Adolescents with depression identified in primary care settings often have limited treatment options beyond antidepressant (AD) therapy. We assessed the cost-effectiveness of a brief cognitive behavioral therapy (CBT) program among depressed adolescents who declined or quickly stopped using ADs.
A total of 212 youth with depression were randomly assigned to treatment as usual (TAU) or TAU plus brief individual CBT. Clinical outcomes included depression-free days (DFDs) and estimated quality-adjusted life-years (QALYs). Costs were adjusted to 2008 US dollars. Incremental cost-effectiveness ratios (ICERs) comparing CBT to TAU were calculated over 12- and 24-month follow-up periods.
Youth randomly assigned to CBT had 26.8 more DFDs (P = .044) and 0.067 more QALYs (P = .044) on average compared with TAU over 12 months. Total costs were $4976 less (P = .025) by the end of the 24-month follow-up among youth randomly assigned to CBT. Total costs per DFD were −$51 (ICER = −$51; 95% confidence interval [CI]: −$394 to $9) at 12 months and −$115 (ICER = −$115; 95% CI: −$1090 to −$6) at 24 months. Total costs per QALY were −$20 282 (ICER = −$20 282; 95% CI: −$156 741 to $3617) at 12 months and −$45 792 (ICER = −$45 792; 95% CI: −$440 991 to −$2731) at 24 months.
Brief primary care CBT among youth declining AD therapy is cost-effective by widely accepted standards in depression treatment. CBT becomes dominant over TAU over time, as revealed by a statistically significant cost offset at the end of the 2-year follow-up.
Brief primary care cognitive behavioral therapy (CBT) has been shown to be an effective option for youth with depression who decline antidepressants. CBT may be a cost-effective alternative to usual care (typically antidepressant treatment).
In this study, we demonstrate that brief primary care CBT is a cost-effective treatment option for adolescents with depression and likely generates cost savings over 2 years.
Adolescent depression is most frequently identified and treated in primary care settings.1,2 Antidepressant (AD) medications are the usual course of treatment for depressive disorders, and few other options are typically available to adolescents and their families in primary care.3,4 However, despite evidence of AD efficacy, as many as 50% of families who are identified by a medical professional as having a child with depression decline to begin AD therapy.3 Among those who do begin AD treatment, nearly half discontinue treatment before reaching a minimum therapeutic duration, for such reasons as side effects, lack of benefit, and cost.5,6 From this incongruity, it is suggested that a large proportion of adolescents with depression may remain untreated or may be receiving suboptimal levels of treatment.
In the current study, we provide an economic evaluation of a clinical trial designed to test the effectiveness of brief, primary care–based cognitive behavioral therapy (CBT) as an alternative treatment of depression among adolescents unreceptive to AD therapy. Although primary care CBT has been tested previously with adolescents with depression, this is the first clinical trial in which the effects of CBT among patients who had previously declined or quickly discontinued AD treatment were examined. In the trial, it was found that CBT was superior in reducing depressive symptoms as compared with a treatment-as-usual (TAU) control condition, with a number needed to treat of 6 at the immediate posttreatment and of 10 for the time points between the 12- and 24-month follow-up periods for the primary outcome of depression recovery.7 The economic evaluation described in this article is a cost-effectiveness analysis of the clinical trial from a societal perspective, comparing CBT to TAU over a 1-year follow-up and over the full 2-year course of the study.
Methods
Participants
Participants included 212 adolescents, aged 12 to 18, who were members of a large United States not-for-profit health system. Eligible participants were recruited between 2006 and 2010 and were diagnosed with depression by their primary care provider and either refused or rapidly discontinued AD treatment. Qualifying youth had to have a diagnosis of major depression obtained via the Children’s Schedule for Affective Disorders and Schizophrenia.8 Exclusion criteria included current AD use; bipolar disorder or any psychotic diagnosis; mental retardation, autism spectrum disorder, or pervasive developmental disorder; imminent risk of suicide; and history of receiving more than 8 sessions of CBT for depression. Participants were allowed to continue or initiate any TAU service. Eligible adolescents were randomly assigned to TAU alone or CBT plus TAU. The institutional review board at Kaiser Permanente Northwest approved the study.
Intervention
The intervention has been described in detail elsewhere.7 Briefly, the CBT program included a 2-module acute session in which addressing unrealistic thinking and/or behavior modification was the focus. Youth chose their beginning module and could stop after completion if they met recovery criteria. Continuation calls (up to 6) were provided as needed. CBT was delivered by master’s-level therapists, and implementation was monitored through biweekly supervision.
