NEJM Catal Innov Care Deliv. 2026;7(s1):CAT250256. doi: 10.1056/CAT.25.0256. Epub 2026 Mar 18.
ABSTRACT
Electrocardiogram (ECG) interpretation is fundamental to cardiovascular care, yet persistent human variability and escalating diagnostic demands hinder its reliability and scalability. Despite the remarkable scientific advancements of artificial intelligence (AI) in health care - including its demonstrated capacity to outperform humans in certain predictive tasks - few systems have successfully bridged the gap between technical potential and sustainable real-world clinical transformation. To address these challenges, this article presents the development and implementation of the Synergistic Human-AI Partnership for ECG analysis system (SHAPE), an AI-enabled ECG system at Fuwai Hospital - a leading cardiovascular center located in Beijing, China. SHAPE is designed to enhance ECG interpretation accuracy and efficiency and also pioneers a model for meaningful human-AI collaboration in clinical practice. SHAPE comprises five interconnected platforms: the Benchmark Labeling Platform generates gold-standard reference data for model validation through a rigorous two-stage expert adjudication process based on national consensus guidelines; the Model Development Platform enables iterative model development and deployment; the Clinical Collaboration Platform embeds AI seamlessly into routine clinical workflows; the Education and Evaluation Platform benchmarks performance across human and AI interpreters; and the Visualization and Management Platform guides oversight and continuous improvement. Together, these components form a closed-loop architecture designed to evolve through real-world feedback and clinician engagement. Since its launch in 2022, SHAPE has yielded substantial clinical and operational gains. Even with a 50% reduction in dedicated reporting staff (from six to three between 2022 and 2025), per-physician productivity nearly doubled, driving the average daily interpretation volume from 272 to 504 reports. Concurrently, diagnostic accuracy - evaluated against independent gold-standard samples - improved significantly, from 96.76% pre implementation (2021) to 98.58% post implementation (2024) (P<0.001). Clinical reliance on AI grew robustly, with the proportion of reports incorporating AI outputs increasing from 84.79% in the second year (2023) to over 95% by mid-2025, a statistically significant increase (P<0.001). These achievements were enabled not only by technical advances, but by deliberate workflow design that nurtured human-AI coevolution - from AI as a learner to AI as a catalyst for system-level improvement. Based on their experience with SHAPE, the authors suggest that AI's greatest potential lies not in automation alone, but in thoughtful human-AI collaboration, underscoring the importance of shared goals and feedback-driven design. As the platform scales beyond Fuwai Hospital, it offers a replicable blueprint for AI integration to enhance care quality and clinician capacity.
PMID:42456003 | DOI:10.1056/CAT.25.0256