Introduction. There are questions on whether the effectiveness of Beck's cognitive behavioral therapy (CBT) on depression in suicidal patients depends on factors early in the CBT process.

Purpose. The causal mediation method of Ten Have et al. (2007) is extended with an interaction to identify early post randomization modifiers of the CBT effect on follow-up depression in suicide attempters.

Sample. 120 recent suicide attempters were randomized to either cognitive therapy or usual care to prevent future suicide attempts during 18 months of follow-up (Brown et al., 2005).

Method. 1- and 3-month hopelessness and social problem orientation measures were examined as effect modifiers of the CBT effect on 6-month depression with a Rank Preserving Model (RPM; Joffe et al., 1998; Ten Have et al., 2007) that included an interaction term between CBT and a post randomization factor. Comparisons were made with a standard interaction regression model, relying on the strong assumption of no unmeasured confounding of the post randomization moderator-outcome relation. The RPM approach does not make such an assumption but does rely on other potentially testable assumptions.

Results. The data analysis showed statistically significant effect modification under the RPM, whereas the standard regression model showed less significant effect modification. Simulation results showed unexpectedly good results for the standard regression model under unmeasured confounding. The RPM performed slightly better with respect to bias of the stratified CBT effect, but worse for the interaction effect. The RPM expectedly showed greater variability than the standard regression model in terms of the mean square error.

Discussion. Both the RPM and standard regression methods performed well in simulations, whereas there was some disagreement between the methods for the data analysis. The RPM can be used as a check for unmeasured confounding when using the standard regression model.