Differential Effects of Brief CBT Versus Treatment as Usual on Posttreatment Suicide Attempts Among Groups of Suicidal Patients

Globally, approximately 800,000 people die by suicide each year (1). In the United States, suicide is the tenth leading cause of death, and deaths from suicide reached a 30-year high in 2015 (2). For each suicide death, there are an estimated 25 suicide attempts (3). Although the suicide rate in the U.S. general population has steadily risen over the past decade, a much faster rise has occurred among U.S. military personnel (4).

Unfortunately, few treatments and prevention strategies have demonstrated efficacy for reducing suicidal behaviors (5). Of these treatments, brief cognitive-behavioral therapy (CBT) was found to be associated with a 60% reduction in suicide attempts among U.S. military personnel (6), results that matched outcomes in nonmilitary samples (7). Two recent trials testing even briefer treatments—the three-session Attempted Suicide Short Intervention Protocol (8) and the single-session crisis response plan (9)—provided even more support for psychological treatments that directly target two key mechanisms contributing to suicidal behavior: emotion regulation and cognitive flexibility (10).

Although accumulating evidence supports the efficacy of brief, suicide-focused interventions for the prevention of suicidal behaviors, the restriction of enrollment to very high-risk patient subgroups (that is, those who had made a suicide attempt in the recent past) in several of these studies (7,8) has led to questions about the relative efficacy of these treatments across the broader spectrum of suicide risk. Currently, it is not known whether suicide-specific treatments are differentially efficacious for patient subgroups with varying levels of suicide risk. Additional studies are needed to examine the relative efficacy of these treatments for a range of patient subgroups.

This study entailed a secondary analysis of a randomized clinical trial comparing the efficacy of brief CBT and treatment as usual for the prevention of suicidal behaviors among U.S. Army personnel (6), a population with an especially high risk of suicide. The primary objective was to examine treatment effects across patient subgroups characterized by varying levels of suicide risk. It was first hypothesized that multiple empirically distinct and clinically meaningful patient subgroups would be identified via latent-class analysis. It was not hypothesized a priori how many latent classes would be identified. It was also hypothesized that brief CBT would reduce posttreatment suicide attempt rates during the 24-month follow-up period across all patient subgroups.

Methods

Participants

Participants included 176 active duty U.S. military personnel stationed at a large Army base in the western United States. Participants were predominantly male (87%, N=152) and ranged in age from 19 to 44 (mean±SD=27.5±8.3). Self-identified racial-ethnic distribution was 71% (N=124) white, 13% (N=23) black, 2% (N=3) Asian, 2% (N=4) Pacific Islander, 5% (N=8) Native American, 8% (N=14) “other,” and 22% (N=38) Hispanic or Latino. Participants had served in the military a mean of 5.7±4.5 years (range 0–25) and had deployed a mean of 1.6±1.3 times (range 0–8). Rank distribution was 73% (N=128) junior enlisted (E1 to E4), 23% (N=40) noncommissioned officer (E5 to E6), 3% (N=6) senior noncommissioned officer (E7 to E9), and <1% (N=1) warrant officer.

Procedures

A full description of the procedures, including a CONSORT chart, has been published elsewhere (6). In summary, soldiers were referred to the study for determination of eligibility if they reported active suicide ideation in the past week or a suicide attempt in the past month. To maximize generalizability, the only exclusion criterion was the inability to complete the informed consent process because of a psychiatric or general medical condition (for example, intoxication, mania, or psychosis). Soldiers meeting eligibility criteria were randomly assigned to either brief CBT or treatment as usual by using a computerized simple randomization algorithm. Brief CBT was provided by two female clinical social workers with varying levels of practice experience (more than 15 years of licensure and less than one year of licensure). All therapy sessions were video recorded for fidelity monitoring, and both therapists participated in weekly supervision with one of the treatment developers (CJB). Participants completed follow-up assessment interviews at three, six, 12, 18, and 24 months postbaseline. Follow-up assessments were conducted by an independent evaluator who was blind to treatment condition. Study procedures were approved by the Madigan Army Medical Center’s Institutional Review Board.

