPT-M with placebo; (4) medication clinic visits with active imipramine therapy; and (5) medication clinic visits with placebo. Of the 230 patients recruited during the acute treatment phase, 128 patients ultimately entered the maintenance therapy phase (26 in treatment 1, 25 in treatment 2, 26 in treatment 3,28 in treatment 4, and 23 in treatment 5). Of the 102 patients who failed to reach the maintenance therapy phase, 46 (45%) failed to respond fully to the IPT and imipramine combination treatment and 21 (21%) developed intolerable side effects. Of the 128 patients who did reach the maintenance therapy phase, only 22 (17%) failed to complete the 3-year protocol (mostly due to noncompliance). These 22 patients were treated as censored observations. It is important to note that the patients who entered the maintenance phase are special in the sense that they responded to the acute phase treatments and were able to remain stable on these treatments for a period of time before being randomized. Therefore they are not necessarily representative of all patients with major depressive disorders [25] and inferences from these data should therefore not be generalized beyond this category of patient.

MAINTENANCE DEPRESSION

TREATMENT

AND THE RECURRENCE

OF

Frank et al. [21] use survival analysis to assess the clinical effectiveness of the different maintenance treatments in delaying or preventing the recurrence of depression, where the event of interest is the first recurrence of depression

21

Cost Utility Analysis of Maintenance Treatment for Recurrent Depression

r

I-

Fr;-M & Drug

....-.. - - -

IPT-Bn IPT-M 8 Placebo Placebo

-------------

____------I------

L________

I

‘1

.-.

I I_____ I I ______---_

1

L

:

I

L____________-

L

0

50

100

150

Time (in Weeks)

Figure 1 Kaplan-Meier curves for time until recurrence. after randomization to maintenance therapy. Figure 1 displays the KaplanMeier product limit estimator of the survival, or “time until recurrence,“curves. These curves indicate the proportion of subjects at a given time t who have not yet had a recurrence in each maintenance treatment. The Drug and the IPT-M & Drug maintenance treatments have the largest proportion of subjects who are recurrence-free over the entire 3-year experiment, with approximately 77% of the patients surviving 3 years without a recurrence; the IPT-M & Placebo and the IPT-M maintenance treatments are the next most effective, with approximately 28% surviving 3 years without a recurrence. Finally, only approximately 10% of those on Placebo remained stable for 3 years. As reported in Frank et al. 1211, based on the Mantel-Cox test, the differences between the IPT-M & Drug and Drug groups, and between the IPT-M & Placebo and IPT-M groups were not significant (p =0.80 and p =0.65, respectively). The IPT-M & Drug and Drug treatments perform significantly better than either of the IPT-M & Placebo or IPT-M treatments (p =0.0007, p =0.0026, ~7 =0.0059, and p =0.0144, respectively). In addition, both the VT-M & Placebo and the IPT-M treatments are more effective than the Placebo (p =0.1038 and p =0.0444, respectively). A careful examination of the Kaplan-Meier curves suggests that after a period of time in maintenance, the risk of a recurrence diminishes markedly.

22

M.S. Kamlet et al. Table 1

Maximum

Likelihood

Estimates of IT; and bi (and Standard

Group

=TTi

Errors)

Pi

IPT-M

0.3214 (0.104)

0.0223 (0.006)

Drug

0.7547 (0.088)

0.0605 (0.023)

Placebo

0.1023 (0.074)

0.0333 (0.008)

IPT-M & Drug

0.6746 (0.165)

0.0135 (0.013)

IPT-M & Placebo

0.2331 (0.114)

0.0198 (0.006)

This observation is consistent with a heterogeneous population for each maintenance group, composed of a proportion whose response to maintenance treatment is such that the probability of a recurrence during the 3 years of the maintenance phase is negligible, and the remaining proportion, who continue to face positive probability of recurrence throughout the maintenance phase. To capture this feature of the data, we posit a parametric “mixture” model for the distribution of the time until recurrence for each treatment group of the form [26,271: S;(t) =7Fi + (1 - n,) exp(-bi

* t) 0 IT,

51, f >O, I*i >O

(1)

In this model ITSrepresents the proportion of subjects in maintenance treatment i who are “cured” in the sense that they are not at risk of a recurrence of depression while on maintenance treatment i (ITSis also called a “surviving fraction”) and IA;I represents the average time until recurrence for patients in maintenance treatment group i who do not belong to the surviving fraction and hence are at risk of a recurrence of depression. Table 1 displays the maximum likelihood estimates of 71i and pi for each treatment group. The plots of the resulting estimated time-until-recurrence curves are presented in Fig. 2. Consistent with the conclusions from Fig. 1, the Drug and IPT-M & Drug maintenance treatments are the most effective in preventing or delaying the recurrence of a depressive episode with estimated surviving fractions of 0.75 and 0.67, respectively, and estimated average recurrence times for those not in the surviving fraction of 17 and 91 weeks, respectively. They are more effective than IPT-M and IPT-M & Placebo, which have estimated surviving fractions of 0.32 and 0.23, respectively, and estimated average recurrence times for those not in the surviving fraction of 45 and 51 weeks, respectively. Both of the latter maintenance treatments are more effective, in turn, than the Placebo alone, which has an estimated surviving fraction of only 0.10 and an estimated average recurrence time for those who recur of 30 weeks. We note that the mixture model allows us to capture more of the relevant features of the data, such as the qualitative difference in distributions of the time until recurrence between patients on the Drug and IPT-M & Drug maintenance treatments which the estimated Kaplan-Meier time-untilrecurrence curves seem to suggest. By specifying a surviving fraction in the

Cost Utility Analysis of Maintenance Treatment for Recurrent Depression

23

---

-----_

1

0

50

100

150

Time (in Weeks)

Figure 2

Exponential with surviving fraction mixture model. mixture model it is possible to estimate the proportion of patients who do not experience a recurrence during the course of the study. We see that more patients on the Drug maintenance treatment tend to fall in the surviving fraction, but that those who do have a recurrence in this group do so relatively quickly. On the other hand, a slightly lower proportion of patients in the IPT-M & Drug maintenance treatment group fall in the surviving fraction, but for those who do have a recurrence of depression, their expected time until recurrence is much longer than for those patients in the Drug alone group. Thus, under the mixture model, while IPT-M and IPT-M & Placebo remain clinically indistinguishable, for the average patient one might now clinically prefer VT-M & Drug to Drug alone. We will use the mixture model to describe the time between recurrences of depression in the cost-utility analysis.

