ABSTRACT
Aim
Prior studies of ICU admission timing often use broad categories (day vs. night), potentially masking hour‐level effects. We examined hour‐by‐hour ICU admission timing and mortality, with attention to nighttime periods and routine handovers.
Methods
We conducted a single‐center retrospective observational study in a closed ICU with 24/7 on‐site intensivist coverage, using prospectively registered data from the Japanese Intensive Care Patient Database. Adults (≥ 18 years) admitted to the E‐ICU between April 2017 and July 2025 were included. Admission hour (0–23) was modeled as a continuous exposure using restricted cubic splines (reference 14:00). The primary outcome was 24‐h mortality and the secondary outcome was ICU mortality. Multivariable logistic regression was adjusted for demographics, APACHE II, invasive mechanical ventilation at admission, comorbidities, diagnostic categories, admission source, and post‐cardiac arrest status. Bootstrap resampling was used to estimate 95% confidence intervals. Additional analyses examined handover, nighttime, and 24‐h periodic patterns.
Results
Among 5724 patients, 24‐h and ICU mortality were 3.4% and 10.6%, respectively. Compared with 14:00, adjusted odds of 24‐h mortality were higher in the early morning. The early‐evening estimate was imprecise, with the 95% CI including unity. In contrast, ICU mortality showed no clear temporal pattern. Handover comparisons showed no clear mortality association; nighttime admission was not associated with higher mortality, and cosinor analysis was nonsignificant.
Conclusion
Hour‐by‐hour analysis showed higher adjusted odds of 24‐h mortality in the early morning, whereas evidence of an increase in the early evening was inconclusive. Direct comparisons of handover periods showed no clear association with mortality.
Keywords: 24‐h mortality, admission timing, handover, ICU mortality, intensive care unit
Hour‐by‐hour modeling showed a local elevation in 24‐h mortality in the early morning, whereas the early‐evening estimate was inconclusive. Direct analyses of handover periods showed no clear association with mortality, and nighttime admission was not associated with higher 24‐h mortality.
1. Introduction
Intensive care units (ICUs) operate continuously, and critically ill patients require timely evaluation regardless of the time of day [1, 2, 3]. Whether the timing of ICU admission influences patient outcomes has been debated for many years. Several studies have reported higher mortality among patients admitted during nighttime or weekends, even after adjusting for illness severity [4, 5, 6, 7, 8, 9]. In contrast, some reports from intensivist‐staffed ICUs have found no mortality difference between daytime and nighttime admissions [1, 2, 3], and the evidence remains inconsistent [10].
Handover periods may represent another point of vulnerability [11, 12]. Patients admitted around shift changes may experience delays in initial assessment or treatment, and previous studies have suggested an association between admissions during these periods and worse outcomes [11, 12, 13, 14, 15]. Mortality may be influenced not only by patient characteristics but also by organizational factors that vary throughout the day, including staffing, workload, resource availability, and care transitions. Identifying time periods associated with higher mortality may therefore provide clues regarding factors that warrant further investigation. However, broad classifications such as daytime versus nighttime may obscure localized temporal variations.
In this study, we investigated the association between ICU admission time and 24‐h mortality in a closed ICU with continuous 24/7 on‐site intensivist coverage. We also examined whether mortality differed during nighttime and routine handover periods.
2. Materials and Methods
2.1. Study Design, Setting, and Participants
This single‐center retrospective observational study analyzed prospectively registered data from the Emergency–Intensive Care Unit (E‐ICU) at Kobe City Medical Center General Hospital, a tertiary care center with a closed 14‐bed ICU staffed by full‐time intensivists and dedicated critical care nurses. Intensivist staffing is 4–5 physicians during daytime hours (08:00–17:30) and one to two overnight (17:30–08:00), with medical handovers at 07:30–08:00 and 16:30–17:30. Nursing coverage is 18 nurses during the daytime shift (08:45–17:00) and eight overnight (17:00–08:45), with nursing handovers at 08:15–08:45 and 16:30–17:00. Figure 1 summarizes hourly staffing and handover periods.
FIGURE 1.
Staffing Structure and Handover Periods in the Emergency–ICU Across a 24‐Hour Cycle. This figure illustrates the hourly staffing pattern of intensivists and nurses in the Emergency–Intensive Care Unit (E‐ICU) at Kobe City Medical Center General Hospital. Intensivist staffing consists of 4–5 physicians during the daytime shift (08:00–17:30) and 1–2 physicians overnight (17:30–08:00). Nursing coverage consists of 18 nurses during the daytime shift (08:45–17:00) and eight nurses overnight (17:00–08:45). Handover periods are shown as vertical shaded bands: Intensivist handovers occur at 07:30–08:00 and 16:30–17:30, and nurse handovers occur at 08:15–08:45 and 16:30–17:00. Hourly staffing levels are displayed within each colored band. The horizontal axis represents the hour of the day from 00:00 to 24:00.
