BACKGROUND: Discrete-choice experiments (DCEs) are increasingly conducted to quantify risk tolerance by computing maximum acceptable risk (MAR) for improvements in efficacy or other benefits gained from new medical treatments. To compute MARs from DCE data, respondents are asked to make choices under uncertainty between treatments. Specific treatment-related harms are included in the choice questions as probabilistic adverse events (AEs), and choice variation with the probability of these outcomes is assumed to indicate their effect on the expected utility of treatments. With a limited number of comparisons between profiles, calculation of MARs requires understanding how outcome probabilities that are not explicitly considered in the DCE can change the value of medical technologies. This study aims to examine how various assumptions on the expected disutility of these excluded probabilities can result in different MAR measures.
METHODS: We summarize commonly used empirical specifications for the expected disutility of AEs and derive the resulting MAR functions. We then discuss an empirical application on treatments to delay bone metastases in oncology patients with solid tumors.
RESULTS: A total of 187 respondents completed the DCE. Results show the impact of making various assumptions about the expected disutility of AEs, and the resulting MAR values for specific health benefits. As expected, different assumptions resulted in variations in MAR values for specific health benefits. Even with small differences in MAR measures, our results suggest that the assumptions evaluated here can lead to different conclusions about the acceptability of a medical technology.
CONCLUSION: Results show possible systematic variations in MARs caused by the assumed form of the effect of changes in the probability of AEs. Furthermore, we find that different assumptions can lead to different conclusions about the acceptability of a medical technology, even when MAR distributions overlap. This result suggests that researchers should evaluate the assumptions they are making for these effects and use sensitivity analysis to evaluate the robustness of risk-tolerance measures from stated-preference data.