Cost-effectiveness of an Intervention to Prevent Depression in At-Risk Teens FREE

Depression is common among adolescents, with a point prevalence between 3% and 8%.¹ By age 18 years, as many as 25% of adolescents have had at least 1 depressive episode.² Depressive disorders in children and teens increase the risk of illness, interpersonal problems, and psychosocial difficulties that persist long after the episode,³ and adolescents who experience depressive episodes have an increased risk of substance abuse and suicidal behavior.^4- 6 Adults with depression have increased health care costs,⁷ and successful depression treatment may decrease these costs for adults⁸ and children.⁹

Recent research^10- 14 indicates that some groups are at much higher risk of developing depression, including children and adolescents with a depressed parent and individuals who report significant subsyndromal depressive symptomatology (without meeting full DSM criteria). Preventing depression in adolescents could decrease the chance of premature death, increase the quality of life and productivity of teens and their families, and reduce health care costs for these teens.

Evidence is emerging that psychosocial interventions can prevent depression^15- 17 in adolescents, and prevention interventions targeted at high-risk groups have recently had favorable results.^16- 17 Our group has described a successful group cognitive behavioral intervention to prevent depression episodes in at-risk adolescents.¹⁸ Teens in the study had 2 significant risk factors: (1) they were offspring of depressed parents and (2) they had significant subsyndromal symptoms and/or a past episode of depression.

Adoption of evidence-based interventions to prevent depression will depend on the balance of the clinical benefits and costs. Yet few studies have examined the economic impact of prevention interventions for any mental health problems, and we are not aware of any cost-effectiveness analyses of depression prevention in adolescents that used randomized clinical trial data. This information could help decision makers assess the relative value of alternative interventions for adolescent depression.

This report presents the cost-effectiveness of a recent prevention trial conducted with the subsyndromal adolescent offspring of parents treated for depression. The randomized controlled trial in a large, group-model health maintenance organization (HMO) examined the ability of the intervention to prevent progression to future episodes of major depression. This article presents an incremental cost-effectiveness analysis of the group cognitive behavioral intervention relative to usual care, from the societal perspective, for 1 year after the intervention.

The randomized clinical trial (RCT) conducted by Clarke et al¹⁸ is described elsewhere. Briefly, the RCT recruited participants from Kaiser Permanente Northwest, an HMO with about 410 000 members. The HMO’s Human Subjects Committee approved all study procedures. The RCT used the HMO databases to identify parents of teenagers who had had at least 2 dispensations of an antidepressant medication and/or mental health visits within the past year. Of these cases, medical chart reviews confirmed that 3935 parents also had a depression diagnosis and/or symptoms. Each parent’s physician mailed introductory letters to those they judged appropriate for the study (n = 2995). Study staff then called parents for a brief screen of study criteria and asked adolescent offspring about participating in the study. Interested families were invited for an intake evaluation at the research center.

Interviews were completed with 481 parents and 551 adolescents. This assessment confirmed the parent’s diagnosis of depression and assessed adolescent psychiatric diagnoses, symptoms, and psychosocial functioning. Parents were assessed with the Family Schedule for Affective Disorders and Schizophrenia.¹⁹ Teens were grouped into clinical groups based on their depressive symptoms and determination of DSM-III-R^20- 21 diagnoses; details on all interviewed subjects are reported elsewhere.²² This analysis focuses on a medium depression group (n = 123 [25.9%]), which was called the subsyndromal group.¹² These teens reported a previous depression episode or subdiagnostic levels of depressive symptoms that were insufficient to meet full criteria for a DSM-III-R affective diagnosis (Center for Epidemiologic Studies Depression Scale score, ≥24).¹⁶ Teens who met the criteria for the subsyndromal group and agreed to participate were randomized to receive either the prevention intervention program or usual care.

