The U.S. Food and Drug Administration (FDA) issued a public health advisory on Oct. 15, 2004, ordering manufacturers of all antidepressant drugs to adopt a boxed warning to alert doctors of a greater risk of suicidal thinking and behavior in children and adolescents being treated with these medications (1).
The aim of the present study was to place the concerns raised by the FDA and other regulatory agencies (2) in context. We report on the incidence and predictors of emergent suicidality in a randomized psychotherapy treatment trial of adolescent depression in which no subjects received pharmacotherapy but were otherwise comparable to participants in previous clinical trials of depression in adolescents used to assess risk of emergent suicidality (1, 2).
Subjects were participants in a clinical trial of psychotherapy for adolescent depression. All had normal intelligence, were between 13 and 18 years of age, had a DSM-III-R diagnosis of major depressive disorder, and had an intake Beck Depression Inventory (3) score of 13 or higher 13 (4). Subjects with ongoing physical or sexual abuse, psychosis, or bipolar, obsessive-compulsive, eating, or substance abuse disorder were excluded. Of 122 eligible participants, 107 were recruited.
The subjects’ median socioeconomic score was 40 (class IV) (4), and 81 (75.7%) were girls. Approximately 22%, 32%, and 21% of the patients had comorbid dysthymia, anxiety, and disruptive disorders, respectively. About one-third of the patients were recruited through advertisement, and the rest were recruited from a child psychiatry outpatient clinic in a university setting. Once the patient and family had given informed consent, they were randomly assigned to receive cognitive behavior therapy, systemic behavioral family therapy, or nondirective supportive therapy. Treatment consisted of two phases: 1) an active phase, in which patients received 12 to 16 sessions provided in 12 to 16 weeks, and 2) a booster phase, in which patients received two to four sessions in as many months. In this report, we focus exclusively on the active phase. Four of the randomly assigned participants never came for treatment and were not included in any of the present analyses.
On the basis of the assumption that interview, rather than self-report, is the best way to assess suicidality, we focus on the 88 participants who at the intake interview denied clinically significant suicidality during the week before evaluation. This interview used the Schedule for Affective Disorders and Schizophrenia for School-Age Children (K-SADS) (5). Clinically significant suicidality was defined as a rating of 4 or higher on the K-SADS suicide ideation item, corresponding to suicidal ideation with a plan or an actual suicide attempt.
Patients were classified regarding history of suicidality before the week preceding intake according to all available interview data from the K-SADS and the Suicide Circumstances Schedule (6), a semistructured interview for the assessment of precipitants, suicidal intent, and lethality of a suicidal episode. Twenty-two patients were classified as nonsuicidal; they had no past history of suicidal ideation or behavior. Sixty-six patients had past but not current suicidality, and 19 of these patients (28.8%) reported a history of suicide attempt. Self-reported thoughts of killing oneself in the week before intake were measured by using item 9 of the Beck Depression Inventory (3), dichotomized as present or not present; information on this item was missing for two subjects.
The primary outcome variable was the emergence of suicidality during the 12 to 16 weeks of active treatment. Emergent suicidality, coded on a 5-point scale (0=no suicidality, 1=ideation without plan, 2=ideation with plan, 3=suicide gesture [i.e., episode in which a subject had suicidal intent and means at hand but did not attempt suicide], 4=suicide attempt), was assessed at each session by the treating therapist.
Potential demographic predictors of emergent suicidality included age, race, sex, referral source (clinical versus advertisement), family constellation, and socioeconomic status. Intake clinical predictors included interview and self-rated suicidality, comorbid disorders, age at onset, duration and severity of depressive episode, functional status, level of cognitive distortion, hopelessness, family discord, and parental depressive symptoms (4).
Factors that distinguished the groups were examined by using standard parametric and nonparametric statistics. Time to emergent suicidality was compared between groups with Kaplan-Meier survival analyses, and differences were tested by using the log-rank statistic. The survival time began with the first treatment session following the intake assessment and continued until a suicidal event occurred, withdrawal from the study due to either dropping out or entering open treatment, or the end of the trial. Alpha was set at 0.05 (two-tailed) for all comparisons.
