ERROR STANDARD DEVIATIONS OF FIVE HEALTH UTILITY INDEXES ACROSS THE RANGE OF HEALTH: IMPLICATIONS FOR ASSESSING RESPONSIVENESS

PURPOSE : We investigate whether commonly used health-related quality-of-life indexes differ with regard to measurement error and responsiveness to changes in health across their ranges from poor to good health.

METHODS: HUI2, HUI3, EQ-5D, SF-6D and QWB-SA were administered in a cohort of patients at 1 and 6 months post cataract surgery (n=181), and to a population-based sample of non-institutionalized adults in the United States (n= 3557).  Error standard deviations (ESDs) were estimated for each index from the repeated measures in the cohort. A two step modeling process utilizing the availability of 5 indexes was applied to the population sample to estimate the ESDs. Following categorical factor analysis of the 5 indexes  to obtain a combined measure of “health”, the mean and ESD of each index were modeled  across this measure. Graphs were produced showing index responsiveness at different levels of health on original preference scored scales and relative to ESD.

RESULTS: Repeated measures led to estimated ESDs  0.15,0.11, 0.10, 0.071 and 0.081, respectively. Population-based approximate ESDs were 0.070  for SF-6D and 0.087 for QWB-SA, but were smaller than repeated measures based ESDs for the HUI2, HUI3 and EQ-5D. The ESD varied considerably across index levels for HUI2, HUI3 and EQ-5D, but not much for SF-6D and QWB-SA . Estimated ESDs from the population sample showed the ESDs for HUI2, HUI3 and EQ-5D to be particularly high for health levels 1-2 population standard deviations below average, while the ESD for SF-6D was highest around average health, and the ESD of QWB-SA was fairly constant.  Responsiveness on the utility scale was somewhat parallel to the trend in ESD for each index. This resulted in responsiveness relative to standard deviation being fairly constant, but low, for the SF-6D, EQ-5D and QWB-SA, and high but declining across health levels below average for the HUI2 and HUI3. Responsiveness was low for all indexes at levels of health above average.

CONCLUSIONS: Error standard deviations differ between indexes and across health. The differences in error standard deviation lead to different conclusions as to the health range where an index is most responsive dependent on whether responsiveness is judged on the original or error standard deviation scale. Low responsiveness above average health may reflect true health utility or lack of index sensitivity.