This investigation constitutes a retrospective observational analysis utilizing the Medical Information Mart for Intensive Care IV (MIMIC-IV) database (version 3.0, updated July 23, 2024). MIMIC-IV represents a comprehensive, single-center repository encompassing clinical data from patients admitted to the intensive care unit (ICU) of a tertiary care hospital in Boston, Massachusetts, USA. The database comprises hospitalization records of 94,458 adult patients (≥18 years) admitted to the ICU between 2008 and 2022.

All data within the database have undergone rigorous de-identification procedures, ensuring the anonymity of individual patients. Consequently, this study does not qualify as human subjects research under current regulatory guidelines, and the use of de-identified health information obviates the requirement for patient consent. The development and maintenance of the MIMIC-IV 3.0 database received ethical approval from the Institutional Review Boards of the Massachusetts Institute of Technology (MIT, Cambridge, Massachusetts) and Beth Israel Deaconess Medical Center (BIDMC). Author Lamei Ouyang obtained authorized access to the MIMIC-IV 3.0 database following successful completion of the requisite Human Subject Research course (certification number: 64058594).

This study enrolled patients aged ≥18 years diagnosed with COPD according to established diagnostic criteria (14), who underwent invasive mechanical ventilation and received EN during their ICU stay. The analysis was restricted to patients experiencing their first ICU admission. Exclusion criteria comprised: (1) age <18 years at ICU admission; (2) ICU stay duration <48 h; (3) receive invasive mechanical ventilation <48 h; (4) no enteral nutrition admission was received during the ICU stay; and (5) started enteral nutrition > 7 days after ICU admission (Figure 1 illustrates the patient flow diagram).

Comprehensive patient data were systematically collected, encompassing demographic characteristics (age, sex, BMI, and ethnicity) and physiological parameters recorded within the initial 24 h of ICU admission, including heart rate, respiratory rate, body temperature, mean arterial pressure, blood glucose levels, and 24-h urine output. Laboratory analyses included arterial blood gas parameters (pH, partial pressure of oxygen [PO₂], partial pressure of carbon dioxide [PCO₂], PaO₂/FiO₂ ratio, lactate), hematological indices (white blood cell count [WBC], hemoglobin, platelet count), coagulation profile (activated partial thromboplastin time [APTT]), and biochemical parameters (creatinine, blood urea nitrogen, and electrolyte levels [chloride, calcium, potassium, sodium]). For variables with multiple measurements within 24 h, mean values were calculated.

Interventional data captured within the first 24 h included administration of vasopressors, implementation of continuous renal replacement therapy, invasive arterial pressure monitoring, and placement of peripherally inserted central catheters. Disease severity was quantified using validated scoring systems: sequential Organ Failure Assessment (SOFA), Glasgow Coma Scale (GCS), Acute Physiology Score III (APS III), and Charlson Comorbidity Index (CCI). Documented comorbidities included congestive heart failure, cerebrovascular disease, liver disease, diabetes mellitus, renal disease, and cancer.

In accordance with the most recent clinical guidelines for nutritional support in critically ill patients, participants were stratified into two distinct cohorts: the early enteral nutrition (EEN) group, defined by the initiation of EN within 48 h following ICU admission, and the delayed enteral nutrition (DEN) group, characterized by EN initiation beyond 48 h post-ICU admission (15).

The normality of continuous variables was assessed using both the Kolmogorov-Smirnov and Shapiro-Wilk tests. As all continuous variables exhibited non-normal distributions, they were expressed as medians with interquartile ranges (IQRs), and comparisons between the two groups were performed using the Mann-Whitney U test. Categorical variables were presented as proportions and analyzed using the chi-square test.

To mitigate potential confounding factors, propensity score matching (PSM) was employed before comparing outcomes between the EEN and DEN groups (16). A propensity score was calculated for each patient using a logistic regression model including 39 potential baseline prognostic and risk factors in Table 1, and then individuals were matched using a 1:1 nearest neighbor matching approach with a caliper width of 0.2 standard deviations of the propensity score logit. Post-matching balance was evaluated using standardized mean differences (SMDs), with an SMD threshold of >0.1 indicating imbalance (17, 18). All PSM analyses were conducted using R software (version 4.4.1).

The primary outcome was 28-day mortality. Secondary outcomes included 60-day mortality, ICU mortality, the incidence of ventilator-associated pneumonia (VAP), hospital length of stay (LOS hospital), ICU length of stay (LOS ICU), and duration of invasive mechanical ventilation. For both primary and secondary endpoints, categorical variables were compared between groups using the Chi-square test, while continuous variables were compared using the Mann-Whitney U test. Subsequently, the 28-day survival, ICU survival and 60-day survival were estimated using Kaplan-Meier (K-M) survival curves, and the hazard ratio between two groups was estimated using the Cox proportional hazards model. Subgroup analyses were performed to identify populations potentially benefiting from EEN, stratified by age, sex, lactate level, BMI, and PaO₂/FiO₂ ratio. The associations between EEN and 28-day mortality were quantified using univariate logistic analysis, with results visualized in forest plots. Furthermore, to delineate independent associations, we performed distinct sensitivity analyses by excluding patients with cancer diagnoses, non-white individuals, or patients not receiving vasoactive agents. These analyses aimed to examine the reliability and applicability of our results across diverse patient groups and clinical scenarios. Variables exceeding 30% missingness were excluded from the analysis. Then, missing data for covariates were addressed using multiple imputation via the MICE package in R (Supplementary Figure S1). The primary analysis model was applied to each of 20 imputed dataset, and estimates were pooled using Rubin's rules to derive final parameter estimates with standard errors accounting for missing data uncertainty. All statistical analyses were conducted in R (version 4.4.1), with a two-sided p-value < 0.05 considered statistically significant.

