Objectives: When incorporating treatment effect estimates derived from a random-effect meta-analysis it is tempting to use the confidence bounds to determine the potential range of treatment effect. However, prediction intervals reflect the potential effect of a technology rather than the more narrowly defined average treatment effect. Using a case study of robot-assisted radical prostatectomy, this study investigates the impact on a cost-utility analysis of using clinical effectiveness derived from random-effects meta-analyses presented as confidence bounds and prediction intervals, respectively.

Methods: To determine the cost-utility of robot-assisted prostatectomy, an economic model was developed. The clinical effectiveness of robot-assisted surgery compared with open and conventional laparoscopic surgery was estimated using meta-analysis of peer-reviewed publications. Assuming treatment effect would vary across studies due to both sampling variability and differences between surgical teams, random-effects meta-analysis was used to pool effect estimates.

Results: Using the confidence bounds approach the mean and median ICER was €24,193 and €26,731/QALY (95%CI: €13,752 to €68,861/QALY), respectively. The prediction interval approach produced an equivalent mean and median ICER of €26,920 and €26,643/QALY (95%CI: -€135,244 to €239,166/QALY), respectively. Using prediction intervals, there is a probability of 0.042 that robot-assisted surgery will result in a net reduction in QALYs.

Conclusions: Using prediction intervals rather than confidence bounds does not affect the point estimate of the treatment effect. In meta-analyses with significant heterogeneity, the use of prediction intervals will produce wider ranges of treatment effect, and hence result in greater uncertainty, but a better reflection of the effect of the technology.