Abstract / Summary
Abstract Background. Electrocardiographic prediction in critical illness has relied on single parameters measured once — most often the corrected QT interval — with modest and inconsistent results. Whether quantitative parameters spanning the whole cardiac cycle (depolarization through repolarization), measured from every ECG of an ICU stay, carry independent prognostic and mechanistic information is unknown. Methods. We linked MIMIC-IV v3.1 to the MIMIC-IV-ECG waveform subset and built a cohort of 14,319 adult ICU stays without known coronary artery disease, atrial fibrillation, or pacing (25,112 analyzable ECGs from a 72,166-ECG linked pool). We assembled a whole-cycle parameter panel per ECG: machine-derived intervals and axes (heart rate, PR, QRS duration, QT/QTc, frontal QRS-T angle) and waveform-measured amplitudes from a validated median-beat engine (ST amplitude, J-to-Tpeak amplitude, QRS voltage, ST-T area, Tp-Te). Analyses: (1) 48-hour landmark Cox models for 28-day mortality; (2) linear mixed models of day-by-day trajectories by survival; (3) new-onset repolarization abnormalities as exposures for ECG-specific outcomes (new atrial tachyarrhythmia, new myocardial injury) with cause-specific and Fine–Gray models; (4) within-patient difference-in-differences ECG responses to ICU interventions, with amiodarone as a positive control. Results. The spatial QRS-T angle was the strongest whole-cycle predictor of 28-day mortality: per-SD HR 1.125 (95% CI 1.068–1.185; P < 0.0001) for the 48-h maximum, with dose-response across strata (90–120° vs < 60°: HR 1.23; ≥120°: HR 1.35), independent of age, sex, SOFA, lactate, and MAP. QTc was null (HR 0.95; P = 0.08), as were QRS duration and T-wave amplitudes; J-to-Tpeak amplitude (HR/SD 1.121; P < 0.0001) and low QRS voltage (HR 0.905; P < 0.01) added independent signal. Trajectories separated from day 1 (mixed-model survival main effect + 15.3°; P = 1×10⁻³⁴) without subsequent divergence (day×survival P = 0.40), and adding within-stay change to the first ECG gave no incremental discrimination (ΔAUC − 0.0005) — the first ECG carried essentially all prognostic information for mortality. For ECG-specific outcomes, by contrast, new-onset abnormalities were strongly predictive: new-onset ST depression → new atrial tachyarrhythmia HR 3.10 (1.51–6.36) and → new myocardial injury HR 2.69 (1.32–5.49); new-onset QRS-T angle ≥ 90° → atrial tachyarrhythmia HR 3.77 (2.03–7.01); QTc ≥ 500 ms predicted nothing. In the pharmacodynamic analysis, amiodarone initiation produced the expected + 16.4 ms QTc prolongation (P = 0.0006) and − 15.6 bpm heart-rate reduction (P < 0.0001), validating the framework; norepinephrine initiation was followed by + 5.7 ms QTc (P < 0.05), and renal replacement therapy initiation by + 6.4 ms QTc, + 7.5° QRS-T angle, and + 3.6 ms QRS widening. Conclusions. In critical illness, spatial depolarization–repolarization mismatch — not repolarization duration — is the dominant ECG prognostic axis. It is measurable from the first ECG, static information suffices for mortality prediction, but new-onset electrical abnormalities precede arrhythmia and myocardial injury with 3–4-fold hazard ratios, and the same panel reads out drug and organ-support effects in vivo. Whole-cycle ECG dynamics are a quantitative, reproducible, and clinically actionable monitoring layer for the ICU.