Introduction

Sepsis is defined as a systemic response to infection accompanied by some degree of organ dysfunction. Despite advances in the management of severe sepsis—including time-sensitive antibiotic administration, fluid resuscitation, and, in some cases, transfer to specialized referral centers for definitive treatment—in-hospital mortality remains alarmingly high, ranging from 25 to 30%1. To reduce complications and improve survival, numerous studies have investigated the pathogenesis and outcomes of sepsis from multiple perspectives. One critical factor is the initial hours following sepsis onset, which is crucial for the management of critically ill patients; timely and appropriate treatment during this period significantly influences outcomes2. Ideally, intensive care unit (ICU) staffing and resources should be consistently adequate throughout the 24 h period, when sepsis patients are admitted. Indeed, research has demonstrated improved outcome with dedicated intensivists, optimized intensivist staffing patterns, and adequate nursing numbers3,4. However, the availability and quality of personnel and technological resources often differ between daytime and nighttime hours. Moreover, during nighttime hours, staffing levels are typically reduced, and some diagnostic and therapeutic procedures may be delayed or postponed until regular office hours. Similar with this, a retrospective analysis indicated that patients admitted to ICUs in Australia after hours and on weekends had higher observed and risk-adjusted mortality compared with those admitted at other times5. Some scholars have reported that nighttime admission was associated with increased crude hospital mortality; however, this association may be explained by differences in case mix.

Any variation in mortality linked to the timing of hospital admission could have substantial implications for healthcare policy, insurance, and workforce planning. Nevertheless, the association between the timing of ICU admission and clinical outcomes in sepsis patients remains unclear. Therefore, this study aimed to elucidate the association between admission timing and mortality in sepsis using a large critical care public database and to identify potential underlying mechanisms.

Methods

Study design and data source

We conducted a retrospective cohort study using the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Developed jointly by the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC), MIMIC-IV contains granular clinical data for more than 40,000 patients admitted to ICU, including demographics, vital signs, laboratory results, medications, and diagnostic codes. One author (Tianqi Shen) completed the Collaborative Institutional Training Initiative (CITI) program and passed the “Conflict of Interest” and “Data or Specimens Only Research” modules (certification ID: 70551437), and was thereby granted access to the database. All reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines6.

Study population

All adult patients meeting the Sepsis-3 criteria who were admitted to the ICU in the MIMIC-IV database were screened for eligibility7. The inclusion criteria were as follows: (1) aged ≥ 18 years; (2) an ICU length of stay ≥ 24 h; (3) first ICU admission only. The exclusion criteria were as follows: (1) discharge from hospital or death within 24 h of ICU admission; (2) lack of critical variable data (e.g., medication information) or outcome data (mortality).

Exposure and variables

The exposure of interest was the hour of ICU admission. Admissions occurring between 08:00 and 17:59 were classified as daytime, and those occurring between 18:00 and 07:59 the following morning were classified as nighttime. We collected the following covariates, all recorded within the first 24 h of ICU admission (repeated values were averaged): 1. General characteristics: sex, age, race, and weight; 2. Vital signs: mean blood pressure (MBP), body temperature, and oxygen saturation; 3. Comorbidities: Charlson Comorbidity Index (CCI), malignant cancer, dementia, myocardial infarction, congestive heart failure (CHF), cerebrovascular disease (CVD), chronic obstructive pulmonary disease (COPD), rheumatic disease, and renal disease. 4. Laboratory tests: chloride, calcium, potassium, sodium, blood urea nitrogen (BUN), creatinine, white blood cell count (WBC), platelet, hemoglobin, international normalized ratio (INR), and activated partial thromboplastin time (APTT). 5. Severity scores: Sequential Organ Failure Assessment (SOFA), Glasgow Coma Scale (GCS), Systemic Inflammatory Response Syndrome (SIRS), Oxford Acute Severity of Illness Score (OASIS), and Logistic Organ Dysfunction System (LODS). 6. Others: total fluid intake, urine output, and mechanical ventilation.

