We used data from MIMIC-IV database [version 3.0 (20, 21)] for this retrospective cohort study. MIMIC-IV comprised detailed clinical information of patients spanning from 2008 to 2022 by two in-hospital database systems: a customized hospital-wide electronic health record (EHR) and an intensive care unit (ICU). The author Fengzhen Huang successfully passed the Collaborative Institutional Training Initiative (CITI) exam and obtained the permission to access the database (Record ID: 63858817 and 63858818). The database utilization was approved by the Institutional Review Boards of the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center. The data has been de-identified and informed consent was thus not necessary.

MIMIC-IV database contained a total of 364,627 patients, 546,028 admissions, and 94,458 icustays. The sepsis was defined as a suspected infection combined with an acute increase in Sequential Organ Failure Assessment (SOFA) score ≥ 2 (22). Patients who met both of the following criteria could be defined as SAE: (1) patients with sepsis; (2) patients having a score on the Glasgow Coma Scale (GCS) < 15 or having delirium (23). GCS has been used to identify SAE and distinguish it from sepsis. It was assessed prior to sedation or surgery for the relevant patients. Delirium was diagnosed according to the International Classification of Diseases codes (ICD) version 9 or 10. Among these SAE patients, the inclusion criteria included: (1) First ICU admission; (2) ICU stay duration >24 h; (3) Age ≥ 18 years. The exclusion criteria included: (1) Patients with primary brain injury, including traumatic brain injury, epilepsy, brain tumor, meningitis or encephalitis, and cerebrovascular disease; (2) Patients with mental disorders; (3) Patients with neurodegenerative diseases; (4) Patients with alcohol or drug abuse; (5) Patients with metabolic encephalopathy, hepatic encephalopathy, hypertensive encephalopathy, hypoglycemic coma, and other liver or kidney diseases affecting consciousness; and (6) Patients with severe electrolyte imbalances or glycemic disturbances, including hyponatremia (<120 mmol/L), hypernatremia (>150 mmol/L), hypoglycemia (<54 mg/dL), hyperglycemia (>180 mg/dL); (7) Patients with insufficient data (missing data >20%).

The patient information were extracted using Structured Query Language (SQL). The following variables were collected: (1) demographic information: age, gender, race, weight; (2) laboratory indexes within the first day of hospitalization: wbc, hemoglobin, platelet, glucose, sodium, potassium, calcium, chloride, creatinine, aniongap, bun, partial thromboplastin time (PTT), prothrombin time (PT), po2, pco2, base excess; (3) mean values of vital signs within the first day of ICU: Saturation of Peripheral Oxygen (Spo2) mean, systolic blood pressure (SBP) mean, diastolic blood pressure (DBP) mean, heart rate mean, respiratory rate (Resp rate) mean, temperature mean; (4) disease severity scores at the time of ICU admission: Sepsis-related Organ Failure Assessment (SOFA), Glasgow Coma Scale (GCS), Acute Physiology Score III (APSIII), Objective Structured Assessment of Intensive Care Skills (OASIS), Charlson Comorbidity Index (CCI); (5) comorbidities diagnosed via ICD codes 9 or 10: hypertension, diabetes, heart failure, cardiovascular disease, chronic obstructive pulmonary disease (COPD); and (6) therapeutic interventions during hospitalization: continuous renal replacement therapy (CRRT), vasoactive agent, ventilation.

Participants were divided into two groups: no aspirin group and aspirin group. Aspirin exposure was defined as receiving at least one dose of any form of aspirin during hospitalization. In addition, we divided SAE patients into the following four groups to investigate the association between the clinical outcome and different aspirin exposure timing: aspirin pre-ICU group, aspirin in-ICU group, aspirin pre-ICU and in-ICU group, and no aspirin group. We also compared the outcome between the low-dose aspirin group (<300 mg/d) and high-dose aspirin group (> = 300 mg/d).

The primary outcome was 30-day mortality. The secondary outcomes included in-hospital mortality, 60-day mortality, 90-day mortality, 180-day mortality, length of hospital stay and gastrointestinal hemorrhage. The complication of gastrointestinal hemorrhage was diagnosed according to ICD-9 or 10.

All baseline variables had fewer than 20% missing data, which was randomly imputed using the “mice” package in R software. Since the skewed distribution of continuous variables, the comparisons were made by Mann–Whitney U test and presented in the form of the median and inter-quartile range (IQR). Categorical variables were compared by the chi-square test and expressed as numbers and percentages.

