Depression response by sleep medication group
The use of trazodone was associated with a much lower rate of depression response compared with those who were treated with no sleep medication (15.2% vs. 50.4%, χ2=14.7, df=1, p<0.01). When participants were classified by CBT assignment, trazodone was associated with lower rates of response among those assigned to combination therapy (11.8% vs. 59.0%, χ2=13.6, df=1, p<0.01), with a non-significant trend in the same direction for those assigned to monotherapy (18.8% vs. 41.6%, χ2=3.1, df=1, p=0.08). Trazodone use was associated with lower rates of response in those assigned to either an SSRI (11.1% vs. 51.5%, χ2=10.4, df=1, p=0.001), or venlafaxine (20.0% vs. 49.3%, χ2=4.7, df=1, p=0.03). Within the SSRI group, none of those cotreated with trazodone and either fluoxetine (0% vs. 56.1%, Fisher's exact test [FET], p=0.001) or paroxetine (0% vs. 41.9%, FET, p=0.27) responded, whereas 40% of those cotreated with citalopram responded (vs. 56% of those who received citalopram but no sleep medication, FET, p=0.64).
In contrast, those treated with other, nontrazodone sleep medication had rates of response at least comparable to those who did not receive any sleep medication (60.0% vs. 50.4%, χ2=0.85, df=1, p=0.36). This finding was true in those who received combination therapy (70.0% vs. 59.0%, FET, p=0.74), mono-therapy (53.3% vs. 41.6%, χ2=0.76, df=1, p=0.38), an SSRI (50.0% vs. 51.5%, χ2=0.01, df=1, p=0.91), or venlafaxine (77.8% vs. 49.3%, FET, p=0.17).
Characteristics of sleep medication groups
The three sleep medication groups showed an overall difference for the baseline insomnia measure (F2,321=3.9, p=0.02, ), but post-hoc pair-wise contrasts escaped statistical significance.
Comparison of the Three Sleep-Medication Groups on Baseline Measures and Characteristics*
Compared with those receiving no sleep medications, patients who received trazodone were 4.8 times (95% CI: 1.4–16.0; p
=0.01) more likely to be female (trazodone can contribute to priapism—painful, chronic erections—and therefore should not typically be prescribed to males), and 5.2 times as likely to be from 2 of the 6 study sites (95% CI: 2.3–11.5; p
=0.01). They also were also more likely than patients with no sleep medication to have had a history of nonsuicidal self-injury (NSSI; 75.8% vs. 32.6%, χ2
=23.4, df=1, p
<0.001), higher mean SIQ-Jr scores, (52.4±22.8 vs. 40.9±22.2, t
2.8, df=301, p
=0.006), and were more likely to have had history of physical or sexual abuse (42.4% vs. 23.0%, χ2
=5.9, df=1, p
=0.015). There was no significant association between initial treatment assignment in TORDIA and use of any specific sleep medication (χ2
=1.03, df=2, p
=0.60). There was no significant association between the 12-week treatment completion rate and the type of sleep medication (no medication, 68.8%; trazodone, 66.7%; other sleep medications; 76.0%; χ2
=0.66, df=2, p
Participants receiving sleep medications other than trazodone were more likely to have a history of suicide attempt compared with youth receiving no sleep medication (44.0% vs. 20.4%, χ2=7.4, df=1, p=0.007). When compared with those on trazodone, patients on other sleep medications were more likely to be male (36.0% vs. 9.1%, χ2=6.3, df=1, p=0.01).
Family conflict, as reported by the parent, was significantly associated with receipt of sleep medications (F2,315=2.9, p=0.05); however, none of the pair-wise comparisons was statistically significant.
Effect of sleep medication on outcome after controlling for baseline differences
Because there were baseline differences associated with the use of trazodone that might have influenced outcome, we conducted a logistic regression analysis of depression outcomes by each of the sleep medication groups controlling for baseline insomnia, self-reported depression, suicidal ideation, history of suicide attempt and NSSI, history of physical or sexual abuse, and family conflict, in addition to age and gender, site, and medication and CBT treatment assignment. Even after adjustment, those who used trazodone were six times less likely to have responded (adjusted odds ratio [OR]=0.16, 95% CI: 0.05–0.50, p=0.001). Baseline sleep symptoms were not significantly associated with response (OR=1.09, 95% CI: 0.95–1.26, p=0.22). Moreover, there was no significant interaction between use of either category of sleep medication and baseline insomnia score (p=0.62), assignment to CBT (p=0.57), or type of antidepressant (p=0.31). Even after excluding the seven participants who received other sleep medications in addition to trazodone, trazodone was still significantly associated with a lower rate of adequate response compared with the no sleep medication group (adjusted OR=0.07, 95% CI: 0.01–0.33, p=0.001).
On average, trazodone was used for 5.1 weeks (SD=4.4, median=3.0, range: 1–12 weeks). The duration of trazodone treatment was not significantly associated with lower depression response rates (p=0.55). In addition, the maximum trazodone dose used was not associated with lower depression response rates (p=0.20). Further, using logistic regression, the week during which trazodone was started was not significantly associated with weekly depression response ratings (p=0.40), failing to support the alternative explanation that trazodone was most often initiated as youth were deteriorating.
Side effects and adverse events by sleep medications
The type of sleep medications was significantly associated with gastrointestinal (GI) symptoms as measured by SEFCA (p=0.002). There were also differences by sleep medication group with respect to the frequency of NSSI (p=0.05), self-harm (defined as suicidal ideation, suicide attempt, or NSSI; p=0.003), irritability (p=0.005), and GI (p=0.02) treatment-emergent adverse events. Post-hoc comparisons showed that, compared with those receiving no sleep medications, patients who received trazodone had higher rates of GI (15.2% vs. 3.6%, FET, p=0.01), self-harm (42.4% vs. 18.5%, χ2=10.1, df=1, p<0.001), and irritability-related adverse effects (18.2% vs. 3.6%, FET, p=0.003). There were no differences by trazodone exposure with respect to other side effects or adverse events, and no association between other sleep medications and any side effects or adverse events. Controlling for baseline suicidal ideation, history of suicide attempt and NSSI, the above-noted baseline differences, demographics, and treatment assignment, trazodone was still associated with increased risk of self harm (OR=3.0, 95% CI: 1.1–7.9, p=0.03), but other sleep medications were not (OR=0.5, 95% CI: 0.1–2.6, p=0.53).
Post-hoc, within the venlafaxine group, co-treatment with trazodone was positively associated with self-harm events (46.7% vs. 19.7%, χ2=5.69, df=1, p=0.017) and irritability (26.7% vs. 2.8%, FET, p=0.003). Similarly, within the fluoxetine group, those cotreated with trazodone, were more likely to experience adverse events, (90% vs. 47%, FET, p=0.015), including self-harm (50% vs. 18.2%, FET, p=0.039), and GI symptoms (30% vs. 1.5%, FET, p=0.006) ().
Comparison of the Three Sleep-Medication Groups with Respect to Side Effects and Adverse Events*
Use of sleep medication and change in sleep symptoms
The improvement in sleep symptoms was evaluated as a function of sleep medication group. Although sleep symptoms significantly improved over time (p<0.001), there was no sleep medication group by time interaction (p=0.88). That is, the rate of improvement among those receiving either trazodone or the other sleep medication was parallel to those on no sleep medication ().
Insomnia scores at baseline and weeks 12 for the three insomnia medication groups (error bar represent±2 standard deviations). Shaded=baseline; white=12 weeks.