Kobe City Medical Center General Hospital participates in the Japanese Intensive Care Patient Database (JIPAD), a nationwide ICU registry that prospectively collects standardized clinical and physiological data using uniform definitions. The present study was a secondary analysis of data prospectively collected and registered by our institution for submission to JIPAD and included only patients admitted to our E‐ICU. Details of JIPAD and its data collection procedures have been reported previously [16].
All adults (≥ 18 years) admitted to the E‐ICU between April 1, 2017, and July 31, 2025, were eligible. We excluded patients with missing data on ICU admission hour, 24‐h vital status, or ICU discharge status. For multivariable analyses, we performed a complete‐case analysis excluding patients with missing covariates.
2.2. Variables
The exposure was E‐ICU admission hour. Admission time was extracted from JIPAD and converted to an hourly value (0–23) by rounding to the nearest hour (e.g., 23:30–00:29 coded as 00:00; 00:30–01:29 as 01:00; …; 22:30–23:29 as 23:00). Admission hour was modeled as a continuous variable using restricted cubic splines.
The primary outcome was 24‐h mortality (all‐cause death within 24 h of E‐ICU admission). The secondary outcome was ICU mortality (all‐cause death before E‐ICU discharge). In‐hospital mortality was additionally evaluated as an exploratory outcome in analyses of handover and daytime‐versus‐nighttime admission periods. Outcomes were obtained from JIPAD discharge status fields.
Covariates included in the multivariable models were age, gender, APACHE II score, invasive mechanical ventilation at admission (yes/no), comorbidities (respiratory failure, metastatic cancer, heart failure, liver failure, liver cirrhosis, maintenance dialysis, immunosuppression, lymphoma, acute myeloid leukemia, and multiple myeloma), ICU admission diagnosis category, admission source, and post‐cardiac arrest status. Age and APACHE II were treated as continuous variables. Invasive mechanical ventilation at admission was defined as invasive ventilation initiated at the time of E‐ICU admission. Comorbidities were coded as present or absent based on the corresponding JIPAD registry fields. ICU admission diagnoses were grouped into respiratory, neurologic, gastrointestinal, sepsis, trauma, metabolic, cardiovascular, and other categories; the other category included musculoskeletal or dermatologic, other internal medicine, genitourinary, hematologic, and obstetric or gynecologic diagnoses. Admission source was categorized as emergency department, operating room, direct interhospital transfer, or other in‐hospital locations. Post‐cardiac arrest status was defined according to the corresponding JIPAD registry field indicating whether the patient had undergone resuscitation for cardiac arrest before E‐ICU admission.
2.3. Statistical Analysis
Baseline characteristics are presented as medians (interquartile ranges) for continuous variables and counts (percentages) for categorical variables. Absolute numbers of admissions and deaths and crude mortality rates were summarized by admission hour and by five clinically defined time blocks: late night (00:00–07:29), morning handover (07:30–08:45), core daytime (08:46–16:14), evening handover (16:15–17:30), and evening/night (17:31–23:59). Patient characteristics were also summarized across these time blocks.
The primary analysis assessed the association between ICU admission hour and mortality using multivariable logistic regression. Admission hour was modeled as a continuous exposure using restricted cubic splines to allow for non‐linear associations, with 14:00 as the reference. Five knots were placed at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles of the admission‐hour distribution. Outcomes were 24‐h mortality (primary) and ICU mortality (secondary).
Models were adjusted for age, gender, APACHE II score, invasive mechanical ventilation at admission (yes/no), comorbidities (respiratory failure, metastatic cancer, heart failure, liver failure, cirrhosis, maintenance dialysis, immunosuppression, lymphoma, acute myeloid leukemia, and multiple myeloma), ICU admission diagnosis category, admission source, and post‐cardiac arrest status. Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) across the 24‐h cycle were derived from the spline models. We estimated 95% CIs using 5000 parametric bootstrap replications to generate confidence bands for predicted risk across admission hours. The distribution of admission times was visualized using a violin plot. The distributions of death hours among patients who died within 24 h and those who died in the ICU were also visualized using violin plots. Death time was defined as the time at which death was confirmed by a physician.
Subgroup analyses for 24‐h mortality examined effect modification by invasive mechanical ventilation at admission (yes/no) and age group (18–64, 65–84, ≥ 85 years) by including interaction terms between spline‐modeled admission hour and each subgroup variable; interactions were evaluated using Wald tests. We also performed a stratified analysis restricted to patients with APACHE II ≥ 30, applying the same spline model. Subgroup models adjusted for the same covariates as the primary analysis, except for the stratifying variable.
As a sensitivity analysis, admission hour was additionally modeled as a 24‐level categorical variable (0–23) with 14:00 as the reference category, using the same covariates. To account for the circular nature of clock time, we also fitted a cosinor logistic regression model for 24‐h mortality that included sine and cosine terms with a 24‐h period and was adjusted for the same covariates as the primary analysis.
