Cost-effectiveness of Clozapine in Patients With High and Low Levels of Hospital Use

Background  This study examined the relationship between pretreatment hospital use and the cost-effectiveness of clozapine in the treatment of refractory schizophrenia.

Methods  Data from a 15-site randomized clinical trial were used to compare clozapine with haloperidol in hospitalized Veterans Affairs patients with refractory schizophrenia (n=423). Outcomes were compared among those with many days in the hospital use (hereafter, high hospital users) (n=141; mean=215 psychiatric hospital days in the year prior to study entry) and those with few days in the hospital use (hereafter, low hospital users) (n=282; mean=58 hospital days). Analyses were conducted with the full intention-to-treat sample (n=423) and with crossovers excluded (n=291).

Results  Clozapine treatment resulted in greater reduction in hospital use among high hospital users (35 days less than controls, P=.02) than among low users (21 days less than controls, P=.05). Patients taking clozapine also had lower health care costs; after including the costs of both medications and other health services, costs were $7134 less than for controls among high hospital users (P=.14) but only $759 less than for controls among low hospital users (P=.82). Clinical improvement in the domains of symptoms, quality of life, extrapyramidal symptoms, and a synthetic measure of multiple outcomes favored clozapine in both high and low hospital user groups.

Conclusions  Substantial 1-year cost savings with clozapine are observed only among patients with very high hospital use prior to initiation of treatment while clinical benefits are more similar across groups. Cost-effectiveness evaluations, and particularly studies of expensive treatments, cannot be generalized across type of use groups.

COST-EFFECTIVENESS analysis has become an increasingly important part of the evaluation of new medical interventions.^1,2 Experimental studies have demonstrated that costly mental health interventions such as intensive case management^3,4 and clozapine pharmacotherapy⁵ can generate savings sufficient to offset the cost of the treatment itself (primarily by reducing hospital use), or that the value of clinical improvement equals the additional cost.⁶

A criterion that has often been used to select patients for studies of costly services is a recent history of high service use because such patients naturally offer the greatest opportunity for savings.^4,5,7-10 Although these studies explicitly target high-cost patients, they are often cited as evidence of the general cost-effectiveness of the evaluated treatments.⁸ Several studies suggest that findings from studies of patients with many days of hospital service use (hereafter, high hospital users) may not be generalizable to patients with fewer days of hospital service use (hereafter, low hospital users).^11-13

Controlled trials have shown that clozapine is more efficacious than conventional pharmacotherapies in the treatment of refractory schizophrenia.^5,14-17 However, clozapine is also considerably more expensive than conventional medications⁸ in part because it requires weekly blood monitoring. Although 2 experimental studies have indicated that these higher costs may be offset by reduced hospital use, both of those studies focused on high hospital users.^5,7 Nonexperimental studies of the cost of clozapine treatment have also focused on high hospital users.^8-10 Since most patients with schizophrenia are not hospitalized at all during a given year and hospital stays continue to decline, the generalizability of studies of high hospital users to more typical patients with refractory schizophrenia needs evaluation.

In this study, we compare patterns of service use, cost, and outcome among subgroups of inpatients defined by their prestudy levels of hospital uses.

This study was a 3-year prospective, double-blind experiment in which patients at 15 Veterans Affairs (VA) medical centers were randomly assigned to clozapine treatment or haloperidol and treated for 12 months.

The study was targeted at patients with refractory schizophrenia with at least 30 days' hospitalization for schizophrenia during the previous year and no more than 364 days.⁵ Clinical eligibility criteria included (1) a DSM-III-R¹⁸diagnosis of schizophrenia on the Semi-structured Clinical Interview for Diagnosis¹⁹; (2) refractoriness defined as persisting psychotic symptoms despite 2 adequate treatment trials; (3) severe symptoms indicated by scores on standard scales of symptom severity and functioning^20,21; and (4) clinical determination of serious social dysfunction for the past 2 years.

After completing baseline assessments, patients were randomly assigned, within centers, to treatment with clozapine (100-900 mg/d) or haloperidol (5-30 mg/d). Dose adjustments were made as clinically indicated, using 12 fixed-dosage levels. Patients in the haloperidol group also received benztropine mesylate (2-10 mg/d) for extrapyramidal symptoms (EPS) while patients receiving clozapine were given a matching benztropine placebo. To maintain the blind, patients in the haloperidol group also had weekly blood counts as required for clozapine treatment. A structured treatment planning procedure was used to provide appropriate psychosocial treatments to all study subjects.²²

Health care costs were estimated by multiplying the number of units of service by estimated unit costs for each patient. Unit costs for VA general psychiatry and substance abuse inpatient and outpatient care, including group treatment and day hospital care (ie, those who stay all day) were estimated for each medical center, using cost data from the local Cost Distribution Report (CDR) and computerized workload data.²³ Costs of VA medical, surgical, domiciliary, and nursing home care and for various types of nonmental health outpatient treatment were estimated using the national average Cost Distribution Report costs.

