Early versus delayed enteral nutrition in patients with sepsis: a propensity score-matched cohort study at a tertiary hospital in Hebei, China

Abstract

Objectives

Enteral nutrition (EN) constitutes a critical therapeutic intervention in the management of patients with sepsis; however, the optimal timing for its initiation remains uncertain. This study sought to evaluate the impact of early versus delayed EN on mortality rates and clinical outcomes among adult patients with sepsis in intensive care units (ICU).

Design

Retrospective cohort study using propensity score matching (PSM) methodology.

Setting

A tertiary hospital ICU in Hebei Province, China, from 2015 to 2024.

Participants

This retrospective analysis involved adult ICU patients with sepsis (meeting Sepsis 3.0 criteria) from January 2015 to December 2024 who began EN within 7 days of admission. Patients were classified into early (within 2 days) or delayed (2–7 days) EN groups based on when nutrition was initiated.

Primary and secondary outcome measures

Primary outcomes were 28-day and 60-day mortality, with secondary outcomes including hospital/ICU stay length, mechanical ventilation duration and nutrition-related complications.

Results

A study of 2205 patients compared early EN (EEN) in 1500 patients (68.0%) with delayed EN in 705 patients (32.0%). After PSM, the EEN group showed a higher mortality risk at 28 days (HR 1.44, 95% CI 1.08 to 1.92) and 60 days (HR 1.45, 95% CI 1.11 to 1.89), confirmed by multivariable Cox regression and inverse probability weighting. EEN patients also had more gastric retention (OR 1.77, 95% CI 1.14 to 2.79). The increased mortality risk was notably pronounced in younger male patients with a body mass index of less than 24 kg/m², abdominal infections, those with severe sepsis and patients necessitating vasopressor support at a norepinephrine equivalent of ≥0.1 µg/kg/min. Sensitivity analyses supported these findings.

Conclusions

Our retrospective analysis of 2205 propensity score-matched patients with sepsis found that EEN is linked to higher short-term mortality, especially in younger males, those with severe illness, abdominal infections or needing moderate to high vasopressor support. This suggests that personalised nutritional timing might be better than universal early feeding. However, these findings are preliminary and need confirmation through randomised controlled trials.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Propensity score matching was used to minimise selection bias and to balance baseline characteristics between the early enteral nutrition and delayed enteral nutrition groups.

• The relatively large sample size, derived from a real-world clinical environment, potentially enhances the generalisability of the results.

• The generalisability of the findings from this single-centre retrospective study may be constrained.

• Residual confounding from unmeasured variables may have influenced results.

• Time-dependent confounding might be present, as clinical and laboratory data obtained at intensive care unit admission may not accurately reflect the patients’ condition at the initiation of enteral nutrition.

Introduction

In the domain of critical care medicine, sepsis continues to be a predominant cause of both morbidity and mortality, posing substantial challenges to global public health. Recent global estimates suggest that there are 48.9 million cases of sepsis and 11.0 million sepsis-related deaths every year.¹ Furthermore, those who survive often suffer from enduring sequelae.^{2 3} Within the intensive care unit (ICU), sepsis is a major determinant of patient mortality.⁴ Despite progress in early detection and management strategies, the incidence and mortality rates associated with sepsis persist at distressingly elevated levels.⁵

In patients with sepsis, appropriate nutritional therapy has the potential to enhance clinical outcomes, bolster immune function, facilitate wound healing and decrease the risk of infections.^{6 7} Sepsis triggers a series of metabolic alterations, including glycogen depletion, increased lipolysis and muscle protein catabolism.^{8 9} These metabolic disturbances heighten the risk of nutritional deficiencies, further impeding recovery.¹⁰ Within this context, enteral nutrition assumes a multifaceted role: it not only supplies essential calories and nutrients but also preserves intestinal integrity by reducing dysbiosis and preventing increased intestinal permeability, thereby mitigating systemic inflammatory responses.¹¹

Nonetheless, the optimal timing for initiating enteral nutrition support in patients with sepsis remains a topic of ongoing scholarly debate. While some studies advocate for early nutritional intervention, citing potential benefits, current guidelines endorsing early implementation are not substantiated by robust evidence.¹² Moreover, early enteral nutrition (EEN) is associated with an elevated risk of certain complications. For example, research has indicated that the premature introduction of enteral nutrition may result in abdominal discomfort or gastrointestinal complications, particularly in patients with high-risk profiles.¹³ Unstable haemodynamics can further complicate the initiation of enteral nutrition, potentially leading to visceral ischaemia and increasing the demand on systemic oxygen delivery.¹⁴ Additionally, early implementation of enteral nutrition may be associated with a heightened risk of non-occlusive mesenteric ischaemia, a rare but life-threatening complication.¹⁵

Given the ongoing controversies surrounding these issues, this study conducted a retrospective cohort analysis using data from a single tertiary hospital’s electronic medical records to assess the benefits and risks of EEN in sepsis patients.

