OBJECTIVES:

Utility values are used in health technology assessment to measure the health-related quality of life impacts of new products, typically taken from a single source. In contrast, meta-analysis synthesizing all available information is a requirement for clinical data. Consequently, utility values currently used to inform health technology assessments lack consistency by ignoring data from the other sources. This analysis presents a standard frequentist meta-regression and a novel and Bayesian approach to the synthesis of utility values.

METHODS:

A literature review for all published utility data in multiple myeloma was conducted, in conjunction with analysis of patient registries across all stages of disease (2,445 patients, over 9,000 completed EQ-5D questionnaires), and of a clinical trial including 669 patients. This information was then synthesized using two distinct approaches – frequentist meta-regression and Bayesian statistical modelling. These approaches were compared in terms of the results produced, internal validity, and efficiency of estimation.

RESULTS:

The systematic review identified 13 papers giving 27 utility values across multiple lines of treatment including some values not linked to a specific disease stage. Analysis of the two datasets produced 9 further values. Both frequentist and Bayesian meta-regression produced similar overall results; low utility on diagnosis (0.53), increasing to approximately 0.65 on first treatment then decreasing with each subsequent treatment class to approximately 0.50 after four classes of treatment. In all analyses, strong evidence was found to suggest an association between stem cell transplant and an increase of 0.06 in patient utility.

CONCLUSIONS:

Both Bayesian and frequentist approaches produced internally consistent utility estimates across the treatment pathway. However, the Bayesian approach more accurately represents the uncertainty in the clinical data, and allows non stage specific utilities to be used as prior beliefs. This exemplifies how Bayesian analyses can be performed using a simple and flexible framework.