To directly evaluate clinically meaningful admission periods, we compared admissions during exact physician handovers, exact nurse handovers, any physician or nurse handover, and periods extending 30 or 60 min before and after handovers with admissions outside the corresponding periods. We also compared daytime (07:30–17:29) with nighttime (17:30–07:29) admissions. The original unrounded admission time was used for these analyses. Multivariable logistic regression models evaluated 24‐h, ICU, and in‐hospital mortality and were adjusted for the same covariates as the primary analysis. Analyses of in‐hospital mortality were restricted to patients with available hospital discharge status.
All analyses were performed using Python 3.11.7 with standard scientific libraries (NumPy, pandas, scikit‐learn, statsmodels, and SciPy). Effect estimates are presented with 95% CIs.
2.4. Ethics
This study was approved by the Institutional Review Board of Kobe City Medical Center General Hospital (approval number: zn251014). The requirement for written informed consent was waived because this was a retrospective observational study using anonymized data, and an opt‐out approach was implemented in accordance with institutional policy.
3. Results
Between April 1, 2017, and July 31, 2025, 5762 patients were admitted to the E‐ICU. After excluding 36 patients aged < 18 years and two patients with missing sex data, 5724 adults were included in the final analytic cohort (Figure 2). No patients were excluded due to missing ICU admission hour, 24‐h vital status, or ICU discharge status.
FIGURE 2.
Patient flow chart of the study population. Of 5762 patients admitted to the E‐ICU during the study period, 36 were excluded due to age < 18 years and two patients with missing gender data. All remaining patients had complete data for ICU admission hour, 24‐h mortality, and ICU discharge status, yielding a final cohort of 5724 patients for analysis.
Baseline characteristics are summarized in Table 1. The median age was 71 years (interquartile range [IQR], 57–81), and 60.3% were male. The median APACHE II score at admission was 20 (IQR, 14–26), and 36.3% received invasive mechanical ventilation at admission. The most common admission diagnoses were respiratory, neurologic, gastrointestinal, sepsis‐related conditions, and trauma.
TABLE 1.
Baseline characteristics of the study population.
| Characteristic | Total (n = 5724) |
|---|---|
| Age (year), median (IQR) | 71 (57–81) |
| Male, n (%) | 3454 (60.3%) |
| APACHE II, median (IQR) | 20 (14–26) |
| Chronic organ insufficiency | |
| Respiratory failure, n (%) | 267 (4.7%) |
| Metastatic cancer, n (%) | 208 (3.6%) |
| Heart failure, n (%) | 248 (4.3%) |
| Liver failure, n (%) | 140 (2.4%) |
| Cirrhosis, n (%) | 197 (3.4%) |
| Maintenance dialysis n (%) | 283 (4.9%) |
| Immunosuppression, n (%) | 308 (5.4%) |
| Lymphoma, n (%) | 59 (1.0%) |
| AML/MM, n (%) | 47 (0.8%) |
| Admission classification | |
| Respiratory, n (%) | 1770 (30.9%) |
| Neurologic, n (%) | 1384 (24.2%) |
| Gastrointestinal, n (%) | 567 (9.9%) |
| Sepsis, n (%) | 464 (8.1%) |
| Trauma, n (%) | 448 (7.8%) |
| Metabolic, n (%) | 320 (5.6%) |
| Cardiovascular, n (%) | 313 (5.5%) |
| Musculoskeletal/Dermatologic, n (%) | 164 (2.9%) |
| Other internal medicine, n (%) | 118 (2.1%) |
| Genitourinary, n (%) | 113 (2.0%) |
| Hematologic, n (%) | 44 (0.8%) |
| Obstetrics/Gynecology, n (%) | 19 (0.3%) |
| Admission source | |
| Emergency department, n (%) | 3303 (57.7%) |
| Operating room, n (%) | 1027 (17.9%) |
| Direct interhospital transfer, n (%) | 676 (11.8%) |
| Other in‐hospital locations, n (%) | 718 (12.5%) |
| Mechanical ventilation at admission, n (%) | 2079 (36.3%) |
| Post‐cardiac arrest, n (%) | 284 (5.0%) |
| 24‐h mortality, n (%) | 196 (3.4%) |
| ICU mortality, n (%) | 608 (10.6%) |
| In‐hospital mortality, n (%) | 1187/5639 (21.0%) |
Note: Values are presented as median (interquartile range) or n (%). Percentages are rounded to one decimal place. Other in‐hospital admission locations include the ward, HCU, other ICU, and CCU. In‐hospital mortality was calculated among patients with available hospital discharge status (n = 5639).
