Authors' reply

Birchwood et al make two points that require clarification. First, their statement that our findings from studies with high methodological rigour, particularly masking, imply that cognitive-behavioural therapy (CBT) has small but by no means negligible effects on positive and total symptoms ‘broadly in line with the National Institute for Health and Care Excellence (NICE) review and particularly that of Wykes et al,’ seems to us questionable. Wykes et al ¹ reported an effect size of 0.37 for positive symptoms, which reduced slightly to 0.31 in masked studies. This latter value was four times larger than the value of 0.08 we found for masked studies of positive symptoms. Ratings of bias were made for the studies included in the 2009 NICE guideline; ^2,3 however, no analyses excluding low-quality studies or otherwise examining methodological rigour were actually carried out.

Second, Birchwood et al’s argument that a finding of significant heterogeneity among studies implies that CBT is effective in certain subgroups of patients is not formally correct. It could simply mean that there are systematic differences in effect size between studies at high and low risk of bias. Tending to support this latter interpretation, in our meta-analysis of positive symptoms there was no significant heterogeneity in either the masked (n = 20, effect size 0.08, I ² = 0%, Q = 18, P = 0.49) or unmasked studies (n = 8, effect size 0.57, I ² = 23%, Q = 9, P = 0.24) when they were considered separately. Heterogeneity was also not significant in the masked studies of overall symptoms (n = 20, effect size 0.15, I ² = 25%, Q = 25, P = 0.15), although it remained significant in the unmasked studies (n = 10, effect size 0.62, I ² = 71%, Q = 31, P<0.001).

Byrne argues that our findings are limited by not considering follow-up data. We presume he is arguing here for a ‘delayed action’ effect of CBT, as found in the 2000 study of Sensky et al ⁴ and an early meta-analysis by Pilling et al. ⁵ However, the meta-analyses carried out for the 2009 NICE guideline ² provide only lukewarm support for such a view: the pooled effect sizes for overall symptoms were 0.27, 0.23, 0.40 and 0.19 at end of treatment, 6 months’, 12 months’ and 12-18 months’ follow-up respectively, when CBT was compared with standard care; they were 0.13 at end of treatment and 0.18 at 12 months when CBT was compared with other active treatments.

Among the other issues raised, whether there is evidence for a dose effect for CBT seems to us essentially imponderable, since none of the 50+ published randomised controlled trials to date has manipulated dose or duration of the intervention. Such an effect would also likely be difficult to detect using meta-analytic methods, given the many other sources of variation among the existing studies. With respect to whether or not CBT should be considered a ‘quasi-neuroleptic’, we simply note that CBT was originally developed for and continues to be promoted as a treatment for positive symptoms.