A STATE-TRANSITION DEPRESSION

MODEL

FOR ANALYZING

RECURRENT

The analysis in the previous section speaks to the clinical efficacy of different maintenance treatments in delaying and preventing recurrence of depression. To analyze these treatments from a cost effectiveness perspective, however, it is necessary to translate the clinical results into their implications

24

MS. Kamlet et al.

for the overall health and, more generally, the quality of life of the individuals being treated. It is also necessary to consider the economic costs associated with each treatment. We integrate the quality of life, length of life, and cost data through a cost-utility analysis. In order to conduct a cost-utility analysis of the maintenance treatments considered here, costs and health outcomes must be considered in a dynamic context since they occur continually over the life of a subject. To do this, a simple state-transition model is presented for recurrent depression. For a given maintenance treatment i, an individual at any given time can be in one of three states: (A) not depressed and under the specified maintenance treatment; (B) depressed; or (C) dead. Death, in turn, can be due to natural causes or due to suicide caused by depression. In this context, an individual can move from state A to state B-a recurrence; from state A to state C-death by natural causes; from state B to state A-stabilization; and from state B to state C-death by suicide. The amount of time the individual is in state A and B is stochastic. Given that a state transition is to occur, we assume a Markovian model whereby the probabilities of transitions from state A to state B, state A to state C, state B to state C, and state B to state A are independent of the past history of the individual. The time line of analysis for a given patient begins when a patient enters the maintenance treatment. At this point, the patient is by definition in state A, having successfully responded to treatment during the preliminary stabilization phase. We assume that the individual remains stable for a period of time, Y. She then has a recurrence of depression (i.e., a move from state A to state B). There is some probability, p, that she commits suicide during this depressive episode (i.e., moves from state B to state C). If not, she remains depressed and receives treatment for the depression for a period of time, X, after which she stabilizes again (i.e., moves from state B to state A). This process continues until a patient either commits suicide or dies of other causes. In this framework, maintenance treatment is intended to increase Y, the time interval between recurrences. It may also shorten the length and dampen the magnitude of future episodes. The literature offers no direct evidence on these latter possibilities, however, and to be conservative we will restrict the impact of maintenance therapy to its effect on Y. Quality of life in our model depends on the presence or absence of depression and on the specific maintenance treatment. To be sure, the relationship between quality of life and the specific pattern of depressive episodes is much more complex. The effects of a depressive episode on an individual’s social functioning often persist long after the episode itself, particularly in marital and close interpersonal relationships 1281. Therefore, if maintenance treatment reduces the frequency of recurrences, it may well improve the quality of life for the individual even during periods of stabilization. But, again, to be conservative we will consider maintenance treatment only as it affects the frequency of episodes. Adopting this perspective, let + be the quality of life when depressed. +I is between 0 and 1, where 0 is death and 1 is perfect health. (b measures the severity of a given health state in terms of the overall quality of life experi-

Cost Utility

Analysis

of Maintenance

Treatment

for Recurrent

Depression

25

enced by the individual. Quality of life as used here encompasses the effects of depression not only on health narrowly conceived, but also on the individual’s social functioning, societal role function (e.g., worker, homemaker, etc.), and leisure. The closer 6 is to 1, the less severe the impact of depression on an individual’s quality of life. More technically, 4 reflects a Von NeumannMorgenstern (VNM) utility rating 1291. It can be interpreted in several ways. Perhaps the simplest interpretation of its use in a cost-utility analysis is in terms of trade-offs between the length and the quality of life. As an example of this interpretation, suppose the quality of life of a given chronic illness is rated at 0.9. In terms of its use in cost-utility analysis, this means that the individual would be indifferent between living 10 years with the chronic illness and living only 9 years but in perfect health. As indicated above, there may be an impact of a maintenance treatment on the quality of a patient’s life even when he or she is not depressed. Let l3, be the quality of life when stable and on maintenance treatment i. As with +, pi also reflects VNM utility and is between 0 and 1, where 0 is death and 1 is perfect health. The direct costs of maintenance treatment considered here include the time of the health professionals involved in the treatment, the use of office space, the cost of any medication and lab work, and the patient’s time. Let D, represent the direct cost per unit time associated with maintenance treatment i while the patient is stable. There are also, of course, medical costs associated with treating depressive episodes. Let C represent the direct cost per episode, where C is assumed to be independent of the maintenance treatment. While the state-transition model presented here is simple, it captures the key health, quality-of-life, and economic aspects required for a cost-utility analysis of maintenance treatment. At the same time, it makes a variety of simplifying assumptions beyond those that have already been highlighted. We assume that individuals who have a recurrence stabilize if they do not commit suicide, although, in fact, some remain chronically depressed. Once stabilized, however, our model does specify a positive probability that the patient (if not belonging to the “surviving fraction”wil1 experience another recurrence of depression. In addition, the model assumes that there are only two states of mental well-being-healthy and depressed-with no intermediate ground or levels of severity. In fact, many individuals do not return to full quality of life after a depressive episode but may improve only to a certain level of functioning. Additionally, we assume that individuals who are depressed seek professional treatment unless they commit suicide. In fact, many do not. Weissman and Myers [30] suggest that fewer than one quarter of the patients who have clinically significant depressive symptoms ever see a mental health professional. However, our assumption probably lends a conservative bias to our overall conclusion. This is suggested by the poor performance of those patients who receive the placebo maintenance treatment, which is still more support than not seeing a mental health professional at all. The state-transition model allows us to derive a formula for a cost effectiveness ratio for a given maintenance treatment relative to the placebo control group. For a given patient with recurrent depression who is stabilized and who at time t =0 begins maintenance therapy, i, let Hi represent the

26

MS. Kamlet

et al.

amount of time over the remainder of the patient’s life in which she is alive and nondepressed (i.e., healthy). Let Bi be the number of future depressive episodes she experiences. Finally, recall that X is the length of each depressive episode in which she does not commit suicide, and is assumed to be nonstochastic. Then, without discounting (see below), the direct costs associated with maintenance treatment over the course of the patient’s life are: HiDi + B,C

(2)

In a similar fashion, the expected number of quality-adjusted beginning maintenance treatment, again without discounting, $B,X + pJYi

life years after is: (3)

Any consideration of the cost-effectiveness of a given maintenance treatment must be in the context of some comparison treatment. In our case, we begin with the comparison maintenance treatment being “placebo.” That is, we are interested in the incremental costs and health outcomes for a patient with recurrent depression who has been treated for a depressive episode (as described above) and stabilizes, and who then receives a given maintenance treatment to prevent the recurrence of an episode relative to that patient receiving only a placebo for maintenance. The direct costs associated with a subject experiencing the placebo treatment are: (41