Clinical Outcomes
Participants were assessed at study entry (baseline) and at 6, 12, 26, 52, 78, and 104 weeks of follow-up. Independent evaluators blinded to each randomly assigned condition assessed the adolescents’ current depressive symptoms and severity with the Children’s Depression Rating Scale-Revised.9 Following other cost-effectiveness analyses of depression treatment trials, we used the Children’s Depression Rating Scale-Revised from all follow-up points to create summary measures of depression symptomology, including a measure of depression-free days (DFDs), over the 12- and 24-month follow-up periods for the economic analyses.10,–14 Days during follow-up were categorized into depression-free, partial-depression, or full-depression days. To calculate depressive symptoms for each day over the follow-up period, we used linear weighting to interpolate between the nondepressed and fully depressed thresholds and assign an estimated depression value to each day in the follow-up interval. The number of DFDs was the total number of days in the interval minus days with significant depressive symptoms. As used in previous work, this approach allows researchers to capture the effects of the intervention, including both elevated-symptom days and days in a full-depressive episode.14
We transformed DFDs into quality-adjusted life-years (QALYs) with preference weights assigned to depression, reflecting the value of different health or disease states derived from empirical studies.12,15 For example, DFDs are typically assigned a utility weight of 1.0 (full health), whereas days in a depressive episode are estimated to have a lower weight, such as 0.6. In empirical studies, it has been indicated that depression is associated with a decrease in health-related quality of life of 0.2 to 0.6.11,12,15 On the basis of these previous reports, we used 0.4 as the decrease in preference weight for the base case analysis. We bracketed sensitivity analyses with a more conservative decrease of 0.3 and a less conservative decrease of 0.5.14
Cost Outcomes
Costs were enumerated and summed across 12- and 24-month follow-up periods. All costs were adjusted to 2008 US dollars to align with study enrollment, which spanned from 2006 to 2010, by using the Bureau of Labor Statistics Consumer Price Index for medical care services.16
Intervention Service Costs
Accounting records provided costs for payroll, facilities and overhead, and goods and services. CBT therapists estimated their total resource use to complete the intervention tasks. We included costs of CBT sessions and therapist time related to travel, setup, out-of-session contacts, and general administrative duties. In addition, we included costs of supervision, training, and materials. We excluded research-specific costs.
Usual Care Costs
The Child and Adolescent Services Assessment provided data on patients’ usual care and mental health services use.17 The Child and Adolescent Services Assessment was designed to collect information on a wide variety of mental health services that may be received in a range of settings, including inpatient, outpatient, and informal services.17 We administered the 31-item Child Health Services Screen along with the Detailed Child Services Form. The latter provided details on the characteristics of the encounter, including intensity and frequency of the service along with medications dispensed.
To estimate health care costs, we applied unit costs developed for other studies.10,–12 We applied these unit costs to outpatient encounters, social services, dispensed medications, and inpatient stays. Variation in estimated costs came from the intensity and frequency of outpatient encounters and social services, the type of provider seen, the type and quantity of dispensed medications, and the type and length of inpatient stays. At baseline, services were assessed for the previous 3 months. At follow-up assessments, service use was assessed since the previous evaluation. We combined youth and adolescent reports and selected the parent report if the service was duplicative.
Family Costs
Following recommended guidelines, we estimated time costs that parents spent taking youth to services.18 To value parent time, we created profiles from published research on parent time spent on the intervention, TAU services, travel, and waiting.10,–12,14 Reported income was used to value time for employed parents, and state-specific minimum wage was used for unemployed parents.
