Response to comments made by assessors
 

Clinical assessor

1. ‘On p8, and in Fig 1, it is stated that of 62 relevant reports, 56 were excluded and 16 remained! Also on p8, (results, para2) tens studies are mentioned in which exercise is compared with no treatment – on p10 and in Table 1 eleven studies are described’


72 reports were retrieved (56 excluded 16 included). 11 studies compared exercise to no treatment. These errors have been corrected.

2. ‘It may be helpful to state clearly that there is genuine doubt about the persistence of any benefits beyond the end of the treatment’

1. ‘Data….were independently extracted by two reviewers using a structured form. Discrepancies between the authors were agreed…(Page 6). Presumably the reviewers and authors are one and the same?’


Yes we are and this is now made clear in the report.

2. ‘The discrepancy between the 62 retrieved reports, the 56 excluded studies, and the 14 included studies needs to be clarified’


See point 1 above

3. ‘When both variables (‘abstract’ and ‘follow-up’) were combined in the model between study variance was reduced to zero (page 10). Does this statement imply that the variance was exactly zero?’


The statistic output for the model with both variables included gives Tau² = 0.000 This is rounded to three decimal places i.e. it will not be exactly zero but very close. We have changed the wording in the report to make this clearer.

4. ‘Length of follow-up in the ‘Reuter’ study is given as 10 weeks in Table 1, but 8 weeks in Figure 2……’

1. Please explain briefly how you allowed for the fact that different cut offs were used for the Beck Depression Inventory, and how this is reflected in the effect sizes.


Baseline depression severity for the whole sample was included in the meta-regression model. For 8 of the 10 studies this was available as a mean Beck Depression Inventory (BDI) score for the whole sample in the reports. For the remaining two the mean values of the other studies was imputed. In fact baseline depression did not account for any of the variance suggesting that despite the different cut offs used depression levels in each study were probably similar. The main point is that the BDI was not developed as a tool to diagnose depression and the use of different cut offs illustrates this. Using non clinical subjects in these studies may limit their generalisation to clinical populations. A sentence has been added to the discussion to clarify this.

2. We weren’t sure why you had included all of the studies in figure 2, given that you had already shown that pooling then was not really valid?


Figure 2 we feel really nicely illustrates the meta-regression results. We also feel that it is appropriate to pool the studies and do the meta-regression model, we would agree with the referee that this is one of the ‘particular strengths’ of this study. We have not pooled all studies in this model just those comparing exercise to no treatment. The ‘outlying’ of the two abstracts can be clearly seen and the tendency for effects to move across the page from left to right illustrates the effect of length of follow-up. Traditionally in meta-analysis the approach has been to pool data that is heterogeneous using random effects models without any further investigation into what lies underneath the heterogeneity and what this suggests about the true size of the effect. Meta-regression analysis as illustrated here shows the value of looking more closely at heterogeneity. Though table 2 gives pooled summaries for different publication types and lengths of follow-up we feel the full model as illustrated by figure 2 very powerfully demonstrates the value of this approach.

3. We wondered whether the paper you cite as reference 5 described a randomised controlled trial of expressive writing. If the study was not an RCT, can you really state that creative writing helps to reduce depressive symptoms?’