Propensity score matching (PSM) was used to balance the baseline characteristics between the no aspirin group and aspirin group (24). We performed PSM by nearest neighbor matching, with a caliper width of 0.05 and patients matched at a ratio of 1:1. The balance of the variables was assessed by the standardized mean difference (SMD), and it was considered balanced if the SMD < 0.1 between the two groups.

The Cox proportional hazards model, with hazard ratio (HR) and 95% confidence interval (CI), was applied to assess the association between aspirin therapy and mortality risk of in-hospital, 30-day, 60-day, 90-day, and 180-day. The Kaplan–Meier survival curves were generated and analyzed by the log-rank test. The multivariate models were adjusted for various variables. E-value analysis was used to evaluate the potential influence of unmeasured or unknown confounding factors. This analysis aimed to evaluate the extent of confounding effects needed to eliminate the observed association between aspirin administration and mortality risk (25). Subgroup analysis was applied to explore potential differences in the effects of aspirin therapy on clinical outcomes across these various groups. In addition, we conducted a assessment between the aspirin exposure timing and mortality rate. We also did a evaluation between the dosage of aspirin and mortality rate. All statistical analysis were conducted using R software (v4.4.0), with statistical significance determined by a two-tailed p-value <0.05.

Figure 1 illustrated the flowchart for selecting SAE patients. In total, 41,296 patients were diagnosed with sepsis, and a subset of 19,662 patients met the SAE criteria in the MIMIC-IV database. Subsequently, 4,707 eligible patients were recruited after the inclusion and exclusion criteria. Participants were divided into two groups: no aspirin group, which included individuals who did not receive aspirin treatment (n = 2,252), and the aspirin group, which included those who received aspirin during hospitalization (n = 2,455). As a retrospective study, there were notable differences in most baseline characteristics between the two groups. Compared to the no aspirin group, patients in the aspirin group were older, heavier, and had a higher proportion of males. In the aspirin group, the hemoglobin, BUN, and PO2 levels were higher, while PT was shorter, compared to the no aspirin group. The WBC, potassium, calcium, chloride, creatinine, and PTT in both groups were within the normal range. The aspirin group also exhibited a higher prevalence of chronic comorbidities (hypertension, diabetes, heart failure, cardiovascular disease), as well as higher OASIS and CCI scores. Propensity score matching (PSM) was employed to alleviate the interference from confounders. A total of 2,518 patients were screened, with 1,259 in each group after PSM. The Standardized Mean Differences (SMD) for all variables were < 0.1, and all p-values were > 0.05, indicating that the baseline characteristics were well balanced between the two groups (Table 1). Figure 2 depicted the difference of baseline characteristics between the two groups before and after PSM.

We constructed three cox proportional hazards models adjusted for potential confounding factors. Interestingly, the in-hospital, 30-day, 60-day, 90-day, and 180-day mortality rate in the aspirin group was consistently significant lower than that in the no aspirin group. Table 2 summarizes the detailed associations between the aspirin exposure and mortality rate. Specifically, compared to the no aspirin group, the in-hospital mortality decreased 41% before PSM (HR = 0.59, 95%CI [0.49 ~ 0.70], p < 0.001) and 44% after PSM (HR = 0.56, 95%CI [0.46 ~ 0.69], p < 0.001), the 30-day mortality decreased 38% before PSM (HR = 0.62, 95%CI [0.53 ~ 0.73], p < 0.001) and 36% after PSM (HR = 0.64, 95%CI [0.54 ~ 0.77], p < 0.001), the 60-day mortality decreased 34% before PSM (HR = 0.66, 95%CI [0.58 ~ 0.76], p < 0.001) and 33% after PSM (HR = 0.67, 95%CI [0.57 ~ 0.78], p < 0.001), the 90-day mortality decreased 31% before PSM (HR = 0.69, 95%CI [0.61 ~ 0.79], p < 0.001) and 31% after PSM (HR = 0.69, 95%CI [0.60 ~ 0.81], p < 0.001), and the 180-day mortality decreased 27% before PSM (HR = 0.73, 95%CI [0.64 ~ 0.82], p < 0.001) and 27% after PSM (HR = 0.73, 95%CI [0.63 ~ 0.83], p < 0.001) in the aspirin group. These results indicated that aspirin administration was an independent protective factor for in-hospital, 30-day, 60-day, 90-day, and 180-day mortality in SAE.

Kaplan–Meier curves showed that SAE patients who received aspirin therapy had significantly higher survival rate at 30-day, 60-day, 90-day, and 180-day, compared to those who did not receive aspirin (log-rank test: all p < 0.001) (Figure 3). These results emphasized the potential benefit of aspirin in improving survival outcomes for SAE patients.