Abbreviations: AML, acute myeloid leukemia; APACHE, acute physiology and chronic health evaluation; CCU, coronary care unit; HCU, high care unit; ICU, intensive care unit; MM, multiple myeloma.
Overall, 196 patients (3.4%) died within 24 h, and 608 patients (10.6%) died before ICU discharge. In‐hospital mortality was 21.0% among the 5639 patients with available hospital discharge status.
When admissions were grouped into five clinically defined time blocks, crude 24‐h mortality was numerically highest during the morning handover period (10/141 [7.1%]), compared with 3.0%–3.7% across the other four time blocks. Crude ICU mortality ranged from 9.3% to 12.8%, and in‐hospital mortality ranged from 18.6% to 26.9%. The proportion of patients receiving invasive mechanical ventilation at admission and the distribution of admission diagnosis categories and admission source varied across time blocks, whereas median APACHE II scores were similar (19–20) (Table S1).
Restricted cubic spline models showed a non‐linear association between ICU admission hour and 24‐h mortality. Compared with 14:00, adjusted odds showed local elevations in the early morning and early evening, with a morning peak around 07:00 (aOR, 1.95; 95% CI, 1.09–3.60) and an evening peak around 19:00 (aOR, 1.37; 95% CI, 0.76–2.45) after adjustment for demographics, illness severity, comorbidities, diagnosis category, admission source, and post‐cardiac arrest status (Figure 3A). The evening estimate was less precise, with the confidence interval including unity. For ICU mortality, the association was attenuated; adjusted odds showed modest elevations in similar periods, but confidence intervals largely overlapped unity across the 24‐h cycle (Figure 3B). Admission times were relatively evenly distributed across the day (Figure 3C).
FIGURE 3.
Association between ICU admission hour and mortality outcomes. Restricted cubic spline (RCS) models with five knots were used to evaluate the association between ICU admission hour and (A) 24‐h mortality and (B) ICU mortality, with 14:00 as the reference. Solid black lines indicate adjusted odds ratios (aORs) with shaded areas representing 95% confidence intervals, while dashed blue lines indicate unadjusted odds ratios with corresponding 95% confidence intervals. Models were adjusted for age, gender, APACHE II score, invasive mechanical ventilation at admission, comorbidities, ICU admission diagnosis category, admission source, and post‐cardiac arrest status. Shaded vertical bands indicate physician and nurse handover periods. (C) The lower panel shows the distribution of ICU admission hours.
In subgroup interaction analyses, there was no evidence of effect modification by invasive mechanical ventilation at admission (P for interaction = 0.57; Figure 4A) or by age group (18–64, 65–84, ≥ 85 years; P for interaction = 0.81; Figure 4B). In patients with APACHE II ≥ 30, the non‐linear association between admission hour and 24‐h mortality remained similar to that in the overall cohort (Figure 4C).
FIGURE 4.
Subgroup analyses of ICU admission hour and 24‐h mortality. Restricted cubic spline (RCS) models with five knots were used to evaluate the association between ICU admission hour and 24‐h mortality across subgroups, with 14:00 as the reference time. (A) Interaction between ICU admission hour and invasive mechanical ventilation at admission. (B) Interaction between ICU admission hour and age group (18–64 years, 65–84 years, and ≥ 85 years). (C) Comparison between the overall cohort and patients with high illness severity (APACHE II score ≥ 30). Solid lines represent adjusted odds ratios (aORs) derived from multivariable logistic regression models. Models were adjusted for age, gender, APACHE II score, invasive mechanical ventilation at admission, comorbidities, ICU admission diagnosis category, admission source, and post‐cardiac arrest status, except for the stratifying variable in the corresponding subgroup analysis. The horizontal dashed line indicates an odds ratio of 1.0.
As a sensitivity analysis, modeling admission hour as a 24‐level categorical variable yielded findings consistent with the spline‐based analysis (Figure 5; Table 2). The adjusted cosinor model did not show clear evidence of a 24‐h periodic association with 24‐h mortality (p = 0.26; Figure S1).
FIGURE 5.
Sensitivity analysis treating ICU admission hour as a categorical variable. Adjusted odds ratios (aORs) for 24‐h mortality were estimated using multivariable logistic regression with ICU admission hour modeled as a 24‐level categorical variable (0–23 h), using 14:00 as the reference. Points represent adjusted odds ratios, and error bars represent 95% confidence intervals. The dashed horizontal line denotes an aOR of 1.0, and the vertical dotted line indicates the reference hour (14:00). Models were adjusted for age, gender, APACHE II score, invasive mechanical ventilation at admission, comorbidities, ICU admission diagnosis category, admission source, and post‐cardiac arrest status, consistent with the primary spline‐based analysis.
TABLE 2.
