Few randomized trials evaluate the cost-effectiveness of treatments for mental illnesses. This is especially true for disorders seen in childhood (1), making the Domino et al. study of the cost-effectiveness of treatments for adolescent depression in this issue of the Journal particularly valuable (2). Families, clinicians, and policymakers must balance potential benefits against costs when making treatment decisions. The Treatment for Adolescents With Depression Study (TADS) compared the effects of cognitive-behavioral therapy (CBT), an antidepressant medication (fluoxetine), and the combination of these treatments to a baseline in which adolescents received only placebo. Placebo and fluoxetine, either alone or in combination with CBT, were prescribed by physicians blind to the identity of the pill but who encouraged patients and their families about the positive effects of treatment. CBT, regardless of whether received alone or in combination with fluoxetine, was administered by a separate group of therapists. The study was a randomized, masked clinical trial conducted at 13 sites across the United States. Data from stage I of TADS (after 12 weeks of treatment) showed 71% of adolescents had improvement in symptoms with combination therapy (as measured by the Children’s Depression Rating Scale—Revised), while 61% improved with fluoxetine alone. Only 43% of those receiving CBT alone improved, which was not significantly better than those receiving placebo (35%) (3). Domino et al. used these data from stage I for their analyses.
The authors expressed outcomes in quality-adjusted life years (QALYs), a unit of outcome often incorporated in health care research. The average QALY achieved by patients in this study was 0.16 years, reflecting about 22 days of relatively low symptoms in the first 12 weeks of treatment. The authors transformed their primary outcome measure (change on the Children’s Depression Rating Scale—Revised) into depression-free days and QALYs using assumptions based on the adult psychiatric literature. These transformations enabled comparison of the cost-effectiveness of treatments in this study to findings from other studies. Costs in the study were intended to be comprehensive measures of social costs of treatment and included treatment costs (in- and out-of-protocol) as well as time and travel for caregivers. In concordance with accepted methodology in the literature, Domino et al. present their findings in the form of incremental cost-effectiveness ratios. Incremental costs and effects are measured relative to a control, which in this case is the placebo group. The key findings, shown in Table 4 of Domino et al., are the incremental cost-effectiveness ratios for fluoxetine relative to placebo ($61 per unit decrease on the Children’s Depression Rating Scale—Revised) and for combination treatment relative to placebo ($249 per unit decrease). Other results follow from these findings, including sensitivity analyses on cost assumptions and transformation of Children’s Depression Rating Scale—Revised scores into depression-free days and QALY measures. In all other analyses, the incremental cost-effectiveness ratio of fluoxetine was reported as 4 to 5 times more cost-effective than combination therapy.
The gold standard of clinical efficacy research is comparison to placebo. For TADS, the initial report of the 12-week data provided important information on the advantage of fluoxetine and CBT in reducing depression, as well as the disadvantage of adverse events such as anxiety, insomnia, and possible suicidality. These advantages and disadvantages are best judged in comparison with a placebo group. The investigators’ foresight in including a placebo group and, more importantly, the willingness of patients and their families to accept the possibility of placebo treatment resulted in invaluable data which will help guide the choice of treatment for future patients (3). However, a placebo baseline, rather than no treatment or treatment as usual, is problematic for a cost-effectiveness analysis, because the choice of comparison group affects the incremental cost-effectiveness ratios for the treatment of interest. Incremental costs and incremental effectiveness can be interpreted in a straightforward way when a new treatment is compared with the alternative of an older treatment. Interpretation of an incremental cost-effectiveness ratio with placebo, as characterized in TADS, is not straightforward, because the costs of placebo treatment are an artifact of the research design. In terms of effects, it is clear that when compared with placebo, the incremental clinical gains of fluoxetine (for example) are due to the active treatment effects of the drug. But what is the meaning of the incremental cost of an active treatment compared with placebo when substantial costs attributed to the placebo exist only because of the research design?
