Rapid Responses to:

PAPERS: Elspeth Guthrie, Navneet Kapur, Kevin Mackway-Jones, Carolyn Chew-Graham, James Moorey, Elizabeth Mendel, Federica Marino-Francis, Sarah Sanderson, Clive Turpin, Gary Boddy, Barbara Tomenson, and George C Patton Randomised controlled trial of brief psychological intervention after deliberate self poisoning • Commentary: Another kind of talk that works? BMJ 2001; 323: 135 [Abstract] [Full text]

Rapid Responses: Submit a response to this article

Rapid Responses published:

Other samples needed, but why bigger ?

Axel Ellrodt   (20 July 2001)

Pragmatic trial causes disappointment

Chris Metcalfe   (25 July 2001)

Other samples needed, but why bigger ? 20 July 2001
Axel Ellrodt, American Hospital of Paris Neuilly sur Seine, France Send response to journal: Re: Other samples needed, but why bigger ?	Dr Patton correctly suggests that psychological intervention should be assessed in the evidence based arena. It is important that the impressive results of Guthrie's study be replicated in other locations, but "bigger" samples of patients are not mandatory. Statistical tests are designed to take sample size into account. A statistical difference is a statistical difference whatever the sample size. Large samples are more likely to help find out clinically subtle but statistically robust differences.
Pragmatic trial causes disappointment 25 July 2001
Chris Metcalfe, MRC / NHS Eastern Region Training Fellow MRC Biostatistics Unit, Cambridge Send response to journal: Re: Pragmatic trial causes disappointment	EDITOR- The article by Guthrie et al presents an exemplary evaluation of the role of a brief psychodynamic intervention in the management of deliberate self poisoning. The trial is described as a pragmatic one, with the effects of the psychodynamic intervention being compared to usual treatment. The authors acknowledge that such a comparison may bias the results due to an effect of the increased contact with nurse therapists in the intervention group which is not related to the intervention itself. I would like to suggest a further potential source of bias: disappointment in the usual treatment group at not being randomised to the new treatment programme. Following a detailed description of the trial and the new treatment programme, those participants randomised to the control group are told they cannot receive the new treatment and are sent back to the usual care services. 62% of those randomised to the usual care group had a history of deliberate self harm and are likely to have had previous, and at least partly unsuccessful, contact with those services. The effect of the randomisation on the control group can only be surmised, but disappointment and a consequent negative effect on the self report measures is certainly possible. If this was the case then the study has failed to give us a pragmatic estimate of the effect of the intervention; the estimate has become contaminated by the research process. It would be interesting to know whether the baseline evaluation was completed before or after the participants learnt the outcome of the randomisation. A second difficulty in interpretation is not discussed by Guthrie et al. Strong evidence of a reduction in suicidal ideation is obtained for the six month assessment but not for the earlier end of treatment assessment. It is feasible that the beneficial effects of the intervention increased and were not merely maintained during the six month follow up period. This would be a distinctly unusual outcome however, and such a temporal pattern of effect was not hypothesised. In fact without such a hypothesis it may have been more appropriate to conduct a single significance test for each of the two self report scales in turn, with post-treatment data consisting of a single mean of the post-treatment assessments for each patient. Interpretation of the results would have been assisted by the provision of further information. Firstly, 60% of the intervention group completed the intervention programme. Can a particularly strong effect of the intervention be demonstrated for these individuals, otherwise there is a leap of faith in allocating the observed difference between randomised groups to the treatment itself. Secondly, baseline means for those people completing the six month assessment, presented separately for the two treatment groups, would have assisted in the interpretation of the covariance analysis of the two self report measures. The points give further cause for Guthrie et al's own cautious interpretation. However, to end on a positive note, this study provides the motivation, and a great deal of useful information, for further research. One final point: is the standard deviation of age really 1.5 years?