Statistical Analyses
Analyses were conducted on an intention-to-treat basis. As described elsewhere, the needed sample size for the current study was calculated for the primary outcome of the clinic trial.7 We estimated incremental cost-effectiveness ratios (ICERs), or the mean difference in cost divided by the mean difference in clinical outcome (DFDs, QALYs), from a societal perspective for the 12- and 24-month periods post–random assignment. We used nonparametric bootstrapping with 1000 replications for hypothesis testing and to construct bias-corrected, accelerated confidence intervals (CIs).19 We used the net benefit regression method to examine cost-effectiveness across a range of willingness-to-pay levels.20,21 After clinical outcomes analyses, we did not adjust estimates of clinical effects for baseline factors; cost estimates were adjusted for baseline total costs, and model-based estimates were calculated for the typical (median baseline total cost) participant.7
To represent uncertainty in estimates, we first used the bootstrapped observations to estimate a 95% CI around the average ICERs. We then created a scatter plot of the bootstrapped incremental cost-and-effect pairs to construct an incremental cost-effectiveness plane, divided into 4 quadrants and centered at 0.21
Planned sensitivity analyses included the following: (1) evaluating impacts of alternate definitions of depression weights for the QALY calculations; (2) examining the impact of removing inpatient services because such services are high cost and may be rare, which may interfere with interpretation of estimated effects; and (3) evaluating impacts of potentially overly influential outliers.22 For the latter sensitivity analysis, we applied a conservative approach to identify overly influential outliers and winsorized cost data that met both the distributional and influence criteria.23
Of the 212 participants, 30 (14.2%) missed the 12-month and 26 (12.3%) missed the 24-month assessment. There were no statistically significant differences in missing data by study condition. With Stata (StataCorp, College Station, TX), we imputed missing data using multiple imputation with chained equations.24 We included baseline demographics and all nonmissing values of costs and outcomes in the imputation process. We created 5 imputation data sets and combined estimates so that SEs reflected the variability introduced by the imputation process.24,25
Results
Baseline demographics and clinical characteristics are reported in Table 1. We found no statistically significant differences between CBT and TAU. Details of recruitment and compliance with the study protocol are reported elsewhere.7
Baseline Demographics
| CBT + TAU (N = 106) | TAU Control (N = 106) | ||||
|---|---|---|---|---|---|
| Mean or n | SD or % | Mean or n | SD or % | P | |
| Youth demographics | |||||
| Age | 14.8 | 1.7 | 14.5 | 1.8 | .148 |
| Gender: % female | 72 | 67.9% | 73 | 68.9% | .883 |
| Ethnicity: % Hispanic | 18 | 17.0% | 16 | 15.1% | .708 |
| Race: % minority | 10 | 9.4% | 15 | 14.2% | .287 |
| Pubertal Development Scale | 3.2 | 0.7 | 3.2 | 0.7 | .768 |
| Parent demographics | |||||
| Socioeconomic status | 39.6 | 11.5 | 42.1 | 13.0 | .208 |
| Annual household income | 63 354 | 26 478 | 64 861 | 28 802 | .738 |
| CBT + TAU (N = 106) | TAU Control (N = 106) | ||||
|---|---|---|---|---|---|
| Mean or n | SD or % | Mean or n | SD or % | P | |
| Youth demographics | |||||
| Age | 14.8 | 1.7 | 14.5 | 1.8 | .148 |
| Gender: % female | 72 | 67.9% | 73 | 68.9% | .883 |
| Ethnicity: % Hispanic | 18 | 17.0% | 16 | 15.1% | .708 |
| Race: % minority | 10 | 9.4% | 15 | 14.2% | .287 |
| Pubertal Development Scale | 3.2 | 0.7 | 3.2 | 0.7 | .768 |
| Parent demographics | |||||
| Socioeconomic status | 39.6 | 11.5 | 42.1 | 13.0 | .208 |
| Annual household income | 63 354 | 26 478 | 64 861 | 28 802 | .738 |
Usual Care Service Use and Costs
In Table 2, we describe the usual care services used in each study condition in addition to the summation of costs during the 12- and 24-month follow-up periods. Approximately two-thirds of both patients receiving CBT (67.7%) and patients receiving TAU (69.2%) used some form of usual care services in the 12-month follow-up period, and ∼4 out of 5 youth in both CBT (82.1%) and TAU (80.2%) used services in the 24-month follow-up period. Patterns of use were generally similar, except patients receiving CBT had fewer hospitalizations compared with those receiving TAU (1.1% vs 8.8% during 12 months, and 4.4% vs 12.1% during 24 months). Unadjusted average financial and time costs of nonprotocol services were lower in CBT (mean = $553, SD = $1698) compared with TAU (mean = $2756, SD = $10 259) across 12 months and lower in CBT (mean = $2141, SD = $5552) compared with TAU (mean = $7354, SD = $24 403) across 24 months. The average cost of the CBT program was $641 (SD = $229). Statistical comparisons of aggregate costs are presented in the Cost-Effectiveness Analysis section.
