Case Study Highlights Need for AI Workflow Integration
A real-world deployment of the TREWS AI-driven sepsis alert system demonstrated that predictive models are only effective if they trigger timely and actionable interventions within a clinician's existing workflow. The case study underscores that the success of clinical AI depends heavily on seamless EHR integration and user-centered design. This is reinforced by research calling for fairness and explainability to be built into clinical AI models to improve trust.
- One independent validation study of Epic's sepsis prediction model, which is used by hundreds of U.S. hospitals, found it performed with an area under the curve of 0.63, which was significantly lower than the 0.76 to 0.83 range reported by Epic. This study also found the model identified only 7% of sepsis cases that clinicians missed, while still generating alerts for 18% of all hospitalized patients, contributing to alert fatigue. - In contrast, a multi-site study of the TREWS system at five hospitals showed it reduced sepsis-related deaths by nearly 20%. The system detected sepsis an average of six hours earlier than standard methods, and caregivers evaluated 89% of all alerts. - Common complaints from frontline nurses about Epic's EHR include excessive time spent on data entry, which reduces time with patients, and discrepancies between the EHR and medication dispensing systems that can lead to errors. Some nurses report that algorithmic tools for determining patient acuity and staffing levels are often inaccurate and not reflective of actual care needs. - To transition into nursing informatics, the American Nurses Credentialing Center (ANCC) offers the Informatics Nursing Certification (RN-BC). Eligibility typically requires a BSN, two years of RN experience, 30 hours of informatics continuing education, and a minimum of 2,000 hours of practice in informatics nursing within the last three years. - The 21st Century Cures Act, implemented by the Office of the National Coordinator for Health Information Technology (ONC), mandates that healthcare providers give patients secure, electronic access to all of their health information at no cost. This rule also prohibits "information blocking" to improve data flow between different EHR systems, a practice for which some vendors have been criticized. - Interoperability standards like HL7 FHIR (Fast Healthcare Interoperability Resources) are crucial for new AI and decision support applications. FHIR uses modern web standards to allow different systems to exchange specific data points, rather than entire documents, enabling more flexible and real-time data sharing for applications like clinical decision support embedded directly into the EHR workflow. - Epic is expanding its AI capabilities with tools to help nurses draft end-of-shift notes, summarize a patient's stay for discharge, and automatically queue up orders based on conversations with patients. The company is also developing its own foundational models, called Cosmos AI, to build predictive tools for clinical decision-making. - ICU experience is highly valuable for an informatics role as it provides a deep understanding of complex clinical workflows, high-stakes decision-making, and the real-world impact of technology on patient care. This background is critical for designing and implementing IT systems that genuinely support, rather than hinder, critical care delivery.
