This study was a retrospective observational study utilizing the Medical Information Mart for Intensive Care IV database. (MIMIC-IV version 2.2 was most recently updated on January 5, 2023) (24). The MIMIC-IV is a large, single-center database that contains data on patients who were admitted to the intensive care unit (ICU) of a large tertiary care hospital in Boston. The database contains 73,181 hospital admissions for adult patients (18 years of age and older) who were admitted to the intensive care unit from 2008 to 2019.

All information in this database has been deidentified, making it impossible to identify individual patients. As a result, this study is not classified as human subject research and does not require consent from the patients owing to unidentified health information. The MIMIC-IV 2.2 database’s creation was approved by institutional review boards at both the Massachusetts Institute of Technology (MIT, Cambridge, MA) and BIDMC. The author, Fuchao Xu, was granted access to the MIMIC-IV 2.2 database after completing the Human Subject Research course (Certification Number: 52712098).

Patients over 18 years of age with a sepsis diagnosis who received enteral nutrition during their stay in the intensive care unit were considered for inclusion. The sepsis diagnoses complied with the definition of sepsis 3.0 (1). We analyzed only the data from the initial admission to the Medical Intensive Care Unit (MICU) and Surgical Intensive Care Unit (SICU) for each patient. Patients who spent less than 72 h in the ICU were excluded.

In our study, we collected objective patient information, including age, sex, body mass index (BMI), care unit, and race. Additionally, we recorded vital signs taken within the first 24 h of ICU admission, specifically heart rate, respiratory rate, mean arterial pressure, temperature, glucose levels, use of vasopressors, and total urine output during the first 24 h. The laboratory results included pH, arterial oxygen partial pressure (PO2), arterial carbon dioxide partial pressure (PCO2), Pao2/Fio2 ratio, lactate, white blood cell (WBC) count, hemoglobin levels, platelet count, creatinine, blood urea nitrogen, serum albumin, and blood electrolyte levels (chlorine, calcium, potassium, sodium). For variables recorded multiple times during the first 24 h, we selected the values related to the greatest disease severity. Interventions within the initial 24 h comprised invasive mechanical ventilation, continuous renal replacement therapy, invasive arterial pressure monitoring, and a peripherally inserted central catheter. The severity scores during the first day of admission into the ICU were evaluated by the Acute Physiology Score III (APS III), Oxford Acute Severity of Illness Score (OASIS), Logistic Organ Dysfunction System (LODS), Sequential Organ Failure Assessment (SOFA), and Charlson Comorbidity Index (CCI). Comorbidities, such as congestive heart failure, chronic pulmonary disease, mild liver disease, diabetes, chronic kidney disease, and cancer, were recorded. The data extraction code is available on GitHub.¹ PostgreSQL tools (version 15) were used for all data extractions (25).

We included only patients with sepsis who were admitted to the MICU or SICU for the first time. The patient flow diagram is presented in Figure 1. The following exclusion criteria were applied: (1) received enteral nutrition in different ICUs; (2) no enteral nutrition admission during the ICU stay; (3) patients admitted to a care unit other than the medical ICU and surgical ICU; (4) age at ICU admission <18 years; (5) age at ICU admission >89 years (for all patients older than 89 years, the database was adjusted for their age, so we excluded them); (6) ICU stay <72 h; (7) Sepsis diagnosed >48 h after ICU admission; (8) Received parenteral nutrition; (9) Underwent gastrointestinal surgery. (We examined whether septic patients underwent gastrectomy (ICD-9-CM codes 43.5–43.9, ICD-10-PCS codes 0DB60ZZ, 0DB70ZZ), bowel resection (ICD-9-CM codes 45.7x, 45.8x, ICD-10-PCS codes 0DBB0ZZ, 0DBC0ZZ), or gastrointestinal anastomosis (ICD-9-CM codes 44.x) procedures based on their respective diagnostic codes). (10) Started enteral nutrition >7 days after ICU admission; and (11) died within 7 days after ICU admission (to avoid immortal time bias).

By the newly published guidelines for sepsis and nutrition guidelines for critically ill patients. We divided the patients who started EN within 3 d after ICU admission into the early EN group and those who started EN 3–7 d after ICU admission into the delayed EN group (9, 26–29).

We employed the Kolmogorov–Smirnov test and the Shapiro–Wilk test to analyze continuous variables for normal distribution. Continuous variables are represented using either the mean ± standard deviation or the median (interquartile range), depending on their distribution. Categorical variables are presented as proportions. Appropriate statistical tests, such as the t test, analysis of variance, and the Mann–Whitney U test, were used for comparisons. The χ2 test was used for comparing categorical variables.

A propensity score–matching method was applied to compare the outcomes between the EEN and DEN groups (30). Propensity score matching (PSM) was performed to balance the baseline characteristics between the EEN group and the DEN group, so we used a logistic regression model to calculate the propensity score for each patient. In our study, 1:1 nearest neighbor matching was applied with a caliper width of 0.05. After PSM, standardized mean differences (SMDs) were used to evaluate the balance of characteristics between the two groups. A variable can be considered imbalanced between groups if its SMD is greater than 0.1 (31, 32). The implementation of the propensity-matching analysis is in R (version 4.3.1).

