Rethinking learning from incidents in healthcare: a Safety-II model integrating work-as-imagined, work-as-done, and work-as-experienced.

Healthcare organizations increasingly investigate patient safety incidents using systems thinking. However, even investigations that claim to adopt a systems approach frequently revert to linear, event-focused reasoning concluding on deviations and recommending further training. A foundational question remains unresolved: when an incident review identifies a "deviation," does this represent exceptional behavior or the system's typical mode of operation that generally succeeds under varying conditions but occasionally fails? Without this distinction, explanations for incidents tend to focus on individual non-compliance rather than understanding the everyday adaptations that enable care delivery in a complex adaptive system. Consequently, incident reviews that promote system learning often continue to recommend retraining, reminders, and reinforced compliance. In this Perspective, we propose a Safety-II model for incident review structured around three analytical lenses: work-as-imagined, work-as-done, and work-as-experienced-the last capturing what only patients and families can witness, longitudinally, across settings and professional boundaries. We introduce a prevalence-indexed analysis of work-as-done, in which longitudinal electronic medical record and operational data are used to establish whether a flagged practice represents exception or norm, while remaining transparent about the gap between documented and enacted work. We position this contribution alongside recent advances in incident analysis and patient partnership, and set out concrete proposals for regulators, healthcare organizations, and researchers.