Li Xiaodi, Rajaganapathy Sivaraman, Hu Xinyue, Feng Jingna, Li Jianfu, Yu Yue, Fiero Phil, Boroumand Soulmaz, Larsen Richard, Liu Xiaoke, Tao Cui, Zong Nansu
Department of Artificial Intelligence and Informatics, Mayo Clinic, Rochester, MN, USA.
Mayo Clinic Platform, Mayo Clinic, Rochester, MN, USA.
medRxiv. 2025 Mar 24:2025.03.19.25324271. doi: 10.1101/2025.03.19.25324271.
Randomized controlled trials (RCTs) provide the highest level of clinical evidence but are often limited by cost, time, and ethical constraints. Emulating RCTs using real-world data (RWD) offers a complementary approach to evaluate the treatment effect in a real clinical setting. This study aims to replicate clinical trials based on Mayo Clinic Platform (MCP) electronic health records (EHRs) and emulation frameworks. In this study, we address two key questions: (1) whether clinical trials can be feasibly replicated using the MCP, and (2) whether trial emulation produces consistent conclusions based on real clinical data compared to the original randomized controlled trials RCTs.
We conducted a retrospective observational study with an adaption of trial emulation. To assess feasibility, we applied a refined filtering method to identify trials suitable for emulation. The emulation protocol was carefully designed on top of the original RCT protocol to balance scientific rigor and practical feasibility. To minimize potential selection bias and enhance comparability between groups, we employed propensity score matching (PSM) as a statistical adjustment method.
Based on our predefined search criteria targeting phase 3 trials focused on drug repurposing for heart failure patients, we initially identified 27 eligible trials. After a two-step manual review of the original eligibility criteria and extraction of the patient cohorts based on MCP visualizer, we further narrowed our selection to the WARCEF trial, as it provided an adequate sample size for the emulation within the MCP. The experiment compares the WARCEF trial and a simulation study on Aspirin vs. Warfarin. The original study (smaller sample) found no significant difference (HR = 1.016, p < 0.91). The simulation (larger sample) showed a slightly higher HR (1.161) with borderline significance (p < 0.052, CI: 0.999-1.350), suggesting a possible increased risk with Warfarin, though not conclusive.
RCT emulation enhances real-world evidence (RWE) for clinical decision-making but faces limitations from confounding, missing data, and cohort biases. Future research should explore machine learning-driven patient matching and scalable RCT emulation. This study supports the integration of RWE into evidence-based medicine.
随机对照试验（RCT）提供了最高水平的临床证据，但常常受到成本、时间和伦理限制。利用真实世界数据（RWD）模拟RCT提供了一种补充方法，用于在真实临床环境中评估治疗效果。本研究旨在基于梅奥诊所平台（MCP）电子健康记录（EHR）和模拟框架复制临床试验。在本研究中，我们解决两个关键问题：（1）使用MCP是否能够可行地复制临床试验，以及（2）与原始随机对照试验（RCT）相比，试验模拟基于真实临床数据是否能得出一致的结论。
我们进行了一项采用试验模拟的回顾性观察研究。为评估可行性，我们应用一种改进的筛选方法来识别适合模拟的试验。在原始RCT方案基础上精心设计模拟方案，以平衡科学严谨性和实际可行性。为尽量减少潜在的选择偏倚并增强组间可比性，我们采用倾向得分匹配（PSM）作为统计调整方法。
基于我们针对专注于心力衰竭患者药物重新利用的3期试验的预定义搜索标准，我们最初识别出27项符合条件的试验。在根据MCP可视化工具对原始纳入标准进行两步人工审查并提取患者队列后，我们将选择范围进一步缩小至WARCEF试验，因为它为MCP内的模拟提供了足够的样本量。该实验比较了WARCEF试验以及阿司匹林与华法林的模拟研究。原始研究（样本量较小）未发现显著差异（风险比=1.016，p<0.91）。模拟研究（样本量较大）显示风险比略高（1.161），具有临界显著性（p<0.052，置信区间：0.999 - 1.350），表明华法林可能增加风险，尽管尚无定论。
RCT模拟增强了用于临床决策的真实世界证据（RWE），但面临混杂、数据缺失和队列偏倚等限制。未来研究应探索机器学习驱动的患者匹配和可扩展的RCT模拟。本研究支持将RWE整合到循证医学中。
