When deciding how to treat their patients, physicians must consider risk-benefit trade-offs between possible good and bad clinical outcomes. When designing clinical trials to evaluate new treatments, a practical approach that reflects this common medical practice is to elicit numerical utilities from the physicians planning the trial that quantify the desirability of each possible clinical outcome a patient may experience. In this talk, I will discuss how this may be done to design Bayesian sequentially adaptive early phase trials. After some preliminary remarks on Bayesian statistics, three illustrations will be presented. These include early phase trials involving (1) dose-finding for radiation therapy of pediatric brain tumors, (2) optimizing sedative dose in premature infants who must be intubated to treat respiratory distress syndrome, and (3) constructing a utility surface for two event times to jointly optimize dose and schedule in stem cell transplantation.