Sensitivity analysis using ICU admission hour as a categorical variable.
| ICU admission hour | Admissions, n | 24‐h deaths, n (%) | Adjusted OR vs. 14:00 | 95% CI |
|---|---|---|---|---|
| 0 | 325 | 12 (3.7) | 1.81 | 0.50–6.57 |
| 1 | 257 | 2 (0.8) | 0.22 | 0.03–1.58 |
| 2 | 239 | 7 (2.9) | 0.71 | 0.15–3.29 |
| 3 | 187 | 5 (2.7) | 1.04 | 0.23–4.80 |
| 4 | 199 | 7 (3.5) | 1.28 | 0.30–5.51 |
| 5 | 160 | 7 (4.4) | 2.07 | 0.46–9.21 |
| 6 | 119 | 5 (4.2) | 1.67 | 0.34–8.08 |
| 7 | 144 | 8 (5.6) | 3.37 | 0.85–13.34 |
| 8 | 105 | 8 (7.6) | 2.90 | 0.68–12.31 |
| 9 | 131 | 7 (5.3) | 1.79 | 0.42–7.68 |
| 10 | 130 | 9 (6.9) | 4.55 | 1.11–18.56 |
| 11 | 192 | 4 (2.1) | 1.78 | 0.37–8.51 |
| 12 | 188 | 7 (3.7) | 3.31 | 0.83–13.14 |
| 13 | 224 | 9 (4.0) | 1.40 | 0.34–5.70 |
| 14 (ref) | 235 | 4 (1.7) | 1.00 | 1.00–1.00 |
| 15 | 325 | 8 (2.5) | 0.72 | 0.18–2.84 |
| 16 | 344 | 10 (2.9) | 1.04 | 0.28–3.88 |
| 17 | 317 | 10 (3.2) | 1.39 | 0.37–5.28 |
| 18 | 340 | 14 (4.1) | 1.96 | 0.54–7.04 |
| 19 | 344 | 16 (4.7) | 2.65 | 0.76–9.25 |
| 20 | 319 | 8 (2.5) | 1.37 | 0.35–5.32 |
| 21 | 316 | 10 (3.2) | 1.47 | 0.39–5.50 |
| 22 | 291 | 12 (4.1) | 2.11 | 0.58–7.71 |
| 23 | 293 | 7 (2.4) | 0.94 | 0.23–3.84 |
Note: Values are presented as n or n (%), as appropriate. Adjusted odds ratios (aORs) were estimated using multivariable logistic regression with ICU admission hour treated as a 24‐level categorical variable (0–23), with 14:00 as the reference category. Models were adjusted for the same covariates as the primary spline‐based analysis. Bold values indicate the reference category (14:00).
Abbreviations: CI, confidence interval; ICU, intensive care unit; OR, odds ratio.
In direct comparisons of clinically defined handover periods with admissions outside the corresponding periods, no clear associations with 24‐h mortality were observed for physician handovers, nurse handovers, any exact handover, or handover windows extended by 30 or 60 min; adjusted ORs ranged from 0.98 to 1.27, with all 95% CIs including unity. Similar findings were observed for ICU and in‐hospital mortality. Nighttime admission was also not associated with higher 24‐h mortality compared with daytime admission (aOR, 1.01; 95% CI, 0.70–1.44), with similar findings for ICU mortality (aOR, 0.91; 95% CI, 0.74–1.13) and in‐hospital mortality (aOR, 0.91; 95% CI, 0.78–1.07) (Table S2).
Death confirmation times were broadly distributed across the 24‐h cycle, without apparent clustering at specific times (Figure S2).
4. Discussion
In this cohort study using an hour‐by‐hour analysis of ICU admission timing, 24‐h mortality was not simply higher during nighttime compared with daytime. Compared with 14:00, the spline analysis showed higher adjusted odds of 24‐h mortality in the early morning. Although adjusted odds also increased in the early evening, this estimate was imprecise and its 95% CI included unity. Direct comparisons of handover periods showed no clear associations with mortality. This granular approach identified temporal patterns that may be obscured when using broader time categories. This apparent discrepancy with previous studies may partly reflect differences in temporal resolution, because broad daytime–nighttime classifications may average out localized hour‐level variations. Consistent with this interpretation, our direct comparison of nighttime versus daytime admission showed no clear association with 24‐h mortality.
Handover has been identified as a high‐risk process for patient safety [15, 17, 18, 19]. Communication failures contribute substantially to medical errors and adverse events [11, 12, 20, 21], many of which occur during care transitions [11, 12, 18, 21]. Care transitions may therefore be one possible explanation for the temporal variation observed in our study, although our results do not demonstrate a direct effect of handover. Prior studies also suggest that nighttime ICU admission is not necessarily associated with increased mortality after adjustment for illness severity [2, 3, 4, 5, 10], and that quality of care may vary across the 24‐h cycle beyond a simple daytime–nighttime dichotomy [6, 7, 21, 22, 23].