The primary outcome was 28-day mortality. Secondary outcomes included 60-day mortality, ICU mortality, length of stay in the ICU (LOS ICU), and the incidence of stage AKI. Kaplan–Meier curves were used to compare the survival of the two groups at 28 and 60 days before and after the propensity-matching procedure. Subgroup analysis was used to identify the specific population that may be more likely to benefit from EEN. We performed subgroup analyses according to age, sex, type of ICU, SOFA score, and lactate. Single-factor logistic analysis and multifactor logistic analysis were used to evaluate the association between early enteral nutrition and 28-day mortality in different subgroups. All subgroup results are presented in a forest plot. For missing data, utilize the MICE package in R to perform multiple imputation on the dataset (33) (Supplementary Figure S1). All the statistical analyses were carried out using R (version 4.3.1), and a p value of <0.05 (two-sided) was used to indicate statistical significance.

The MIMIC-IV database included 32,971 adult patients diagnosed with sepsis, from which our study cohort comprised 1,111 patients (Figure 1). Among these, 786 (70.7%) patients were classified into the early enteral nutrition (EEN) group, initiating enteral nutrition within 3 days after admission, while 325 (29.3%) patients were assigned to the delayed enteral nutrition (DEN) group, commencing enteral nutrition between 3 and 7 days after admission. Characteristics of patients in the EEN and DEN groups are summarized in Table 1. Notably, patients in the DEN group demonstrated significantly higher lactate levels (2.80 [1.70, 4.60] vs. 2.20 [1.50, 3.69]; p<0.001) and elevated disease severity scores, including LODS score (7.00 [6.00, 10.00] vs. 7.00 [5.00, 9.00]; p<0.001), and APS III score (66.00 [51.00, 82.00] vs. 57.50 [44.00, 74.00]; p<0.001) compared to the EEN group. Within the initial 24 h following ICU admission, the EEN group exhibited a higher likelihood of requiring invasive mechanical ventilation (631 (80.3%) vs. 226 (69.5%); p<0.001) and first-day urine output (1382.50 [770.50, 2233.75] vs. 1113.00 [455.00, 1955.00]; p<0.001).

Before propensity score matching (PSM), 28-day mortality was significantly lower in the early enteral nutrition group compared to the delayed enteral nutrition group (161 (20.5%) vs. 88 (27.1%); p = 0.018) (Table 2). The univariate Kaplan–Meier survival curve for 28 days also indicated that the early enteral nutrition group had a longer survival time (HR = 0.719, 95% CI: 0.546–0.947; p = 0.019) (Figure 2A). After PSM, all standardized mean differences were less than 0.1, indicating similar distributions of baseline variables in both groups (Supplementary Table S1). Following propensity matching, 28-day mortality was 4.2% lower in the early enteral nutrition group compared to the delayed enteral nutrition group, but this difference was not statistically significant (65 (22.4%) vs. 77 (26.6%); p = 0.247). The 28-day Kaplan–Meier curve post-propensity matching echoed the propensity-matched result (HR = 0.821, 95% CI: 0.591–1.141, p = 0.240) (Figure 2B). Additionally, we conducted multifactor logistic regression analysis to validate the results of propensity matching, incorporating age, gender, and covariates with p-values <0.05. The results of the multifactor logistic analysis (OR = 0.778 (0.559–1.084); p = 0.138) were consistent with those of propensity matching. Propensity matching suggested that early enteral nutrition did not significantly reduce 28-day mortality in sepsis patients compared with delayed enteral nutrition.

Before propensity matching, 60-day mortality was lower in the early enteral nutrition group than in the late enteral nutrition group (216 (27.5) vs. 113 (34.8); p = 0.017) (Table 2). The univariate Kaplan–Meier survival curve also showed longer survival in the early enteral nutrition group (HR = 0.744, 95% CI: 0.585–0.946, p = 0.017) (Figure 2C). The EEN group had shorter length of ICU stay (8.3 [5.2, 12.3] vs. 10.0 [7.5–14.2]; p < 0.001) and a lower incidence of stage 3 AKI (333 (42.4) vs. 184 (56.6); p < 0.001). After propensity matching, we found no statistically significant differences in 60-day mortality (95 (32.8) vs. 99 (34.1); p = 0.725) between the EEN and DEN groups. The 60-day Kaplan–Meier curve after propensity matching was consistent with the result after propensity matching (hazard ratio = 0.935, 95% CI: 0.705–1.239, p = 0.639) (Figure 2D). Multivariate logistic analysis was performed to verify the results of propensity matching and found that 60-day mortality (aOR = 0.859, 95% CI: 0.629–1.173, p = 0.339) in both groups were consistent with the post-PSM results. However, the EEN group still had a shorter length of ICU stay (8.3 [5.2, 12.4] vs. 110.1 [7.5, 14.2]; p < 0.001) (Figure 3) and a lower incidence of severe kidney injury (143 (49.3) vs. 161 (55.5); p = 0.030) than the DEN group after propensity matching (Figure 4).

We conducted several subgroup analyses based on propensity-matched data to explore the relationship between early enteral nutrition and 28-day mortality across different subgroups of sepsis patients. Patients were categorized based on sex, age, ICU type, SOFA score, and lactate levels. Single-factor logistic analysis and multifactor logistic analysis were utilized to assess the association between early enteral nutrition and 28-day mortality within various subgroups. The results for all subgroups are illustrated in a forest plot (Figure 5). Our findings indicate that early enteral nutrition decreased the 28-day mortality rate in patients with lactate levels ≤4 mmol/L (aOR = 0.579, 95% CI: 0.361–0.930, p = 0.024). The COX survival analysis-adjusted Kaplan–Meier survival curve is depicted in Figure 6.