Authors' reply

Marlowe notes that the primary outcome of our trial was relapse and comments that it is surprising, therefore, that it was not analysed in more detail. McKenna et al attempt to analyse the relapse data further. Neither Marlowe nor McKenna et al appear to understand the inferential problems raised by the lack of full or partial remission in a considerable proportion of the patients in this trial. The number with full or partial remission is itself an outcome of the trial (i.e. it is a post-randomisation measure). Those who have shown no recovery are excluded from the relapse data that Marlowe and McKenna et al present. In fact, twice as many people show no recovery in TAU as in CBT (18:9). The data reported by Marlowe and McKenna et al are therefore not a causal effect of randomisation (i.e. not an intention-to-treat effect). Because of this problem, we used months in full or partial remission as our primary indicator of outcome for which a formal intention-to-treat analysis is presented. This analysis and also a further examination of total days in hospital and number of admissions very clearly demonstrate that CBT, family intervention and TAU do not differ. We also reported fully on deaths and other adverse events and found no differences (the only completed suicide was in TAU). We are therefore not at all convinced by the suggestion that psychological intervention might be detrimental. Indeed, we infer on the basis of the results of this trial and of numerous meta-analyses (e.g. Pfammatter et al, ¹ Pilling et al ² and Wykes et al ³ ) that CBT and family intervention are beneficial for certain populations for a range of outcomes.

With respect to the point raised by Marlowe on the effects of having a carer on a psychological intervention, we are of course very aware of the Hogarty et al study, ^4,5 which we also discuss. It reported mixed findings. Our point here concerned the apparently beneficial effect of having a carer on CBT, which has not been examined before.