Despite this general agreement, evidence remains mixed regarding the impact of admission during handover periods on mortality. Some studies have reported no association, including propensity score–matched analyses, 21 and higher mortality during nighttime or early morning has sometimes been attributed primarily to higher baseline illness severity rather than care‐process deficiencies [1, 3, 10]. Similarly, in our study, direct comparisons of admissions during handover periods, including extended handover windows, showed no clear association with mortality. However, the limited number of 24‐h deaths within the handover exposure groups resulted in imprecise estimates, and these findings cannot distinguish a true absence of association from insufficient precision to detect a modest association.
ICU admission timing may reflect not only temporal differences in patient characteristics and clinical trajectories but also organizational factors, including staffing, workload, fatigue, resource availability, and care transitions [7, 8, 9, 17, 20, 21, 23]. Although the direct handover comparisons showed no clear associations, the hour‐by‐hour analysis provided a more detailed assessment of temporal differences than broad time categories and helped identify periods warranting further investigation. However, the present study could not identify which specific patient‐related, clinical, or organizational factors might explain the observed temporal differences.
The association between admission hour and outcomes was most evident for 24‐h mortality and was attenuated for ICU mortality, suggesting that vulnerability related to admission timing is concentrated in the initial phase of ICU care. In addition, direct comparisons of handover periods showed no clear association with in‐hospital mortality. One possible explanation is that, as the clinical course progresses, subsequent optimization of care may mitigate early vulnerabilities and reduce the relative influence of admission timing on later outcomes [3, 6, 8, 20, 21, 24]. Subgroup analyses showed similar temporal patterns across patient subgroups, suggesting that the observed temporal variation was not confined to a specific patient group.
Although improving handover quality is an important component of patient safety, our findings do not establish handover itself as the cause of the observed temporal variation. Larger multicenter studies with enough events are first needed to determine whether handover periods are associated with mortality and whether the observed temporal patterns are reproducible. If these findings are confirmed, subsequent studies should investigate the patient‐related, clinical, and organizational factors underlying these patterns, including admission processes, staffing, workload, and care transitions.
Several limitations should be acknowledged. This single‐center retrospective study is subject to residual confounding, and local staffing structures and workflows may limit generalizability; however, the design enabled detailed hour‐by‐hour analyses to detect fine‐grained patterns.
We did not directly assess handover quality or measure staffing levels, occupancy, or workload at admission. Information on DNAR orders or other treatment‐limitation decisions was not available, and their potential contribution to the observed temporal variation in early mortality could not be evaluated. Detailed information on pre‐ICU delays, postoperative status beyond admission source, and the timing of family arrival was also unavailable, and these unmeasured factors may have contributed to residual confounding. Therefore, we could not determine the mechanisms underlying the observed temporal variation. Although the spline analysis showed local elevations in 24‐h mortality, direct comparisons of handover periods showed no clear association with mortality. However, the number of 24‐h deaths within the individual handover exposure groups was limited, resulting in wide confidence intervals and limited statistical precision. Therefore, these findings cannot distinguish a true absence of association from insufficient precision to detect a modest association. The cosinor analysis did not demonstrate a significant 24‐h periodic pattern. These findings therefore do not establish a direct effect of handover. The association was less apparent for ICU mortality, and no clear association with in‐hospital mortality was observed in the direct comparisons, suggesting that the temporal variation was most evident for early mortality. Finally, causal inference is not possible.
5. Conclusion
In this single‐center cohort study, hour‐by‐hour ICU admission timing showed a non‐linear pattern in relation to 24‐h mortality, with higher adjusted odds in the early morning and inconclusive evidence of an increase in the early evening. No clear daytime–nighttime difference was observed, and direct comparisons of handover periods showed no clear association with mortality. The temporal variation was less apparent for ICU mortality. As a first step, larger multicenter studies with sufficient numbers of events are needed to assess the reproducibility of these temporal patterns and the association between handover periods and mortality. If these findings are confirmed, subsequent studies should investigate the underlying patient‐related, clinical, and organizational factors, including admission processes, handover processes, staffing, and workload.
Ethics Statement
This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Kobe City Medical Center General Hospital (approval number: zn251014). Because this was a retrospective observational study using anonymized data from the Japanese Intensive Care Patient Database (JIPAD), the requirement for written informed consent was waived. Instead, an opt‐out approach was implemented in accordance with institutional policy.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Figure S1: Cosinor analysis of ICU admission time and 24‐h mortality. The association between ICU admission time and 24‐hour mortality was evaluated using an adjusted cosinor logistic regression model with sine and cosine terms representing a 24‐hour period. The solid line represents the adjusted odds ratio, with 14:00 as the reference, and the shaded area represents the 95% confidence interval. Vertical shaded bands indicate physician and nurse handover periods. The model was adjusted for the same covariates as the primary restricted cubic spline analysis.
