Following patient questions across clinical and evidence services with large language model assistance

Abstract

Preprint. Not peer reviewed. For people with multimorbidity, completing an evidence task does not ensure that the question raised in care has been answered. We propose requirements for following an individual patient question across clinical and evidence services. A continuing record would preserve the patient's priorities and clinical context, distinguish evidence limitations from barriers to care, identify an accepted task owner, and document the return and discussion of a reviewed response. We specify conditions for closing and reopening a question and illustrate the record with a hypothetical clinical example. The proposal builds on existing evidence services and remains untested. Large language models could assist selected transitions, but the effects of organising a service and adding model assistance need separate comparisons. Evaluation should examine whether reviewed answers return to care, alongside source support, patient understanding, safety, workload and outcomes. The record offers a concrete object for clinical informatics evaluation rather than evidence of improved care.