JMIR Publications

ABSTRACT

Predictive medicine is increasingly moving beyond population-level risk estimation towards individualized treatment-effect prediction, causal modelling, multimodal artificial intelligence, and emerging digital-twin systems capable of representing patient-specific alternative health trajectories. These developments may make adverse outcomes appear increasingly foreseeable, modifiable, and avoidable, potentially strengthening inferences about patient responsibility when a predicted-better course was not pursued. Yet greater computational precision does not collapse the normative distance between modelled modifiability, reasonable avoidability, warranted responsibility, and legitimate responsibility-sensitive care. This Viewpoint examines the justificatory transitions required as patient-specific predictive information moves from representing an unrealized health alternative to supporting stronger moral or institutional conclusions. We develop a Gated Relevance Framework that distinguishes direct clinical and deliberative uses of prior-choice information from a responsibility-sensitive pathway involving three gates: from modelled modifiability to reasonable avoidability, from reasonable avoidability to warranted responsibility, and from warranted responsibility to legitimate responsibility-sensitive institutional use. The framework shows that prior-choice information may remain clinically or deliberatively relevant without supporting responsibility attribution, and that even warranted responsibility does not itself justify differential care. We further identify normative relevance migration as a downstream failure mode in which information legitimately generated or used for one purpose is reused for a normatively more demanding purpose without the additional justification that the new use requires. Because digital infrastructures allow predictive information to persist, travel across contexts, and be recombined in later decisions, system design should preserve inferential status, make decision-relevant purpose explicit, support contextual reassessment, and govern responsibility-sensitive downstream reuse. Responsible predictive medicine therefore requires not only reliable prediction, but responsible inference about what predictive information may legitimately justify in subsequent care.