Outcomes

The primary endpoint was the difference in 30-day all-cause mortality between patients admitted during daytime and nighttime hours. Secondary outcomes included 90- and 180-day all-cause mortality and the incidence of sepsis-related complications: sepsis-associated encephalopathy (SAE)8, sepsis-induced coagulopathy (SIC)9, sepsis-induced acute lung injury (S-ALI)10, acute kidney injury (AKI), and continuous renal replacement therapy (CRRT).

Statistical analysis

Continuous variables were expressed as the median and interquartile range (IQR) given their non-normal distribution. Categorical variables were described as the number and percentage (%). Daytime and nighttime admission groups were compared with the Mann–Whitney U test for continuous data and the χ2 test for categorical data.

Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for 30-, 90- and 180-day mortality, with daytime admission as the reference category. Logistic regression was used to examine sepsis-related complications. We constructed sequential models: first adjusting for demographics, then adding comorbidities, laboratory values, and severity scores, to isolate the association between nighttime admission and these outcomes. Results were reported as odds ratios (ORs) with 95% CIs.

Propensity score matching (PSM) was performed using a logistic model that included all baseline covariates. Using 1:1 nearest-neighbor matching without replacement and a caliper width of 0.1 standard deviations of the logit score, we assembled the matched cohort that formed the analytic sample for this study. Balance after matching was assessed with the absolute standardized mean difference (SMD); an absolute SMD < 0.10 indicated adequate balance. Subgroup analyses for the primary outcome were performed by sex, age, renal disease, MBP, hemoglobin, and SOFA score, with interaction p-value calculated across subgroups. Sensitivity analysis was conducted by calculating the E-value. All analyses were performed using the R statistical package (version 4.4.2), with a two-tailed p-value < 0.05 considered statistically significant.

Ethics approval and consent to participate

This study was conducted in strict accordance with the principles of the Declaration of Helsinki (1964) and its subsequent amendments. The data were sourced from the MIMIC database (https://mimic.physionet.org), a large, anonymized public database whose development was approved by the IRBs of the MIT and BIDMC. The requirement for informed consent was waived due to the database’s complete anonymization and de-identification of all patient information.

Results

Baseline characteristics

The flowchart of participant selection is shown in Fig. 1. A total of 25,437 patients with sepsis were ultimately included in the analysis. The baseline characteristics are summarized in Table 1. Among these patients, 11,946 patients (46.96%) were admitted during daytime hours (daytime group), and 13,491 (53.04%) were admitted at night (nighttime group). Compared with the nighttime group, patients in the daytime group had a higher proportion of males and White individuals, more frequent use of mechanical ventilation, higher SOFA scores, and elevated levels of chloride, total fluid intake, and urine output. In contrast, the nighttime group exhibited higher MBP, body temperature, BUN, creatinine, platelet count, and hemoglobin levels, along with a greater prevalence of malignant cancer and higher GCS and OASIS scores.

Fig. 1
Fig. 1
Full size image

The flowchart of participant selection.

Table 1 Basic characters between daytime and nighttime administration before and after PSM.
Full size table

To adjust for potential confounders, 1:1 PSM was performed, resulting in 8,704 well-balanced patients in each group. After matching, no statistically significant differences in baseline characteristics were observed between the two groups (Table 1).

