Smith et al state that there is increasing concern among clinicians about the association between second-generation antipsychotics and diabetes. ^{Reference Smith, Hopkins, Peveler, Holt, Woodward and Ismail1}

It is interesting then that while commenting on the lack of systematic reviews and meta-analyses that support this concern, the authors go on to investigate not the relationship between starting antipsychotics and developing diabetes, but the relative risk of developing diabetes between groups of patients commenced on first-generation and second-generation antipsychotics. It is questionable whether this meta-analysis addresses, in any clinically meaningful way, the risk of developing diabetes after starting an antipsychotic, whether second or first generation. This would appear to be more usefully addressed by looking at the absolute risk.

The authors report on the difficulties in finding high-quality trials to include in their study. This is illustrated by the inclusion of only 11 trials out of an identified 1974. Smith et al then go on to outline their own criteria for a study to be considered of ‘high quality’. These criteria include a prospective design and at least 1 year of follow-up recorded. It is of note then that of the 11 studies eventually included in the analysis, only 3 were prospective. Furthermore, of these 3 prospective trials, none was longer than 3 months. All trials included in the review could, therefore, be classified as low quality. The test for heterogeneity between studies, applied by the authors, further illustrates the highly significant methodological heterogeneity between studies.

We would suggest that given the overall poor quality of studies found in the review there seems to be no rationale for going on to conduct a meta-analysis. One common pitfall of any meta-analysis is that if you put only poor-quality data in, you will get poor-quality data out. Consequently, this meta-analysis would seem to add little to the current evidence base with regard to antipsychotics and diabetes, except, perhaps, the confirmation that the studies on this subject are heterogeneous and generally of poor quality.

If one does want to consider whether a significant relationship exists between antipsychotic use and diabetes, or a metabolic syndrome, then the CATIE study ^{Reference Lieberman, Stroup, McEvoy, Swartz, Rosenheck and Perkins2} would seem to provide reasonably robust evidence that such a relationship does exist. This large, randomised, prospective study, carried out over a period of 18 months, has data collected at baseline and following the introduction of antipsychotic, and demonstrates clinically and statistically significant adverse changes in blood glucose, weight and cholesterol. This is particularly the case for those patients commenced on olanzapine.