Treatment Conditions

Participants in both treatment conditions received treatment as usual from military mental health clinicians and nonmilitary mental health clinicians in the local community. Treatment as usual included a combination of individual psychotherapy, psychopharmacology, group therapy, substance abuse counseling, marriage therapy, and support groups. Participants in the brief CBT condition received 12 individual sessions of brief CBT, a transdiagnostic outpatient psychotherapy that directly targets deficits in emotion regulation skills and cognitive rigidity, two key factors underlying suicidal behaviors. Brief CBT is organized to be delivered in three phases: emotion regulation skills training (five sessions), cognitive restructuring skills training (five sessions), and relapse prevention (two sessions). Participants randomly assigned to brief CBT were not restricted from receiving treatment as usual (for example, medication, group therapy, and family counseling). As was previously reported by Rudd and colleagues (6), treatment utilization was similar across both groups, indicating that brief CBT largely “replaced” individual psychotherapy provided as part of treatment as usual but did not replace or alter other forms of treatment.

Measures

A suicide attempt was defined as a behavior that is self-directed and deliberately results in injury or the potential for injury to oneself, for which there is implicit or explicit evidence of suicidal intent (11). Suicide attempts were assessed with the Suicide Attempt Self-Injury Interview (SASII) (12). The SASII is a reliable and valid clinician-administered interview that measures various characteristics of self-directed violence (for example, method, medical severity, and intended outcome). Recent suicide ideation during the past week was assessed with the Scale for Suicide Ideation (13). Depression was assessed with the Beck Depression Inventory–Second Edition (14). Posttraumatic stress symptoms were assessed with the Posttraumatic Stress Disorder Checklist (15). Hopelessness was assessed with the Beck Hopelessness Scale (16). Perceived burdensomeness and thwarted belongingness were assessed with the Interpersonal Needs Questionnaire (17). Lifetime trauma exposure was assessed with the Life Events Checklist, and psychiatric treatment history (current and past) was assessed with the Cornell Service Index (18).

Data Analysis

To identify patient subgroups at baseline, latent-class analysis was used with the 19 items of the Scale for Suicide Ideation selected as indicator variables. Because Scale for Suicide Ideation items contain three ordinal response options, items were specified as categorical variables, and robust maximum-likelihood estimation was used. The following fit statistics were examined to compare the results of latent-class analysis with two to four classes extracted: sample size–adjusted Bayesian information criterion, entropy, and Vuong-Mendell-Rubin adjusted likelihood ratio test values. Smaller Bayesian information criterion values indicate relatively better model fit, whereas entropy values above .80 indicate an acceptable level of separation between identified subgroups. Statistically significant Vuong-Mendell-Rubin likelihood ratio tests indicate better fit for the current model relative to the model with one fewer class, whereas nonsignificant tests indicate no incremental improvement in fit relative to the simpler model. Once the classes were extracted, participants were assigned to their most likely class, and then baseline characteristics of each class were compared by using chi-square tests (for categorical variables) and generalized linear models (for continuous variables). To identify differences in treatment effects across classes, the Kaplan-Meier method and Cox regression were used, the latter of which is robust to missingness and aligns with the intent-to-treat principle. Mplus 7.4 software (19) was used for the latent-class analysis, and SPSS 23 software was used for comparisons of patient outcomes by baseline characteristics and Cox regressions.

Results

Baseline scores on the Scale for Suicide Ideation ranged from 0 to 33 (mean=9.6±8.6), indicating a range of self-reported suicide risk from low to high, with a moderate level of suicide risk severity on average. Scale for Suicide Ideation scores were not correlated with age, gender, race, ethnicity, deployment history, or history of suicide attempts and did not differ between treatment groups.

