Sedation trajectories and circadian rhythm disruption in critically ill adults: an observational cohort study with external validation using MIMIC-IV and eICU-CRD

Background Whether ICU sedation strategies are associated with circadian rhythmicity of vital signs, and whether such disruption mediates delirium, remains unclear. Methods In a retrospective cohort study of MIMIC-IV ( n = 13,592 surgical and mixed medical–surgical ICU patients), with supportive external validation in eICU-CRD ( n = 12,004), multivariate dynamic time warping (DTW)-based clustering classified four sedation trajectories over 48 h. Rhythmicity parameters of four vital signs were quantified from available ICU-stay recordings using Lomb–Scargle, wavelet, entropy, and cosinor methods. Marginal structural models estimated associations, and exploratory mediation analysis examined indirect pathways. Results Sustained high-dose propofol showed the weakest vital-sign rhythmicity. Compared with this group, the early high-dose propofol and propofol-to-dexmedetomidine transition groups had higher spectral power for heart rate (β = 0.033 and 0.015) and temperature (β = 0.055 and 0.020; all P < 0.001). With CAM-ICU as the primary delirium measure, temperature wavelet power formed an exploratory indirect pathway (proportion mediated 23.5%, 95% CI 16.9–31.0%; Group 2 vs. 3), though this ranged from 0% to 34% across assumptions. In eICU-CRD, temperature-derived parameters were directionally consistent (Cohen's d 0.263 to 0.426). Conclusions ICU sedation trajectories were associated with vital-sign rhythmicity parameters. Temperature-derived rhythmicity was consistent with an exploratory indirect pathway to delirium, but estimates depended on analytic assumptions. These proxies should not be interpreted as direct measures of central circadian function.