Methods

Study design

This investigation was a propensity score-matched, retrospective, observational cohort study conducted at Hebei Medical University Second Hospital. The reporting of this cohort study adhered to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines.¹⁶ Given the retrospective design and the anonymity of the data, the requirement for obtaining written informed consent was waived.

Setting and participants

Electronic medical records of patients diagnosed with sepsis and admitted to the ICU at Hebei Medical University Second Hospital between January 2015 and December 2024 were collected for analysis. The diagnosis of sepsis was based on the Sepsis 3.0 criteria.¹⁷ Data analysis was conducted exclusively for each patient’s initial ICU admission. The exclusion criteria were as follows: (1) patients who received enteral nutrition therapy in the general ward prior to ICU admission; (2) patients who did not receive enteral nutrition therapy during their ICU stay; (3) patients younger than 18 years at the time of ICU admission; (4) patients with an ICU stay of less than 72 hours; (5) patients with a history of gastrointestinal surgery, excluding those with a history of gastrectomy, intestinal resection or gastrointestinal anastomosis; (6) patients who received parenteral nutrition; (7) patients who commenced enteral nutrition more than 7 days after ICU admission and (8) patients who died within 7 days of ICU admission, to mitigate immortal time bias.

Eligible patients were stratified into two cohorts: the EEN group, comprising individuals who commenced enteral nutrition within 2 days of ICU admission, and the delayed enteral nutrition (DEN) group, consisting of those who initiated enteral nutrition between 2 and 7 days post-ICU admission. The patient flow diagram is depicted in figure 1.

Figure 1. Patient flow diagram for the study. DEN, delayed enteral nutrition; EEN, early enteral nutrition; ICU, intensive care unit.

Outcomes

The principal outcomes comprise the hospital mortality rates at 28-day and 60-day post-ICU admission. Secondary outcome measures include the duration of hospitalisation, length of ICU stay, duration of mechanical ventilation and complications related to enteral nutrition, such as diarrhoea, gastric retention and gastrointestinal bleeding.

Data collection

Data were retrospectively extracted from the Hospital Information System (HIS), encompassing objective patient demographics, including age, gender, and body mass index (BMI). In addition, vital signs recorded within the 24 hours preceding ICU admission were compiled, specifically heart rate (HR), respiratory rate (RR), mean arterial pressure, and body temperature. Laboratory parameters included white cell count, haemoglobin levels, platelet count (PLT), lymphocyte count (LYM), C reactive protein, serum albumin, serum prealbumin, creatinine, blood urea nitrogen (BUN), alanine aminotransferase, total bilirubin and serum electrolyte concentrations (sodium, Na; potassium, K; calcium, Ca; bicarbonate), as well as prothrombin time, fibrinogen, arterial partial pressure of oxygen, lactate, procalcitonin, total cholesterol, triglycerides, glucose, glycated haemoglobin and interleukin-6 (IL-6). Infection sites were categorised according to anatomical location, including pulmonary, abdominal, skin and soft tissue and other sites. The severity of disease was evaluated using established scoring systems, including the Acute Physiology and Chronic Health Evaluation (APACHE II),¹⁸ Sequential Organ Failure Assessment (SOFA)¹⁹ and Simplified Acute Physiology Score,²⁰ modified Nutrition Risk in the Critically III (mNUTRIC)²¹ and Charlson Comorbidity Index (CCI).²² Recorded comorbidities encompassed conditions such as congestive heart failure, chronic obstructive pulmonary disease, liver disease, diabetes, kidney disease and cancer. Documented interventions within the first 24 hours of admission to ICU comprised invasive mechanical ventilation (IMV), continuous renal replacement therapy (CRRT) and the administration of vasopressor agents, with detailed records of the maximum vasopressor doses administered within this period. The norepinephrine equivalent (NEE) score is defined as a formula that calculates the dosage of vasopressors relative to the norepinephrine dosage (μg/kg/min), excluding inotropic agents.²³

NEE=norepinephrine dose(μg/kg/min)+epinephrie dose(μg/kg/min)+0.01×dopamine dose(μg/kg/min)+0.06×phenylephrine dose(μg/kg/min)+2.5×vasopressin dose(U/min) 0.0025×angiotensin II dose(ng/kg/min)+10×terlipressin dose(μg/kg/min)+0.2×methylene blue dose(mg/kg/hour)+8×metaraminol dose(μg/kg/min)+0.02×hydroxocobalamin dose(g)+0.4×midodrine dose(μg/kg/min)

All the previously mentioned variables were derived from the initial measurements taken within the first 24 hours following ICU admission to maintain temporal consistency.

Missing values and outliers of variables

Numerous variables exhibited missing values, as illustrated in online supplemental figure 1A. Variables with more than 20% missing data were excluded from the analysis. The remaining dataset underwent multiple imputation using the MICE package in R, resulting in the generation of five complete datasets after 20 iterations. Rubin’s Rules were employed to pool the estimates, thereby ensuring robust overall estimates. To assess the consistency of the distributions, kernel density curves of the observed and imputed values were overlaid, as shown in online supplemental figure 1B. Continuous variables were evaluated for extreme observations, with values falling below the 1st percentile or exceeding the 99th percentile identified as outliers. These outliers were subsequently winsorised to the corresponding 1st or 99th percentile values, respectively.

