| Sumario: | The intensive care unit is a high-stakes, information-intensive environment requiring nurses to make rapid and accurate decisions. This study aimed to elucidate the cognitive and neural mechanisms underlying nurses' risk decision-making under time pressure and complex clinical demands. Thirty ICU nurses participated in a computer-based multitasking experiment simulating concurrent medical multitasking scenarios, with twenty-one valid datasets analyzed. Participants performed priority judgments under high- and low-risk conditions while EEG signals were continuously recorded. Event-related potential components and oscillatory activities across δ, θ, α, and β frequency bands were analyzed. Gaussian Hidden Markov Models were used to characterize cognitive state transition dynamics aligned to task events. Risk decision-making emerged as a multi-stage, dynamically coordinated process involving four distinct cognitive patterns: monolithic stability progression, compulsory path lock-in, multi-path flexible convergence, and flow separation and premature convergence. Correct decisions were associated with enhanced low-frequency oscillations (δ, θ) and stable HMM transitions, reflecting efficient integration and adaptive cognitive control. In contrast, incorrect decisions exhibited early perceptual inefficiency, unstable state transitions, and premature cognitive closure under high-risk conditions. This study is the first to identify four distinct dynamic cognitive patterns of risk decision-making in a simulated ICU multitasking context. The findings indicate that decision accuracy is closely linked to coordinated state-transition dynamics rather than isolated neural activations, highlighting the importance of adaptive cognitive control in clinical judgment. Although the present findings are exploratory, they may provide a preliminary reference for future research on brain-machine collaboration in clinical nursing contexts. In particular, future work could examine how EEG-decoded cognitive states might be incorporated as input information for robot-assisted systems to characterize nurses' cognitive intentions during risk tasks. Further studies with larger samples and in more realistic clinical settings are needed to validate the model's robustness and generalizability.
|
|---|