We read Slee et al's ^{Reference Slee, Garnefski, van der Leeden, Arensman and Spinhoven1} article with interest and concern. We believe there are major biases towards the treatment arm of this study which may invalidate their conclusions. Furthermore, our experience of working in a liaison psychiatry team receiving more than 1500 self-harm referrals a year leads us to question the applicability of the intervention given the characteristics of the study group.

At the outset, there are more participants in the treatment-as-usual (TAU) group shown to be depressed and this difference reaches statistical significance from the first follow-up at 3 months and gradually grows with each follow-up. Hence, it can be argued that the difference in outcome is a mere difference in depression and anxiety, which we know respond well to cognitive–behavioural therapy (CBT). Moreover, as the authors themselves admit, there was a trend from the beginning of higher suicidal cognitions in the TAU group, which assumed statistical significance from the first follow-up at 3 months. Furthermore, the authors have not attempted to match the extra time spent with participants in the CBT group with a similar amount of therapist/contact time in the TAU group. Masking (as acknowledged) of follow-up assessments was not undertaken. Therapists in the treatment group very actively pursued participants; this may have been the active ingredient rather than CBT. Sending postcards alone as an intervention significantly reduces the frequency of hospital-treated self-poisoning events. ^{Reference Carter, Clover, Whyte, Dawson and D'Este2} All these factors bias the results in favour of the treatment group. Despite these biases, the reported benefit in reducing self-harm was marginal and only statistically significant at 9 months, with questionable clinical significance.

The participants in this study differ very significantly from the individuals seen after self-harm by routine liaison psychiatry services. The self-harm definition used was very wide, including punching and head banging, which are not usually defined as self-harm by clinicians and not proven to be associated with higher suicide risk, unlike self-poisoning and self-cutting. No data are reported on the proportion of self-harm in the study which was of this milder nature. Right from the recruitment phase, participants with alcohol and drug misuse were eliminated. This clearly skews the population enormously since a very high proportion of our patients have comorbid issues. The treatment group in particular lost eight individuals before CBT was started, and all assessments and therapy sessions were then completed. We contend that this was a highly motivated and selected group likely to benefit from the intervention, and unrepresentative of the clinical population.

Short-term interventions for self-harm have not generally proved significant when explored in large-scale studies. ^{Reference Crawford and Kumar3} It is therefore crucial that small randomised trials of CBT or other interventions are carefully designed to minimise bias, and we feel this study fell short of the design and reporting standards we would expect. We are also concerned that high-profile publication of such studies may lead to unwarranted implementation of interventions whose effect is unproven, and whose opportunity costs are great.