B&

where it is assumed that there are no direct costs during stabilized periods and where the subscript “Pl” stands for “placebo.” The expected number of quality-adjusted life years after beginning maintenance treatment, again without discounting, is: (5)

W&,X + &I

where it is assumed that the quality of life while stable and undergoing the placebo treatment, pis equal to 1. Thus, without discounting, the formula for the cost effectiveness ratio of maintenance treatment i relative to the placebo is: [HiDi + (Bi - B,,)Cl/I$(Bi - B,,)X + PiHi - HP11

(6)

Note that the numerator gives the difference in cost while the denominator gives the difference in the quality of life. Hence, the units are dollars per to derive similarly a cost quality-adjusted life year. It is straightforward effectiveness ratio for one maintenance treatment i compared to another maintenance treatment i: [HiDi - HiDj + (Bi - Bi)C]/[~

(Bi - Bj)X + PiHi - PiHi].

(7)

In most cost-utility analyses, future costs and benefits are “discounted,” SO that future costs and health outcomes do not count as heavily as current costs and health outcomes [18]. The above formulas, however, are derived assuming no discount rate. If one wishes to discount future costs using a discount factor rl and to discount future health/quality-of-life outcomes using a dis-

Cost Utility Analysis of Maintenance Treatment for Recurrent Depression

27

count factor r2, then it is necessary for computing Hi and B,X to index the time f at which an individual is stable or depressed, respectively. If an individual is stable for some duration from f to f + A, rather than this “counting” as an increment of Ato H,, it now counts as Ae-‘1’. Notationally, let H,(f) be an indicator variable, with the value of 1 if the individual is stable at time f and 0 if the individual is depressed or dead at time f. Then H,, as it would be used in Eq. (6), becomes: Hi = J H,(f)e-rltdf

(8)

Similarly, direct costs experienced while depressed and health/quality-of-life outcomes when depressed and when stable must be discounted in this fashion as well (by discount factors Y, for costs and r2 for health outcomes). Hi and B, in Eqs. (6) and (7) are stochastic. It is analytically difficult to derive explicitly the expected value for such cost effectiveness ratios. The problem becomes almost intractable in the presence of discounting. To alleviate these problems, we chose to solve for the expected value of the ratios in Eqs. (6) and (7) through computer simulation. This approach, known as Monte Carlo analysis, provides a flexible and straightforward alternative to deriving an explicit analytic expression for these expected values. To conduct the Monte Carlo analysis, the distribution for Yi, the time to recurrence under maintenance treatment i, is specified and an individual’s life experience under this maintenance treatment is simulated, with results for Y for each period of stabilization drawn from this distribution. By doing this a sufficiently large number of times, very accurate estimates can be made of the expected costs and health outcomes associated with given maintenance treatments. We simulate the life of a specific patient, namely, a 40-year-old female, N times keeping track of quantities such as the number of depressive episodes during each lifetime. Averages of such quantities (along with standard deviations) are then easily calculated. We chose N =lOOO as both the resulting size of the standard deviations and computing time were reasonable. Since the Monte Carlo analysis keeps a running track of costs and health outcomes over time for each simulation, it is very easy to account for discounting by simply weighing each cost and health outcome by the appropriate amount (based on the time of occurrence and the discount rate being employed) and converting it to a present value equivalent.

ASSIGNING NUMERICAL VALUES PARAMETERS OF THE MODEL

TO THE EMPIRICAL

The various parameters defined above must be assigned numerical values. To estimate the probability of death from natural causes (conditional on age and gender), we use a Gompertz distribution 1311.The hazard rate function at age x for the Gompertz distribution has two parameters, R and a, and is written: h,(x) =R eax

(9)

To obtain estimates for the parameters of the Gompertz distribution, we can write Eq. (9) as log&(x)1 =log R + ux. This allows us to regress the observed

28

MS. Kamlet et al. log hazard rate for mortality on age in order to estimate R and a. An estimate of the hazard rate for mortality is obtained from life tables for the U.S. population (Vital Statistics of the United States, 1987). For each Monte Carlo simulation we draw from the relevant Gompertz distribution a value indicating when a patient would die from natural causes if she had not committed suicide when depressed prior to this date. For the illustrative analysis presented in the next section, we assume that the patient is a 40-year-old female and use the estimated values R =0.0008 and u =0.0673. The parametric time-to-recurrence distributions for Y for each maintenance treatment are taken directly from the estimated distributions derived in the section on maintenance treatment and recurrence of depression using the mixture model. As reported in the section on sensitivity analysis, we also conduct a wide range of sensitivity analyses concerning the parameters of these distributions. To assign direct costs, IPT-M involves one therapy session each month. Each session lasts 60 min. Therapist time at Western Psychiatric Institute and Clinic, where the sessions are held, is billed at $95 per hour. This is used as a first approximation of the costs of therapists and of the administrative overhead involved in the session. In addition to these costs are the time costs to the patient, which, including transportation, amount to approximately 2 hours per month. Most empirical studies place an hour of transportation time at 60-80% of an average hour of pay, or between $4.95 and $6.60 in 1986 dollars. Including $2 for parking or bus fare, this leads to an opportunity cost of about $14 per visit for the patient, and a total direct cost of approximately $110 per month or $1320 per year for IPT-M. Imipramine drug therapy is estimated to cost approximately $600 per year. This is based on an estimated cost of $0.30 per 50 mg of imipramine, an average daily dose of 200 mg/day, and $lbO/year associated with blood work and physician visits. It is difficult to assign a utility value for 4, the quality of life associated with being depressed. However, several studies in the literature provide some guidance about reasonable values. Sackett and Torrance 1321 presented brief scenarios of various chronic health states to individuals in the general public. They used the “time trade-off” method to calibrate the VNM utility associated with each health state. One of the chronic illnesses considered was a 3-month episode of depression. On a scale where 1 represented full health and 0 represented death, the mean daily utility for depression was 0.44, with a standard error of 0.024. It is interesting to note the severe diminution of quality of life that depression represents in this rating. Compared with other 3-month illnesses, the mean daily utility for depression was lower than that for tuberculosis or dialysis. It also was rated as lower than having a kidney transplant, a mastectomy for injury, or a mastectomy for breast cancer. Torrance et al. 1331 provide a multiattribute scale of the utility of health states, based again on preferences of the general public. Their scale involves the multiplicative interaction of scales for physical functioning, role functioning, and social-emotional functioning. One of the social-emotional functioning categories is: “Being anxious or depressed some or a good bit of the time and having very few friends and little contact with others.” This condition,