The prevention program²³ was an abbreviated version of an adolescent depression treatment program²⁴ that had been tested previously.^25- 26 The intervention consisted of 15 one-hour cognitive behavioral therapy (CBT) sessions for groups of 6 to 10 adolescents. The CBT groups were led by a master’s-level therapist trained in the approach and were conducted at the HMO clinic offices. Details of the program are reported elsewhere.¹⁸

All teens could initiate or continue any services normally provided by the HMO and/or outside services, including specialty mental health care and antidepressant medication. No additional services were provided to the usual-care control group, but no services usually available were limited in any way.

We include direct and indirect costs of the intervention.²⁷ Direct costs include intervention sessions and all usual-care services (both HMO and outside services). Indirect treatment costs include teen and parent time and travel costs for obtaining usual-care and intervention services. All costs are valued in 2000 US dollars. We did not use discounting because the analysis time frame was 1 year after study enrollment.

We estimated the total cost of intervention services from clinical trial records and study staff estimates. We divided intervention costs into fixed and variable costs. We allocated fixed costs across all randomized intervention participants and allocated variable costs according to each participant’s use of intervention services. Information for the intervention cost estimates were collected throughout the trial. Study and HMO accounting records provided payroll costs, cost of facilities and overhead, and information on purchases of goods and services. Study staff used time sheets and written records of intervention activities to estimate the time to complete each intervention task. For example, the intervention therapists kept logs of time (in minutes) spent speaking with participants outside of intervention sessions. Study staff also reported use of capital equipment, space, and supplies needed to produce the intervention.

We included all costs of conducting the intervention, including costs of running the CBT groups, therapists’ training, and all session materials (workbooks, handouts, etc). We also included the identification and outreach costs. Research-specific costs, such as randomization costs, were excluded.

We created comprehensive profiles of usual-care HMO services from the available electronic HMO data. These data, used in numerous previous studies, very accurately represent services paid for by the HMO. Services include all outpatient visits, including mental health specialty and other medical care; all inpatient care; drug utilization; laboratory tests; and radiology procedures. We supplemented HMO data with monthly mailed surveys asking participants to report any non-HMO services that they received for their depression symptoms.

For the HMO services, we estimated costs by applying unit costs developed and tested in previous studies^28- 31 to the HMO utilization measures. These final cost variables represent HMO expenditures. For non-HMO services, we applied local market unit costs to create final cost variables (unit cost details are available on request).

We estimated family costs on the basis of patient utilization data and other study information collected during the trial. From these data, we created profiles of teen and parent time spent for the intervention, usual-care services, travel to services, and waiting. Study records contained the number of sessions each participant attended; sessions lasted 1 hour. The intervention therapists also collected data on time spent with participants individually either in person or over the telephone. Because the trial was not originally designed to collect family cost information, we did not have the amount of time teens and parents spent traveling and waiting. On the basis of information from study staff, we estimated that parents brought teens to intervention sessions about 50% of the time, with adolescents providing their own transportation for the remainder. We estimated travel time by using information on participant’s residence and locations of health services. We used information from the HMO to estimate average usual-care appointment and waiting times.³² For nonprotocol services, we used estimates of visit times and transportation costs from local information and from published research when local information was not available.^32- 36

Economic experts have suggested several approaches to valuing study participant or patient time.^27,37- 38 Wages have been widely used as a proxy value for time spent in interventions or lost from work because of illness.²⁷ However, this approach may overvalue patient costs when earnings do not accurately reflect the amount of production lost to society.^37,39- 40 Although methods that incorporate information about community economic conditions (eg, unemployment, replacement costs of workers) might calculate patient costs more accurately,^37- 38 we were not able to use these methods because of data limitations. We also wanted the ability to compare our work with similar studies^{33- 34,41- 42}; therefore, this study used wages to value family members’ time. Because the RCT did not collect information on teen and parent wages, we priced teen and parent time by using national data on hourly wages for teens and parents in the same geographic region.^43- 44 Other studies of the cost-effectiveness of mental health programs have used this approach.⁴⁵