The mean cost of the placebo treatment arm ($1,433) was 84% of the cost of the fluoxetine arm and 43% of the combination arm (Table 3 of Domino et al.). In the incremental cost-effectiveness ratio calculation, incremental effects of fluoxetine are thus figured in relation to only 16% of the costs of treatment with the drug, making the ratio lower (and much more favorable) than it would be compared against no treatment (i.e., no effects and no costs). Benchmarking combination therapy against placebo also makes its incremental cost-effectiveness ratio more favorable, but because 57% of combination therapy costs are incremental over placebo, the effect on the ratio for combination therapy is proportionally less. Computation of incremental cost-effectiveness ratios relative to a costly placebo improves the overall cost-effectiveness of all interventions and exaggerates the difference in cost-effectiveness between drug alone and combination treatment.
Domino et al.’s chosen measure of treatment effects, QALYs, measures both quantity (life expectancy) and quality of life years. Quality values are gleaned from estimates made by one sample of respondents and then extrapolated to the sample under consideration, ideally a sample with similar characteristics (e.g., using ratings from adult women with depression to construct QALYs for adult women with depression). To determine the effect of an intervention on quality of life over time, one multiplies the extra years obtained through the intervention by the health utility assigned to living in a particular condition (roughly, the quality of those extra years). One can then determine the cost per QALY and compare various interventions based on their relative costs for providing an added QALY. However, using data from adults to impute QALYs for depressed adolescents may be an unreliable guide to true benefits from treatment.
To evaluate treatment alternatives fully, all of the costs and effects of a treatment should have a chance to emerge. Domino et al. used the data available to them, outcomes and costs at 12 weeks, but many clinicians would consider this a very short period of time in which to make a decision about the utility of an intervention, especially if the disorder being studied is likely to result in longer periods of symptoms or if the effects of the intervention may not be seen until later. Indeed, a recent publication of follow-up TADS data reveals a somewhat different outcome. At 36 weeks, combination therapy continued to show better outcomes (86% improvement on the Children’s Depression Rating Scale—Revised), but improvement with CBT alone (81%) was similar to fluoxetine alone (81%) (4).
In addition, another important potential outcome measure, suicidal ideation, appears to be more prominent in those who are taking fluoxetine alone. If a longer study period showed even more improvement with CBT than fluoxetine and the incidence of suicidal ideation and problematic medical side effects became more prominent in those patients taking fluoxetine alone, then calculations for QALYs would change considerably and cost-effectiveness could change in favor of CBT. This highlights the problem of basing cost-effectiveness decisions on short-term data for conditions likely to be chronic and in need of long-term interventions, which have variable effects over time on health and functional status.
Recent studies from outside the United States also indicate uncertainty about the relative cost-effectiveness of various interventions for depressed adolescents. Although there are differences in the participants and study design, a recent British study of depression in adolescents reported no difference in the effectiveness of fluoxetine alone compared with a combination of fluoxetine and CBT (5). An analysis of the cost-effectiveness of the interventions at 28 weeks showed no advantage to combination therapy versus fluoxetine alone (6). However, an Australian study of children and adolescents reported that CBT alone was more cost-effective than antidepressant medications (7). Thus, it appears this issue remains unresolved.
Domino and colleagues have provided an important addition to the continuing development of the field of cost-effectiveness studies in children with mental illness. Interim results from one study cannot be expected to resolve all the questions that parents, clinicians, and policymakers have about which interventions are the most cost-effective for depressed adolescents. Further results from TADS and a critical assessment of findings from other studies will bring us closer to useful answers. Some of us were involved in the development of research recommendations issued almost a decade ago in a report from the National Institute of Mental Health (NIMH) (8). We noted then the critical need for the development and improvement of measures and methods to determine the cost-effectiveness of clinical interventions. Today more than ever, NIMH and other federal research agencies should expand efforts to ensure the field develops methods to assess these critical questions of cost-effectiveness. Bridging research and practice requires providing policymakers with reliable estimates of the relative cost-effectiveness of treatment alternatives.