OBJECTIVES: Health-related quality-of-life, often reported as utility scores, is an important endpoint in the evaluation of healthcare interventions. Yet the analysis of utility data is complicated by its distributional properties, which raise numerous statistical challenges. As there is no consensus on the most appropriate regression for analyzing utility data, the aim of this paper was to develop a framework to evaluate common regression methods.

METHODS: We reviewed electronic databases (e.g., PubMed) and websites of HTA agencies to identify the most used regression approaches for analyzing utility data. A conceptual framework was then created to illustrate the ability of each regression model to handle the unique distributional features of utility data (i.e., skewed, multimodal, bounded, heteroscedastic, time dependent, with individual or group effects). Using this framework, we examined the strengths and limitations of each approach.

RESULTS: Conventional methods, such as ordinary least squares regression, are widely used to analyze utility data despite violation of key assumptions surrounding normality and independence. While there are advantages in using more complex models, such as mixed effects models, this comes at the cost of untestable assumptions. Some regression assumptions are more robust to violation (e.g., normality of observations) while others are managed at the study design-level (e.g., independence of observations). When selecting a regression approach, ensuring a balance between feasibility, interpretability and statistical correctness is critical.

CONCLUSIONS: Health technology assessment is a process of making statistical inference from clinical, health-related quality of life, and economic data so that decision-makers can assess the value of new technologies. Inappropriate statistical analysis can result in unreliable estimates of cost-effectiveness that fail to provide accurate and robust information to inform resource allocation decisions. This framework provides an overview of the ability of common regression models to analyze utility data, with the aim of supporting evidence-based decision-making.