OBJECTIVES: Traditional stated-preference models with fixed effects assume that individuals behave similarly. However, empirical evidence has shown that individuals’ preferences are often diverse. Hierarchical Bayesian models that include random effects provide individual-specific utilities to account for heterogeneity. This research studies oncologists’ choices about various pharmaceutical therapies for patients who have metastatic breast cancer. METHODS: In this discrete choice experiment conducted in Lima, Peru, each of 113 oncologists was presented with 11 choice tasks (each consisting of four scenarios of therapies plus the NONE option) and asked to pick the best choice. The attributes included Treatment Scheme, Patient Recovery Status, Treatment Length, Cost, and Risk Factors. Hierarchical Bayesian methods were used in this multinomial logit conjoint analysis to account for heterogeneity in preferences. RESULTS: Treatment Scheme, Recovery Status, and Risk Factors showed impact on the choices. On average, treatments with shorter periods of follow-up medication were preferred, and these oncologists tended to choose therapies that would have a better recovery status (0.19 with a 95% HPD credible interval [0.06, 0.33]). More importantly, Risk Factors had a large influence: the utility estimates of all risk factors were all negative (cardiovascular disease –1.21 [–1.56, –0.86], thromboembolism –1.45 [–1.81, –1.11], arterial hypertension –1.44 [–1.78, –1.11]). Cost did not play a role, probably because the respondents were doctors (not patients) and the study dealt with metastatic breast cancer. Several entries in the covariance matrix of random effects were large, indicating diversity in preferences. CONCLUSIONS: Oncologists had diverse preferences in response to breast cancer therapies. Heterogeneity is an important aspect of the study, and ignoring its presence would lead to incorrect inferences. This finding has implications on clinical trials and research: hierarchical Bayesian models with random effects provide solutions to create individual-level utilities to account for heterogeneity.