Human Error risk modeling in anesthesia for improving Patient safety: a hybrid CREAM–Bayesian Network approach

Human error contributes to safety risks in anaesthesia under varying cognitive and contextual conditions. This study developed a hybrid CREAM–Bayesian Network (BN) framework to assess and prioritise human-error risks across anaesthesia tasks. The workflow was decomposed into five task groups and 23 subtasks. Cognitive Failure Probabilities (CFPs) were estimated using extended CREAM. A hierarchical BN integrated normalised CFP and sensitivity through Noisy-MAX. GeNIe was used for inference and sensitivity analysis, with robustness assessed across α = 0.30–0.90. Recovery had the highest aggregated CFP (0.0816), while ‘Carrying out special care for critical patients” (R1) had the highest individual CFP (0.0796). Sensitivity analysis identified R1, ‘Examination of the patient for artificial respiration and blood gases” (BP2), ‘Preparing drugs and other necessary equipment for anesthesia” (AP1), and ‘Helping the anesthesiologist in various matters, including anesthesia and resuscitation” (AP4) as influential subtasks. The hybrid framework supported task-level prioritisation by integrating CFP and model-based sensitivity. The modelled Total Error probability was 0.049979 at α = 0.70 and remained approximately 0.05 across the tested scenarios. These findings represent model-based estimates, not observed clinical error rates or causal effects.