DO INDIVIDUALS' UTILITIES FOR DIVERSE HEALTH STATES TEND TO HAVE A CONSISTENT RANK WITHIN A POPULATION?

Purpose: To assess the consistency of the rank of individuals’ utilities for diverse health states among their peers’ responses.                                                                                     

Methods: We evaluated the stability of the rank order of respondents’ stated preferences, compared to other respondents, across 5 health states. We elicited time trade-off (TTO) values for monocular blindness, deafness, binocular blindness, deafness with binocular blindness, and current shortness of breath (dyspnea) among 314 subjects having diseases associated with dyspnea.  All subjects reported minimal current dyspnea symptoms. Respondents anticipated and received a monetary bonus for accurately interpreting survival curves and for passing undisclosed internal consistency checks, such as binocular blindness having lower utility than monocular blindness.  The top 20% of performers were identified and analyzed with and without the other 80%. We calculated the percentile of each respondent’s TTO for each health state, the mean and standard deviation of the 5 percentiles for each respondent, and pairwise-correlations between percentiles for the 5 states. We generated uncorrelated “control” data by independently permuting the original TTO scores for each of the 5 states. We repeated this ten times and the resulting control subjects were pooled before calculating percentiles. The mean and standard deviation for the percentiles were calculated for each control subject.

Results: Median and interquartile range [IQR] TTO values in the real data were: Monocular blindness, 0.946 [0.825-0.992]; Deafness, 0.825 [0.556-0.945]; Binocular blindness, 0.600 [0.300-0.900]; Deaf-blindness, 0.450 [0.100-0.725]; Current dyspnea, 0.940 [0.733-0.996]. Correlations ranged from 0.206 (Deaf-Blindness vs Current dyspnea) to 0.608 (Binocular blindness vs Deaf-blindness). The distribution of percentile means was significantly less centralized for real data than for control data (48.754 [33.415 – 66.585] versus 50.382 [41.370 – 59.013], p<0.001). The standard deviation in percentiles was greater in the control data than the real data, with 91% of the permuted subjects having standard deviations greater than the real data median (p<0.001). Percentile distributions for top performers and all real subjects were similar.

Conclusions: The percentiles of these individual’s utilities for these states was relatively stable, across all percentiles. Whenever this pattern holds, the percentiles of an individual’s utilities for familiar states could provide a first-order estimate of that individual’s placement within a distribution of utilities elicited for other states. This could facilitate utility estimates for scenarios involving unfamiliar health states and multimorbidity.