Abstract
<title>Abstract</title> <p>Human error in aviation often reflects latent fluctuations in attention and workload that are not captured by conventional performance measures. Practical operator-state monitoring requires physiologically grounded markers that are interpretable and suitable for real-time use. We present a proof-of-concept pipeline that links global EEG “engagement indices” to a continually updated crash hazard (the instantaneous crash risk) in a high-fidelity helicopter simulator. Twelve aviators flew a challenging route; five trajectories ended in terrain collisions. A context-aware alignment procedure matched non-crash to crash flights in a low-dimensional state space (latitude, longitude, cumulative distance, elapsed time), reducing confounds from route differences. Penalized time-varying Cox models incorporating global EEG indices and terrain elevation estimated instantaneous crash hazard. Within this multivariable model, an alpha-normalized index (1/α) showed a negative association with crash hazard, consistent with sustained external engagement. A simple Kalman state-space model applied to model-implied hazards produced smoothed latent risk trajectories that diverged between crash and non-crash flights roughly one minute before impact. Although based on a modest sample with unstable leave-one-subject-out generalization, these results illustrate how low-dimensional EEG metrics can be embedded in survival and state-space frameworks to generate interpretable, time-resolved risk signals for future neuroadaptive flight-deck applications.</p>