We used the primary clinical outcome data from the RCT (episodes of depression and depression symptom ratings) to create summary measures over time that could be converted into utility-based outcomes. Following a widely used approach to cost-effectiveness of depression treatment,^33- 36,46 we developed measures of depression-free days (DFDs). For our main analysis, we incorporated information about the depression episodes and symptom information from the Center for Epidemiologic Studies Depression Scale score collected at each assessment. We identified depression episodes based on DSM-IV criteria for major depression evaluated at each clinical assessment. We summed across the clinical assessment periods to get the 12-month total days in a full depression episode. We also used data from each assessment to estimate days with elevated symptoms occurring outside of days in a full depression episode. This method estimates elevated symptom days during an interval between 2 assessments or between an assessment and a depression episode if one occurred. Each day in the interval is assigned a value by means of linear interpolation of clinical ratings at the beginning and end of the interval. We assigned weights to each day with elevated depression symptoms that was not in a depression episode. A Center for Epidemiologic Studies Depression Scale score of 21 or higher indicated elevated symptoms; weights increased (from 0.25 to 0.75) with higher scores. We then summed the number of DFDs during the 12-month period. This approach captures a more complete picture of the effects of the intervention, including both elevated symptom days and days in a full depression episode, and similar methods were used in previous work.^33- 36,46

To compare the cost-effectiveness of this intervention with that of others, we transformed the DFDs into quality-adjusted life-years (QALYs) by using utility weights assigned to depression from the literature. Transition from fully symptomatic depression to full remission is associated with a health utility improvement between 0.2 and 0.6.^47- 52 On the basis of previous reports, we used 0.4 for the base-case analysis.^33- 36

We conducted intent-to-treat analyses. We removed 1 significant outlier from the analysis—a study participant with a severe congenital chronic physical illness that led to multiple non–mental health hospitalizations. Clinical effects (DFDs) and cost variables were modeled by means of ordinary least squares regression, controlling for baseline patient differences that could remain after randomization. This method, used in several similar cost-effectiveness studies,^33- 36,46 can more precisely estimate outcomes than simple analysis of variance.

The raw cost data indicated a skewed distribution of health care costs. Following the Briggs and Gray approach,⁵³ we examined the distribution of costs before and after nonparametric bootstrapping and found that the bootstrap estimate of the sampling distribution closely approximated a normal distribution. Thus, we used nonparametric bootstrap methods with a single model. This method avoids the difficulties of transformation and retransformation in traditional 2-part models.^54- 55

Confidence intervals (CIs) for the clinical effects, cost measures, service use measures, and incremental cost-effectiveness ratios were derived by nonparametric bootstrapping methods with 1000 replications by means of the bias-corrected and accelerated method.^56- 58 Adjusted differences between the intervention and usual-care groups were estimated by means of ordinary least-squares regression models with bootstrap interval estimates; all analyses were adjusted for baseline characteristics including age, sex, race, months of health plan enrollment, baseline depression severity, and comorbidity. Hypotheses tests for the clinical and cost outcomes were based on the significance of the group variable in the bootstrapped multiple regression equations.^59- 60

In addition to the base-case analysis, we evaluated several models to examine how sensitive the cost-effectiveness results were to our assumptions. We conducted 1-way sensitivity analyses on the clinical and cost variables, estimating the cost-effectiveness ratio by using each bound of the 95% CI for each variable. We conducted 2 additional sensitivity analyses of clinical effects. First, we used a more conservative method to estimate DFDs, excluding days with elevated symptoms that did not meet full criteria for depression diagnosis. Next, we examined the sensitivity of our cost-effectiveness estimates to the utility weights used when calculating QALYs by using a more conservative utility weight (0.20) from the literature.^47- 51 We also analyzed the sensitivity of our cost estimates. First, we calculated incremental cost-effectiveness without family time and travel costs. We also estimated the incremental cost-effectiveness from the HMO perspective, including only HMO costs. Finally, we examined the sensitivity of our cost-effectiveness estimates to our assumption that parents would attend 50% of teen visits. This sensitivity analysis examined how our incremental cost-effectiveness estimates changed as we varied this assumption between 0% and 100%.

