*Purpose: Peri-operative safety metadata in transplant surgery remain limited. Lack of standardisation and limited models of failure analysis, as well as non-intuitive error taxonomies, deter data collection. Consequently, transplant surgical services cannot monitor and measure perioperative hazards or adverse events in real-time. This leads to an inability to respond to performance variation or intervene in imminent crises.

Aim: To develop a pragmatic model and algorithm for point-of-care reporting and analysis of peri-operative safety events, amenable to machine learning processing.

*Methods: Phase I: A prototype model was derived from a narrative review of the safety literature across several high-reliability industries. Phase II: Focus groups with healthcare and aviation global safety experts (N=11) provided content and format modifications of the model. Phase III: Semi-structured expert interviews (N=5) refined the model by addressing investigator queries, provided implementation strategies and translated the model into algorithmic form. Qualitative data from Phases II and III was processed using thematic analysis according to COREQ guidelines. Phase IV: The final model was tested on a cohort of real-world peri-operative incidents by two independent safety analysts. Inter-rater agreement was determined using Cohen’s kappa coefficient.

*Results: 133 literature sources were reviewed to create a three-part crisis model. Failures were shown to arise from four categories of threat (systems, culture, operator skills, operator condition). These triggered one of three failure mechanisms: 1. Performance errors (problems related to diagnosis, planning and execution); 2. Awareness and monitoring errors; 3. Rescue errors (inability to contain, correct or prevent known failures). Concomitant occurrence of all three error types led to patient, staff or institutional harm. Subsequent expert inputs through focus groups and interviews resulted in two diagnostic algorithms for determining error cause and effect, respectively. Application of the algorithms to 51 clinical incidents demonstrated feasibility (mean diagnostic time of 128 seconds per event analysed); good inter-rater agreement (k score >0.75 in 8 out of 10 categories assessed); and pragmatic clinical implication for transplant care.

*Conclusions: Surgical safety events in transplantation can be systematically reported and analysed at the point of care using intuitive binary algorithms applied to real-world clinical data. This may facilitate the creation of transplant peri-operative meta-data and subsequent data-mining opportunities.

« Back to 2022 American Transplant Congress