Ordinal relationship of composite time trade-off and discrete choice experiment preference data

OBJECTIVES: Commensurability between the preference data derived from composite time trade-off (cTTO) and discrete choice experiment (DCE) is unknown. There is considerable information asymmetry associated with cTTO and DCE data; the single comparable property common to both cTTO and DCE, testable at the individual respondent level, is limited to the directional preference of health state A over health state B. Thus, this study aimed to investigate the ordinal relationship between cTTO and DCE data generated by individual respondents. METHODS: A survey, using the EQ-VT platform, was conducted, where each participant answered a set of 10 cTTO and 12 DCE questions. Data was collected from members of the Swedish general public. Analyses focused on the agreement between implied ordinal rankings of pairs of states based on individual's cTTO values, and their explicit ranking when presented as a DCE pair. The observed disagreement was further compared with theoretical benchmarks estimated as the probability of choosing an unexpected state due to random errors under the Random Utility Theory. RESULTS: The analysis sample consisted of 226 interviews. Amongst all compared pairs, the total agreement was 65%; of the remaining 35%, the pairs with tied cTTO values made up 22%. Stratifying by differences in cTTO scores of pairs (Δ cTTO scores), agreements did not continue to increase with increasing Δ cTTO scores-staying around 90% for subgroups with larger Δ cTTO scores. Compared to benchmarks, divergences of two methods were present around pairs with tied values and those involving health states worse than dead. CONCLUSION: cTTO and DCE preference data each possesses their own attributes, shaped by each unique procedural element. Combining such preference data needs a clear rationale based on explicit grounds. The findings of the current study suggest that there exist some methodological divergences between cTTO and DCE. To be able to safely combine both data and to maximize the potential of the hybrid model, first these divergences should be carefully reviewed and justified.