Of 88 depressed subjects who denied current suicidality on interview, 11 (12.5%) developed suicidality during treatment (10 had ideation without a plan; one made a suicide attempt). With respect to timing of emergent suicidality, eight (72.7%) of 11 episodes occurred within 3 weeks of beginning treatment. There were no completed suicides during the study period.
The majority (N=53 [61.6%]) of the 86 subjects classified as nonsuicidal endorsed item 9 of the Beck Depression Inventory, reporting that they had thoughts of killing themselves. All 10 patients classified as nonsuicidal who developed emergent suicidality during active treatment endorsed this item, reporting suicidal thoughts at intake, compared with 43 (56.6%) of the 76 nonsuicidal patients who did not become suicidal during treatment (p=0.01, Fisher’s exact test). Of the 10 cases of emergent suicidality, none was identified by interview alone, four were identified by self-report alone, and six were identified by both interview and self-report. There was no significant difference in the rate of emergent suicidality between patients with (N=66) and without (N=22) a past history of suicidality at intake (nine [13.6%] versus two [9.1%]) (p=0.72, Fisher’s exact test). Among patients with a past history of suicidality, there was a nonsignificant trend for a previous suicide attempt to predict emergent suicidality, both in terms of rate (five [26.3%] of 19 versus four [8.5%] of 47) (p=0.11, Fisher’s exact test) and in terms of time to emergent suicidality (log-rank χ2=3.75, df=1, p<0.06).
Subjects with and without emergent suicidality were similar with regard to treatment assignment, age, sex, race, socioeconomic status, and referral by advertisement (7). With respect to clinical predictors, patients who developed emergent suicidality had lower overall scores on the Children’s Negative Cognitive Errors Questionnaire (7) (mean=67.1, SD=18.3, versus mean=84.0, SD=18.5) (t=2.71, df=85, p=0.008), reflecting greater cognitive distortion, and higher Beck Depression Inventory scores (not including item 9) (mean=30.5, SD=8.9, versus mean=21.8, SD=6.9) (t=3.62, df=84, p=0.001). However, the only significant predictor of emergent suicidality in a multivariate Cox regression model was self-reported suicidality on Beck Depression Inventory item 9.
In this clinical trial, which enrolled subjects similar to those enrolled in pharmacotherapy clinical trials, rates of emergent suicidality in patients receiving psychotherapy but no pharmacotherapy were comparable to rates observed in antidepressant trials (1, 2). Self-reported suicidality in the week before intake predicted the onset of emergent suicidality to a much greater extent than did interview-rated suicidality, indicating that self-report may be a necessary component to the assessment of adolescent suicidal risk.
These findings raise methodological issues for the design and interpretation of psychotherapy and pharmacotherapy treatment trials of depression in young patients. Notably, emergent suicidality, even in those patients who did not report suicidality during the intake interview, occurred fairly commonly. In the case of psychotherapy, the emergence of suicidality is probably not easily attributable to treatment. In our study, the detection of emergent suicidality could be increased by its specific and systematic assessment, whereas in previous clinical trials of depression among young patients, adverse events were reported by participants or observed by investigators. Self-reported suicidality in the week before intake predicted the onset of emergent suicidality to a much greater extent than did interview-rated suicidality, treatment assignment, cognitive distortions, and depression severity. Therefore, it is important to assess intake suicidality by self-report and to consider balancing treatment groups on this key predictor of emergent suicidality.
Received July 19, 2004; accepted Sept. 24, 2004. From Western Psychiatric Institute and Clinic, University of Pittsburgh Medical Center. Address correspondence and reprint requests to Dr. Brent, Western Psychiatric Institute & Clinic, Child and Adolescent Psychiatry, 3811 O’Hara St., BFT 311, Pittsburgh, PA 15213-2592; firstname.lastname@example.org (e-mail). Supported by NIMH grants T32 MH-18951, R01 MH-46500, P30 MH-66371, and K01 MH-69948. The authors thank their colleagues who aided in the conduct of this study and, above all, the patients and families for their participation in this study.