Unadjusted Service Use and Cost (2008 US Dollars) by Randomly Assigned Condition Through 12 and 24 Months
| Use and Cost Categorizations | Through 12 mo | Through 24 mo | ||
|---|---|---|---|---|
| CBT | TAU | CBT | TAU | |
| Nonprotocol service use: % of youth endorsing use | ||||
| Inpatient mental health days | 1.1 | 8.8 | 4.4 | 12.1 |
| Inpatient alcohol or drug days | 0 | 1.1 | 1.1 | 3.3 |
| Counseling or medication management visits | 55.6 | 51.6 | 72.2 | 61.5 |
| Day hospital days | 0 | 0 | 1.1 | 3.3 |
| Alcohol or drug treatment visits | 0 | 3.3 | 0 | 3.3 |
| Crisis services | 1.1 | 0 | 2.2 | 1.1 |
| Medical doctor visits | 23.3 | 14.3 | 35.6 | 25.3 |
| Emergency department visits | 4.4 | 3.3 | 11.1 | 8.8 |
| AD medications | 10.0 | 7.7 | 18.9 | 16.5 |
| Stimulant medications | 2.2 | 3.3 | 4.4 | 3.3 |
| Other psychotropic medications | 2.2 | 0 | 2.2 | 2.2 |
| Any school services | 21.1 | 28.6 | 35.6 | 39.6 |
| Juvenile correction contact | 0 | 3.3 | 3.3 | 4.4 |
| Any nonprotocol services use | 67.8 | 69.2 | 82.1 | 80.2 |
| Average total cost, 2008 US $, mean (SD) | ||||
| Nonprotocol costs | ||||
| Service use cost | 450 (1545) | 2215 (8740) | 1578 (4533) | 5905 (20 098) |
| Family costs | 103 (262) | 541 (1734) | 563 (1318) | 1448 (4590) |
| Total | 552 (1698) | 2756 (10 259) | 2141 (5552) | 7354 (24 403) |
| Intervention costs | ||||
| CBT program costs | 587 (206) | NA | 587 (206) | NA |
| Intervention family costs | 55 (37) | NA | 55 (37) | NA |
| Total | 641 (229) | NA | 641 (229) | NA |
| Total cost (nonprotocol + intervention) | 1221 (1730) | 2756 (10 259) | 2811 (5550) | 7354 (24 403) |
| Use and Cost Categorizations | Through 12 mo | Through 24 mo | ||
|---|---|---|---|---|
| CBT | TAU | CBT | TAU | |
| Nonprotocol service use: % of youth endorsing use | ||||
| Inpatient mental health days | 1.1 | 8.8 | 4.4 | 12.1 |
| Inpatient alcohol or drug days | 0 | 1.1 | 1.1 | 3.3 |
| Counseling or medication management visits | 55.6 | 51.6 | 72.2 | 61.5 |
| Day hospital days | 0 | 0 | 1.1 | 3.3 |
| Alcohol or drug treatment visits | 0 | 3.3 | 0 | 3.3 |
| Crisis services | 1.1 | 0 | 2.2 | 1.1 |
| Medical doctor visits | 23.3 | 14.3 | 35.6 | 25.3 |
| Emergency department visits | 4.4 | 3.3 | 11.1 | 8.8 |
| AD medications | 10.0 | 7.7 | 18.9 | 16.5 |
| Stimulant medications | 2.2 | 3.3 | 4.4 | 3.3 |
| Other psychotropic medications | 2.2 | 0 | 2.2 | 2.2 |
| Any school services | 21.1 | 28.6 | 35.6 | 39.6 |
| Juvenile correction contact | 0 | 3.3 | 3.3 | 4.4 |
| Any nonprotocol services use | 67.8 | 69.2 | 82.1 | 80.2 |
| Average total cost, 2008 US $, mean (SD) | ||||
| Nonprotocol costs | ||||
| Service use cost | 450 (1545) | 2215 (8740) | 1578 (4533) | 5905 (20 098) |
| Family costs | 103 (262) | 541 (1734) | 563 (1318) | 1448 (4590) |
| Total | 552 (1698) | 2756 (10 259) | 2141 (5552) | 7354 (24 403) |
| Intervention costs | ||||
| CBT program costs | 587 (206) | NA | 587 (206) | NA |
| Intervention family costs | 55 (37) | NA | 55 (37) | NA |
| Total | 641 (229) | NA | 641 (229) | NA |
| Total cost (nonprotocol + intervention) | 1221 (1730) | 2756 (10 259) | 2811 (5550) | 7354 (24 403) |
NA, not applicable.