Results of Latent-Class Analysis

Fit statistics for the three latent-class analysis models are reported in Table 1. Bayesian information criterion and entropy values declined as the number of classes increased, suggesting improved fit combined with only a marginal decline in separation between assigned groups. Results of the Vuong-Mendell-Rubin likelihood ratio tests suggested that although the three-class model was significantly better than the two-class model, the four-class model did not improve upon the three-class model. Therefore, the more parsimonious three-class model was selected. The average latent-class probabilities were 97% for class 1, 99% for class 2, and 97% for class 3, suggesting clear separation of groups.

Differences in item response patterns on the Scale for Suicide Ideation across the three classes are summarized in Table 2. On most items, there were moderate to large differences in item responses across the three classes, as reflected by Cramer’s V statistics. The three items with the relative largest between-group differences were the intensity of the wish to die (V=.74), severity of active suicide ideation (V=.59), and expectation for making a suicide attempt (V=.58). Members of class 1, which constituted 44% (N=77) of the sample, almost exclusively endorsed the lowest response options on these items: 89% endorsed “I have no wish to die,” 92% endorsed “I have no desire to kill myself,” and 100% endorsed “I do not expect to make a suicide attempt.” Members of class 2, which constituted 33% (N=57) of the sample, rarely endorsed these lowest response options: 40% endorsed “I have a weak wish to die,” and 59% endorsed “I have a moderate to strong wish to die”; 64% endorsed “I have a weak desire to kill myself,” and 28% endorsed “I have a moderate to strong desire to kill myself”; and 79% endorsed “I am unsure that I shall make a suicide attempt,” and 9% endorsed “I am sure that I shall make a suicide attempt.” Finally, members of class 3, which constituted 23% (N=41) of the sample, showed a mixed endorsement pattern: 98% endorsed “I have a weak desire to kill myself,” 29% endorsed “I have no desire to kill myself,” and 71% endorsed “I have a moderate to strong desire to kill myself”; 89% endorsed “I do not expect to make a suicide attempt.” Overall, these patterns suggest that members of class 1 endorsed relatively mild intent and did not expect to make a suicide attempt, members of class 2 endorsed moderate to strong intent and expected to make a suicide attempt, and members of class 3 expressed ambivalence about suicide but did not expect to make a suicide attempt.

Demographic and Clinical Characteristics Associated With Each Subgroup

Overall, the three classes were similar to each other across all demographic variables except black racial identity, which was significantly more common in class 2 than in class 3 (Table 3). However, the three classes significantly differed with respect to psychiatric symptom severity. Class 1 reported the least severe symptoms, class 2 reported the most severe symptoms, and class 3 fell in the middle. In combination with the results noted above, class 1 was labeled as low severity, class 2 as high severity, and class 3 as moderate severity.

Follow-Up Suicide Attempt Rates Across Subgroups, by Treatment Condition

Of the 176 participants who completed the initial assessment, 152 (86%) met eligibility criteria and were enrolled in the clinical trial. No differences between classes were found with regard to treatment assignment, total number of prescribed medications, or rates of use of antidepressants, antipsychotics, anticonvulsants, and benzodiazepines. There also were no differences by class and treatment group with respect to the total number of individual therapy sessions attended and the total days of psychiatric inpatient admission during follow-up. These results suggest that the overall amount and type of treatment received by participants within each class was comparable across brief CBT and treatment as usual.

The proportion of participants in each treatment group who made a suicide attempt during the follow-up period, by class, is summarized in Table 4. Overall rates of suicide attempts during the follow-up period across the three groups were similar and statistically nonsignificant (low severity, 15%; moderate severity, 10%; and high severity, 25%). Rates did not differ across classes in brief CBT; however, rates differed significantly across classes in treatment as usual (χ²=7.86, N=76, df=2, p=.020). When treatment differences by class were considered, suicide attempt rates during follow-up significantly differed in the high-severity class (log-rank χ²=7.19, N=55, df=1, p=.007) but not in the low-severity or the moderate-severity classes. Hazard ratios indicated that the between-group difference was very large and statistically significant in the high-severity class (that is, a 79% relative reduction in attempts associated with brief CBT). This difference remained statistically significant even when the false discovery rate method was used to correct for multiple comparisons (20). The between-group difference in the low-severity class was large (that is, a 55% relative reduction in attempts associated with brief CBT) but not statistically significant. In the moderate-severity class, the between-group difference was small and not statistically significant.