Figure S2: Distribution of death confirmation times among patients who died within 24 h and in the ICU. Violin plots show the distribution of physician‐confirmed death times across the 24‐hour cycle among patients who died within 24 h of ICU admission (n = 196) and among all patients who died in the ICU (n = 608). Green shaded areas indicate physician handover periods (07:30–08:00 and 16:30–17:30), and purple shaded areas indicate nurse handover periods (08:15–08:45 and 16:30–17:00).
Table S1: Patient characteristics and crude mortality according to clinically defined ICU admission time blocks.
Table S2: Direct comparisons of mortality outcomes according to handover periods and daytime versus nighttime ICU admission.
Acknowledgments
The authors thank the staff of the E‐ICU of Kobe City Medical Center General Hospital for their dedicated clinical care and for their contribution to data collection for the JIPAD registry.
Data Availability Statement
The datasets analyzed during the current study are not publicly available because they are derived from the Japanese Intensive Care Patient Database (JIPAD), access to which is restricted to participating institutions. Data are available from the corresponding author upon reasonable request and with permission of JIPAD.
References
- 1. Meynaar I. A., van der Spoel J. I., Rommes J. H., et al., “Off Hour Admission to an Intensivist‐Led ICU Is Not Associated With Increased Mortality,” Critical Care 13, no. 3 (2009): R84, 10.1186/cc7904. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Arulkumaran N., Harrison D. A., and Brett S. J., “Association Between Day and Time of Admission to Critical Care and Acute Hospital Outcome for Unplanned Admissions to Adult General Critical Care Units: Cohort Study Exploring the ‘Weekend Effect’,” British Journal of Anaesthesia 118, no. 1 (2017): 112–122, 10.1093/bja/aew398. [DOI] [PubMed] [Google Scholar]
- 3. Brunot V., Pouline M. F., Boivin J. M., et al., “Impact of ICU Admission During Off Hours on Patient Outcome: A Single‐Center Survey,” PLoS One 11, no. 5 (2016): e0154940, 10.1371/journal.pone.0154940. [DOI] [Google Scholar]
- 4. Cavallazzi R., Marik P. E., Hirani A., et al., “Association Between Time of Admission to the ICU and Mortality: A Systematic Review and Meta‐Analysis,” Chest 138, no. 1 (2010): 68–75, 10.1378/chest.09-3018. [DOI] [PubMed] [Google Scholar]
- 5. Galloway M., Hegarty A., McGill S., Arulkumaran N., Brett S. J., and Harrison D., “The Effect of ICU Out‐of‐Hours Admission on Mortality: A Systematic Review and Meta‐Analysis,” Critical Care Medicine 46, no. 3 (2018): 290–297, 10.1097/CCM.0000000000002837. [DOI] [PubMed] [Google Scholar]
- 6. Zajic P., Bauer P., Rhodes A., et al., “Time of Day and Its Association With Risk of Death and Chance of Discharge in Critically Ill Patients: A Retrospective Study,” Scientific Reports 9 (2019): 12533, 10.1038/s41598-019-48947-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Kuijsten H. A. J. M., Brinkman S., Meynaar I. A., et al., “Hospital Mortality Is Associated With ICU Admission Time,” Intensive Care Medicine 36 (2010): 1765–1771, 10.1007/s00134-010-1918-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Langston M. C., Subedi K., Fabrizio C., Wimmer N. J., Choudhry U. I., and Urrutia L. E., “An Examination of Cardiovascular Intensive Care Unit Mortality Based on Admission Day and Time,” Heart & Lung 50, no. 5 (2021): 579–586, 10.1016/j.hrtlng.2021.02.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Laupland K. B., Shahpori R., Kirkpatrick A. W., and Stelfox H. T., “Hospital Mortality Among Adults Admitted to and Discharged From Intensive Care on Weekends and Evenings,” Journal of Critical Care 23, no. 3 (2008): 317–324, 10.1016/j.jcrc.2007.09.001. [DOI] [PubMed] [Google Scholar]
- 10. Hall A. M., Stelfox H. T., Wang X., et al., “Association Between Afterhours Admission to the Intensive Care Unit, Strained Capacity, and Mortality: A Retrospective Cohort Study,” Critical Care 22 (2018): 97, 10.1186/s13054-018-2027-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. McCarthy S., Motala A., Lawson E., and Shekelle P. G., “Use of Structured Handoff Protocols for Within‐Hospital Unit Transitions: A Systematic Review From Making Healthcare Safer IV,” BMJ Quality and Safety 34 (2025): 680–690, 10.1136/bmjqs-2024-018385. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Robertson E. R., Morgan L., Bird S., Catchpole K., and McCulloch P., “Interventions Employed to Improve Intrahospital Handover: A Systematic Review,” BMJ Quality and Safety 23 (2014): 600–607, 10.1136/bmjqs-2013-002309. [DOI] [PubMed] [Google Scholar]
- 13. Shahian D., McEachern K., Rossi L., et al., “I‐PASS Handover System: A Decade of Evidence Demands Action,” BMJ Quality and Safety 30 (2021): 769–774, 10.1136/bmjqs-2021-013314. [DOI] [PubMed] [Google Scholar]