Clinical Outcomes
In Table 3, we show the adjusted mean and differences in costs and clinical outcomes used to calculate ICERs. Means and differences are adjusted to the typical study participant where applicable. Youth in CBT had 26.8 more DFDs (P = .043) and 0.067 higher quality-adjusted life-years based on depression-free days (DFD-QALYs) (P = .043) compared with those in TAU over the 12 months after study entry. Across the full 24-month follow-up period, the youth in CBT had an additional 43.3 DFDs on average compared with those in TAU (but the difference was not statistically significant [P = .078]) and an additional 0.109 QALYs on average for QALYs (but again, the difference was not statistically significant [P = .078]).
Adjusted Means, Differences, and ICERs Through 12 and 24 Months
| CBT (N = 106), Mean (SEa) | TAU (N = 106), Mean (SEa) | Differences (CBT − TAU), Mean (SEb) | ICER (95% CIc) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Cost, $ | Effect | Cost, $ | Effect | Cost, $ | Effect | Cost per DFD, $ | Cost per DFD-QALY, $ | ||||
| DFDs | DFD-QALYs | DFDs | DFD-QALYs | DFDs | DFD-QALYs | ||||||
| 12 mo post–random assignment | |||||||||||
| Base case (N = 212) | 1903 (752) | 230.8 (10.3) | 0.665 (0.026) | 3268 (796) | 204.0 (10.9) | 0.598 (0.027) | −1365 (906) | 26.8* (13.2) | 0.067* (0.033) | −51 (−394 to 9) | −20 282 (−156 741 to 3617) |
| Sensitivity analyses | |||||||||||
| Less conservative QALYs (0.5) | 1903 (752) | NA | 0.620 (0.030) | 3268 (796) | NA | 0.553 (0.031) | −1365 (906) | NA | 0.067* (0.034) | NA | −19 970 (−154 330 to 3561) |
| More conservative QALYs (0.7) | 1903 (752) | NA | 0.632 (0.029) | 3268 (796) | NA | 0.567 (0.030) | −1365 (906) | NA | 0.065* (0.032) | NA | −20 604 (−159 229 to 3674) |
| Exclude inpatient services | 1226 (98) | 230.8 (10.3) | 0.665 (0.026) | 610 (95) | 204.0 (10.9) | 0.598 (0.027) | 616*** (108) | 26.8* (13.2) | 0.067* (0.033) | 23 (10 to 199) | 9156 (3955 to 79 390) |
| 24 mo post–random assignment | |||||||||||
| Base case (N = 212) | 3655 (1978) | 500.7 (21.0) | 1.427 (0.053) | 8631 (2103) | 457.4 (19.7) | 1.319 (0.049) | −4976* (2225) | 43.3 (24.6) | 0.109 (0.062) | −115 (−1090 to −6) | −45 792 (−440 991 to −2731) |
| Sensitivity analyses | |||||||||||
| Less conservative QALYs (0.5) | 3655 (1978) | NA | 1.427 (0.053) | 8631 (2103) | NA | 1.319 (0.049) | −4976* (2225) | NA | 0.109 (0.062) | NA | −45 079 (−433 989 to −2688) |
| More conservative QALYs (0.7) | 3655 (1978) | NA | 1.427 (0.053) | 8631 (2103) | NA | 1.319 (0.049) | −4976* (2225) | NA | 0.109 (0.062) | NA | −46 528 (−448 222 to −2774) |
| Winsorize highly influential outliers | 3462 (1761) | NA | 1.427 (0.053) | 7808 (1905) | NA | 1.319 (0.049) | −4345* (1963) | NA | 0.109 (0.062) | −100 (−999 to −8) | −39 992 (−398 919 to −3254) |
| Exclude inpatient services | 1886 (168) | 500.7 (21.0) | 1.427 (0.053) | 1278 (203) | 457.4 (19.7) | 1.319 (0.049) | 607** (183) | 43.3 (24.6) | 0.109 (0.062) | 14 (2 to 194) | 5588 (860 to 77 012) |