The Cost Distribution Report is a facility-by-facility accounting record that identifies total expenditures and unit costs of all VA health care services. It uses systematic accounting procedures to distribute both the immediate costs of providing services (personnel services and supplies) and other facility costs (administration, building maintenance, engineering service, equipment depreciation, etc) to each major type of service. Since this was a national multisite study, costs were adjusted for differences in VA locality pay. The opportunity cost of capital was estimated using methods presented elsewhere.²⁴ An additional adjustment for variations in the intensity of inpatient treatment was made by costing the first 3 days at 1.3 times average costs, and the days following the 10th day at 0.97 of the average cost. Days in restraints or seclusion were estimated at an additional $300 per day.

Non-VA health care costs were minimal (<2% of all costs) and were estimated on the basis of a recent study that compared VA with non-VA costs in various communities.²⁵

The VA health service utilization data were derived from the VA's national computerized workload systems.

Use of non-VA services was evaluated through monthly patient interviews and validated using treatment records from non-VA providers.

The costs of study medications were estimated at VA pharmacy cost levels for clozapine therapy of $2.13/100 mg and for haloperidol of $0.02/5 mg. Costs of blood draws, white blood cell counts, and additional pharmacy processing were added to the weekly cost of clozapine treatment ($14 per week).

Cost data were summarized at the individual patient level and analyzed from the perspective of the overall health care system (ie, including both VA and non-VA health care). We also present data on psychiatric inpatient days and psychiatric inpatient costs. Previous reports showed that nonhealth care costs (ie, the costs of disability payments, criminal justice system costs, family burden, and productivity) were equivalent to only 1% to 2% of health care costs,⁵ and since they were not significantly related to hospital days, they were excluded from these analyses.

Symptom outcomes were assessed with the Structured Clinical Interview for Positive and Negative Syndrome Scale for Schizophrenia.²⁶ Social functioning and quality of life were evaluated with the Heinrichs-Carpenter quality of life scale, a clinician-rated scale of social functioning and severe behavioral deficits.²⁷ Medication side effects were assessed with the Simpson-Angus scale for EPS and several other measures not reported here.²⁸ Change in these measures was scored as the difference between the final 12-month interview and the baseline scores.

The primary clinical outcome measure is an integrated measure of improvement based on multiple measures of symptoms, social relationships, community functioning, and side effects, which represents improvement in quality-adjusted life-years (QALYs), a scale that ranges from 0 (worst possible quality of life, approximating death) to 1 (perfect health). On this scale, explained at length elsewhere,²⁹ an improvement of 100% represents the equivalent of adding 1 year of good-quality life.

To compare groups that represent clinically meaningful difference patterns of inpatient hospital use during the year prior to study entry, we compared the impact of clozapine in the highest tertile of hospital users (n=141; mean=215 days; SD=77; median, 196; range, 116-360 days) with those in the 2 lower tertiles (n=282; mean, 58 days; SD, 27 days; median, 52; range, 8-116 days). This division was selected in preference to a median split because the baseline utilization data were modestly skewed and the one-third or two-third stratification increased the difference between the groups with limited loss of statistical power.

First we compared low hospital users with high hospital users on a wide range of baseline measures using χ² tests for categorical measures and t tests for continuous measures. Although randomization was not conducted within these strata, we then compared patients assigned to the clozapine regimen with those assigned to haloperidol on the same measures, within each group.

Comparison of treatment groups and treatment effects between high and low hospital users was conducted using hierarchical linear modeling.³⁰ The hierarchical linear modeling technique may be applied to data in which individual measurements cannot be considered independently because they are clustered within larger groups (eg, patients clustered within distinct VA hospitals). One must use this type of modeling rather than standard linear regression because the correlation among individual patient observations will bias variance estimates and augment type I error. The PROC MIXED procedure of the SAS (R) software system (SAS Institute, Cary, NC) was used for these analyses.

