Cost-Effectiveness of Preventing Depression Among At-Risk Youths: Postintervention and 2-Year Follow-Up

Participants were 316 youths, ages 13–17 (mean±SD=14.8±1.4). Eligibility criteria were having current subsyndromal depressive symptoms (≥20 on the Center for Epidemiological Studies-Depression scale (18), a prior depressive episode in full remission for at least 2 months, or both. All youths had a parent with current or prior depression. Eighty percent of youths had at least one prior depressive episode; the mean number of prior episodes did not differ significantly by group (CBP=1.4±1.0; usual care=1.3±0.8). Youths were excluded if they currently were taking a therapeutic dosage of an antidepressant, they had received more than eight sessions of CBT for depression, or they or their biological parent had a diagnosis of bipolar I disorder or schizophrenia. Participants were recruited at Vanderbilt University, Nashville, Tennessee; University of Pittsburgh, Pittsburgh; Kaiser Permanente Northwest, Portland, Oregon; and Judge Baker Children’s Center and Children’s Hospital, Boston. Institutional review boards at each site approved the study. Parents and youths provided written informed consent and assent, respectively. Study recruitment began in August 2003 and ran through February 2006.

Youths were randomly assigned to CBP or usual care. CBP consisted of eight weekly, 90-minute sessions and six monthly continuation sessions led by a mental health professional. CBP taught cognitive restructuring and problem-solving skills and additional skills during continuation sessions. All youths in CBP or usual care could seek mental health services postrandomization. Thus youths were assigned to either usual care alone or usual care plus CBP (CBP, hereafter). Additional details about the study design are reported elsewhere (15).

At baseline, independent evaluators assessed youths’ current and lifetime depressive disorders and other diagnoses (for example, substance use disorders) with the Schedule for Affective Disorders and Schizophrenia for School-Age Children (19). Current depressive symptoms and severity were evaluated with the 17-item Children’s Depression Rating Scale–Revised (CDRS-R) (20) and the Longitudinal Interval Follow-up Evaluation (LIFE) (21) at 3-, 9-, 21-, and 33-month follow-ups. The LIFE provides a continuous assessment of symptoms and impairment since the previous assessment, rated on a 6-point Depression Symptom Rating (DSR) scale. The primary clinical outcome was a probable or definite episode of depression (that is, DSR ≥4) for at least 2 weeks. Parents’ current and past depressive episodes were evaluated with the Structured Clinical Interview for DSM-IV (22).

Following other CEAs of depression interventions, we created a measure of depression-free days (DFDs) (11, 23–25) by using CDRS-R scores from all follow-up points to categorize DFDs, days with some depression but not meeting full criteria, and days in a depressive episode. 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 interval; days with fewer depression symptoms were weighted less. DFDs were the total number of days in the interval minus days with depressive symptoms. This approach captured the effects of the intervention, including both days with some depressive symptoms and days in a full depressive episode (24, 25). DFDs based on weekly DSR scores yielded similar results. [Additional details are provided in an online supplement to this article.]

We transformed DFDs into quality-adjusted life years (QALYs) with utility weights derived from empirical studies using large populations that allow stable estimation of the impact of depression on quality of life (24, 26). Specifically, each day in the interval received a QALY weight. A DFD with no depressive symptoms was assigned a utility weight of 1.0 (full health). Days in a full depressive episode were estimated to have a lower weight, reflecting a decrease in quality of life during a depressive episode. Depression has been associated with a decrease in quality of life of 0.2–0.6 (24, 26). We used 0.4 as the decrease in utility weight for the base-case analysis, such that a day in a full depressive episode would be assigned a weight of 0.6. We also explored a range of weights in sensitivity analyses.

Accounting records provided costs for payroll, facilities, and overhead. CBP leaders estimated time to complete tasks and use of equipment, space, and supplies. We included costs of CBP sessions, time leaders spent with individuals by phone or in person, supervision, training, and materials. We excluded research-specific costs (for example, randomization).

We administered the Child and Adolescent Services Assessment (CASA) (27) to gather data on services received outside the study by youths in both CBP and usual care. The CASA captures a wide range of services, including inpatient mental health, counseling, medication management, crisis services, and mental health counseling in nonhealth settings (for example, schools). We applied unit costs developed from the Truven Health MarketScan Commercial Claims Database for medical services; for social services, we used unit costs developed for other studies (11, 25). At baseline, youths and their parents separately reported on services that the youth received in the previous 3 months. At each follow-up, we assessed youths’ service use since the previous evaluation. Table 1 lists the mental health and other services that youths in CBP or usual care used, the percentage of youths receiving each service, and the frequency (mean) of services used.

Following recommended guidelines (28), we estimated family costs for the time parents spent in taking youths to services. To value parent time, we created profiles based on published research on parent time spent in services, travel to services, and waiting (11, 25).