Cost Utility Analysis of Maintenance

Treatment for Recurrent Depression

29

by full physical functioning and full functioning in terms of self-care and being able to work, yields a utility value of 0.45, with a standard error of 0.053, where 1 corresponds to a life at full health and 0 to death shortly after birth. Insofar as the emotional functioning of chronically depressed individuals is at least as bad as the category description, alone with less than full physical, self-care, and occupational functioning, this 0.45 rating can be seen as an upper boundary for recurrent depression. Kaplan and Anderson [34] summarize 15 years of research on the development of the Quality of Well-being (QWB) Scale. This scale, based also on preferences of the public toward health states, uses an additive set of subscales for mobility, physical activity, and social activity, weighted by a severity scale based on symptom and problem complexes. It too ranges from 0 to 1. The authors indicate that individuals with depression have a QWB score of either 0.68 or 0.61, depending on whether they are outpatients or inpatients. Wells et al. 1351 report on the functioning and well-being of depressed patients. Depressed individuals were assessed in terms of physical, role, and social functioning as well as perceived overall current health. On scales ranging from 0 to 100, depressed individuals had ratings of 78, 81, and 73 for physical, social, and role functioning, respectively. This was a lower rating on all three scales than for other chronic illnesses such as hypertension, diabetes, arthritis, gastrointestinal problems, and back problems. On the overall health scale, depressed individuals scored 58.7 on a 0 to 100 scale, lower than the above chronic conditions and also lower than advanced coronary artery disease and angina. Taken as a whole, the literature on the quality of life of depressed individuals provides a relatively consistent picture in which depression is judged to be among the most severe chronic illnesses. We choose a utility value of 0.45 as a baseline value for 4, the quality of life associated with being depressed. We also choose values of 0.3 and 0.7 as “reasonably pessimistic” and “optimistic” values. (Here and throughout, “reasonably optimistic” is defined as a parameter value that makes depression less bad-and therefore makes maintenance treatments less cost effective-and conversely for “reasonably pessimistic.“) It is also difficult to assign a value to ~7,the probability per episode that someone with major depression will commit suicide. As indicated above, it has been estimated that some 60% of suicides have clinically significant depression as the primary psychiatric disorder. Stoudemire et al. [2] estimate some 20,000 deaths per year that are secondary to depression. Assuming a 2%, 6-month prevalence rate for depression, this is about a 0.0025 chance of suicide per depressive episode. It is unclear as to whether such a rough calculation is applicable to recurrent depression. Nonetheless, 0.0025 seems a reasonable base case value. We use 0.001 and 0.01 as the optimistic and pessimistic range for this variable. Stoudemire et al. 121 estimate the direct costs of depression to be about $2.1 billion in the aggregate for the United States for 1980. Using a 20/o,6-month prevalence rate, this implies about $250 (in 1980 dollars) per episode in direct costs. If, for example, two thirds of those who are depressed seek some form of treatment within 6 months, this would be about $375 (in 1980 dollars) in direct costs per treated episode, or about $500 per episode in current dollars.

accompanied

30

M.S. Kamlet et al. Such overall averages are, again, somewhat misleading. Different choices of provider lead to different direct costs. The direct costs are probably less in the general mental health care sector than in the specialty mental health care sector. Moreover, interviews with experienced health professionals suggest that this $500 figure is quite low. Nonetheless, to be conservative, we choose $500 per episode as our base case value, using $1500 and $200 per episode as our range. For these cost figures and for others used in the analysis, we assume that real costs remain fixed over time and thus all dollar figures are to be interpreted as real 1991 dollars. Finally, it also very difficult to assign a value to l3, the utility associated with imipramine drug treatment. Of the 230 patients who entered the study during the acute treatment phase, 18 developed side effects intolerable enough to drop out before reaching the continuation phase. Of the 128 patients who entered the maintenance phase, only 4 dropped out due to significant side effects. These results are consistent with the hypothesis that imipramine is well tolerated by the majority of patients but that side effects are pronounced for a small subset of patients. As described above, a subset of patients on maintenance drug therapy experience side effects. In very rare instances these side effects are sufficiently severe as to be a significant deterrent to the patient remaining on the maintenance therapy. The magnitude of these side effects, however, maintains hard to calibrate and varies somewhat across patients. As a result, the analyses of the Drug and the IPT-M & Drug maintenance treatments assign a range of values for /3,with cost effectiveness ratios computed for each of these values. Table 2 summarizes the parameter values used in the empirical analysis. Before preceding with that analysis, it is necessary to note that while costutility analysis has evolved over time toward a well-defined set of methodological procedures for analyzing the efficacy of health interventions from an economic perspective, several controversies remain in the literature 1361. One concerns the use of discount rates. The issue is not only what discount rate to use but also whether health outcomes and costs should be discounted or only the latter. We choose several discount rates-O%, 3%, and 5%-and examine the results obtained when discounting both costs and health outcomes by these rates. Another controversy in the cost-utility analysis literature occurs when, as is the case here, the impact of a health intervention involves more than direct medical costs and more than health narrowly conceived. Here, there are impacts of maintenance treatment on the social functioning of the individual, on his work, and on his amount of leisure. As discussed in Kamlet 1361, these impacts can in principle either be measured in monetary terms (using, for instance, a “willingness-to-pay”basis for determining the monetary magnitudes) and included in the costs of the health intervention (the numerator of the cost-utility analysis cost effectiveness ratio), or measured in terms of their quality-of-life impact and included in the quality-of-life measure (the denominator of the cost effectiveness ratio). We include direct medical costs and the money equivalent of leisure devoted to treatment (measured in terms of the willingness of the individual to pay for the time involved) as costs in the numerator of the cost effectiveness ratio. We measure the remaining impacts, including the so-called indirect costs of the illness on ability to work and

Cost Utility

Analysis

Table

2

of Maintenance

Treatment

for Recurrent

Depression

Parameter Values for Cost-Utility Analysis Treatments for a 40-Year-Old Female Variable

D DRUG D Placebo D IPT-M&DR

C P

Baseline

31

of Maintenance

Value

$1320/year $1320/year $600 /year

W/year $1920/year 1 1 1 Variable values Variable values 0.4 year 0.45 $500/episode 0.0025

(0.7,0.8,0.9,1) (0.7,0.8, 0.9,1)

where D =direct cost per unit time of maintenance treatment while patient is stable X =duration of depressive episode + =quality of life while depressed p =quality of life while stable under maintenance treatment C =direct costs of treating a depressive episode p =probability of suicide per episode IPT-M =interpersonal therapy maintenance treatment DRUG =imipramine drug maintenance treatment IPT-M&DR =imipramine drug and interpersonal therapy maintenance treatment IPT-M&Placebo =interpersonal therapy and placebo drug pill Placebo =placebo drug pill, no interpersonal therapy

productivity at work, in terms of quality of life in the denominator of the cost effectiveness ratio. We do, however, reexamine the results of our analysis from several of the other approaches that might be adopted on this issue.