Primary outcomes

Kaplan–Meier curves for 30-day survival in both the full and PSM cohorts are shown in Fig. 2A, B. Log-rank tests revealed that 30-day survival was significantly higher in the daytime group than in the nighttime group (before PSM: p < 0.001; after PSM: p = 0.039). In the PSM cohort, 1,704 (19.58%) patients in the daytime group and 1,850 (21.25%) in the nighttime group died within 30 days of admission (crude HR [95% CI]: 1.101 [1.031–1.175], p = 0.004) (Tables 2, 3). Stepwise multifactorial Cox regression models were constructed to examine 30-day mortality risk, comparing daytime versus nighttime admissions. These included four sequential models that progressively adjusted for baseline characteristics, vital signs, laboratory values, comorbidities, illness-severity scores, and other variables. After adjustment for all covariates, nighttime admission remained associated with an increased 30-day mortality risk (adjusted HR [95% CI]: 1.081 [1.012–1.154], p = 0.021) (Table 3). Similar findings were observed in the full cohort. After full adjustment, the nighttime group exhibited a significantly increased risk of death (n = 3,003, 22.26%) compared with the daytime group (n = 1,798, 15.05%) (adjusted HR [95% CI]: 1.219 [1.148–1.293], p < 0.001) (Table 2, Fig. 2A).

Fig. 2
Fig. 2
Full size image

Kaplan–Meier curves for short- and long-term survival in both full and PSM cohorts.

Table 2 Mortality rates and adverse events between daytime and nighttime administration before and after PSM.
Full size table
Table 3 Primary and secondary outcomes in multiple regression model after PSM.
Full size table

Secondary outcomes

In PSM cohort, 90- and 180-day mortality risks remained significantly higher in the nighttime group than in the daytime group. The 90-day mortality rate was 25.99% (2,262 patients) in the daytime group versus 28.21% (2,455 patients) in the nighttime group (crude HR [95% CI]: 1.108 [1.048–1.172], p < 0.001). For 180-day mortality, the corresponding figures were 2,594 (29.80%) versus 2,842 (32.65%) (crude HR [95% CI]: 1.123 [1.067–1.181], p < 0.001) (Tables 2, 3). After multivariable adjustment, these associations persisted: 90-day mortality (adjusted HR [95% CI]: 1.084 [1.025–1.147], p = 0.005) and 180-day mortality (adjusted HR [95% CI,]: 1.098 [1.043–1.155], p < 0.001) (Table 3).

Consistent findings were observed in the full cohort. The nighttime group had significantly higher risks of 90-day mortality (2,408 [20.16%] vs 3,955 [29.32%]; adjusted HR [95% CI]: 1.226 [1.165–1.291], p < 0.001) and 180-day mortality (2,784 [23.30%] vs 4,521 [33.51%]; adjusted HR [95% CI]: 1.217 [1.162–1.275], p < 0.001) (Fig. 2C). Furthermore, analysis of the full cohort revealed a distinct temporal pattern in the association between admission time and mortality across different follow-up periods, with a lower trend of mortality risk observed between 07:00 and 13:00 (Fig. 3). Additionally, an analysis comparing mortality between morning (8:00–12:59) and afternoon admissions (13:00–17:59) in the daytime group was conducted. The results demonstrated that morning admission was associated with lower mortality compared to afternoon admission (Supplementary table 1).

Fig. 3
Fig. 3
Full size image

Admission numbers and short- and long-term mortality rates among sepsis patients by time of admission.

Regarding sepsis-associated adverse events, the PSM cohort showed a significantly higher incidence of both SAE and S-ALI in the nighttime group compared to the daytime group. SAE occurred in 3,970 patients (45.61%) in the daytime group and 4,436 patients (50.97%) in the nighttime group (crude OR [95% CI]: 1.239 [1.168–1.315], p < 0.001) (Tables 2, 3). For S-ALI, the corresponding figures were 1,608 (18.47%) versus 1,951 (22.41%) (crude OR [95% CI]: 1.275 [1.184–1.373], p < 0.001) (Tables 2, 3). After adjustment for all covariates, these associations remained significant (SAE: adjusted OR [95% CI]: 1.294 [1.198–1.398], p < 0.001; S-ALI: adjusted OR [95% CI]: 1.257 [1.158–1.363], p < 0.001) (Table 3). By contrast, the incidences of SIC, AKI and CRRT among patients were similar between the two groups (Table 3).

