在急性后期与长期照护中实施AI辅助多重用药审查:基于4Ms框架的执业护士主导工作流
该文描述了一种由执业护士主导、与4Ms框架对齐的工作流,在急性后期与长期照护中利用人工智能仅作初步信号生成与分诊,所有有临床意义的信号须经人工核实,高风险情形须升级至药师审查。两个临床场景示例展示了该工作流的可行性。
要点
- AI仅用于初步信号生成与分诊,所有临床有意义信号须人工核实后方可行动。
- 高风险情形如QT风险组合、强CYP450相互作用、利托那韦相关相互作用等须升级药师审查。
- 工作流由新症状、新医嘱、出院药物核对、肾功能变化等常规临床触发因素激活。
核心结论
- 该工作流为在急性后期与长期照护中整合AI至用药安全提供了务实且可迁移的模型,同时保留临床医生问责、跨专业保障与结构化记录。
研究设计与样本
- 本文描述了一种执业护士主导、与4Ms框架对齐的工作流,AI仅用于初步信号生成与分诊,随后须经权威药物信息资源核实并在预设高风险情形下由药师审查。
主要发现
- AI支持的筛查利用患者特异性输入生成简短的优先候选用药安全问题列表,包括可能的药物相互作用、潜在不适当用药、药物不良事件及肾功能剂量关注。
- 高风险场景包括QT风险组合、强细胞色素P450相互作用、利托那韦相关相互作用、肾功能剂量不确定性、窄治疗指数药物及衰弱患者中的极端多重用药,须升级至药师。
- 联合临床示例在2个场景中展示了可行性:识别可能引起唾液分泌过多的药物因素及管理奈玛特韦/利托那韦治疗期间的精神药物相互作用风险。
关键数值与推荐
- 工作流由常规临床触发因素激活,包括新症状、新药物医嘱、出院药物核对、启动可能相互作用的治疗、老年综合征、肾功能变化及计划性定期用药审查。
临床怎么用
- 在急性后期与长期照护中,可以AI生成优先用药安全问题列表作为初步筛查,但所有临床有意义信号须经人工核实,高风险情形须升级药师审查。
局限性
- 本文为工作流描述,仅通过2个临床场景示例展示可行性,未报告系统性的有效性或安全性结局数据。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Polypharmacy in post-acute and long-term care (PA/LTC) is common and is associated with falls, delirium, hospitalization, functional decline, and mortality. In these settings, clinicians must identify drug-drug interactions (DDIs), potentially inappropriate medications (PIMs), renal dosing concerns, and medication contributors to new symptoms under time pressure and during frequent care transitions. We describe a nurse practitioner-led, 4Ms-aligned workflow in which artificial intelligence (AI) is used only for preliminary signal generation and triage, followed by mandatory verification using authoritative medication information resources and pharmacist review in predefined high-risk situations. The workflow is activated by routine clinical triggers, including new symptoms, new medication orders, hospital discharge reconciliation, initiation of potentially interacting therapies, geriatric syndromes, renal function change, and scheduled periodic medication review. AI-supported screening uses patient-specific inputs to generate a brief prioritized list of candidate medication safety issues, including possible DDIs, PIMs, adverse drug events, and renal dosing concerns. All clinically meaningful signals require human verification before action. High-risk scenarios, including QT-risk combinations, strong cytochrome P450 interactions, ritonavir-related interactions, renal dosing uncertainty, narrow therapeutic index medications, and extreme polypharmacy in frail patients, require pharmacist escalation. A combined clinical example demonstrated feasibility in 2 scenarios: identification of a likely medication contributor to hypersalivation and management of psychotropic interaction risk during nirmatrelvir/ritonavir treatment. This workflow offers a pragmatic and transferable model for integrating AI into medication safety in PA/LTC while preserving clinician accountability, interprofessional safeguards, and structured documentation.
引用
- Romem A, Givon D. Operationalizing AI-Assisted Polypharmacy Review in Post-Acute and Long-Term Care: A 4Ms-Based, NP-Led Workflow. J Am Med Dir Assoc. 2026. PMID: 42407225. doi:10.1016/j.jamda.2026.106304
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。