Lutgens et al's interesting paper ^{Reference Lutgens, Gariepy and Malla1} describes the results of their meta-analytic study on the effect of psychosocial interventions on negative symptoms for people with psychosis. Despite commending the aims of the study we have some methodological reservations on the results presented.

We believe that the studies included are only a partial representation of the research conducted on the therapeutic modalities considered. For example, we have recently completed a meta-analysis on the effect of cognitive remediation on negative symptoms. ^{Reference Cella, Preti, Edwards, Dow and Wykes2} Our study had a similar time frame to Lutgens et al's, and the same participant inclusion criteria. Our search retrieved 45 eligible studies, compared with only 16 retrieved by Lutgens et al, in their neurocognitive therapies category. We believe that this is due to their search strategy, which included the term ‘negative symptom’ and therefore retrieved only studies with this term in the abstracts. This had two effects: it was more likely to retrieve studies reporting positive findings; and when investigating interventions not specifically designed to target negative symptoms it missed a large body of studies across all the therapeutic modalities considered.

The nature of the control condition is also important when considering effect sizes. Lutgens et al conflated passive with active control conditions. Active control conditions for one study (e.g. cognitive remediation) were then considered active treatment conditions in subsequent analyses. We also noted some overlap in the therapy groups considered. Both art and music and exercise therapy included dance-based interventions. The miscellaneous category adds to the limited clarity of the category definitions by considering comprehensive ‘care packages’, such as in Garety et al, ^{Reference Garety, Craig, Dunn, Fornells-Ambrojo, Colbert and Rahaman3} which include medication management and allocation to a psychosocial intervention among a number recommended by clinical guidelines (i.e. family therapy or cognitive-behavioural therapy (CBT)). These limitations, in our view, make it difficult to reliably compare effect sizes from the intervention groups considered.

We also wish to point out some methodological considerations that may limit the accuracy of the results reported First, it appears that the authors considered only end-of-therapy data in estimating effect sizes. This does not account for relative change. In other words, this method considers symptom reduction of a hypothetical 3 points on a negative symptoms scale to be equivalent in individuals entering the study with an initial score of 5 or of 23. The importance of taking into account baseline levels in meta-analysis is clear and it is considered best practice. ^{Reference Achana, Cooper, Dias, Lu, Rice and Kendrick4} There is also evidence that the DerSimonian-Laird method has limitations compared with methods using restricted maximum likelihood estimators. ^{Reference Veroniki, Jackson, Viechtbauer, Bender, Bowden and Knapp5}

Last, it is unclear how the authors considered the treatment that participants received as part of treatment as usual (TAU). They state: ‘Compared with TAU, 59% (10/17) of studies reported CBT to be more effective at the end of treatment’. From this statement one might assume that participants received either TAU (e.g. medication) or CBT. In all likelihood, studies compared CBT + TAU with TAU only.