Temporal distribution of recorded intensive care unit deaths across the 24-hour cycle: a 15-year retrospective hospital-based study.

backgroundWhether recorded deaths in intensive care units (ICUs) follow a uniform temporal distribution throughout the 24-hour cycle remains uncertain. Previous studies have primarily compared broad day-night periods, potentially overlooking finer temporal patterns. We investigated the temporal distribution of recorded ICU deaths using both predefined time intervals and continuous-time modeling.

objectiveTo characterize the temporal distribution of recorded intensive care unit deaths across the 24-hour cycle using continuous-time and interval-based analyses.

methodsThis retrospective observational study included all recorded ICU deaths occurring between 1 January 2010 and 30 June 2025 at a tertiary referral hospital. Exact clock time of death, age, sex, and ICU type were extracted from the hospital information system. The primary analysis evaluated the temporal distribution of recorded ICU deaths using generalized additive models (GAMs) with cyclic cubic regression splines. The cyclic spline specification preserved continuity across midnight and accommodated the periodic nature of the 24-hour cycle.

resultsA total of 9,889 recorded ICU deaths were analyzed. Continuous-time analysis demonstrated a significant non-uniform temporal distribution of recorded deaths across the 24-hour cycle (global Wald χ² = 101.88, df = 12, p < 0.001). The fitted generalized additive model identified the highest relative frequency of recorded deaths during the late-morning period (approximately 11:15), whereas the lowest relative frequency occurred during the early overnight hours (approximately 03:36). Secondary interval-based analyses also demonstrated significant temporal heterogeneity (three-interval analysis: χ² = 60.3, p < 0.001; four-interval analysis: χ² = 65.3, p < 0.001), although the corresponding effect sizes were small (Cramer's V = 0.055 and 0.047, respectively). Calendar-period sensitivity analyses indicated that the overall temporal pattern remained broadly consistent despite substantial expansion of the ICU system over the study period.

conclusionsRecorded ICU deaths were not uniformly distributed across the 24-hour cycle. Continuous-time modeling identified significant temporal clustering that would not have been fully captured using conventional interval-based analyses alone. Because denominator data (e.g., ICU occupancy, patient census, admissions, and the population at risk) were unavailable, the present study was designed to evaluate the temporal distribution of recorded ICU deaths rather than temporal variation in mortality risk. The observed temporal differences were statistically significant but quantitatively modest.