In the principal analyses, hierarchical linear modeling was used to examine the interaction of treatment group (clozapine vs haloperidol) and preentry use group (low hospital users vs high hospital users) in predicting 12-month use and cost totals, controlling for baseline values of the dependent measures and modeling hospital as a random effect. t Tests were used to evaluate treatment group differences within use strata and to compare effects across treatment strata. Comparable analyses were conducted on clinical improvement (ie, change). In addition, linear regression was used to examine the effect of the interaction of preentry hospital days and treatment group on costs.

Least square means adjusted for site differences and baseline values are presented in Table 1, Table 2, and Table 3. An α level of P=.05 was used as the criterion of significance.

During the course of the 12-month follow-up, some patients stopped taking study medication because of lack of efficacy or adverse effects and switched to other treatments.⁶ During the entire 12-month trial, 83 (40%) of 205 patients assigned to the clozapine regimen were unblinded and discontinued use of the study medication and crossed to a standard antipsychotic medication (including haloperidol). In addition, 157 (72%) of the 218 patients receiving haloperidol therapy were unblinded, and 49 (22%) received clozapine treatment for 4 or more weeks during the trial and were considered "crossovers."

Since the inclusion of crossover patients in the analysis of long-term outcomes dilutes our evaluation of the effectiveness of clozapine, 2 sets of analyses are presented: first on the entire sample of patients (the intention-to-treat sample) (N=423) and then on a subgroup with crossovers excluded (n=291). In the later analyses, we retain patients receiving haloperidol therapy who switched from blind treatment to other conventional antipsychotic medications, but not those who switched to clozapine treatment because it is well demonstrated that other conventional medications have equivalent efficacy to haloperidol while clozapine has been shown in many studies to have superior efficacy to haloperidol. Excluding all patients who discontinued blind treatment generates similar results, but the sample that remains is small and unbalanced and we therefore have not presented these data. We thus present a sensitivity analysis in which the intention-to-treat sample is a conservative, lower-bound evaluation of clozapine's effectiveness, while the crossovers-excluded analysis represents a more liberal upper-bound evaluation.

While high hospital users were older and less likely to be Hispanic than low hospital users, there were no differences with other sociodemographic or baseline clinical measures (Table 1). As intended, high hospital users had almost 4 times as many hospital days in the year before study entry as low hospital users, and their costs during the 6 months prior to study entry were over twice as great. There were no differences between these groups in clinical change at 1 year in symptoms or quality of life, but high hospital users showed greater improvement in EPS.

Comparison of patients assigned to the clozapine regimen with those assigned to the haloperidol one within each use group revealed no significant differences.

In the intention-to-treat sample, in the low hospital user group, during the year after randomization, patients receiving clozapine experienced 121 days of psychiatric hospitalization compared with 142 days for the controls, for a clozapine-related reduction of 21 days (Table 2). Among high hospital users, in contrast, clozapine-treated patients experienced 186 days of psychiatric hospitalization while the controls had 221 days, a difference of 35 days. The figure to the right (fourth column in Table 2)(−14) shows the difference in clozapine effect between high and low hospital users (a difference of 14 days). Although this differences is substantial, the interaction term is not significant.

The second section in Table 2 shows that treatment with clozapine resulted in a significant inpatient cost saving of $6865 (−15%) among low hospital users, and a significant saving of $13,235 (−18%) among high hospital users, a net difference of $6370 favoring the use of clozapine in high hospital users.

Data on all health care costs, presented in the lowest section in Table 2, show that when nonpsychiatric inpatient costs, outpatient professional costs, and medication costs are added to the inpatient psychiatric costs, treatment with clozapine saved only $759 (1.4%) among low hospital users, but $7134 (9.9%%) among high hospital users. Clozapine savings among high hospital users were almost 10 times greater than those observed among low hospital users.

Analysis of the effect on total health care costs of the interaction of preentry hospital days and treatment group assignment showed a nonsignificant interaction (regression coefficient for the interaction_3,419, −20.22, P=.53), suggesting that for every additional preentry hospital day, clozapine saves an additional $20.²³

With crossovers excluded, clozapine reductions of inpatient use and cost are similar in trend but, unexpectedly, smaller in magnitude than in the intention-to-treat sample (right-hand section of Table 2). More importantly, when we examine total health care costs, clozapine treatment is found to be significantly more expensive, by $4140 (8.0%), among low hospital users, but it still saves $4806 (−7.2%) among high hospital users. This result is explained, in part, by the weaker clozapine effect on hospital use when crossovers are excluded, but more importantly, by the greater difference in medication and outpatient clinical costs in this subsample, since these patients took clozapine for a full year.