We estimated incremental cost-effectiveness ratios (ICERs) from a health and public services perspective for 9- and 33-month postintake evaluations, computing the ratio in each instance as the mean cost difference between CBP and usual care divided by the mean difference in clinical outcome (DFD or QALY). In the base-case analysis, outcomes and costs beyond 12 months were discounted at a rate of 3%. Costs were adjusted for inflation. We used the net benefit regression method (29) with ordinary least-squares regression to examine cost-effectiveness; we confirmed the robustness of parametric tests by using nonparametric bootstrapping (30) with a single model with 1,000 replications and the bias-corrected and accelerated method (31). We performed hypothesis tests and estimated adjusted differences between groups by using ordinary least-squares regression models with bootstrap interval estimates. We adjusted all analyses for baseline CDRS-R scores, baseline costs, age, sex, race-ethnicity, and socioeconomic status.

Following recent work on highly influential observations (32), we examined all outliers, considering whether any observation met the strictest criteria for possible removal. Criteria for removal were observations above the 99th percentile for total costs and whose removal from analyses would influence parameter estimates by a factor of 1.25; both criteria well exceeded the suggested threshold for removal of .15 (32). On the basis of these criteria, in each analysis a highly influential observation was removed from the base-case analyses. We also conducted subanalyses with the outlier included to test for robustness.

To represent uncertainty in estimates, we first used bootstrap observations to estimate a 95% confidence interval (CI) around the average ICERs. We then created a scatterplot of bootstrapped cost-and-effect pairs to construct a cost-effectiveness plane (33) divided into four quadrants, centered at zero (Figure 1). If the majority of the cost-effect observations are located in the SE quadrant (less costly/more effective), then the CBP intervention is “dominant” over the alternative. Similarly, if most observations are in the NW quadrant (more costly/less effective), then the intervention is “dominated” by the alternative.

We conducted sensitivity analyses to examine the robustness of the results. We examined how results might change with different utility weights and with the inclusion of the highly influential observation. On the basis of the literature, we ran a model limited to outpatient costs because utilization of higher levels of care (for example, inpatient) is rare and likely not affected by a short-term intervention (34).

Of the 316 youths in the sample, approximately 85% (N=269) completed the 33-month assessment, 69% (N=218) completed all evaluations, and 98% (N=310) completed at least one follow-up assessment. Sample retention did not differ by group or site (15), demographic characteristics, entry characteristics, or depression measures (15). We imputed missing data by using Stata multiple imputation with chained equations (35, 36). We included baseline demographic characteristics and all nonmissing values of costs or outcomes at all time points in the imputation models. We created five imputation data sets and combined estimates so that standard errors reflected the variability introduced by the imputation process.

The percentage of youths who experienced depressive episodes was smaller in the CBP group than in usual care over 9 months (21% versus 33%, p=0.03) (15) and over 33 months (37% versus 48%, p=0.04) (17). Compared with youths in usual care, at 9 months youths in CBP had 12 more DFDs (p=0.02) and .018 more QALYs (p=0.02). Two years later (that is, at 33 months), youths in CBP had 40 additional DFDs (p=0.05) and .045 more QALYs (p=0.05), compared with those in usual care. DFDs increased over time for both groups (p<0.001), but the proportion of DFDs for youths in CBP was significantly greater at each follow-up point.

Table 1 presents descriptive estimates of services not related to the study, by treatment group. The pattern of service use was similar at both 9 and 33 months. Some differences were noted in the group means for specific services. On average (mean±SD), the cost of the CBP intervention, separate from use of nonstudy services, was $591±$286.

Table 2 presents the ICERs, a measure of the estimated cost per DFD and QALY. At 9 months, the estimated cost was $35 per DFD and $24,558 per QALY. At 33 months, the cost was $15 per DFD and $12,787 per QALY.

Figure 1 presents the cost-effectiveness planes for the base-case analysis and the subgroup analyses. Each point in the scatterplots represents a cost-clinical outcome pair for a single bootstrap replication. Most observations indicated that CBP had better outcomes and higher costs, 98% of observations at 9 months and 77% of observations at 33 months. In addition, at 33 months, 23% of observations had better outcomes and lower costs.

Figure 2 presents the cost-effectiveness acceptability curves for QALYs at 9 and 33 months for the base-case analysis and subgroup analyses. The curve represents the probability that CBP was cost-effective compared with usual care over a range of dollar amounts that a decision maker might pay for an additional outcome (for example, a QALY). For example, if a decision maker can pay $50,000 per QALY, the probability that CBP is cost-effective compared with usual care at 9 months is 85%, and at 33 months it is 89%.

Table 2 also presents data on the subgroup analyses. Elsewhere, we reported that time to onset of depressive episodes varied depending on whether the parent was currently depressed at baseline (15). Therefore, we examined CEAs by subgroup following the net benefit regression method (29). CBP had a higher net benefit for youths whose parents were not depressed at baseline (Table 2). At 9 months, youths whose parents were not depressed at baseline had estimated ICERs of $15 per DFD or $10,498 per QALY. At 33 months, estimated ICERs were $15 per DFD, or $13,620 per QALY. For youths whose parents were in a depressive episode at baseline, CBP had higher cost with no more effective outcomes than usual care at both 9 and 33 months.

Sensitivity analyses and secondary analyses examined whether CEA estimates were different when various parameters were used [see online supplement]. Results indicated some variation in cost per QALY, with a range of $10,229 to $53,383 per QALY.