OBJECTIVES: Discrete-choice experiments (DCE) in health often assume the studied alternatives are perfectly substitutable. This may be inappropriate when evaluating treatment options that represent fundamentally different interventions, and can potentially lead to biased estimates of the value of medical technologies. We evaluated the impact of violating this assumption and the effectiveness of options commonly used to correct the issue.

METHODS: Adult members of a US online panel completed a web-enabled DCE survey (N=811) designed to elicit preferences for reducing the risk of future health problems with surgery or medication (set_1), only surgery (set_2), or only medication (set_3). Respondents were asked to imagine they had an unspecified gene variant associated with a health problem. Severity and likelihood of the health problem was varied experimentally across respondents. Three random-parameters logit (RPL) models were used to estimate preferences when respondents evaluated set_1, set_2, and set_3. Odds of choosing surgery/medication relative to watchful waiting were calculated for the three modeling groups, and compensating variation (CV) was calculated to measure the expected benefit of making an intervention available to respondents.

RESULTS: Odds of choosing surgery (0.54) and medication (0.68) in set_2 and set_3 were not statistically different from odds for interventions in set_1 (0.77 surgery, and 0.69 medication). CV for interventions varied with the likelihood and severity of the health problems and were different when treatment types were presented alone and together. For example, CV ranged between $15 and $99 when considering only medications, and between $8 and $12 when surgery and medications were presented together. Differences in ranges were statistically significant.

CONCLUSIONS: Although lack of statistically-significant differences in the odds of choices suggest that surgeries/medications may be excluded without compromising the valuation of medications/surgeries, having only one type of intervention in the DCE questions was associated with greater CV. None of usual solutions resolved the difference in CV.