To help evaluate the cost-effectiveness results, we created a cost-effectiveness acceptability curve,^61- 62 which presents the probability that an intervention would be deemed cost-effective at different maximum monetary values for a 1-unit increase in clinical outcome. Specifically, we used the bootstrapped multiple regression results to calculate the proportion of the time that the intervention was cost-effective for potential maximum dollar values, ranging from $0 to $50, that a decision maker might pay for an additional DFD.

Of 123 subsyndromal teens identified as eligible for the prevention trial, 94 agreed to be randomized to either intervention or control groups. The 2 groups did not differ on rates of current and past psychiatric disorder at baseline. The treatment group had more minority participants (17.8% vs 4.1%; P = .03) and slightly higher baseline Child Behavior Checklist depression scores (8.8 vs 6.8; P = .04) but did not differ on any other key measures at baseline. The mean total health care cost for the year before the intervention was higher in the control group, but not significantly so (mean, $1816 vs $1289; P = .59). Table 1 reports selected baseline characteristics of randomized teens; detailed group comparisons are available elsewhere.¹⁸

Beneficial clinical effects of the intervention have been detailed elsewhere.¹⁸ For the incremental cost-effectiveness analysis, we used clinical outcome data from the trial and published utility weights to estimate DFDs and QALYs (Table 2). Intervention participants reported significantly fewer DFDs (P = .001), with an average of 53 fewer depressed days in the year after intake than control participants. This translated into a significant increase in QALYs for the intervention group, with an average increase in QALYs of 0.059 for the intervention group compared with controls.

Given that the clinical effects of the intervention were significantly better than usual care, we next examined the economic impact of the intervention. The average cost per participant of delivering the group CBT intervention was $1632 (Table 3). Identifying at-risk teens and recruiting families to the intervention accounted for about 65% of costs. Recruitment included $11 233 to conduct chart reviews of HMO paper medical records to identify at-risk teens. Outreach costs included time that HMO providers spent attending informational sessions about the program, screening medical records to provide assent to contact families, and reviewing and signing letters to participant families; these activities cost about $15 200. The final outreach steps were creating and sending invitation letters to parents of at-risk youth, calling them on the telephone to screen for the intervention, and conducting a brief in-person intake assessment; these activities cost about $18 780. The costs were about $25 184 for running the CBT groups, which includes teen CBT and parent information group sessions, out-of-group telephone contact with teens, group leader training, and supervision of group leaders. Overhead costs (including space and capital costs) are included in the estimates and were about 28% of the total intervention costs.

Table 4 presents patterns of service use by group and type of service for the 12 months after the intervention. Teens in both groups reported use of services in a variety of sectors outside the HMO, including schools, specialty mental health services, and family counseling. Multiple regression results indicated that the intervention participants used significantly (P≤.05) fewer services in 7 of the 13 categories of service use. Intervention participants used significantly more services in 4 of the 13 categories. Generally, in categories where the control group had significantly more service use, the magnitude of difference was greater, with the control group using as much as 15 times more of a service than teens who received the intervention. Intervention participants went to an average of 9.383 (95% CI, 7.867-11.000) CBT group intervention visits.

For the 12 months after the intervention, the multiple regression results (Table 5) indicated that the intervention participants incurred fewer costs on nonprotocol services (P = .01). On average, the intervention group incurred less cost for all types of nonprotocol services, although only the difference in HMO–other medical expenditures was statistically significant at the .05 level or less. The mean protocol services cost was $1632 (95% CI, $1554-$1714) per intervention participant. Total direct costs, the sum of nonprotocol and protocol services costs, was not significantly different between the groups (P = .83).

