Assessing the performance of utility techniques in the absence of a gold standard

BACKGROUND. Utility techniques are the most commonly used means to assess p atient preferences for health outcomes. However, whether utility techniques produce valid measures of preference has been difficult to determine in th e absence of a gold standard. OBJECTIVE TO introduce and demonstrate two methods that can be used to eval uate how well utility techniques measure patients' preferences. SUBJECTS AND DESIGN. Patients treated for advanced prostate cancer (n = 57) first ranked eight health states in order of preference. Four utility tech niques were then used to elicit patients' utilities for each health state. MEASURES. The rating scale, standard gamble, time trade-off, and a modified version of willingness-to-pay techniques were used to elicit patients' uti lities. Technique performance was assessed by computing a differentiation a nd inconsistency score for each technique. RESULTS. Differentiation scores indicated the rating scale permitted respon dents to assign unique utility values to about 70% of the health states tha t should have received unique values. When the other techniques were used, about 40% or less of the health states that should have received unique uti lity scores actually did receive unique utility scores. Inconsistency score s, which indicate how often participants assign utility scores that contrad ict how they value health states, indicated that the willingness-to-pay tec hnique produced the lowest rate of inconsistency (10%). However, this techn ique did not differ significantly from the rating scale or standard gamble on this dimension. CONCLUSIONS. Differentiation and inconsistency offer a means to evaluate th e performance of utility techniques, thereby allowing investigators to dete rmine the extent to which utilities they have elicited for a given decision problem are valid. In the current investigation, the differentiation and i nconsistency methods indicated that all four techniques performed at sub-op timal levels, though the rating scale out-performed the standard gamble, ti me trade-off, and willingness-to-pay techniques.