OBJECTIVES: To (a) provide a comprehensive account of the various sources of uncertainty affecting HRQoL 'value sets'; (b) identify methods which have been explored to account for uncertainty, and (c) highlight gaps in methods and reporting, and their implications for associated analyses such as the estimation of Quality Adjusted Life Years (QALYs) used in cost-effectiveness analysis (CEA).

METHODS: Uncertainty affecting HRQoL values were classified by source (e.g. sampling, study design, stated preference methods, analysis and modelling value sets). We describe how values combine with health state descriptions to create uncertainty in QALY estimates. A scoping review was undertaken to establish what methods have been developed to handle uncertainty around HRQoL value sets. We identify a standard method for calculating standard errors for predicted values that is rarely applied to the statistical models used to identify the value sets for instruments.

RESULTS: Many factors drive uncertainty in HRQoL values, including choices about stated preference methods and study design, sampling procedures, respondent and interviewer effects, and approaches taken to modelling values. Just nine studies were identified that attempt to quantify uncertainty around modelled HRQoL value sets; all focus on statistical uncertainty around estimated parameters.

CONCLUSIONS: Despite 50 years of research on HRQoL valuation, the associated uncertainty around values has been largely ignored and remains under-reported. This conveys a false sense of precision around the values reported for instruments such as EQ-5D, and means that uncertainty is not adequately taken into account in QALY calculations and CEA. Few papers report methods for addressing uncertainty around HRQoL value sets; these tend to focus specifically on statistical uncertainty arising from modelling. Many of the other sources of uncertainty we identified remain under-explored. We present one method for reporting standard errors around modelled health state values, using information readily generated for existing value sets, that should become standard practice.