Can we derive an 'exchange rate' between descriptive and preference-based outcome measures for stroke? Results from the transfer to utility (TTU) technique

Abstract

BACKGROUND: Stroke-specific outcome measures and descriptive measures of health-related quality of life (HRQoL) are unsuitable for informing decision-makers of the broader consequences of increasing or decreasing funding for stroke interventions. The quality-adjusted life year (QALY) provides a common metric for comparing interventions over multiple dimensions of HRQoL and mortality differentials. There are, however, many circumstances when--because of timing, lack of foresight or cost considerations--only stroke-specific or descriptive measures of health status are available and some indirect means of obtaining QALY-weights becomes necessary. In such circumstances, the use of regression-based transformations or mappings can circumvent the failure to elicit QALY-weights by allowing predicted weights to proxy for observed weights. This regression-based approach has been dubbed 'Transfer to Utility' (TTU) regression. The purpose of the present study is to demonstrate the feasibility and value of TTU regression in stroke by deriving transformations or mappings from stroke-specific and generic but descriptive measures of health status to a generic preference-based measure of HRQoL in a sample of Australians with a diagnosis of acute stroke. Findings will quantify the additional error associated with the use of condition-specific to generic transformations in stroke. METHODS: We used TTU regression to derive empirical transformations from three commonly used descriptive measures of health status for stroke (NIHSS, Barthel and SF-36) to a preference-based measure (AQoL) suitable for attaching QALY-weights to stroke disease states; based on 2570 observations drawn from a sample of 859 patients with stroke. RESULTS: Transformations from the SF-36 to the AQoL explained up to 71.5% of variation in observed AQoL scores. Differences between mean predicted and mean observed AQoL scores from the 'severity-specific' item- and subscale-based SF-36 algorithms and from the 'moderate to severe' index- and item-based Barthel algorithm were neither clinically nor statistically significant when 'low severity' SF-36 transformations were used to predict AQoL scores for patients in the NIHSS = 0 and NIHSS = 1-5 subgroups and when 'moderate to severe severity' transformations were used to predict AQoL scores for patients in the NIHSS > or = 6 subgroup. In contrast, the difference between mean predicted and mean observed AQoL scores from the NIHSS algorithms and from the 'low severity' Barthel algorithms reached levels that could mask minimally important differences on the AQoL scale. CONCLUSION: While our NIHSS to AQoL transformations proved unsuitable for most applications, our findings demonstrate that stroke-relevant outcome measures such as the SF-36 and Barthel Index can be adequately transformed to preference-based measures for the purposes of economic evaluation.