Medication-related clinical decision support is commonly judged through alert counts, override rates, acceptance rates, or rule-level technical performance. These measures describe system activity but do not directly establish whether preventable medication harm has been reduced. This Current Opinion argues that medication-safety signals should instead be assigned to response modes according to the combination of potential harm, immediacy, contextual uncertainty, persistence, interpretive requirements, and responsibility for clinical action. Evidence from interruptive alerts, noninterruptive decision support, pharmacist-mediated review, and post-implementation optimization suggests that no single delivery mode is appropriate for all medication risks. Highly consequential and time-sensitive hazards may justify interruption when signals are sufficiently specific, whereas lower-immediacy or context-dependent signals may be better managed through silent surveillance followed by contextual reassessment. Pharmacist review can add an interpretive layer when rule correctness does not establish patient relevance, and escalation requires explicit transfer of responsibility rather than simple forwarding of another notification. We therefore propose a risk-adaptive medication-safety model in which signals can migrate among surveillance, interruption, pharmacist review, and clinical escalation as risk and uncertainty change. The model is conceptual rather than validated. Its usefulness will depend on local data quality, staffing, workflow, patient heterogeneity, system configuration, and prospective evaluation using patient-safety and balancing outcomes rather than alert-volume metrics alone.
