OBJECTIVES:

A core element of considering cost-effectiveness is the clinical efficacy or effectiveness of an intervention relative to an appropriate comparator. In the absence of a statistically significant difference in clinical effect, guidelines generally suggest the use of a cost-minimisation analysis. However, it is apparent that economic evaluations are carried out and published in cases where there is no statistically significant effect. This study investigated the context of a sample of such studies to identify circumstances in which they may be appropriate.

METHODS:

As part of a review of chronic disease self-management support interventions, systematic reviews of cost-effectiveness studies were carried out for a range of indications. We included studies that carried out cost-effectiveness analyses when no statistically significant effect had been demonstrated. A narrative review approach was adopted.

RESULTS:

We identified 16 published economic evaluations for inclusion. These studies were typically trial-based studies or simulation models with effectiveness data from a single small trial that was possibly under-powered to detect a treatment effect. The absence of a statistically significant effect was not always acknowledged as a limitation in the analyses or as an issue to be considered when interpreting the results.

However, a cost-effectiveness analysis in the absence of a demonstrated treatment effect may help to identify potentially cost-effective treatments for which additional evidence on effect would support decision making.

CONCLUSIONS:

The point estimate of cost-effectiveness is strongly influenced by the mean estimate of clinical effect. Typical methods for conveying parameter uncertainty may not adequately highlight the fact that the intervention has not demonstrated an improvement in outcomes. Cost-effectiveness analyses carried out in the absence of a demonstrated clinical benefit should be treated purely as exploratory analyses, and only considered as a basis for seeking additional data.