Kapur and colleagues provide a brief review of contact-based interventions for self-harm and note their continuing appeal, despite largely unconvincing trial results. ^{Reference Kapur, Cooper, Bennewith, Gunnell and Hawton1} The question the authors should have posed is not ‘How might these interventions work?’ but ‘Why, when participants report that the interventions are of benefit, are trial results so unconvincing?’

The importance of self-harm lies in its strong association with suicide. The ultimate aim of interventions in this area is to prevent suicide, but the rarity of suicide makes it difficult to use as an outcome measure. Of those studies reviewed by Kapur et al, only two used death as an outcome. ^{Reference Motto and Bostrom2,Reference Fleischmann, Bertolote, Wasserman, De Leo, Bolhari and Botega3} The remainder used repetition of self-harm, which is the best available proxy measure. ^{Reference Hawton, Arensman, Townsend, Bremner, Feldman and Goldney4}

Measuring repetition of self-harm is problematic. Hospital-treated episodes represent the standard measure but fail to capture the true pattern of self-harming behaviour, most of which occurs in secret and does not result in hospital presentation. Those who repeatedly self-harm avoid accident and emergency (A&E) departments at all costs and, when forced by the severity of their injuries to present, are adept at concealing the self-inflicted nature of those injuries, resulting in possible miscoding of visits. There is a need for a reliable, user-designed self-report instrument and a better understanding of the relationship between acts of self-harm and hospital visits.

Hospital-treated episodes do not provide a measure of reduction in self-harm; only a measure of reduction in clinical encounters for self-harm. It is debatable whether reducing clinical encounters is a beneficial outcome for this highly vulnerable and hard-to-reach population (repeat self-harmers). Reducing the number of hospital presentations may cut service costs in the short term; it may not save lives.

In a recent pilot study of a text-messaging intervention for self-harm, ^{Reference Owens, Farrand, Darvill, Emmens, Hewis and Aitken5} we had an interesting case. One of the participants reported during the trial that the intervention (a text message) had saved their life by interrupting a suicide attempt and prompting them to call for help instead of taking an overdose. They were conveyed to A&E and treated for very severe lacerations. Partly as a result, their visits to A&E increased during the 6-month pilot trial compared with 6 months prior to entry: a negative result using hospital-treated episodes as a measure. Two further participants told us that a suicide attempt had been interrupted by the timely arrival of a text message and begged to be allowed to continue to use the intervention at the end of the trial, yet standard reporting of the results of the study would not provide convincing evidence of effectiveness.

So why are trial results unconvincing, despite qualitative evidence to the contrary? The low status of qualitative data is one possible reason. Another is that we are measuring outcomes in inappropriate ways. We do not yet understand what outcomes are important to those who engage in repeated self-harm, nor how best to measure them.