This presentation is part of: Oral Concurrent Session A - Patient and Physician Behavior/Preferences 1
CAN WE BRIDGE FURTHER? EVIDENCE OF LIMITED EXTERNAL VALIDITY OF MODELS MAPPING SF-36 TO HEALTH UTILITY
Andrew P. Yu, MA, Yanni F. Yu, MA, and Michael B Nichol, PhD. University of Southern California, Pharmaceutical Economics and Policy, Los Angeles, CA
Purpose: Many studies have attempted to develop a model mapping health status to utility score based on a sample of population where both measures were obtained. However, evidence have shown different models yielded dramatically different estimated utilities. Can this disagreement be partially explained by differences in characteristics of populations where the models were generated? External validity (generalizability) of the model can be limited when subscale coefficients vary with patient characteristics. The purpose of this study is to examine whether a regression model that bridges from health status to utility is sensitive to patient characteristics. Methods: Data included 6923 Southern California Kaiser Permanente members who filled both SF-36 and HUI2 in year 1994-1995. Missing item responses were imputed by MCMC and propensity score method with a missing at random assumption. The linear regression framework mapping from SF-36 subscales to HUI2 utility was used as a base model, which also adjusted for age and gender. Interaction effect between subscales and each of the following ten patient factors was examined: ethnicity, work status, marital status, income level, education, chronic disease score (CDS), diabetes, depression, COPD/asthma and cardiovascular diseases (CVD). To evaluate the individual characteristic effect, these variables were examined separately. Analysis of variance (F-test) was used to test the significance of all interactions. Results: Every patient factor examined in this study had significant interaction effect with at least one SF-36 subscale. The effects of some subscales on utility were more likely to be influenced by patient factors. The coefficient of body pain on utility were significantly associated with marital status, education, income and COPD/asthma; role limitation (emotional) with marital status, education, CDS and CVD; physical functioning with work status, CDS, diabetes and depression; mental health with income and ethnicity; role limitation (physical) with COPD/asthma. When all the patient factors were presented in a model allowing full interaction, interactions were highly significant (F=1.57, d.f.=290, p<0.0001). Conclusion: This study cast doubts on the existence of a reliable mapping model that can be universally applied to any patient sample. In this study, the mapping model appeared to be strongly influenced by respondent’s demographic attributes, which could limit external validity of the model. Therefore, such regression models may not be appropriate to estimate utility in different populations without further adjustment.
