Interventions:
The interventions were clearly reported. The authors did not justify their choice of comparators and there were other relevant interventions, which makes this study only a partial analysis.
Effectiveness/benefits:
The authors stated that equal effectiveness was assumed, but they did not report the clinical statistics. They could have included the effectiveness in the model and evaluated uncertainty using probabilistic sensitivity analysis. Extra-pyramidal symptoms and the risk of coronary heart disease were mentioned, but they were not modelled. Most of the evidence came from a randomised controlled trial, the design of which should have ensured the validity of the clinical inputs, due to its double-blind and randomised approach. Although full details of the clinical trial were not presented, it was reported that patient groups were comparable at baseline, which makes the comparison more robust. The measurement of the clinical effects and their statistical analyses were transparent. The authors did not justify basing the analysis solely on this trial and it is not clear if the best available evidence was used. As equal effectiveness was assumed for the two treatments, the authors used only intermediate disease-specific measures of benefit. However, these will prevent the assessment of the impact of the interventions on patients’ condition and quality of life and will hinder cross-disease comparisons.
Costs:
The costs appear to have reflected the perspective stated. A breakdown of the cost items was not presented and resource quantities were not given separately. The costs were presented as macro-categories due to the sources used (diagnosis-related group data). The cost analysis was well reported with respect to the sources, price year and discounting, but, apart from medication costs, the estimates were not investigated in the sensitivity analysis.
Analysis and results:
The basic model structure was presented in a diagram. The authors stated that they calculated an annual risk of diabetes and yet there was no mention of a state transition model. The synthesis of costs and benefits was appropriate and was carried out using an incremental approach. However, the issue of uncertainty was not sufficiently addressed, as the deterministic sensitivity analysis was restricted to the doses of the medications only. This might limit the generalisability of the findings. This issue of the generalisability of the findings was acknowledged by the authors as one of the limitations to their study.
Concluding remarks:
The study appeared to exclude important health outcomes and comparators. The authors tried to assess the treatment cost-effectiveness in terms of the impact on diabetes only, but this seems to have been inappropriate.