Classification of a continuous risk score into risk levels is common. However, while the absolute risk score is essential, it is arguably unethical to label anyone at 'high, moderate or low risk' of a serious event, simply because management based on a single criterion (e.g. avoiding the target condition) has been determined to be effective or cost-effective at a population level. Legally, mono-criterial risk labeling can inhibit the obtaining of a fully-informed, preference-based consent, since multiple considerations (various benefits and harms) matter to most individuals, not only the single criterion that is the basis of the provided risk category. These ethical and legal challenges can be met by preference-sensitive multi-criteria decision support tools. In this future vision paper, we demonstrate, at a conceptual proof-of-method level, how such decision support can and should be developed without reference to risk-level classifications. The statin decision is used as illustration, without any empirical claims.