Psychometric Comparison of Patient Reported Summary Scores for EQ-5D (EQ-PRSM)

OBJECTIVES: EQ-5D is increasingly being used outside the context of health technology assessment (HTA), where there is no rationale for using value sets to summarize profile data. There is a need for a global, single summary measure based on self-reported data. Previously we developed EQ-5D-5L patient-reported summary scores (PRSM) based on dimension-specific rating scales (RS); and Item Response Theory (IRT) models. Here, we psychometrically compare five EQ-5D-5L PRSM scores.

METHODS: The following PRSM models were compared: RS additive (RSA), RS weighted (RSW), unidimensional Rasch (Rasch), unidimensional 2 parameter IRT (2PL) and multidimensional IRT (MIRT) including items from other instruments. The comparison also included the experience-based VAS value set for Sweden (EXP) and the level sum score (LSS ). Scores were compared in terms of distributional characteristics, convergent validity, discriminatory power, relative efficiency and responsiveness.

RESULTS: RSW and MIRT resulted in the smoothest distributions while EXP demonstrated a large upper gap. Distribution evenness was highest for MIRT, followed by 2PL, and lowest for RSA. Convergent validity was highest for RSW and EXP, and lowest for 2PL . Discriminatory power was highest for MIRT, followed by RSW, and lowest for 2PL and EXP. Relative efficiency showed that MIRT, RSW, Rasch and LSS were able to discriminate better than RSA, while RSW was superior to EXP, and MIRT was superior to RSW. Responsiveness results were mixed.

CONCLUSIONS: RSW and MIRT arguably performed best overall. 2PL performed poorly for mental health patients. EXP performed poorly in terms of relative efficiency and responsiveness. RS and IRT empirically performed superior to LSS; these approaches may also be preferred based on theoretical and conceptual grounds. MIRT however is based on external items and on arbitrary choices. This study proves the usefulness of using PRSM for EQ-5D-5L beyond LSS, for use in population health and health systems applications.