- 14. Farid M., Tsugawa Y., and Jena A. B., “Assessment of Care Handoffs Among Hospitalist Physicians and 30‐Day Mortality in Hospitalized Medicare Beneficiaries,” JAMA Network Open 4, no. 3 (2021): e213040, 10.1001/jamanetworkopen.2021.3040. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Desai S. V., Feldman L., Brown L., et al., “Effect of the 2011 vs 2003 Duty Hour Regulation–Compliant Models on Sleep Duration, Trainee Education, and Continuity of Patient Care Among Internal Medicine House Staff: A Randomized Trial,” JAMA Internal Medicine 173, no. 8 (2013): 649–655, 10.1001/jamainternmed.2013.2973. [DOI] [PubMed] [Google Scholar]
- 16. Irie H., Okamoto H., Uchino S., et al., “The Japanese Intensive Care PAtient Database (JIPAD): A National Intensive Care Unit Registry in Japan,” Journal of Critical Care 55 (2020): 86–94. [DOI] [PubMed] [Google Scholar]
- 17. Martins N. R. S., Martinez E. Z., Simões C. M., Barach P. R., and Carmona M. J. C., “Analyzing and Mitigating the Risks of Patient Harm During Operating Room to Intensive Care Unit Patient Handoffs,” International Journal for Quality in Health Care 37, no. 1 (2025): mzae114. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. van Sluisveld N., Hesselink G., van der Hoeven J. G., et al., “Improving Clinical Handover Between Intensive Care Unit and General Ward Professionals at Intensive Care Unit Discharge,” Intensive Care Medicine 41 (2015): 589–604. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Montoya‐Garrido M. J., Rodríguez‐Suárez C. A., Mateos‐López N., Santiago‐Díaz Y. T., and González‐de la Torre H., “Shift‐To‐Shift Information Transfer: Phenomenological Study of Nurses' Experiences,” JMIR Nursing 8 (2025): e81703. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Appelbaum R. D., Farrell M. S., Hoth J. J., et al., “Handoffs and Transitions of Care in the Intensive Care Unit: An American Association for the Surgery of Trauma Critical Care Committee Clinical Consensus Document,” Trauma Surgery & Acute Care Open 10 (2025): e001677. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Midega T. D., Leite Filho N. C., A. P. Nassar, Jr. , et al., “Impact of Intensive Care Unit Admission During Handover on Mortality: Propensity Matched Cohort Study,” Einstein (São Paulo) 19 (2021): eAO5748. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Bray B. D. and Steventon A., “What Have We Learnt After 15 Years of Research Into the ‘Weekend Effect’?,” BMJ Quality and Safety 26 (2017): 607–610. [DOI] [PubMed] [Google Scholar]
- 23. Ganesan S., Magee M., Stone J. E., et al., “The Impact of Shift Work on Sleep, Alertness and Performance in Healthcare Workers,” Scientific Reports 9 (2019): 4635. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Checkley W., Martin G. S., Brown S. M., et al., “Structure, Process and Annual Intensive Care Unit Mortality Across 69 Centers: United States Critical Illness and Injury Trials Group Critical Illness Outcomes Study (USCIITG‐CIOS),” Critical Care Medicine 42, no. 2 (2014): 344–356. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1: Cosinor analysis of ICU admission time and 24‐h mortality. The association between ICU admission time and 24‐hour mortality was evaluated using an adjusted cosinor logistic regression model with sine and cosine terms representing a 24‐hour period. The solid line represents the adjusted odds ratio, with 14:00 as the reference, and the shaded area represents the 95% confidence interval. Vertical shaded bands indicate physician and nurse handover periods. The model was adjusted for the same covariates as the primary restricted cubic spline analysis.
Figure S2: Distribution of death confirmation times among patients who died within 24 h and in the ICU. Violin plots show the distribution of physician‐confirmed death times across the 24‐hour cycle among patients who died within 24 h of ICU admission (n = 196) and among all patients who died in the ICU (n = 608). Green shaded areas indicate physician handover periods (07:30–08:00 and 16:30–17:30), and purple shaded areas indicate nurse handover periods (08:15–08:45 and 16:30–17:00).
Table S1: Patient characteristics and crude mortality according to clinically defined ICU admission time blocks.
Table S2: Direct comparisons of mortality outcomes according to handover periods and daytime versus nighttime ICU admission.
Data Availability Statement
The datasets analyzed during the current study are not publicly available because they are derived from the Japanese Intensive Care Patient Database (JIPAD), access to which is restricted to participating institutions. Data are available from the corresponding author upon reasonable request and with permission of JIPAD.