Subgroup and sensitivity analyses

We performed subgroup and interaction analyses to examine whether the association between nighttime admission and 30-day mortality varied across patient characteristics. No significant interactions were observed for sex, age, renal disease, mean arterial pressure, haemoglobin, or SOFA score, indicating that the association remained consistent across these strata (Fig. 4).

Fig. 4
Fig. 4
Full size image

Subgroup analysis of the relationship between daytime and nighttime groups in sepsis patients.

To assess the potential impact of unmeasured confounding, we calculated E-values for both the point estimate and the lower limit of the 95% confidence interval for the association between nighttime admission and 30-day mortality (Table 4). The E-value for the point estimate was 1.122, indicating that an unmeasured confounder would need to be associated with both nighttime admission and 30-day mortality by a risk ratio of at least 1.122 to fully explain away the observed association. This suggested that our findings are robust to unmeasured confounding.

Table 4 E-Value for primary and secondary outcomes.
Full size table

Discussion

This study investigated the association between ICU admission timing and mortality in sepsis patients. We included 25,437 patients with sepsis who were categorized into the daytime admission group (n = 11,946) and the nighttime group (n = 13,491). Multifactorial cox regression analysis revealed that nighttime admission was independently associated with increased risks of 30-day, 90-day and 180-day mortality. Furthermore, patients admitted at night exhibited a higher incidence of SAE and S-ALI. These findings were robust to adjustments in statistical methods, refinements of the study population, and risk stratification. In addition, subgroup analyses revealed that the association between nighttime admission and 30-day mortality was consistent across key characteristics, including sex, age, renal disease, mean arterial pressure, haemoglobin, and SOFA score.

The association between nighttime admission and increased mortality has caused a heated debate by several studies. There was a study indicated that nighttime admission to the ICU was not associated with higher mortality rate or prolonged hospital or ICU stay compared with daytime admission11. This finding is consistent with other reports identifying no clear relationship between admission timing and patient survival12,13,14. Furthermore, one study demonstrated that nighttime ICU admissions were not associated with poorer ICU, hospital, or ventilator outcomes—except for mortality—compared with nighttime admissions15. In contrast, other studies have reported higher mortality among nighttime and weekend admissions5,16,17. Specifically, these studies reported higher APACHE II scores, prolonged length of stay, and elevated 28-day mortality risk among nighttime and weekend admissions. The inconsistency across these findings may be attributed to several factors, including differences in study populations, sample sizes, case-mix effects, ICU management strategies (e.g., bed availability, admission threshold, end-of-life care practices), definition of “nighttime”, and the treatment capabilities of the ICU. Therefore, the present analysis leverages data from the MIMIC database, which provides a large sample size that helps mitigate potential biases. Additionally, by focusing specifically on sepsis patients, this study reduces case-mix heterogeneity, thereby enabling a more precise assessment of the relationship between admission timing and outcomes in this specific population.

Even after PSM and subgroup analyses were performed to account for baseline characteristics, the results remained consistent, indicating that the detrimental effect of nighttime admission was independent of these potential confounders. A nine-year cohort study examining mortality associated with nighttime ICU admissions with on-site intensivist coverage found that ICU mortality was significantly higher for patients admitted between 00:00 and 07:5918. Due to the lack of specific staffing data from the BIDMC, we hypothesize that differences in ICU staffing and possible delayed procedures between daytime and nighttime may be contributing factors. Although it is impossible to verify empirically, we speculate that if intensivists are continuously present in the ICU during daytime hours and available on-site when it is necessary during nighttime hours, this would be sufficient to avoid a quality gap during nighttime hours. Further elucidating the intrinsic influence of nighttime admission, a study highlighted the detrimental effect of circadian biorhythm on human performance, particularly toward the end of the night shift19. Because circadian rhythms adapt slowly to rapidly rotating schedules, ICU staff working recurring day–evening–night rotations experience sustained misalignment that increases physical morbidity and mood disturbance, both of which can impair clinical decision-making. This aligns with a study demonstrating that the highest (predicted and observed) mortality in ICU occurred at the end of the night shift (05:00–06:00), when both healthcare workers and patients perform at their worst16. Our results similarly indicated a trend toward lower mortality between 07:00 and 13:00 (Fig. 3). Moreover, morning admission might be associated with lower mortality compared to afternoon admission. Therefore, it is crucial for ICU managers to develop strategies that ensure consistent, high-quality medical care throughout the 24-h cycle.