Time and waiting costs for the families associated with nonprotocol services were significantly lower in the intervention group (P = .05). Families of intervention participants incurred an average of $265 (95% CI, $232-$299) in intervention-related costs. Total indirect costs, the sum of nonprotocol and protocol time and waiting costs to families, was not significantly different between groups. Including all direct and indirect costs, the average total cost of services for intervention participants was about $610 greater than that for usual care; however, this difference was not statistically significant at the .05 level. Although the increased intervention cost was not significant, health systems would have to increase expenditures initially to provide the program. Detailed information on cost-effectiveness could aid in the decision of whether to provide this intervention vs other investments for health improvement.

In the base-case analysis, including teen and parent time and travel costs, the average incremental cost-effectiveness ratio (ICER) was $10 per DFD (95% CI, −$13 to $52) or $9275 per QALY (95% CI, −$12 148 to $45 641) (Table 6).

Sensitivity analyses indicated that the ICER estimate was sensitive to several factors. One-way sensitivity analyses using the 95% CI for costs and clinical effects indicated that if the intervention cost was at the low end of the 95% CI, the ICER would be negative on average (−$13 per DFD or −$11 854 per QALY). If the intervention cost was at the high end of the 95% CI, the intervention would cost on average more than double our base-case estimates ($23 per DFD or $26 266 per QALY). If the clinical effect was at the weaker tail of the 95% CI, the average ICER would be higher than the base-case estimates ($12 per DFD or $34 518 per QALY). If the clinical effect was at the stronger tail of the 95% CI, the average ICER would be lower ($10 per DFD or $3279 per QALY). Using a more conservative method for calculating the clinical effects (including only days in full depression episodes to calculate DFD), we found an average ICER of $23 per DFD or $19 655 per QALY. Using the more conservative utility weight (0.2), we estimated a mean ICER of $20 171 per QALY. When we excluded family costs from the calculation of the base case, we found an average ICER of $9 per DFD or $8419 per QALY. With only HMO-incurred costs, we found an average ICER of $18 per DFD or $16 178 per QALY. Finally, when we examined our assumption about parents’ attendance, the average ICER ranged from $9 per DFD or $8176 per QALY if parents attended no visits, to $12 per DFD or $10 375 per QALY if parents attended all visits.

A cost-effectiveness acceptability curve for the base-case analysis (Figure) shows the intervention in relation to different amounts a decision maker might be willing to pay to increase the number of DFDs in a population. For instance, if a decision maker is willing to pay $20, the probability of CBT being cost-effective is about 75%.

We found that a group CBT intervention with at-risk teens led to more DFDs and cost an average of $1632 per participant to deliver; this cost was partially offset by a statistically significant cost offset in general medical costs, and a close-to-significant cost offset to other sectors (outside the HMO). We need to understand whether the costs to provide the intervention are worth its benefits.

Some commonly cited guidelines^63- 65 indicate that if a new intervention is more effective than existing ones and costs less than $20 000, $50 000, or $100 000 per QALY, it should be adopted. Our base-case analysis ($9275 per QALY) indicated that the intervention is cost-effective on average by any of these standards. Most of the sensitivity analyses also indicate that the intervention is cost-effective on average by any of these standards. In all cases, our base-case and sensitivity analyses indicate that the intervention is cost-effective on average with the criterion of $50 000 per QALY or less.

We could also evaluate this intervention by comparing its cost-effectiveness with that of similar interventions. However, to our knowledge, no other studies have examined the cost-effectiveness of interventions to prevent or treat depression in at-risk teens. We therefore have to compare the cost-effectiveness of this intervention with the cost-effectiveness of other depression treatments for adults. Comparing interpersonal psychotherapy with usual care for depression treatment, Lave and colleagues³³ reported average cost-effectiveness for 2 types of depression treatment at $13 and $18 per DFD for direct costs only and $15 and $25 per DFD for costs including patient time and transportation. Simon and colleagues⁴² reported average cost-effectiveness for systematic depression treatment for high utilizers of general medical care of about $41 per DFD for direct costs and $52 per DFD including patient time cost. Valenstein and colleagues⁴¹ reported average cost-effectiveness of $32 053 per QALY for one-time screening for depression in primary care, $50 988 per QALY for screening every 5 years, and $192 444 per QALY for annual screening. A study comparing cost-effectiveness of quality improvement programs for depression reported average cost per QALY of $9478 to $30 663 (range depending on utility weight selected).³⁴ Another study comparing cost-effectiveness of collaborative care for persistent depression reported cost per DFD of between $21 and $35 depending on the types of costs included.³⁵ Finally, a study comparing cost-effectiveness of collaborative care for depression in a veteran population reported cost per DFD of between $2 and $33 depending on the types of costs included.⁴⁶ Our results are within the same range as these results.

