The selection of the comparators was appropriate as olanzapine and risperidone were two of the three (together with quetiapine) most commonly prescribed antipsychotics for schizophrenia. The two drugs accounted for about 55% of the prescriptions for schizophrenia in Norway. The authors stated that future studies should consider a wider range of antipsychotics.
The clinical inputs were from a review of the literature in standard databases, which should have identified the key studies. Most of the data were from a meta-analysis, but the studies that were included in this synthesis were not reported. It is likely that most of them were clinical trials, but a full description is needed to judge the validity of the clinical inputs. The authors justified their selection of benefit measure, which was relevant to the disease and took account of the patients’ quality of life, but was not comparable with benefit measures for other diseases.
The economic analysis was extensively presented. The cost categories and their sources appear to have been relevant to the stated health care perspective and the Norwegian setting, except for the cost of suicide, which was from a relatively old US study. The authors acknowledged that most of the costs were from a Norwegian study, conducted in 1999, and the resource use might have changed since then. Other details of the analysis, such as the price year and discounting, were clearly reported. The costs were treated stochastically in the probabilistic sensitivity analysis.
Analysis and results:
The results were clearly reported and an incremental approach was used to identify the optimal treatment. The calculation of a cost-effectiveness ratio was not required as olanzapine was dominant. Appropriate methods were used to assess uncertainty, and the results were clearly presented and discussed. The authors compared their results with those of other published studies, which had similar findings. The transferability of the results was not discussed and it is unclear whether they could be relevant to other settings, given the high uncertainty found.
The methods were robust and various areas of uncertainty were considered. The authors’ conclusions appear to be valid.