BACKGROUND: Utility scores for use in cost-utility analysis may be imputed from the SF-36 health instrument using various techniques, typically regression analysis. This paper explored imputation using partial credit Rasch analysis. METHOD: Data from the Assessment of Quality of Life (AQoL) instrument validation study were re-analysed (n = 996 inpatients, outpatients and a community sample). For each AQoL item, factor analysis identified those SF-36 items forming a unidimensional scale. Rasch analysis located scale logit scores for these SF-36 items. The logit scores were used to assign AQoL item scores. The standard AQoL scoring algorithm was then applied to obtain the utility scores. RESULTS: Many SF-36 items were limited predictors of AQoL items; some items from both instruments obtained disordered thresholds. All imputed scores were consistent with the AQoL model and fell within AQoL score boundaries. The explained variance between imputed and true AQoL scores was 61%. DISCUSSION: Rasch-imputed mapping, unlike many regression-based algorithms, produced results consistent with the axioms of utility measurement, while the proportion of explained variance was similar to regression-based modelling. Item properties on both instruments implied that some items should be revised using Rasch analysis. The methods and results may be used by researchers needing to impute utility scores from SF-36 health scores.
BACKGROUND: Utility scores for use in cost-utility analysis may be imputed from the SF-36 health instrument using various techniques, typically regression analysis. This paper explored imputation using partial credit Rasch analysis. METHOD: Data from the Assessment of Quality of Life (AQoL) instrument validation study were re-analysed (n = 996 inpatients, outpatients and a community sample). For each AQoL item, factor analysis identified those SF-36 items forming a unidimensional scale. Rasch analysis located scale logit scores for these SF-36 items. The logit scores were used to assign AQoL item scores. The standard AQoL scoring algorithm was then applied to obtain the utility scores. RESULTS: Many SF-36 items were limited predictors of AQoL items; some items from both instruments obtained disordered thresholds. All imputed scores were consistent with the AQoL model and fell within AQoL score boundaries. The explained variance between imputed and true AQoL scores was 61%. DISCUSSION: Rasch-imputed mapping, unlike many regression-based algorithms, produced results consistent with the axioms of utility measurement, while the proportion of explained variance was similar to regression-based modelling. Item properties on both instruments implied that some items should be revised using Rasch analysis. The methods and results may be used by researchers needing to impute utility scores from SF-36 health scores.
Authors: A E Raczek; J E Ware; J B Bjorner; B Gandek; S M Haley; N K Aaronson; G Apolone; P Bech; J E Brazier; M Bullinger; M Sullivan Journal: J Clin Epidemiol Date: 1998-11 Impact factor: 6.437
Authors: Martijn B A van der Wal; Wim E Tuinebreijer; Monica C T Bloemen; Pauline D H M Verhaegen; Esther Middelkoop; Paul P M van Zuijlen Journal: Qual Life Res Date: 2011-05-20 Impact factor: 4.147
Authors: Benjamin M Davies; Maire McHugh; Ali Elgheriani; Angelos G Kolias; Lindsay A Tetreault; Peter J A Hutchinson; Michael G Fehlings; Mark R N Kotter Journal: PLoS One Date: 2016-08-02 Impact factor: 3.240