PMS67 EXPERIENCE AND PREFERENCE-BASED EQ-5D-3L VALUE SET DERIVED FROM LONGITUDINAL DATA: RESULTS FROM THE INTERNATIONAL COSTS AND UTILITIES OF OSTEOPOROTIC FRACTURES STUDY (ICUROS)

Abstract

Own health state utility values (HSUV, “experience-based utility”) is important for economic valuation and other purposes. There is only one preference-based EQ-5D value set derived from direct patient experience. Furthermore, the impact of interactions between EQ-5D dimensions on HSUV is poorly understood. Therefore we constructed a value set using direct patient experience and explored the impact of interactions between health states. The ICUROS was a multinational prospective observational study on the consequences of fragility fracture. In the study, patients who sustained a fragility fracture completed EQ-5D-3L and Time-trade-off (TTO) questionnaires for five time points: before fracture (recall), within two weeks after fracture, and at 4, 12 and 18 months after fracture. We derived a value set by fitting a random intercept linear model with TTO as the dependent variable and ‘moderate’ (Level 2, “L2”) or ‘severe’ impairment (Level 3, “L3”) in each EQ-5D dimension as dummy variables. We also explored the potential impact of interactions between impairment in dimensions using rigorous Lasso regression with cluster robust standard errors. In total 4,873 patients reported 17,779 EQ-5D and TTO pairs. Mean (SD) age at study enrolment was 71 (11) years and 81% were women. In the random intercept model, all coefficients had the expected sign and were statistically significant (p<0.001). The resulting value set was: 0.92 - 0.03xMobillityL2 - 0.08xMobillityL3 - 0.12xSelf-careL2 - 0.14xSelf-careL3 - 0.02xUsual ActivitiesL2 - 0.03xActivitiesL3 - 0.02xPain/DiscomfortL2 - 0.10xPain/DiscomfortL3 - 0.05xAnxiety/DepressionL2 ´- 0.11xAnxiety/DepressionL3. The Lasso regression indicated that interactions between mobility and self-care, anxiety/depression and self-care, and anxiety/depression and usual activities may be important. This experience-based EQ-5D-3L value set had consistently ordered and statistically significant coefficients, demonstrating face validity. Furthermore, the results show that explicit modelling of interactions may improve the predictive power of EQ-5D value sets. Compared to value sets based on hypothetical health, the decrements associated with impaired health were generally smaller.