In their meta-analysis, ^{Reference Szöke, Trandafir, Dupont, Méary, Schürhoff and Leboyer1} Szöke et al found significant improvements in most neuropsychological variables, along with well-known potential practice effects, and that semantic verbal fluency holds promise as a suitable cognitive endophenotype in schizophrenia. We acknowledge that this review is a step forward, attempting to merge and quantify the evidence from both naturalistic observational studies ^{Reference Rund2} and clinical trials. ^{Reference Woodward, Purdon, Meltzer and Zald3}

We agree with the authors that the current literature is limited by the virtual absence of healthy control groups. Since only 9 out of 53 studies reported longitudinal neurocognitive data for controls, it is difficult to disentangle whether patients' cognitive changes are true improvements or confounded by the non-specific effects of practice-related learning. ^{Reference Goldberg, Goldman, Burdick, Malhotra, Lencz and Patel4} The use of healthy individuals from test–retest studies (‘external controls’) is also problematic and the parallel assessment of controls would rule out the possibility that patients' apparent cognitive stability is not deterioration in disguise, as the authors accurately suggest.

In addition, we would like to highlight other critical issues that may limit the conclusions. First, the authors' choice to lower the minimum study duration to 1 month led to the median test–retest interval being only 4 months, which is shorter than the recommended trial duration to evaluate cognitive changes (e.g. at least 6 months). ^{Reference Buchanan, Davis, Goff, Green, Keefe and Leon5} This also may have biased the review towards short-term clinical trials intended to improve cognitive deficits, especially with second-generation antipsychotics. Had a more stringent and clinically relevant 1-year follow-up cut-off ^{Reference Rund2} been used, only 24 out of 53 studies would have been reviewed. Second, approximately half of the 20 clinical trials previously reviewed ^{Reference Woodward, Purdon, Meltzer and Zald3} are open, and these are mostly clozapine trials of patients who are treatment-resistant. Significant associations between cognitive change and change in negative symptoms are more likely to occur in these studies than in naturalistic follow-up studies. ^{Reference Buchanan, Davis, Goff, Green, Keefe and Leon5} Third, we feel the authors missed an opportunity to include the distinction between patients with first-episode or chronic schizophrenia and geriatric patients as a potential moderator variable. This could have contributed to a better understanding of the probably complex cognitive pathways during the lifespan.

Despite the number of longitudinal neurocognitive reports, less is known than was originally supposed about the course of cognition in schizophrenia. Only a small subset (n = 4) of longitudinal reports have compared these neurocognitive pathways with those of healthy controls over at least 1 year. When reviews ^{Reference Rund2} are not focused on the neurocognitive effects of antipsychotics, stable long-term performances and, in some cases, cognitive gains could be expected, thus offering a rather pessimistic picture for cognitive enhancement. This approach seems more useful for understanding the long-term natural history of cognition in schizophrenia. Conversely, this meta-analysis ^{Reference Szöke, Trandafir, Dupont, Méary, Schürhoff and Leboyer1} relies on short-term data, mostly from clinical trials, ^{Reference Woodward, Purdon, Meltzer and Zald3} and therefore more likely overestimates the potential for cognitive improvement.

In summary, it would be misleading if the conclusions were regarded as a major leap forward instead of as tentative hypotheses awaiting further investigation. Since the possibility remains that the current findings are more valuable to researchers, a new meta-analysis that takes into consideration these and other limitations might be more helpful for clinicians, patients and caregivers.