Our study revealed a higher incidence of SAE and S-ALI in the nighttime admission group, which might account for the poor prognosis observed in these patients. SAE is a common but poorly understood neurological complication of sepsis, which is associated with increased morbidity and mortality. It typically presents as acute encephalopathy, with manifestations ranging from delirium to coma. Previous study had suggested that SAE could be initiated or exacerbated by secondary factors, including environmental influences20. S-ALI is a severe clinical condition characterized by refractory hypoxemia, respiratory distress, and non-cardiogenic pulmonary edema, representing one of the most frequent complications of sepsis21. Therefore, early recognition and management of SAE and S-ALI may be essential for improving sepsis outcomes. Moreover, in present study, the recorded occurrence of SAE and S-ALI were observed after patients admitted to ICU. Therefore, our findings demonstrated sepsis patients admitted to the ICU at night might have higher occurrence rates of SAE and S-ALI compared with those admitted during the day. In other words, when sepsis patients are admitted at night, medical staff should maintain a higher index of suspicion for SAE and S-ALI to facilitate timely intervention and optimize sepsis care.

However, this study has several limitations. First, the data were sourced from the MIMIC database, which includes clinical information of ICU patients from BIDMC in the United States. Although BIDMC contributes a substantial number of patients, the use of a single-center database may introduce selection bias and vary across different healthcare systems. Variations in treatment protocols, ICU admission criteria, and patient demographics across different institutions may limit the external validity of our findings. In addition, specific staffing measures, such as daily nurse-to-patient ratios, as well as times to attendance by the staff intensivist and completion of investigations and consults, were unavailable. Therefore, the explanation of our observations is subject to certain limitations. Future studies should validate these results in multi-center cohorts with diverse patient populations to improve generalizability. Second, given its retrospective observational design, numerous potential confounders required control through PSM or multivariable adjustment. While PSM was applied to mitigate confounding effects, residual bias from unmeasured confounders and selection bias might still persist. However, in critical care, randomized trials evaluating organization aspects of care are often infeasible making observational studies with careful adjustment for case mix essential. Additionally, in the present study, we defined daytime hours as 8:00 to 17:59, whereas other studies had adopted alternative intervals, such as 06:00 to 17:59 or 8:00 to 22:00—depending on their specific contexts5,22,23. And we did not distinguish between weekdays and weekends. Our primary objective was to examine the effect of nighttime admission on the prognosis of sepsis patients, rather than weekend effect. Furthermore, we were able to extract appropriate and complete data from BIDMC. Additionally, distinguishing between weekdays and weekends would require knowledge of holiday and festival policies at BIDMC, as the holiday effect may be similar to the weekend effect. However, such specific data from BIDMC were unavailable. Apart from the uncertain definition of daytime hours, institutional variations in scheduling and staffing were not accounted for and could influence outcomes.

Conclusion

Our analysis demonstrated that nighttime ICU admission among patients with sepsis was significantly associated with increased mortality risk at 30, 90, and 180 days. In addition, sepsis-related adverse events, including SAE and S-ALI, were more frequently observed in patients admitted during nighttime hours. These findings remained consistent after PSM and adjustment for all covariates, suggesting that the observed association was independent of baseline characteristics and likely reflects the detrimental effect of nighttime admission itself. Further investigation through well-designed, multicenter, randomized controlled trials is warranted to confirm our findings and explore additional underlying mechanisms.