Brief CBT and Suicide Risk: Ruling Out Nonspecific Effects of Individual Therapy

To the Editor: We commend Dr. Rudd and coauthors on their important trial (1), published in the May 2015 issue of the Journal, exploring the efficacy of brief cognitive-behavioral therapy (CBT) to reduce suicidal behavior. The authors deserve to be especially applauded for their detailed reporting; this reporting suggests an opportunity to add further value to their study. In Table 4 in their study, a sizable difference is apparent in the amount of individual therapy received by the CBT group compared with the treatment as usual group, especially within the first 3 months. This raises a question of whether some of the efficacy seen in the CBT group is simply due to a nonspecific effect of receiving more individual therapy.

We suggest that Dr. Rudd and coauthors conduct post hoc analyses to investigate the sensitivity of their findings to the amount of individual therapy received. The simplest approach might be to stratify the groups based on individual therapy received in the first 3 months (e.g., ≥12 sessions versus <12 sessions). It would be useful to learn the attempt-free probabilities at 6 months and at 24 months for both the CBT and treatment as usual group within both the ≥12 individual sessions and <12 sessions strata. A second approach would be to build a regression model in which the number of individual therapy visits in the first 3 months was included. This approach would likely be conservative, but it might be useful to provide a lower bound on how much of the treatment benefit is specifically due to CBT. The most complex approach might be to divide the sample into more strata or to match individuals within the CBT and treatment as usual group based on the number of individual therapy sessions received (with some individuals potentially having to be dropped).

One or more of these approaches could be taken to investigate the impact of differences in individual therapy over the entire study period as well. These differences are not statistically significant (p=0.10) but are still sizable.

Because of the sample size, these post hoc analyses would be unlikely to yield statistical significance. The focus would rather be on the sensitivity of the results (i.e., the amount of change observed) to efforts to address imbalances between the treatment groups in the amount of individual therapy received. Depending on the p values observed, it may not be possible to exclude the possibility that differences observed in these secondary analyses arise from random variation. Nevertheless, whatever the findings, these analyses would add to the important contribution made by the Rudd et al. study.