Mapping the Alzheimer’s Disease Cooperative Study-Activities of Daily Living Inventory to the Health Utility Index Mark III

Yin Bun Cheung, Hui Xing Tan, Vivian Wei Wang, Nagaendran Kandiah, Nan Luo, Gerald C. H. Koh, Hwee Lin Wee

To map the Alzheimer's Disease Cooperative Study-Activities of Daily Living Inventory (ADCS-ADL) to the Health Utility Index Mark III (HUI3) in people living with dementia (PWD) and to compare the performance of five methods for mapping. A cross-sectional study of 346 dyads of community-dwelling PWD and family caregiver was carried out in Singapore. ADCS-ADL and HUI3 were rated by the family caregivers. Disease severity ratings and Mini Mental State Examination (MMSE) results were retrieved from medical records. A recently proposed mapping method called the Mean Rank Method (MRM) was described and applied, and the results were compared with regression-based mapping, including ordinary least squares, censored least absolute deviation (CLAD), Tobit and response mapping. The MRM produced a mapped utility distribution that closely resembled the observed utility distribution. The standard deviations (SDs) of the observed and MRM-mapped utility were both 0.340, whereas the SDs of the other mapped utilities ranged from 0.243 (response mapping) to 0.283 (CLAD). Regressing the MRM- and CLAD-mapped and observed utility values upon disease severity and MMSE gave similar regression lines (each P > 0.05). Regressing the other mapped utility values upon the covariates under- (over-) estimated the utility of good (poor) clinical states. However, regression-based mapping methods gave a better fit at the individual level, as measured by root mean square error, mean absolute error and R2. K fold cross-validation gave similar results. The MRM is accurate at the group level. The regression-based mapping methods are more accurate for making individual-level prediction. In addition, CLAD also performed reasonably well at the group level.