Discussion and Conclusions

Although independent clinical trials have supported the efficacy of brief CBT for the reduction of suicidal behavior (6,7), little is known about the efficacy of these treatments among different subgroups of suicidal patients. As expected, this study identified three distinct subgroups of active duty U.S. Army soldiers with recent suicidal thoughts or behaviors. These groups differed markedly with respect to severity of suicidal thinking and psychiatric symptom severity but did not differ with respect to any demographic variable. Whereas the low-severity group was characterized by a strong desire to live, a weak desire to die, and low rates of suicidal planning and preparation, the high-severity group was characterized by a weak desire to live, a strong desire to die, and high rates of suicidal planning and preparation. Finally, the moderate-severity group was characterized by ambivalence—that is, a moderate desire to die combined with a moderate desire to live—and relatively low rates of suicidal planning and preparation. Overall, results converge with previous studies that identified three predominant typologies of suicidal patients characterized by a relatively strong desire to live, a relatively strong desire to die, and ambivalence (21,22). As in that study, psychiatric symptoms were most severe in the high-severity group and least severe in the low-severity group.

The relative effect of brief CBT on suicide attempts during the follow-up period differed across the three patient subgroups, although this finding was largely attributable to variability in outcomes across classes in treatment as usual. Specifically, suicide attempt rates among participants in brief CBT were consistent across all three classes (10% in low severity, 13% in moderate severity, and 10% in high severity) but significantly varied across classes in treatment as usual (21% in low severity, 8% in moderate severity, and 41% in high severity). Therefore, brief CBT yielded more consistent outcomes across all patient subgroups, but outcomes in treatment as usual depended on the patient’s class membership, with the relative worst outcomes among participants who initially presented with more severe levels of suicide risk.

Of note, in treatment as usual, the low-severity class had the next highest rate of suicide attempts during follow-up, and the moderate-severity class had the lowest rate. Our data are unable to account for this pattern, but one possibility is that clinicians providing treatment as usual may have underestimated the level of risk of low-severity participants because of the low rates of endorsement of active suicide ideation among these participants. Thus the clinicians may have approached treatment in different ways for participants in the low-severity and moderate-severity classes. For example, the clinicians may have prioritized patients’ psychiatric conditions in the treatment plans for low-severity participants but focused more explicitly on suicide risk in the treatment plans for moderate-severity participants. More research is needed to understand how clinicians approach treatment with different patient subgroups under treatment-as-usual conditions.

In contrast to the clinicians providing treatment as usual, the clinicians providing brief CBT used a treatment protocol across all patient classes because of the treatment’s emphasis on clinician fidelity, a variable that has been identified as a central characteristic of effective suicide-focused treatments (23). Suicide-focused treatment protocols may lead to more consistent outcomes across patient subgroups because they help clinicians and patients prioritize treatment targets that are most central to the emergence of suicidal behavior. In the case of brief CBT, these central targets are emotion dysregulation and cognitive rigidity, regardless of psychiatric diagnosis and overall risk level (10). By administering brief CBT in a consistent and reliable manner, patients in all subgroups received interventions designed to target these suicide-specific vulnerabilities. Overall, the results reported here support the efficacy of brief CBT for patients spanning the full spectrum of suicide risk.

This study was not without limitations. First, the sample was confined to active duty U.S. Army personnel, which may restrict generalizability to other patient groups. Second, the sample was relatively small, which hindered statistical power. Conclusions based on these results should thus be made cautiously until additional studies can be conducted with larger samples from different clinical populations. Despite these study limitations, the results provide further support for the efficacy of a brief, time-limited outpatient treatment that specifically targets suicide risk in a high-risk population.