The result of Mann–Whitney U test showed that, the median length of hospital stay was significantly longer in the aspirin group compared to the no aspirin group before PSM (8.91 vs. 8.08, p < 0.001) and after PSM (9.10 vs. 8.02, p < 0.001). Additionally, we evaluated the risk of gastrointestinal hemorrhage between the two groups using the chi-square test, which revealed no significant differences both before PSM (p = 0.10) and after PSM (p = 0.37) (Table 3).

The subgroup analysis was applied based on age, gender, hypertension, diabetes, heart failure, cardiovascular disease, CRRT, vasoactive agent, ventilation, SOFA score, and GCS score. The results indicated that aspirin therapy was significantly associated with a reduction in 30-day mortality in the overall population (HR 0.63, 95%CI [0.53–0.75], p < 0.001), as depicted in the Forest Plot in Figure 4. It seemed that aspirin’s protective effect was more pronounced in the following subgroups: age < 60 (HR = 0.42, 95%CI [0.21 ~ 0.86], p = 0.017), male (HR = 0.64, 95%CI [0.50 ~ 0.81], p < 0.001), hypertension (HR = 0.60, 95%CI [0.43 ~ 0.83], p = 0.002), non-diabetes (HR = 0.60, 95%CI [0.48 ~ 0.74], p < 0.001), heart failure (HR = 0.61, 95%CI [0.47 ~ 0.80], p < 0.001), cardiovascular disease (HR = 0.34, 95%CI [0.22 ~ 0.52], p < 0.001), CRRT (HR = 0.14, 95%CI [0.05 ~ 0.40], p < 0.001), ventilation (HR = 0.58, 95%CI [0.42 ~ 0.79], p = 0.001), SOFA>6 (HR = 0.34, 95%CI [0.18 ~ 0.64], p = 0.001), and GCS ≤ 8 (HR = 0.59, 95%CI [0.40 ~ 0.85], p = 0.005). Aspirin reduced 30-day mortality in the subgroup receiving vasoactive agent treatment by 48% (HR = 0.52, 95%CI [0.41 ~ 0.67], p < 0.001), whereas it did not reduced the mortality in those without receiving vasoactive agent treatment. An interaction effect was observed in the SOFA score. Aspirin intervention reduced 30-day mortality by 33% (HR = 0.67, 95%CI [0.56 ~ 0.81], p < 0.001) in patients with SOFA<=6 versus 66% (HR = 0.34, 95%CI [0.18 ~ 0.64], p = 0.001) in those with SOFA>6 (p for interaction =0.034). Our subgroup analysis might contribute to the personalization of aspirin treatment for SAE patients.

We calculated the E-value to assess the association between the mortality and aspirin administration. The result indicated that association between 30-day mortality and aspirin would require an unmeasured confounder with a relative risk greater than 2.50 to account for the observed effect. Therefore, we considered that the residual confounders could explain the observed association, and it appeared that other unknown or unmeasured confounders exerted a relatively smaller effect on 30-day mortality compared to known risk factors. Similarly, the other unknown or unmeasured factors also had a relatively minor impact on in-hospital (E-value>2.97), 60-day (E-value>2.35), 90-day (E-value>2.26), and 180-day mortality (E-value>2.08) (Table 4).

To determine the optimal treatment timing, we divided SAE patients into four groups: aspirin pre-ICU group, aspirin in-ICU group, aspirin pre-ICU and in-ICU group, and no aspirin group. The multifactorial cox proportional hazards models demonstrated that, compared to the no aspirin group, the in-hospital, 30-day, 60-day, 90-day, and 180-day mortality rate of SAE patients in the other three groups decreased significantly (Table 5). As illustrated in Figure 5, Kaplan–Meier curves suggested the survival of SAE patients in pre-ICU group, in-ICU group, and pre-ICU and in-ICU group was higher than that in no aspirin group for 30-day, 60-day, 90-day, and 180-day (log-rank test: all p < 0.05).

To further investigate the effects of different aspirin dosages on mortality, we divided the SAE patients into two groups: low-dose aspirin group (≤300 mg/day) and high-dose aspirin group (>300 mg/day). Then, three cox proportional hazards model was used to evaluate mortality rate. We discovered that patients in the high-dose aspirin group experienced a significantly higher mortality rate compared to those in the low-dose group for in-hospital, 30-day, 60-day, 90-day and 180-day, as detailed in Table 6.

The Kaplan–Meier curves displayed in Figure 6 demonstrated that patients in the high-dose group exhibited a significantly higher mortality rate for 30-day, 60-day, 90-day, and 180-day, compared to those in the low-dose group (log-rank test: all p < 0.05).