In the intention-to-treat sample, after 1 year of treatment the low hospital users showed a significant reduction in symptoms of (−4.1 points on the Positive and Negative Symptom Scale for Schizophrenia), while high hospital users showed a much smaller improvement (−1.1 points) for a difference of 3.0 points in clozapine's effectiveness across groups. Symptom effects at 1 year were markedly attenuated by the crossovers, as suggested previously. However, data from the 3-month interviews, before crossovers were a substantial factor, show the expected significant reduction in symptoms in both groups: by −4.4 points in the low hospital user group (t₃₅₇=2.3; P=.02) and by −5.6 points in the high hospital user group (t₃₅₇=2.1; P=.04).

Similarly, no significant effects were observed in the intention-to-treat sample on the Quality of Life Scale at either 3 months or 1 year, although high hospital users showed 2.44 points less improvement (the sign of the differences on the Quality of Life Scale are opposite to those on the Positive and Negative Symptoms Scale for Schizophrenia). On the measure of EPS, in spite of crossovers, clozapine yields significant improvement, of similar magnitude, in both groups. The measure of QALYs showed significant improvement for low and high hospital users, but greater improvement among high users.

As expected, clinical benefits of clozapine treatment are far more substantial and significant with crossovers excluded (right-hand section of Table 3). In these analyses, therapy with clozapine is associated with significant benefits on virtually all measures with effects of greater magnitude among high hospital users.

As hypothesized, we found a substantial reduction in psychiatric hospital use and costs attributable to clozapine treatment in both high– and low hospital user groups—with consistently greater savings in the high hospital user group than in the low hospital user group. After including the costs of medication and other health care services, reductions in total health care costs attributable to clozapine were over $7134 in the high hospital user group but only $759 in the low hospital user group. These cost findings are consistent with an experimental study of very-long-stay state hospital patients that showed robust reductions of inpatient use among patients treated with clozapine, even in the absence of any measurable effect on symptoms.⁷ Estimates of cost savings among high hospital users in the current study are lower than in some others because the comparison groups in those studies were patients who had discontinued treatment with clozapine^8,9 or other nonexperimental control groups.¹⁰

Because a substantial number of patients crossed over to the other treatment in the course of the trial, we further examined subsample results with crossovers excluded. This was an important comparison because while patients in this sample had full exposure to the benefits of clozapine treatment, they also incurred the greater costs of a full year of expensive treatment. With crossovers excluded, clozapine was associated with savings of $4806 in the high hospital user group; however, in the low hospital user group, costs were $4140 higher for clozapine-treated patients than for controls. We conclude that clozapine is a robust cost saver in high hospital users, but may be either a weak or more expensive cost saver among low hospital users.

This study also confirms clozapine's effectiveness in reducing symptoms and improving quality of life. Even though the cost impact of clozapine treatment is variable, its clinical benefit appears to be more consistent.

In the language of formal cost-effectiveness analysis, clozapine treatment for high hospital users is what is called a "dominant choice" over clozapine treatment for low–hospital users because it results in both lower costs and higher benefits.^1,31,32 Cost-effectiveness ratios cannot be meaningfully calculated in the case of a dominant treatment,^1,31,32 but these results imply that the return on societal health care resource use is maximized by treating high hospital users with clozapine before low hospital users.

This study suggests that there is a pretreatment inpatient cost threshold that patients must exceed to achieve sufficient savings to offset the cost of clozpaine therapy. Our data suggest that in both high and low hospital users, and in both the intention-to-treat and crossovers-excluded samples, treatment with clozapine results in a 15% to 20% savings on inpatient psychiatric care and a $5000 to $9000 increase in the cost of medications plus outpatient mental health care. Combining these figures in a sensitivity analysis we estimate that, at the low end, clozapine treatment would not be cost neutral among patients with inpatient costs of less than $25,000 per year (assuming 20% savings with only $5000 increased costs). At the high end, clozapine therapy would not be cost neutral among patients with inpatient costs of less than $60,000 per year (15% savings with $9000 increased costs). Thus, for patients with less than $25,000 to $60,000 in inpatient costs per year, the increased cost of the clozapine regimen is not likely to be offset by inpatient savings. These estimates are consistent with data from a sample of outpatients with hospital use of only 27 days in the previous year, in which clozapine treatment increased costs by $2300 per patient per year.³³

