3H-5 PREDICTING UTILITY SCORES FOR MYELOFIBROSIS PATIENTS: MAPPING THE MYELOFIBROSIS SYMPTOM ASSESSMENT FORM AND MYELOPROLIFERATIVE SYMPTOM ASSESSMENT FORM TO THE EUROQOL-5D

Tuesday, October 20, 2015: 11:30 AM
Grand Ballroom B (Hyatt Regency St. Louis at the Arch)

Chang Ho Lee, BSc1, Karen E Bremner, BSc2, Paul Grootendorst, PhD1, Jolie Ringash, MD, MSc3, Murray D Krahn, MD, MSc, FRCPC4 and Nicholas Mitsakakis, MSc PhD4, (1)University of Toronto, Toronto, ON, Canada, (2)University Health Network, Toronto, ON, Canada, (3)University of Toronto and the Princess Margaret Cancer Centre (PMCC), Toronto, ON, Canada, (4)Toronto Health Economics and Technology Assessment (THETA) Collaborative, University of Toronto, Toronto, ON, Canada
Purpose:

   Health utility is a preference-based measure of quality of life that informs healthcare resource allocation decisions. Utilities are commonly measured using generic instruments, such as the EuroQol-5D (EQ-5D), that are applicable to diverse disease areas.  The quality of life of patients with myleofibrosis, however, is often measured using disease-specific profile instruments, including the Myelofibrosis Symptom Assessment Form (MF-SAF) and Myeloproliferative Symptom Assessment Form (MPN-SAF).  In order to estimate utilities and compare the quality of life of patients with myleofibrosis to patients with other conditions, we mapped the MFSAF and MPN-SAF instruments to EQ-5D utility scores.

Method:

   174 patients with myelofibrosis completed the MF-SAF, MPN-SAF and EQ-5D by online survey and mail. Patients were recruited from English-speaking countries (Canada, US, UK, and Australia) through a local myelofibrosis patient support group and websites of national and international support groups. We fitted two linear regression models for each myelofibrosis instrument, one using the original EQ-5D scores as a response variable and one which applied the Box-Cox power transformation to EQ-5D scores. Models were selected using stepwise selection and Akaike Information Criterion (AIC) and adjusted R2 criterion approach. Bootstrap was used to validate and assess any overfitting in the models. The predictive accuracy of the models was assessed with Root Mean Square Error (RMSE). Spearman Correlation between the predicted and observed utility scores was calculated to measure predictive ability.   

Result:

   The best-fitting model for both the MF-SAF and MPN-SAF used the power of two-transformed EQ-5D utility scores. Prior to transformation, a small constant (of 0.466) needed to be added to the utility scores to ensure their value was positive. The R2 values were 0.40 and 0.55 for the MF-SAF and MPN-SAF, respectively, and the RMSE were small at 0.46 (MF-SAF) and 0.39 (MPN-SAF)

Conclusion:

   Our mapping algorithm predicts EQ-5D scores from MF-SAF and MPN-SAF scores, thus allowing estimation of utilities for groups of myelofibrosis patients who completed only the MF-SAF or MPN-SAF in clinical studies.  However, as in previous mapping studies, our model is less accurate in predicting utilities for individual patients. We are exploring various statistical techniques with the goal of improving the accuracy of models to map utilities from descriptive instruments. Our next step will be to create a mapping algorithm using different types of regression.