EMPIRICAL

RESULTS

Base Case Results Table 3a displays the empirical results of the Monte Carlo analysis for a 40-year-old female for the different maintenance treatments using a 0% discount rate and the base case parameter values from Table 2. For each maintenance treatment, the table indicates the expected number of QALYs experienced by the individual after beginning maintenance treatment, the expected total direct cost incurred in treating future depressive episodes, the expected life span of the individual, and the expected number of years of life lost due to suicide. We consider different possible levels of the magnitude of side effects from the drug maintenance treatments (p =0.7, 0.8, 0.9, and 1.0). The only resulting difference in outcome is in the expected number of QALYs. In our analysis, each increase of 0.1 in quality of life on a drug maintenance

M.S. Kamlet et al.

32 Table

3a

Empirical Results for the Maintenance Old Female (and Standard Errors)

Expected Values

for a 40-YearIPT-M

Placebo

IPT-M

& Placebo

9.56 (0.33)

15.18 (0.42)

14.00 (0.401

$21,204 ($493)

$34,316 ($524)

$34,176 ($523)

66.88 (0.41)

66.90 (0.41)

66.88 (0.41)

0.031 (0.021)

0.012 (0.012)

0.028 (0.020)

QALY

Total direct costs Years lived Years lost due to suicide Expected Values

Treatments

Drug p =0.7

Drug p =l.O

Dr&IPT-M p =0.7

Dr&IPT-M p =l.O

14.98 (0.33)

21.39 (0.47)

15.61 (0.30)

22.29 (0.43)

$19,573 ($368)

$19,573 ($368)

$48,390 ($756)

$48,390 ($756)

Years lived

66.91 (0.41)

66.91 (0.41)

66.90 (0.41)

66.90 (0.41)

Years lost due to suicide

0.000 (0.000)

0.000 (0.000)

0.015 (0.015)

0.015 (0.015)

QALY Total direct costs

treatment adds approximately two expected QALYs. Table 3a reports results for the most extreme cases, p=O.7 and b=l.O. As the table indicates, IPT-M and IPT-M & Placebo increase an individual’s expected number of QALYs from 9.56 in the Placebo group to over 14, while Drug and IPT-M & Drug increase QALYs to between 15 and 22 (depending on the magnitude of the side effects of the Drug treatment). Virtually all of this marked impact on QALYs from maintenance treatments is attributable to increases in the quality, not the duration, of life. While the probability of suicide is diminished under the maintenance treatments, the overall cumulative probability of suicide from depression is sufficiently low that there is only a modest reduction in the anticipated length of life under the different treatments. Examining direct costs, IPT-M leads to an increase in expected lifetime direct costs form $21,204 to $34,316. Thus, while IPT-M reduces the number of future depressive episodes and reduces as a result the costs of treatment of depressive episodes, this cost reduction is not sufficient to fully offset the direct cost associated with IPT-M itself. In contrast, the direct costs associated with Drug maintenance treatment are $19,573. This is lower than the Placebo and reflects the fact that drug maintenance is sufficiently effective at reducing the number of future depressive episodes that the cost reduction associated with less need to treat depression more than compensates for the direct cost of the drug treatment itself. Drug & IPT-M leads to a direct cost of $48,390, approximately the sum of the expected direct cost of Drug treatment and IPT-M treatment separately.

Cost Utility Analysis of Maintenance

Table 3b

Treatment for Recurrent Depression

Cost Effectiveness Ratios for the Maintenance 40-Year-Old Female

Comparing

Placebo

with Placebo

IPT-M & Placebo

Drug 0.7

Drug 0.8

Drug 0.9

Drug 1

2,333

2,922

A

A

A

A

119

B

A

A

A

A

A

A

A

IPT-M &Placebo Drug 0.7 &IPT-M

4,494

32,730

8,829

Drug 0.8 &IPT-M

3,287

5,311

3,711

Drug 0.9 &IPT-M

2,589

2,884

2,346

Drug 1 &IPT-M

2,136

1,979

1,715

A = maintenance B = a cost-utility QALYs

for a

IPT-M

IPT-M

where

Treatments

33

45,741 40,587

treatment saves money and improves health ratio of $73,715 resulting from decrease in cost and

36,021 32,019

an associated

decrease

in

Table 3b compares the maintenance treatments by considering their cost effectiveness ratios, expressed in terms of !$/QALY [see Eqs. (6) and (711. In each comparison, the difference in QALY between the two treatments is divided by the associated difference in direct cost. So, for instance, the upper-left entry in the table, 2,333 =(34,316 - 21,204)/(15.18 - 9.561, compares IPT-M vs. placebo maintenance treatment and indicates that (1) II’T-M results in an expected increase in costs compared to the placebo; (2) IPT-M results in an expected increase in QALYs experienced; and (3) the expected cost of IPT-M relative to the placebo is $2333 per additional QALY. Consistent with the conclusions from Table 3a, Table 3b indicates that regardless of the severity of the side effects of imipramine (within the broad range considered), Drug maintenance treatment both improves quality of life as reflected in QALYs and leads to a reduction in direct costs as compared to Placebo. Similarly, in the comparison of Drug with IPT-M (and IPT-M & Placebo), Drug leads to both an improvement in health and a reduction in direct costs in seven of the eight comparisons. In the eighth comparison, in which side effects from Drug treatment are assumed to be at their most severe level, there is a substantial reduction in direct cost ($14,743) but a slight reduction in QALYs (0.2). IPT-M and IPT-M & Placebo do not simultaneously reduce direct cost and improve health. They do, however, lead to a very substantial improvement in health relative to modest increases in direct cost. The cost effectiveness ratio for the three comparisons is in each instance less than $3000. This compares very favorably with other health interventions that are considered to be cost-effective. The typical cutoff for a cost effectiveness ratio being judged as indicating a cost-effective health intervention is in the vicinity of $50,000 per QALY 1371.

