Needs and barriers of healthcare teams to implement real-time AI & predictive analytics in cardiothoracic ICU: a field study

Introduction Artificial Intelligence (AI) offers significant opportunities to improve healthcare quality, especially in critical care settings, where timely decisions are essential yet challenging due to the volume of continuously accumulating patient data. AI algorithms can analyze data in real time, summarize key information for busy care teams, and alert clinicians to critical events and important care milestones. However, successful implementation requires first understanding the needs and barriers care teams face, including needs that are currently missing from their workflow, and what obstacles to integration exist. Methods To address these questions, we conducted a field study in a Cardiothoracic Intensive Care Unit (CT-ICU) to evaluate user needs for implementing real-time AI predictive analytics. We observed and interviewed fourteen participants ( n = 14), including nursing staff, advanced practice providers (NP/PAs), cardiovascular critical care attending physicians, and surgeons. Through a qualitative thematic analysis, we identified key considerations for developing and implementing static and real-time AI and predictive tools in the CT-ICU, aligning with the Technology Acceptance Model (TAM) constructs of perceived usefulness and perceived ease of use . Findings Key themes included: (1) the system must have added clinical value to the clinical team and patients, (2) trust in the AI system should be prioritized through co-design processes, transparency in the system, and consistent prediction accuracy, and (3) integration into workflow was critical, including robust onboarding and minimizing alarm fatigue, charting burden, click burden, and technology burden with consideration of personal wellness. Conclusion As a synthesis of our findings, we identify a series of key design insights and suggest a preliminary framework for AI technology development focused on clinical team trust and adoption, the Trusted AI for Clinical Teams (TACT) framework. These findings offer the potential to bridge from AI research to clinical implementation.