| CBT (N = 106), Mean (SEa) | TAU (N = 106), Mean (SEa) | Differences (CBT − TAU), Mean (SEb) | ICER (95% CIc) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Cost, $ | Effect | Cost, $ | Effect | Cost, $ | Effect | Cost per DFD, $ | Cost per DFD-QALY, $ | ||||
| DFDs | DFD-QALYs | DFDs | DFD-QALYs | DFDs | DFD-QALYs | ||||||
| 12 mo post–random assignment | |||||||||||
| Base case (N = 212) | 1903 (752) | 230.8 (10.3) | 0.665 (0.026) | 3268 (796) | 204.0 (10.9) | 0.598 (0.027) | −1365 (906) | 26.8* (13.2) | 0.067* (0.033) | −51 (−394 to 9) | −20 282 (−156 741 to 3617) |
| Sensitivity analyses | |||||||||||
| Less conservative QALYs (0.5) | 1903 (752) | NA | 0.620 (0.030) | 3268 (796) | NA | 0.553 (0.031) | −1365 (906) | NA | 0.067* (0.034) | NA | −19 970 (−154 330 to 3561) |
| More conservative QALYs (0.7) | 1903 (752) | NA | 0.632 (0.029) | 3268 (796) | NA | 0.567 (0.030) | −1365 (906) | NA | 0.065* (0.032) | NA | −20 604 (−159 229 to 3674) |
| Exclude inpatient services | 1226 (98) | 230.8 (10.3) | 0.665 (0.026) | 610 (95) | 204.0 (10.9) | 0.598 (0.027) | 616*** (108) | 26.8* (13.2) | 0.067* (0.033) | 23 (10 to 199) | 9156 (3955 to 79 390) |
| 24 mo post–random assignment | |||||||||||
| Base case (N = 212) | 3655 (1978) | 500.7 (21.0) | 1.427 (0.053) | 8631 (2103) | 457.4 (19.7) | 1.319 (0.049) | −4976* (2225) | 43.3 (24.6) | 0.109 (0.062) | −115 (−1090 to −6) | −45 792 (−440 991 to −2731) |
| Sensitivity analyses | |||||||||||
| Less conservative QALYs (0.5) | 3655 (1978) | NA | 1.427 (0.053) | 8631 (2103) | NA | 1.319 (0.049) | −4976* (2225) | NA | 0.109 (0.062) | NA | −45 079 (−433 989 to −2688) |
| More conservative QALYs (0.7) | 3655 (1978) | NA | 1.427 (0.053) | 8631 (2103) | NA | 1.319 (0.049) | −4976* (2225) | NA | 0.109 (0.062) | NA | −46 528 (−448 222 to −2774) |
| Winsorize highly influential outliers | 3462 (1761) | NA | 1.427 (0.053) | 7808 (1905) | NA | 1.319 (0.049) | −4345* (1963) | NA | 0.109 (0.062) | −100 (−999 to −8) | −39 992 (−398 919 to −3254) |
| Exclude inpatient services | 1886 (168) | 500.7 (21.0) | 1.427 (0.053) | 1278 (203) | 457.4 (19.7) | 1.319 (0.049) | 607** (183) | 43.3 (24.6) | 0.109 (0.062) | 14 (2 to 194) | 5588 (860 to 77 012) |
NA, not applicable.
Delta method.
Bootstrapped SE.
Bias corrected.
P < .05; ** P < .01; *** P < .001.
Cost-effectiveness Analysis
At 12 months post–random assignment, the estimated cost per DFD was −$51 (ICER = −$51; 95% CI: −$394 to $9). Using DFD-QALYs as the clinical outcome, we estimated that the cost per QALY was −$20 282 (ICER = −$20 282; 95% CI: −$156 741 to $3617). At 24 months post–random assignment, the estimated cost per DFD was −$115 (ICER = −$155; 95% CI: −$1090 to $6). Using DFD-QALYs as the clinical outcome, we estimated that the cost per QALY was −$45 792 (ICER = −$10 721; 95% CI: −$440 991 to $2731). A negative ICER estimate can result from either a negative difference in the numerator (cost) or denominator (outcome), with meaningfully different interpretations. With a negative difference in costs, it is suggested that the intervention may be less expensive on average than the comparison condition, which is the case here.