MS. Kamlet et al.

34

The combination of IPT-M & Drug similarly leads to cost effectiveness ratios that are very modest when comparison is made to the Placebo treatment. All of these ratios are well under $5000 per QALY (increasing somewhat as the severity of the side effects of the drug treatment increase, as expected). The combination of IPT-M & Drug leads to cost effectiveness ratios that are also below $5000 when comparison is made to IPT-M and to IPT-M & Placebo, with the exception of the case where the side effects of Drug treatment are assumed to be at their most severe. Even here, the ratios are significantly below the $50,00O/QALY threshold mentioned above. Finally, comparison of IPT-M & Drug with Drug alone is more equivocal. The estimated cost effectiveness ratios are between $32,000 and $46,000. While this is still below the $50,00O/QALY threshold and suggests that modest improvements in health from the combination of IPT-M & Drug may still be cost-effective, the estimated cost effectiveness should not be given undue credence. The estimated ratios are based on very modest differences in the clinical efficacy of IPT-M & Drug combined over Drug alone that were found in the section on maintenance treatment and the recurrence of depression not to be statistically different from one another. It should be noted that the numerator in the cost effectiveness ratios in Table 3b include only “costs,” consisting of the medical care services provided as well as some other relatively minor components such as transportation. The conclusions about the cost effectiveness of the various maintenance treatments as they compare to the Placebo, or no treatment, would only be reinforced further if we were to follow the practice of some studies and include “indirect costs”as well in the numerator of the cost-utility ratios. These indirect costs consist of changes in future earnings experienced by patients under maintenance treatment. While the ability of a patient to be engaged in market work may be defined as part of her “social role function” and therefore a component in overall quality of life, some studies, drawing from the so-called “capital” approach to valuing life and health, single these costs out and combine them with direct cost. Stoudemire et al. 121 estimate $1180 in decreased wages per depressive episode in 1980. Updating this to 1990 dollars, this is $1867 in savings for each depressive episode averted. IPT-M and VT-M & Placebo reduce the number of anticipated future episodes by about 13 compared to Placebo. This implies a savings of about $24,275 and means that IPT-M and IPT-M & Placebo do not result in an increase in cost of approximately $13,000 compared to Placebo, as reported in Table 3a, but rather a decrease in cost of approximately $11,000. This, in turn, would mean that in Table 3b the symbol “A” would replace the $/QALY figures provided since now IPT-M and IPT-M & Placebo represent both an increase in health and a decrease in overall cost. The effect of incorporating indirect costs into the numerator for the analysis of Drug and Drug & Placebo would be even more dramatic, adding a savings of about $54,000 for these maintenance treatments. Sensitivity

Analysis

The results presented above correspond to the base case parameter values and reflect a 0% discount rate. In this section, we will discuss a range of

Cost Utility Analysis of Maintenance

Table 4

Cost Effectiveness Ratios for the Maintenance Female Using “Optimistic” Parameter Values

Comparing

Placebo

with Placebo

Treatments

for a 40-Year-Old

IPT-M

IPT-M & Placebo

3,913

4,379

1,646

398

6,807

A

A

A

9,134

A

A

A

IPT-M IPT-M &Placebo Drug 0.7 &IPT-M

9,817

A

A

Drug 0.8 &IPT-M

5,915

14,283

9,701

Drug 0.9 &IPT-M

4,226

4,656

4,067

Drug 1 &IPT-M

3,294

2,783

2,574

where A = maintenance

35

Treatment for Recurrent Depression

Drug 0.7

Drug 0.8 898

Drug 0.9 617

Drug 1 471

34,333 30,045 26,570 23,965

treatment saves money and improves health

sensitivity analyses that were conducted. We describe these results only briefly here because the main conclusions of the base case are robust to the different sensitivity analyses examined. Table 4 compares the maintenance treatments in terms of $/QALY, using “optimistic” parameter values. That is, the length of a depressive episode (X) is assumed to be 0.2 year, the quality of life while depressed (+) 0.7, the cost of a depressive episode (T) $200/episode, and the probability that a person commits suicide while depressed (p) is taken to be 0.001. The only substantial change is that Drug maintenance treatment does not both improve health and decrease direct costs compared to Placebo. Now there is a very small increase in direct costs comparing Drug maintenance and Placebo. Still, the implied cost-utility ratios are small, indicating that Drug maintenance treatment remains very cost-effective. Table 5 compares the maintenance treatments in terms of $/QALY, using “pessimistic” parameter values. That is, the length of a depressive episode (X) is assumed to be 0.6 year, the quality of life while depressed (+) 0.3, the cost of a depressive episode (ZJ $1500/episode, and the probability that a person commits suicide while depressed (p) is taken to be 0.01. None of the above conclusions is substantially altered. The maintenance treatments simply become increasingly cost-effective. Recall the mixture model used to estimate the time until recurrence for a patient on each of the given maintenance treatments, where a proportion ITSof the population is presumed to have essentially a zero probability of recurrence. This parameter is estimated from the data and therefore subject to uncertainty. However, running the Monte Carlo analysis using the 95% confidence level upper and lower bounds for each 7~~produces no substantial change in the conclusions. Along these same lines, in using the survival, or “time-until-recurrence,” analysis to extrapolate beyond the 3 years of the study, it was assumed that if

M.S. Kamlet et al.

36 Table 5

Cost Effectiveness Ratios for the Maintenance Treatments Female Using “Pessimistic” Parameter Values

Comparing

Placebo

with Placebo

IPT-M 749

IPT-M

IPT-M & Placebo

Drug 0.7

Drug 0.8

Drug 0.9

Drug 1

1,269

A

A

A

A

A

A

A

A

A

A

A

A

A

IPT-M &Placebo Drug 0.7 &IPT-M

for a 40-Year-Old

1,209

3,083

1,106

Drug 0.8 &IPT-M

924

1,204

606

Drug 0.9 &IPT-M

749

749

417

Drug 1 &IPT-M

629

543

318

103,033 90,565

81,821 72,295

where A = maintenance treatment saves money and improves health

in drawing a value for Y an individual happened to be assigned to the surviving fraction for a maintenance treatment (i.e., the subgroup with a zero probability of recurrence), they were to remain “immune” to recurrence throughout their life. Redrawing every 3 years for each patient to determine whether she is part of the surviving fraction for the next 3 years produces no substantial changes in the conclusions. Finally, the analysis presented in the previous section assumes a discount rate of 0% for both costs and health outcomes. Table 6 presents the results of

