Treatment Features Associated with Youth Cognitive Behavioral Therapy Follow-Up Effects for Internalizing Disorders: A Meta-Analysis

Abstract

Our aim was to investigate whether four treatment features (i.e., the inclusion of parental involvement, goal-setting strategies, maintenance/relapse prevention sessions, the addition of booster sessions) were associated with posttreatment and follow-up effect size of youth cognitive behavioral therapies (yCBTs) for anxiety, depression, posttraumatic stress disorder, and obsessive-compulsive disorder in age groups spanning young children to adolescents. We conducted a random-effects meta-analysis of 106 yCBTs tested in 76 randomized clinical trials from the PracticeWise Database to examine average effects of yCBTs posttreatment and at a later follow-up assessment. We coded the use of parental involvement, goal setting, booster sessions, and maintenance/relapse prevention in each yCBT and conducted random-effects meta-regression analyses to investigate whether these treatment features were associated with yCBT effects at posttreatment as well as at follow-up. Overall, yCBTs produced large pre- to posttreatment effects (d = 1.05), 95% confidence interval [0.94, 1.15], and larger pre- to follow-up effects (d = 1.29), 95% confidence interval [1.18, 1.40]. Metaregression results indicated that parental involvement was significantly associated with larger pre- to posttreatment effect sizes as well as pre- to follow-up effect sizes. Booster sessions, goal setting, and maintenance/relapse prevention were not significantly related to effect sizes at posttreatment or follow-up. Parental involvement may be helpful for maximizing long-term effectiveness of yCBT. Future studies should investigate for whom and under what conditions inclusion of yCBT treatment features is related to the durability of treatment gains.

Declaration of interest

Coding performed under an agreement with, and does not necessarily represent the views of, PracticeWise, LLC.

Notes

¹ We used the d_average as opposed to the d_{repeatedmeasure} (a more conservative metric) because calculation of this metric requires the correlation coefficient of pretreatment and posttreatment means, which most studies in our sample do not report. We decided that the costs of dropping studies that did not report this coefficient outweighed the costs of using d_average. Alternatively, we could have substituted a standard r (e.g., r = .70) for all studies that did not report this statistic. However, we decided that using an arbitrary value for the calculation of most yCBT’s effect sizes would not provide additive benefit, especially because it might advantage or disadvantage those few studies that did report a r value for their pre–post measure correlations.

² This arrangement of the steps does not provide an estimate of the unique variance attributable to treatment features. As an additional step, we reran these models flipping Steps 2 and 3, setting treatment features as the last step. Although overall the results appear identical, we found that there was less unique variance explained by treatment features: pre to posttreatment effects, ΔR² analog = 0.06, F(16, 89) = 3.93, p < .001; pre- to follow-up effects, ΔR² analog = 0.05, F(21, 84) = 2.33, p < .003.

Additional information

Funding

This research was supported in part by the National Institute of Mental Health of the National Institutes of Health (Sun, Williamson, T32-MH015750; Sun, F31-MH111199). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.