Interventions:
The reason for selecting the comparators was clear; the proposed hypothetical intervention was compared against no such intervention, which was the usual care in the authors’ setting.
Effectiveness/benefits:
Sources for the clinical inputs appear to have been valid and appropriate. For example, the treatment effect was from a meta-analysis of randomised controlled trials and the epidemiological data were from large cohort studies, conducted in the authors’ setting. The key methods of these source studies (patient population, study design, and sample size) were reported. Some assumptions were needed, mainly for the offset and onset periods for the treatment effect, which had a substantial impact on the cost-effectiveness results. Extensive sensitivity analysis was conducted on these parameters. QALYs were an appropriate benefit measure for capturing the impact of the disease on the patients’ health, but no information on the derivation of the utility values was given.
Costs:
A broad perspective was adopted and a wide range of costs was included. Productivity losses were not considered, due to the age of the patients (65 years). The costs were not fully described; fracture costs were from other publications and the cost items were not reported. No resource quantities were reported. The authors stated that they used the wholesale price for medications, which might have overestimated their costs as discounts are often available to health plans and patients, but it might also have underestimated them for patients reinitiating treatment and switching to non-generic bisphosphonates. The impact of variations in selected inputs was tested in the sensitivity analyses. Details, such as the price year and discount rate, were reported.
Analysis and results:
An incremental approach was used to synthesise the costs and benefits of the two interventions. The uncertainty was satisfactorily investigated and the key results of the sensitivity analyses were presented and discussed. To validate the model results, projections of 10-year and lifetime fracture risks were compared with estimates from the published literature. The authors acknowledged some limitations to their analysis, which mainly related to the need for some key assumptions. The transferability of the results was not explicitly addressed and they might be difficult to transfer to other settings.
Concluding remarks:
The methods were robust and transparent, which ensures the validity of the authors’ conclusions, but the results were very dependent on several model parameters.