This study has several limitations. We examined the effects and costs of this group CBT intervention in a single HMO, with a relatively small group of teens, and therefore we cannot be certain our results would be generalizable to other locations or health care systems. We evaluated cost-effectiveness during 12 months after the intervention. Thus, we cannot provide information on the long-term impact of the intervention. The RCT was not designed to collect complete information on patient costs, so we relied on a combination of information from the trial and published literature to estimate patient costs. This study was designed to use systems similar to those used in cancer screening and other prevention services that use health plan records to identify at-risk groups and then conduct outreach to bring them in for services. We found that identification and outreach were very costly. However, at the time of this study, we had to rely on paper chart review to identify parents of at-risk teens. This health plan has since adopted comprehensive electronic medical records. The identification process would likely be significantly less expensive with the use of electronic methods to identify teens.

To estimate QALYs, we relied on utility weights assigned to depression from published literature. These utility weights were estimated for adults with depression; however, utility weights for teens with depression might be different.⁶⁶ Epidemiologic information on depression indicates that once a teen has had 1 episode of depression, that teen may be at risk for a number of adverse outcomes.^4- 6 Therefore, teens, parents, or communities might value reducing depression in teens more highly than in adults because of the possibility of preventing these adverse consequences and increasing the total lifetime benefit of improved functioning and productivity. In addition, there is ongoing debate about whether QALYs adequately capture mental health outcomes.⁶⁶ We know of no studies that have collected data for estimating the value of mental health treatments for children or teens.

Although we attempted to implement the intervention in a manner that would represent the “real world,” it is likely that our supervision and training standards exceeded usual practice standards. These standards may have led to better outcomes and greater costs than would be experienced in a typical health plan.

Our findings suggest that health plans, and other integrated systems of health care, can intervene to prevent depression in at-risk teens for a cost similar to or more attractive than that of other generally accepted medical interventions. However, these promising results need to be verified by examining the clinical effectiveness and associated costs of this intervention in a larger and more diverse population. Members of this research team and collaborators from 3 other sites are currently replicating this intervention with a larger sample size (planned N = 320) in 4 sites in the United States (Oregon, Tennessee, Pennsylvania, and Massachusetts).

Our results indicate that it is possible in a real-world setting to prevent depression in at-risk teens in a cost-effective manner. At this time, few managed care organizations provide coverage for any type of mental health prevention services.⁶⁷ Previous studies⁶⁸ indicate that managed care organizations and other insurers are often reluctant to adopt or cover new services that might attract persons at risk for mental health problems to their system because of concerns about possibly increasing costs. Changes in the priorities of health systems, changes in the insurance system, or public policy initiatives to provide incentives for implementing depression prevention programs would probably be necessary for this intervention to be adopted in real-world settings.

Correspondence: Frances L. Lynch, PhD, MSPH, Kaiser Permanente Center for Health Research, 3800 N Interstate Ave, Portland, OR 97227-1098 (frances.lynch@kp.org).

Submitted for Publication: March 8, 2004; final revision received March 31, 2005; accepted April 29, 2005.

Funding/Support: This study was supported by grant 1R01-MH51318-01A1 from the National Institute of Mental Health, Bethesda, Md (Dr Clarke).

Acknowledgment: We thank Kimberly Hoagwood, PhD, Lauren Haworth, and Jennifer Coury, MA, for their assistance on this project.