The study by Hallahan et al (Reference Hallahan, Hibbeln and Davis2007) has clinically important implications but before accepting the findings as valid we wish to raise a few points regarding some of the methodological and analytical aspects.

Of the 392 patients initially assessed for eligibility, only 39 (10%) completed the study, a large number (343) having been excluded for various reasons. Although this rigorous selection procedure might have enhanced the internal validity of the findings, we are concerned that the generalisability of the findings in the real-world clinical situation (i.e. external validity) might have been compromised.

Certain sample characteristics merit attention. Apart from mentioning that participants had had at least one lifetime self-harm episode in addition to the index episode, the report does not provide any data on the number, frequency, severity and recency of self-harm episodes. These data are important to characterise the sample and to ensure that they did not differ between the two groups. For example, the risk profile of a 60-year-old patient with two self-harm episodes 10 years apart would be very different from that of a 20-year-old with the previous episode only 10 days prior to the index episode. Furthermore, in patients with borderline and other personality disorders, suicidality and impulsivity can vary drastically over time, even in a single day. Instruments rated every 4 or 6 weeks might not capture the ‘real’ picture. Finally, significantly more participants in the placebo group were single or divorced compared with the active drug group. In view of this significant difference, marital status should have been included in the logistic regression and other analyses.

For analysis of suicidality scores the two groups were compared after categorical classification of values (no suicidal ideation v. presence of any suicidal ideation) to obtain a statistically significant difference. For all other variables of interest mean scores were compared. When the mean suicidality scores were compared the difference was not statistically significant. Indeed, it is interesting to note that the proportion of self-harm episodes was actually higher during the study period in the patients on active drug (7/22, 38.2%) compared with those in the placebo group (7/27, 25.9%), although the difference was not statistically significant.

Finally, it is not clear what the findings really mean in terms of decrease in ‘surrogate markers of suicidal behaviour’. Hallahan et al discuss the findings in terms of improved mood and well-being, but the logistic regression analysis showed that depression and other psychological measures did not have any effect on the suicidality score. Other surrogate markers such as impulsivity and aggression scores were not significantly different between the two groups.