Table 6

Cost Effectiveness Ratios for the Maintenance Female Using a 5% Discount Rate

Comparing

Placebo

with Placebo

IPT-M

IPT-M & Placebo

2,415

2,975

II’T-M

123

IPT-M &Placebo Drug 0.7 &IPT-M

4,753

65,145

10,829

Drug 0.8 &II’T-M

3,433

5,755

4,012

Drug 0.9 &IPT-M

2,686

3,010

2,462

Drug 1 &IPT-M

2,207

2,038

1,776

Treatments

for a 40-Year-Old

Drug 0.7

Drug 0.8

Drug 0.9

Drug 1

A

A

A

A

22,654

A

A

A

A

A

A

A

33,339

where A = maintenance treatment saves money and improves health

29,148 25,893 23,292

Cost Utility Analysis of Maintenance

Treatment for Recurrent Depression

37

analysis using baseline parameters and applying a 5% discount rate to both cost of treatment and health outcomes. Overall conclusions are not substantively affected.

SUMMARY

AND CONCLUSIONS

Cost-utility analysis is increasingly used as a method for comparing the societal health benefits and costs associated with given health interventions. It offers the potential to compare the cost effectiveness of diverse health interventions using a common metric. However, it has been very difficult to apply cost-utility analysis in mental health care. This is because the etiology of mental illnesses and the clinical efficacy of treatment are often subject to substantial uncertainty. It is also because mental illnesses are often chronic and require dynamic consideration of the impact of a health intervention across the life span of an individual. This paper provides a framework for taking available clinical, economic, and quality-of-life data to examine the cost utility of several maintenance treatments for recurrent depression. As reported in the previous section, we find that such maintenance treatments have their major impact on increases in the quality not the duration of life and it is precisely in these types of situations, in which a health intervention has a substantial impact on morbidity but only a modest impact on mortality, that cost-utility analysis is particularly insightful as an evaluation tool. The most important conclusion of this analysis is that the examined maintenance treatments for recurrent depression are very cost-effective. Imipramine Drug maintenance treatment is cost-effective when compared to Placebo in the strongest sense of the term. It both improves expected lifetime health (taking account of mortality and morbidity effects over the course of a patient’s lifetime) and reduces direct medical costs. This is true even when relatively severe levels of side effects of the Drug maintenance treatment are considered. IPT-M and the combination of VT-M & Drug each improve expected lifetime health when compared to the placebo, although in neither case are expected direct medical costs reduced. Still, the cost of the resulting health improvements, under $3000 per quality-adjusted life year for VT-M and under $5000 per quality-adjusted life year for IPT-M & Drug, are very modest. They are well below the $50,000 per quality-adjusted life year threshold typically invoked in the literature [371 to determine whether a health care intervention is an appropriate use of health care resources. These conclusions regarding the cost effectiveness of IPT-M and IPT-M & Drug maintenance treatment ignore “indirect cost” savings that result from changes in the earnings of individuals experiencing maintenance treatment. If these are taken into account, then IPT-M and IPT-M & Drug, like Drug maintenance treatment, each improves expected lifetime health and reduces costs. These results are quite robust to sensitivity analysis, including incorporating more conservative parameter assumptions and differences in discount rates. In the analysis of the choice between maintenance treatments, the answer is obvious in two cases, whereas in one case the data do not allow for a definitive conclusion. Drug maintenance treatment is clearly cost-effective

38

M.S. Kamlet et al. compared to VT-M. Drug treatment improves expected lifetime quality of life relative to IPT-M and leads to a reduction in direct costs. Similarly, IPT-M & Drug is cost-effective relative to IPT-M alone (except possibly for the most severe of the considered side effects), with, however, a modest increase in direct costs associated with the resulting health improvements. Only in the comparison of IPT-M & Drug to Drug alone do the data now allow for a definitive conclusion, owing to the inability of the clinical data to substantiate whether IPT-M & Drug is more effective in delaying recurrences than Drug maintenance treatment. Before concluding it is important to highlight several caveats of relevance to the analysis. First, the state transition model is clearly stylized and simplified. As discussed in the section on a state-transition model for analyzing recurrent depression, which presents the model, the assumptions made are conservative. If biased at all, we have tried to ensure that they are biased against the benefits of maintenance treatment. Second, another difficulty is in generalizing from a clinical experiment, in which even patients in the placebo group receive unusual attention (and in which patients in other groups are monitored for compliance with their treatment more than may otherwise be the case), to usual care. Thus, our results are only generalizable to patients who are assigned to one of the maintenance treatment schedules described here and who actually adhere to this schedule. Third, the individuals entering the maintenance treatment have already displayed a responsiveness to imipramine treatment in the preliminary stabilization phase. Therefore the costutility analysis is only relevant to this class of patients. Fourth, individuals who dropped out of the study were treated as “right-censored,” in the main analysis. Completely dropping these subjects from the analysis in the section on maintenance treatment and the recurrence of depression does not, however, substantially affect the results. Finally, since rates of relapse tend to be high once depression becomes recurrent, one might argue against the appropriateness of the Markovian transition structure because it seems possible that the number of prior episodes likely influences the time to relapse and transition probabilities. However, Frank et al. 1211 found that the number of previous episodes had no effect on outcome “as defined by time to recurrence.” Since the number of previous episodes varied greatly among the patients (an average of 6.8 with a rather large standard deviation of 7.3), this suggests that the use of the Markovian transition structure is justified. This paper represents a contribution not only in terms of the above substantive conclusions but by developing and estimating a state-transition model that uses Monte Carlo analysis to simulate the path of recurrent depression over a patient’s life in a fashion amenable to cost-utility analysis. This type of analysis holds much promise for serving as the basis for analyzing other types of recurrent illnesses.

From the Mental Health Clinical Research Center, Western Psychiatric Institute and Clinic, University of Pittsburgh, This research was supported in part by National Institute of Mental Health grants MH29618-10 and MH-30915-14. Nancy Paul is a Howard Hughes Medical Institute Predoctoral Fellow.

Cost Utility Analysis of Maintenance

Treatment for Recurrent Depression

39

REFERENCES 1. Myers JK, Weissman

WM, Trischler GL: Six-month prevalence of psychiatric disorders in three communities: 1980-1982. Arch Gen Psychiatry 41:959-67, 1984

2. Stoudemire A, Frank R, Kamlet M, Hedemark N: Depression. In Amler RW, Dull HB eds. Closing the Gap: The Burden of Unnecessary Illness. New York: Oxford University Press, New York, 1987, pp. 65-71 3. Monkoff K, Bergman E, Beck A: Hopelessness, Am J Psychiatry 139L455459,1973

depression, and attempted suicide.

