To the Editor: Segal et al propose a new method — “transfer to utility” (TTU) — to convert clinical and quality-of-life (QOL) trial-based outcome measures to a common metric — quality-adjusted life-years (QALYs) — to compare the cost-effectiveness of different interventions.1 TTU uses regression to map clinical and QOL instruments to a “utility-equivalent scale” that aims to measure strength of preference for health outcomes. In their example, several instruments were administered to osteoarthritis patients. Australian Assessment of Quality of Life (AQoL) scores were regressed on SF-36 subscale scores and QALYs were generated from published SF-36 results by means of a conversion algorithm based on these parameter estimates. The approach is novel, but we question its validity.

Utility and health-related quality of life (HRQOL) are fundamentally different concepts. HRQOL is a standardised multi-dimensional, ordinal measure of the individual’s perception of how disease and treatment affect physical, social and emotional functioning. Measuring utility requires a critical next step: capturing strength of preference for outcomes in a unidimensional interval scale. HRQOL and utility scales have very different interpretations. Any appearance of similarity is superficial. The algebra of scale conversion is easy, but conceptually problematic.

TTU involves a complex trail of estimation and prediction. Ordinal SF-36 subscale scores are used as arguments in regression, imposing interval properties. Statistically valid interpretation of the resulting parameters requires dummy coding. The dependent variable is the AQoL score generated by applying the AQoL algorithm, developed in previous research using a different population2 to respondents’ surveys. The new algorithm from the TTU regression parameters is used to convert published trial-based average SF-36 subscale scores to AQoL scores. Misspecification errors are built into the predicted AQoL scores. Information about variability in outcomes and preferences in intermediate measures, and hence uncertainty around point estimates of predicted AQoL scores, is suppressed.

How plausible are these results? Segal et al report an estimated “utility gain” 12 months after hip surgery of 0.304 (from 0.464 before surgery to 0.767 at 12 months).1 The interpretation is that an average patient undergoing hip surgery for osteoarthritis would be willing to forgo about 40% of their remaining life span for the quality-of-life improvement the surgery would provide.

Apparently, sophisticated quantitative approaches cannot add information about factors not measured in trials. The validity of the approach stands or falls on appropriate statistical methods, data quality and interpretable results. TTU potentially introduces bias, suppresses information relevant to decision-making and may lead decision-makers to place undue trust in point estimates based on heroic assumptions.