Although therapy with clozapine is thus not likely to save money in the care of refractory patients who make little or no use of hospital care (a growing proportion of patients with schizophrenia³⁴), it is still likely to be more effective than conventional treatments for such patients. The pharmacoeconomic decision whether clozapine should be used in such patients must be based on a determination of whether the clinical benefits are worth the additional cost.^1,31,32 Data from our measure of QALYs suggest that clozapine treatment gives patients in the low hospital user group 2% to 3% more QALYs, and by implication 2% to 3% more years of quality-adjusted life. Estimates of the monetary value of a QALY (ie, how much society is willing to pay for a year of healthy life) vary considerably, ranging from $20,000 to $100,000 according to one review,³¹(pp94-95) and from $150,000 to $350,000 according to another.³²(p68) Thus, in dollar terms, a 2% to 3% improvement in QALYs could be valued at from $400 (2% of $20,000) to $10,500 (3% of $350,000). By the low estimate, the improvement attributable to clozapine therapy is not equal to its cost, but by the high estimate the improvement is greater than the cost. Methods for estimating the monetary value of health care gains, although relatively clear in theory, are quite uncertain when applied in practice.^1,31,32 We cannot, therefore, confidently determine whether use of clozapine is a rational, cost-effective treatment choice in patients where limited inpatient savings are to be had. We have, however, illustrated a method for addressing this question and we have presented data suggesting that, given a high valuation of QALY, benefits are likely to equal costs.

The primary methodological limitation of this study is that, as is to be expected in a 12-month trial with severely mentally ill patients, substantial numbers of patients in each condition crossed over to the other treatment. The fact that we obtained consistent trends between the intention-to-treat analysis and the analysis with crossovers excluded adds credibility to our conclusions.

Second, because this study was conducted exclusively with VA patients, who are older and more likely to be males than patients treated elsewhere, we cannot be sure of its generalizability to other health care systems. In addition, cost estimates for VA care are likely to be different from those of other systems.³⁵

It should also be noted that the Food and Drug Administration has recently relaxed the required schedule of blood monitoring for clozapine to once every 2 weeks after the first 6 months of treatment. While this change will reduce the costs of clozapine treatment somewhat, it does not alter our basic findings or interpretations.

A third limitation is that our analysis involved multiple comparisons involving 2 primary outcome measures (societal cost and QALYs) with 2 subgroups (high and low hospital users) and 2 samples (intention-to-treat and crossovers excluded). Using a conservative Bonferroni correction, our α would be reduced to P<.006 and the only statistically significant results are the greater improvement in QALYs among high users. We feel, however, that the pattern of results are more broadly informative and that the differences in cost savings for high as compared with low hospital users are an important addition to our understanding of the relationship of baseline service use and cost-effectiveness.

Several new medications (eg, risperidone, olanzapine, quetiapine) have been introduced that are also considered to be atypical antipsychotic medications, but do not require blood monitoring. These medications, however, are also far more costly than conventional medications, but experimental studies of their total cost impact have not been completed. The general lines of analysis presented here will apply equally well to the pharamcoeconomic evaluation of these agents. If they reduce inpatient service use as effectively as clozapine, substantial savings will be realized among a small number of high-cost patients, but savings will be much harder to achieve in more typical inpatients and inpatients who are not hospitalized at all, even though medication costs may be somewhat lower than with clozapine. Data on the cost-effectiveness of these agents in high hospital users should not be generalized to patient populations with lower-intensity service use.

Accepted for publication March 8, 1999.

This study was supported by the Department of Veterans Affairs Health Services Research and Development Service and the Cooperative Studies program. Clozapine was generously provided by Sandoz Pharmaceutical Corporation, East Hanover, NJ.

We would like to thank Lois Ucas, Jennifer Cahill, and Dennis Thompson of the chairman's office of the Department of Veterans Affairs Cooperative Study Group for data analysis . Marilyn Stolar, MS, assisted with the statistical analysis.

John Grabowski, MD, Detroit, Mich; Denise Evans MD, Augusta, Ga; Lawrence Herz, MD, Bedford, Mass; George Jurjus, MD, Brecksville, Ohio; Sidney Chang, MD, Brockton, Mass; Lawrence Dunn, MD, Durham, NC; John Crayton, MD, Hines, Ill; William B.Lawson, MD, PhD, Little Rock, Ark; Yeon Choe MD, Lyons, NJ; Richard Douyon, MD, Miami, Fla; Edward Allan, MD, Montrose, NY; John Lauriello, MD, Palo Alto, Calif; Michael Peszke, MD, Perry Point, Md; Jeffrey L. Peters, MD, Pittsburgh, Pa; Janet Tekell MD, San Antonio, Tex; Joseph Erdos, MD, PhD, West Haven, Conn.