4. Keller MB, Shapiro RW, Lavori PW, Wolfe N: Recovery in major depressive disorder: analysis with life tables and regression models. Arch Gen Psychiatry 39:905-910, 1982 5. Keller MB, Shapiro RW, Lavori PW, Wolfe N: Relapse in major depressive order: analysis with life table. Arch Gen Psychiatry 39:911-915, 1982

dis-

6. Klerman CL: Long-term treatment of affective disorders. In Lipton MA, DiMascio A, Killam KF, eds. Psychopharmacology: A Generation of Progress. Raven Press, New York, 1978, pp. 1303-1311 7. Coppen A, Noguera R, Bailey J, Burns BH, Swami MS, Hare EH, Gardner Maggs R: Prophylactic lithium in affective disorders. Lancet 2:275-279, 1971

R,

8. Glen AIM, Johnson AL, Shepherd M: Continuation therapy with lithium and amitriptyline in unipolar depressive illness: a randomized double-blind, controlled trial. Psychiatr Med 14:37-50, 1984 9. Prien RF, Kupfer DJ, Mansky PA, Small JG, Tuason VB, Voss CB, Johnson WE: Drug therapy in the prevention of recurrences in unipolar and bipolar affective disorders-a comparison of lithium, imipramine, and a lithium combination. A report of the NIMH Collaborative Study Group. Arch Gen Psychiatry 41:10961104, 1984 10. Klerman GL, Weissman MM, Roundsaville BJ, Chevron ES teds.): Interpersonal Psychotherapy of Depression. Basic Books, New York, 1984 11. DiMascio A, Weissman MM, Prusoff BA, Neu C, Zwilling M, Klerman GL: Differential symptom reduction by drugs and psychotherapy in acute depression. Arch Gen Psychiatry 36:1450-1456,1979 12. Weissman MM: The psychological treatment of depression: evidence for the efficacy of psychotherapy alone, in comparison with, and in combination with pharmacotherapy. Arch Gen Psychiatry 36:1216-1269, 1979 13. Weissman MM: The psychological treatment of depression: update of clinical trials. In Williams JBW, Spitzer RL, eds. Psychotherapy Research: Where Are We and Where Should We Go? Guilford Press, New York, 1984, pp. 89-105. 14. Prusoff BA, Weissman MM, Klerman GL, Roundsaville BJ: Research diagnostic criteria subtypes of depression: their role as predictors of different response to psychotherapy and drug treatment. Arch Gen Psychiatry 37:796-801,198O 15. Klerman GL, DiMascio A, Weissman M, Prusoff B, Paykel ES: Treatment depression by drugs and psychotherapy. Am J Psychiatry 131:186-191, 1974 16. Angst J: Prospective study on the course of affective disorders. Consensus Development Conference Program, 1984, pp. 23-26. 17. Fernstrom MH, Krowinski RL, Kupfer DJ: Chronic weight gain. Psychiatr Res 17:269-273, 1986

imipramine

18. Weinstein MC, Stason WB: Foundations of cost-effectiveness and medical practices. N Engl J Med 296(13):716-721, 1977

of

Presented at the treatment

and

analysis for health

19. Weinstein MC, Fineberg HV: Clinical Decision Analysis. WB Saunders, phia, 1980

Philadel-

20. Drummond MF, Stoddart GL, Torrance GW: Methods for the Economic Evaluation of Health Care Programmes. Oxford University Press, New York, 1987

M.S. Kamlet et al.

40

21. Frank E, Kupfer DJ, Perel JM, Cornes C, Jarrett DB, Mallinger AG, Thase ME, McEachran AB, Grochocinski VJ: Three-year outcomes for maintenance therapies in recurrent depression. Arch Gen Psychiatry 47~1093-1099, 1990 22. Spitzer RL, Endicott J, Robins E: Research diagnostic criteria: rationale and reliability. Arch Gen Psychiatry 35773-782, 1978 23. Hamilton M: A rating scale for depression. 62,196O

J Nemo1 Neurosurg

Psychiatry 23:56-

24. Raskin A, Schulterbrandt J, Reatig N, McKeon JJ: Replication of factors of psychopathology in interview, word behavior and self-report ratings of hospitalized depression. J Nerv Ment Dis 148:87-98, 1969 25. Greenhouse JB, Stangle D, Kupfer DJ, Prien RF: Methodological issues in maintenance therapy clinical trials. Arch Gen Psychiatry 48:313-318, 1991 26. Farewell VT: The use of mixture models for the analysis of survival data with long-term survivors. Biometrics 38:1041-1046, 1982 27. Greenhouse JB, Wolfe RA: A competing risks derivation of a mixture model for the analysis of survival data. Corn Stat Theory Meth 13(25):3133-3154, 1984 28. Bothwell S, Weissman MM: Social impairments four years after an acute depressive episode. Am J Orthopsychiatry 47:231-237, 1977 29. Pliskin JS, Shepard DS, Weinstein status. Oper Res 28:206-224, 1980

MC: Utility functions for life years and health

30. Weissman MM, Myers JK: Rates and risks of depressive symptoms States urban community. Acta Psychiatr Stand 57:219-231,1978

in a United

31. Elandt-Johnson RC, Johnson NL: Survival Models and Data Analysis. John Wiley and Sons, New York, 1980 32. Sackett DL, Torrance GW: The utility of different health states as perceived by the general public. J Chron Dis 31:697-704, 1978 33. Torrance GW, Boyle MH, Horwood SP: Application of multiattribute utility theory to measure social preferences for health states. Oper Res 30(6):1043-1067, 1982 34. Kaplan RM, Anderson JP: A general health policy model: update and applications. J Health Serv Res 23(2):203-235, 1988 35. Wells KB, Stewart A, Hayes RD, Baumam WR, Daniels M, Berry S, Greenfield S, Ware J: The functioning and well-being of depressed patients: results from the medical outcomes study. JAMA 262(7): 914-919,1989 36. Kamlet MS: A Framework for Cost-Utility Analysis of Government Health Care Programs. U.S. Department of Health and Human Services, Public Health Service, 1992 37. Fryback DG, Thornbury 11:88-94, 1991

JR: The efficacy of diagnostic

imaging. Med Decis Mak