In Fig 1, we present the cost-effectiveness planes for both the 12- and 24-month follow-up periods. The points on the scatter plot depict the incremental mean difference in clinical outcome and cost from each of the 1000 bootstrap replications. In Fig 2, we show the base case cost-effectiveness acceptability curves for DFD-QALYs at 12 and 24 months. The probability that an intervention is cost-effective is represented on the vertical axis, and the maximum dollar amount a decision-maker is willing to pay per outcome (eg, QALY) is depicted on the horizontal axis. Included in Fig 2 are vertical markers at $50 000 and $100 000, which are common thresholds for reference.26 Using analyses through month 12, we determined that CBT has a 93%, 98%, and 98% probability of being cost-effective at a willingness to pay of $0, $50 000, and $100 000 per DFD-QALY, respectively. In analyses through month 24, it is demonstrated that CBT has a 97%, 99%, and 98% probability of being cost-effective at a willingness to pay of $0, $50 000, and $100 000 per DFD-QALY, respectively.
Cost-effectiveness planes. Total costs and outpatient costs through 12- and 24-month follow-up periods. A, Total costs through 12 months. B, Total costs through 24 months.
Cost-effectiveness acceptability curves. Total costs and outpatient costs through 12- and 24-month follow-up periods. A, Total costs through 12 months. B, Total costs through 24 months.
Sensitivity analyses were used to test the robustness of our findings across a variety of planned analyses. Alternate utility weights used in the QALY calculation had little impact on results compared with our base case analyses. Specifically, cost per QALY ranged from −$20 604 to −$19 970 and −$46 828 to −$45 079 in analyses of the 12- and 24-month follow-up periods, respectively. Removal of inpatient services caused both the cost per DFD and cost per QALY to become positive. The cost per DFD was $23 and $14 at 12 and 24 months post–random assignment, respectively. The cost per QALY was $9156 and $5588 at 12 and 24 months post–random assignment, respectively. Two participants were identified as being overly influential outliers in analyses of the 24-month follow-up period. Winsorization of their cost data did not change the patterns of results. The cost per DFD was −$100 and the cost per QALY was −$39 992 in the 24-month follow-up period.
Discussion
We found a significant advantage of CBT over TAU in terms of DFDs and QALYs over a 12-month follow-up period, and we found that CBT became dominant over TAU by the end of the 24-month follow-up, as revealed by a statistically significant cost offset. Total costs were lower on average among CBT participants in the 12 months after random assignment, but the difference between CBT and TAU was statistically significant. Although differences in DFDs and QALYs favored CBT compared with TAU in the 24-month follow-up, this advantage was not statistically significant at traditional thresholds.
Cost-effectiveness analyses provide the ability to compare the relative costs and benefits of different interventions as decision-makers consider what programs to implement. Use of QALYs in the calculation of ICERs allows the additional benefit of comparison of interventions across different categories of disease or illness. Although no official metric exists in the United States as to what levels of cost-per-QALY are considered to be cost-effective, $50 000 is a frequently cited ICER threshold.27 With all of our estimates, including sensitivity analyses, falling well below this threshold, strong evidence is provided that the intervention is a cost-effective option for adolescents with depression who declined or rapidly discontinued pharmacotherapy.
In addition, with our estimates we demonstrate lower total costs on average for CBT youth, and suggest lower costs and better clinical outcomes for the CBT condition. We show clear evidence of a significant, positive clinical effect over 12 months and evidence that CBT becomes dominant over TAU by the end of the 24-month follow-up. The primary study was powered for us to detect a main clinical effect, not any differences in economic outcomes, which typically require far larger sample sizes to detect differences. Regardless, because of the fact that total costs were not statistically different between groups over 12 months, we suggest the CBT intervention may be cost neutral in the first year, but because of the statistically significant difference across 24 months, we also suggest the CBT intervention is likely cost saving by the end of the second year.
Cost-effectiveness acceptability curves allowed us to examine probabilities of cost-effectiveness over the range of willingness-to-pay values. Decision-makers willing to pay $50 000 per QALY would have a 98% probability of CBT being cost-effective across 12 months and a 99% probability of CBT being cost-effective across 24 months. As an extreme example, decision-makers who place no value on DFD-QALYs (a willingness to pay equal to $0) would still have a 93% and 97% chance that CBT is cost-effective across 12 and 24 months, respectively. With all of these findings, which are conservative because many have suggested these thresholds may be too low, we suggest the intervention will likely be cost-effective if implemented.26
Our findings are in contrast to previous research on the cost-effectiveness of CBT, in which researchers found CBT was not cost-effective compared with usual care or AD treatment alone.10,28 These differences could be attributable to the composition of our sample’s depression severity and onset as well as the types of nonprotocol services being used by both study groups, which could include a return to or initiation of pharmacotherapy. As a reminder, our sample is composed of youth with an incident episode of depression who either refused or quickly discontinued AD treatment. In addition, differences might be explained by the setting and mode of CBT delivery. All youth in the study were identified through primary care encounters, and youth randomly assigned to CBT received one-on-one counseling sessions in the same primary care clinic.
