OBJECTIVES: Patient Reported Outcome Measures (PROMs) are gaining attention as healthcare system funders increasingly seek value based care. One instrument used to collect PROMs is the Patient-Reported Outcomes Measurement Information System (PROMIS) tool. While PROMIS is used in healthcare systems around the world (including Australia), its results cannot be used to estimate utilities, making it less relevant for economic evaluations. Mapping PROMIS to a multi-attribute utility instrument (MAUI) enables estimation of utilities. Previous studies have mapped PROMIS to EQ-5D-3L, but not to the new EQ-5D-5L , this study aims :1) To map the PROMIS Global 10 to EQ-5D-5L. 2) To use a non-parametric methodology based on machine learning, Classification and Regression Tree (CART) analysis, to map these two instruments and compare its accuracy to traditional methods of mapping

METHODS: An online survey was conducted to collect responses to PROMIS Global 10 and EQ‑5D‑5L from the Australian general population (N=2,032). This analysis first employed a recently developed Australian algorithm to compute utilities and then mapped PROMIS Global 10 results to EQ-5D-5L using CART. This was compared to using linear regression, Tobit, generalised linear model (GLM) and censored regression model (CLAD). The robustness of the analysis was assessed using a range of statistical tests.

RESULTS: Among all the models considered, the CART resulted in predicting the most accurate utilities and lowest MAE, RMSE values. Moreover CART was more accurate in predicting lower utilities.

CONCLUSIONS: The proposed mapping algorithm can be used to predict utilities from PROMIS Global 10 data. Furthermore, this study explored a new approach to mapping, which has not been previously applied. The key strength of CART is its flexibility in terms of pre-specifying the estimation model. CART is a non-parametric method which can handle highly skewed data and does not need model specifications as with traditional regression models.