The study cohort consisted of 1,052 patients, as depicted in Figure 1. Among these, 513 patients (48.76%) were classified into the EEN group, defined by the initiation of EN within 48 h of ICU admission, while 539 patients (51.24%) were assigned to the DEN group, characterized by EN initiation beyond 48 h of ICU admission. The demographic and clinical characteristics of both groups are presented in Table 1. Notably, during the first 24 h of ICU admission, the DEN group demonstrated significantly higher lactate levels (1.60 [1.10, 2.50] vs. 1.36 [1.00, 1.97]; p < 0.001) and elevated disease severity scores, including the Acute Physiology Score III (APS III) (55.00 [45.00, 71.00] vs. 54.00 [42.00, 68.00]; p = 0.038) and Charlson Comorbidity Index (CCI) (6.00 [5.00, 8.00] vs. 6.00 [5.00, 8.00]; p = 0.018). Additionally, the DEN group exhibited a higher prevalence of vasopressor support (451 [83.67%] vs. 353 [68.81%]; p < 0.001), invasive arterial pressure monitoring (428 [79.41%] vs. 322 [62.77%]; p < 0.001), and continuous renal replacement therapy (89 [16.51%] vs. 51 [9.94%]; p = 0.002). After PSM, there was no significant differences in all these baseline characteristics, with all SMDs < 0.10 (Supplementary Table S1).

Prior to PSM, no statistically significant difference in 28-day mortality was observed between the EEN group and the DEN group (145 [28.27%] vs. 179 [33.21%]; p = 0.082) (Table 2). The univariate Kaplan–Meier survival curve for 28 days further confirmed the absence of a significant difference in survival time between the EEN and DEN groups (HR = 0.825, 95% CI: 0.663–1.028; p = 0.086) (Figure 2A). Post-matching analysis revealed a 2.31% reduction in 28-day mortality in the EEN group compared to the DEN group; however, this difference did not reach statistical significance (119 [30.51%] vs. 128 [32.82%]; p = 0.488). The 28-day Kaplan–Meier curve after propensity matching was consistent with the result (HR = 0.921, 95% CI: 0.718–1.182; p = 0.518) (Figure 2B).

Prior to PSM, the EEN group exhibited significantly lower ICU mortality compared to the DEN group (83 [16.18%] vs. 119 [22.08%]; p = 0.015), whereas no significant difference was observed in 60-day mortality (182 [35.48%] vs. 220 [40.82%]; p = 0.075) (Table 2). Univariate Kaplan-Meier analysis for ICU survival (HR = 0.954, 95% CI: 0.720–1.263; p = 0.742) (Figure 2C) and 60-day survival (HR = 0.835, 95% CI: 0.686–1.016; p = 0.071) (Figure 2E) demonstrated no significant intergroup difference. The incidence of VAP did not differ significantly between the EEN and DEN groups (98 [19.10%] vs. 121 [22.45%]; p = 0.182). However, the EEN group showed significantly shorter total hospital stays (13.58 [8.35, 20.37] vs. 16.74 [10.87, 25.86] days; p < 0.001), reduced ICU stays (8.17 [5.17, 14.45] vs. 10.65 [7.44, 17.20] days; p < 0.001), and decreased duration of invasive mechanical ventilation (116.90 [79.58, 233.68] vs. 144.88 [95.47, 260.12] hours; p < 0.001). Following PSM, no significant differences were observed in ICU mortality (67 [17.18%] vs. 83 [21.28%]; p = 0.146) or 60-day mortality (149 [38.21%] vs. 155 [39.74%]; p = 0.660). The ICU Kaplan–Meier curve (HR = 0.978, 95% CI: 0.708–1.350; p = 0.892) (Figure 2D) and 60-day Kaplan–Meier curve (HR = 0.948, 95% CI: 0.757–1.187; p = 0.640) (Figure 2F) post-propensity matching echoed the propensity-matched result. VAP incidence remained comparable between groups (81 [20.77%] vs. 91 [23.33%]; p = 0.388). Nonetheless, the EEN group maintained significantly shorter total hospital stays (14.64 [8.97, 21.46] vs. 16.63 [10.68, 25.58] days; p = 0.001), reduced ICU stays (9.08 [5.39, 15.35] vs. 10.07 [7.24, 16.11] days; p < 0.001), and decreased duration of invasive mechanical ventilation (127.50 [80.55, 247.23] vs. 137.94 [93.05, 266.69] hours; p = 0.023).

Subgroup analyses based on propensity-matched data were conducted to explore the association between early EN and 28-day mortality across subgroups of COPD patients. Stratifications included sex, age, lactate levels, BMI, and PaO₂/FiO₂ ratio. Results revealed that EEN was significantly associated with a reduction in 28-day mortality in patients with PaO₂/FiO₂ > 200 (OR = 0.626, 95% CI: 0.414–0.943; p = 0.026) (Figure 3). Adjusted Kaplan-Meier survival curves from Cox regression analysis are presented in Figure 4.

There were 1,052 patients in the entire cohort. We performed further sensitivity analyses after excluding 140 patients with cancer diagnoses (HR: 0.871; 95% CI: 0.665–1.140; p = 0.314), 357 non-white individuals (HR: 0.842; 95% CI: 0.615–1.153; p = 0.284), and 248 patients not receiving vasoactive agents (HR: 0.959; 95% CI: 0.739–1.246; p = 0.756), respectively, and the results were consistent with the primary outcome (Table 3).