OBJECTIVES:

Health state utility values (HSUVs) are pivotal inputs to health-state defined economic models informing most Health Technology Assessments (HTAs), and thus adoption decisions for novel interventions. Where there are multiple potential candidate HSUVs typically a single value is used, implicitly weighting other candidate HSUVs at 0, and occasionally HSUVs meta-analysed.

METHODS:

Using three case studies in lung cancer, liver cirrhosis, and diabetic retinopathy blindness, a Bayesian ‘Power Prior’ (BPP) approach was implemented to incorporate beliefs on study applicability. The BPP is a technique used in clinical research to allow historical data to supplement contemporary control data in clinical studies. In this implementation older studies, non-UK value sets, and vignette studies were downweighted (but not entirely disregarded). Power prior HSUV estimates were compared with using a single preferred value (SPV), fixed effects meta-analysis (FEMA), and random effects meta-analysis (REMA).

RESULTS:

In the lung cancer example (1 contemporary study and 5 historical HSUVs), the BPP reduced uncertainty (HSUV of 0.74 95% Credible Interval, CrI: 0.73-0.75) compared to the use of only contemporary data (0.75, 95% Confidence Interval, CI: 0.72-0.78) without being affected by studies of low relevance. In cirrhosis, synthesizing 9 HSUVs with low relevance, the BPP reflected the uncertainty in the applicability of the evidence base (BPP 0.71, CrI 0.66-0.75). In diabetic retinopathy blindness, the effect of the BPP was to allow inclusion of multiple studies with lower estimates, reducing the estimate (BPP 0.39, SPV 0.61, FEMA 0.40, REMA 0.48).

CONCLUSIONS:

The concept of the BPP can be adapted for meta-analysing HSUVs, leading to and corresponding implications for economic model results. In particular, downweighting shows wider CI/CrIs relative to other meta-analysis methods but more precise estimates than using a SPV. The implementation of BPP provides a practical mechanism to explicitly incorporate perceived relevance to a decision problem.