The model of CBT delivered in this trial had a maximum of 9 sessions, with the option of pausing or ending therapy partway through. This yielded an average number of 6.4 sessions (median = 7). We chose to test this “lean” model because it was more likely to be adopted in the health care organization. This low number of sessions is in contrast to the typically much longer CBT delivered in many previous trials. In a meta-analysis of CBT for children with depression, researchers found that the average CBT exposure across studies was comparable to ∼13 sessions in the current study.29 The much lower session total for CBT in this study resulted in a much lower cost of delivering this intervention, particularly compared with other cost analyses of other, longer CBT programs. This may have contributed to the more favorable cost outcomes observed in this study. A final possible explanation for differences between our cost-effectiveness findings and those in previous studies is the lack of a clear, clinical benefit of CBT in those studies. In the STAND (Study of Adolescents and Depression) study,30 researchers found a clear, marked benefit of CBT compared with usual care in the primary outcomes and secondary depression-related outcome measures.
Conclusions from this economic evaluation are limited by several factors. First, the study sample did not include nationally representative proportions of racial and ethnic minorities and did not include adolescents without health insurance coverage, both of which reduce the ability to generalize these findings more broadly. Second, we did not include depression-related productivity losses for youth in our calculations of total cost, either in the 24-month study period or beyond. Depression has been shown to correlate with poor educational outcomes, such as not completing high school, which in turn affect lifetime productivity and earning power.31,–33 Improvements in depressive symptomology from CBT found in this study might lower the productivity costs related to adolescent depression; therefore, our reported estimates may be slightly conservative. Third, in the clinical trial we did not collect a preference-based measure of health-related quality of life, which would have allowed richer quantification of the impact of CBT on patients’ well-being. We estimated the effect of CBT on youth QALYs by applying indirect established methods for translating DFDs into QALYs.10,–14 Preference weights for adults with depression were used, which may not mirror the effects of depression on health-related quality of life among adolescents because no published empirical weights for youth exist.34 Fourth, the feasibility of widely implementing a similar primary care–based CBT program would likely be limited by the availability of master’s-level behavioral health care providers qualified to deliver such an intervention.
Conclusions
Many adolescents with depression choose to not initiate or continue AD therapy, which limits their options for depression treatment. When providing alternative treatment options for adolescents with depression, health care providers typically consider (among other factors such as patient preferences) whether the treatment is cost-effective. In this evaluation, it is demonstrated that brief, primary care–based CBT is a cost-effective option for the treatment of depression among adolescents with depression who decline or quickly discontinue pharmacotherapy. Given the enduring negative impact of untreated or undertreated depression and the evidence presented in this study, brief CBT may be a beneficial treatment option for health systems to implement.
Dr Dickerson conceptualized and designed the study, conducted analyses, oversaw data management and quality audits, and drafted the initial manuscript; Dr Lynch assisted with conceptualizing and designing the study and obtaining funding and reviewed and revised the initial manuscript; Dr Leo assisted with analyses, conducted data management and quality audits, and reviewed and revised the manuscript; Drs DeBar and Pearson assisted with the conduct of the study and reviewed and revised the initial manuscript; Dr Clarke helped conceptualize and design the study, obtained funding, directed the conduct of the study, and reviewed and revised the initial manuscript; and all authors approved the final manuscript as submitted.
This trial has been registered at www.clinicaltrials.gov (identifier NCT00523081).
FUNDING: This study was funded by the National Institute of Mental Health (grant R01-MH73918). Funded by the National Institutes of Health (NIH).
Acknowledgments
For their project implementation and data collection contributions, we thank Rebecca Bogorad, MA; Kristina Booker, BS; Alison Firemark, MA; Stephanie Hertert, MEd; Kelly Kirk, BS; Sue Leung, PhD; Alex MacMillan, BS; Kevin Rogers, MA; Nina Scott, MSW; Natalia Tommasi, MA; Bobbi Jo Yarborough, PsyD; and Micah Yarborough, MA. We also thank all the youth and families who participated in our studies.
