Controlling for Site and Baseline Differences
There was no significant association between the time period of the treatment and any demographic characteristics (). The V group, compared with the S group, had higher SCARED total score (32.2 ± 14.9 versus 28.4 ± 16.2, t = −2.0, df = 314, p = .05). In particular, the V group had higher scores on the following subscales: generalized anxiety disorder (9.9 ± 4.1 versus 8.8 ± 5.0, t =−2.1, df = 225, p = .04), separation anxiety disorder (4.2 ± 3.6 versus 3.1 ± 3.0, t =−2.6, df = 160.8, p = .01), and social anxiety disorder (7.3 ± 4.2 versus 6.2 ± 4.2, t = −2.1, df = 313, p = .04). There was no difference between the two groups with respect to any of the other clinical characteristics. Moreover, time of treatment was not associated with either that parental psychopathology (depression and anxiety) or family environment. In addition, there was no difference between the two groups with respect to being switched to an SSRI or venlafaxine (p = .47), or the type of SSRI they were switched to (p = .80). However, there was a significant difference with respect to number of pharmacotherapy visits (median, nine for both groups, Mann-Whitney z =−2.5, p = .01; mean ± SD, S = 7.7 ± 2.3 versus V = 8.3 ± 1.8) but not number of CBT sessions (p = .30). Controlling for demographic characteristics (age, gender, race, and family income), site, treatment assignment, SCARED total scores, and number of pharmacotherapy visits, the V group was 1.8 times (95% CI = 1.03-3.31, p = .04) as likely to show adequate response as the S group. Upon adding interaction terms to the logistic regression model, timing of treatment had no significant interaction with assignment to combination therapy (p = .21), type of antidepressant the participants were switched to (p = .11), or SCARED scores (p = .81).
Demographic and Clinical Characteristics of Study Participants, by Season
To examine the possible effect of seasonal affective disorders, which might interact with latitude, sites were split into northern (Pittsburgh, Brown, and Portland) and southern (Dallas, Los Angeles, and Galveston) sites. Controlling for site, SCARED, and number of pharmacotherapy visits, the V group was 1.7 times (95% CI = 1.03-2.8, p = .04) as likely to show adequate response as the S group. Moreover, there was no significant interaction between timing of treatment and site in predicting outcome (p = .22), indicating that timing of treatment had same effect in the northern and southern sites. In addition, participants were screened for seasonal affective disorder at baseline and subsequent time points, and only two participants were found to have this disorder.
School problems at week 0 were not associated with either response at week 12 (responders: 9.1 ± 4.6 versus non responders: 9.0 ± 4.3, t = −0.2, df = 276, p = .81) or timing of treatment (S: 9.0 ± 4.5 versus V: 9.4 ± 4.4, t =−0.7, df = 264, p = .50). However, responders, as compared with nonresponders, had significantly fewer school problems at week 12 (4.0 ± 3.2 versus 7.2 ± 4.7, t = 5.8, df = 204.1, p < .001). Also, V participants reported fewer school problems than S participants (4.4 ± 3.9 versus 6.1 ± 4.4, t = 2.5, df = 224, p = .01).
In a multivariable logistic regression, adequate response at week 12 was associated with lower school problem scores at week 12 (SAS-SR school difficulty scores <5 versus scores ≥5: OR = 3.2, 95% CI = 1.7-5.8, p < .001) and being assigned to CBT (CBT versus no CBT: OR = 1.8, 95% CI = 1.03-3.3, p = .04), but not to type of pharmacotherapy (SSRI versus venlafaxine: OR = 0.9, 95% CI = 0.5-1.7, p = .83) or timing of treatment (V versus S: OR = 1.5, 95% CI = 0.7-2.9, p = .29). There was no significant interaction between school problems and being assigned to CBT (p = .63) or type of pharmacotherapy (p = .81); however, there was significant interaction between school problems and timing of treatment (p = .01). Upon adding the interaction term between school problems and timing of treatment to the model, the following were significantly associated with adequate response: being assigned to CBT (OR = 1.9, 95% CI = 1.1-3.5, p = .03), lower school problem scores (OR = 4.9, 95% CI = 2.4-10.2, p < .001), V group membership (OR = 3.6, 95% CI = 1.3-10.0, p = .01), and the interaction term of timing by school problems at week 12 (OR = 0.2, 95% CI = 0.05-0.7, p = .01).
To understand this interaction, we examined the relationship between school difficulties in response in both the S and V groups. Among the S youth, those with lower school difficulty scores (scores <5) were more likely to show response at week 12 as compared with participants with higher school difficulty scores, i.e., SAS-SR school difficulty scores ≥5 (68.0% versus 30.5%, χ2 = 23.6, df = 1, p < .001; ). On the other hand, among the V youth, who ended their treatment while out of school, there was no significant association between school problem scores and response to treatment (SAS-SR school difficulty scores <5: 61.8% versus scores ≥5: 63.6%, χ2 = 0.02, df = 1, p = .89; ). In other words, although there was no significant association between timing of treatment and response among participants with low school difficulty (S versus V, 68% versus 61.8%; χ2 = 0.2, df = 1, p = .52), among participants with higher school difficulty (scores ≥5), ending treatment, while in school was associated with lower rates of response as compared with ending treatment while being out of school (S versus V, 30.5% versus 63.3%, χ2 = 8.4, df = 1, p = .004).
Response rate by treatment timing stratified by school problems at week 12. School problems assessed with six items from the Social Adjustment Scale—Self Report.
Sensitivity analyses were conducted to determine whether SAS-SR missing values could have affected our results. The V group had a higher rate of adequate response as compared with the S group regardless of whether the week 12 SAS-SR scores were missing (54.8% versus 32.7%, χ2 = 4.6, df = 1, p = .03) or not missing (62.5% versus 47.1%, χ2 = 4.0, df = 1, p = .047).