Trusting the Algorithm or Trusting the Nurse? Critical Care Nurses' Experiences of Automation Bias and Professional Autonomy in AI ‐Assisted Early Warning

Résumé fourni par la source

BACKGROUND: Artificial intelligence-assisted early warning systems (AI-EWS) are increasingly integrated into critical care, yet little is known about how nurses experience automation bias and negotiate professional autonomy when algorithmic recommendations intersect with bedside judgement. AIM: To explore how critical care nurses experience automation bias and negotiate professional autonomy when using AI-EWS in intensive care. STUDY DESIGN: A qualitative Interpretive Description study was conducted across four hospitals in a regional health cluster in northern Saudi Arabia. Twenty-three purposively sampled critical care nurses with direct experience of a unified electronic health record-integrated AI-EWS participated in virtual semi-structured interviews. Data were analysed using Braun and Clarke's reflexive thematic analysis, and reporting followed the COREQ 32-item checklist. FINDINGS: Four interpretive themes were identified: reading the algorithm through a nursing lens; the architecture of trust and override; responsibility, risk and the weight of disagreement; and autonomy reshaped at the human-algorithm interface. Nurses viewed algorithmic outputs as useful but insufficient, requiring interpretation through bedside assessment and contextual clinical knowledge. Trust was actively calibrated through experience, while override was framed as an accountable exercise of professional judgement. A subset of participants described overriding alerts without formal documentation under workload pressure, highlighting a governance gap at the human-algorithm interface. Disagreement with the algorithm was experienced as professionally exposing yet shaped by relational and organisational conditions, and sustained AI engagement reshaped surveillance practice, clinical authority and professional identity. CONCLUSIONS: AI-EWS reorganise rather than replace nursing judgement. Safe integration depends not only on algorithmic performance but also on governance, training and interprofessional cultures that recognise and protect nurses' interpretive role at the bedside. RELEVANCE TO CLINICAL PRACTICE: Critical care services should establish non-punitive, low-burden pathways for documenting overrides, provide training on trust calibration and discordant alerts, and position nursing leadership centrally in AI governance, implementation and evaluation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.