Improving Medication Safety in Chronic Kidney Disease Using Rule-Based and Artificial Intelligence-Based Clinical Decision Support Systems: A Systematic Review of Randomized Controlled Trials.
完整摘要 · 中文译文
背景：在慢性肾脏病（CKD）中，优化肾脏药物剂量以避免药物毒性至关重要，但处方错误仍很常见。基于规则和基于人工智能/机器学习（AI/ML）工具的临床决策支持系统（CDSS）被用于弥补这一安全差距；然而，其影响仍不确定。 方法：我们对随机对照试验（RCTs）进行了一项PRISMA指南引导的系统评价和荟萃分析，比较基于规则或AI/ML的CDSS与常规治疗对照在成年CKD患者或有CKD相关处方错误风险患者中的效果。主要终点是与CDSS逻辑一致的药物安全终点（适当的肾脏剂量、潜在不适当处方和用药错误）。为处理异质性，我们以结构化最佳证据综合以及按实施方式和工作流程阶段进行的试验水平映射作为荟萃分析的补充。 结果：在符合纳入标准的20项随机对照试验中，6项提供了荟萃分析数据。合并四项试验显示，CDSS改善了适当的肾脏剂量（RR 1.76；95% CI 1.13-2.74），但异质性极大（I 2 = 97%），且95%预测区间（0.75-4.14）跨越了无效值；因此，合并估计值是依赖于情境的平均效应，而非可迁移的效应，在新环境中不能保证获益。支持CDSS的一致效应方向反而来自最佳证据综合。电子健康记录中CKD的记录得到持续改善（RR 1.19；95% CI 1.07-1.32；I 2 = 0%）。目前的证据主要为中断式医嘱录入干预；临床医生依从性范围为17%-74%，原因是警报疲劳、时间限制以及 不明确的系统功能和超控过程。
完整摘要 · 英文原文
BACKGROUND: Optimization of renal drug dosing to avoid drug toxicity is essential in Chronic Kidney Disease (CKD), yet prescribing errors are common. CDSS with rule-based and AI/ML based tools are used to address this safety gap; however, their impact remains uncertain. METHODS: We performed a PRISMA-guided systematic review and meta-analysis of RCTs comparing rule-based or AI/ML CDSS with usual care comparators among adults with CKD or at risk of CKD-related prescribing errors. The primary outcome was a medication safety endpoint aligned with the CDSS logic (appropriate renal dosing, potentially inappropriate prescribing, and medication errors). To address heterogeneity, we supplemented meta-analysis with a structured Best Evidence Synthesis and trial-level mapping by delivery mode and workflow stage. RESULTS: Among 20 RCTs meeting inclusion criteria, 6 provided meta-analytic data. Pooled across four trials, CDSS improved appropriate renal dosing (RR 1.76; 95% CI 1.13-2.74), but heterogeneity was extreme (I 2 = 97%) and the 95% prediction interval (0.75-4.14) crossed the null; the pooled estimate is therefore a context-dependent average rather than a transportable effect, and benefit cannot be assured in a new setting. A consistent direction of effect favoring CDSS came instead from the Best Evidence Synthesis. Documentation of CKD in electronic health records improved consistently (RR 1.19; 95% CI 1.07-1.32; I 2 = 0%). Current evidence was predominantly interruptive order-entry interventions; clinician compliance ranged 17-74% owing to alert fatigue, time constraints, and unclear system function and override processes.