Propensity score matching

Propensity score matching (PSM) serves as an effective technique to mitigate confounding effects in non-randomised clinical trials by equilibrating differences between groups. In this study, propensity scores for the two patient cohorts were estimated using a multivariable logistic regression model. The nearest neighbour matching method was applied for 1:1 matching of eligible patients, with a calliper width set at 0.1. To evaluate the balance of confounding variables, standardised differences were employed, with a significant imbalance defined as a value exceeding 10% (see figure 2). The implementation of the propensity-matching analysis is in R (V.4.5.1).

Figure 2. Covariate balance before and after propensity score matching (A), Propensity score distribution before and after propensity score matching (B). Alb, albumin; ALT, alanine aminotransferase; APACHE II, Acute Physiology and Chronic Health Evaluation II; BMI, body mass index; BUN, blood urea nitrogen; Ca, calcium; CCI, Charlson Comorbidity Index; CRP, C reactive protein; CRRT, continuous renal replacement therapy; Cr, creatinine; EN, enteral nutrition; FIB, fibrinogen; Hb, haemoglobin; HR, heart rate; IMV, invasive mechanical ventilation; K, potassium; LMY, lymphocyte count; MAP, mean arterial pressure; mNUTRIC, Modified Nutrition Risk in the Critically Ill; Na, sodium; NEE, norepinephrine equivalent; PA, serum prealbumin; PaO₂, partial pressure of oxygen; PLT, platelet count; PT, prothrombin time; RR, respiratory rate; SAPS-II, Simplified Acute Physiology Score II; SOFA, Sequential Organ Failure Assessment; TBL, total bilirubin level; WCC, white cell count.

Statistical analysis

Continuous variables were assessed for normality using the Shapiro-Wilk test. Variables exhibiting a normal distribution are reported as mean±SD and were analysed using one-way analysis of variance. Variables not conforming to a normal distribution are presented as median with IQR and were analysed using the Kruskal-Wallis test. Categorical variables are expressed as counts and percentages and were compared using Fisher’s exact test. To address potential multicollinearity, which could compromise the accuracy of multivariable regression analyses, one variable with a lower correlation to the outcome was excluded when two variables exhibited high correlation (Spearman coefficient >0.7) or when the variance inflation factor exceeded 10, as illustrated in online supplemental figures 1C and D.

The association between the timing of enteral nutrition and clinical outcomes: We conducted a Kaplan-Meier analysis to estimate the 28-day and 60-day mortality rates for patients in the EEN and DEN groups. To evaluate the effect of enteral nutrition timing on mortality, we used a multivariable Cox proportional hazards regression model, presenting HRs and 95% CIs. Covariates were selected via Lasso regression, as illustrated in online supplemental figures 2A and B), and included variables such as age, gender, BMI, CRRT, vasopressor use, infection sites, septic shock, CCI, liver disease, renal disease, PLT, PA, BUN, K, FIB, lactate and glucose. These variables were incorporated into the regression model. Additionally, we assessed the relationship between the timing of enteral nutrition and the occurrence of enteral nutrition-related complications, expressing the results as OR with 95% CIs. To enhance the robustness of our findings, we employed PSM, optimal matching methods through inverse probability weighting (IPW), and conducted regression analyses to verify the consistency of the results (covariate balance shown in online supplemental figure 3).

Stratification and interaction analysis: To assess whether associations varied across different subgroups, stratification and interaction analyses were conducted. The subgroups considered included age, gender, BMI, infection sites, SOFA score, APACHE II score, mNUTRIC score, presence of septic shock, utilisation of CRRT, IMV, dosage of vasopressors and lactate levels.

Sensitivity analysis: Two sensitivity analyses were undertaken to ensure the robustness and representativeness of the findings. Initially, an analysis was performed on the primary results after excluding patients with missing data. Furthermore, patients who had been in the ICU for more than 72 hours but died within 7 days were previously excluded to mitigate concerns of immortal time bias. However, recognising that such exclusion could introduce selection bias, a subsequent sensitivity analysis was conducted, this time including patients who were in the ICU for at least 72 hours and died within 7 days.

Given the retrospective nature of this study, we did not perform a priori sample size calculation. However, post hoc power analysis was performed using PASS software (V.25.0.2) with a two-sided log-rank test (Freedman method) at α=0.05 to assess the adequacy of our sample size.

All statistical analyses were conducted using R (V.4.5.1), with a two-sided p value of less than 0.05 employed to denote statistical significance.

Patient and public involvement

Patients or the public were not involved in the design, conduct, reporting or dissemination plans of this research.

Results

Baseline characteristics

Between January 2015 and December 2024, a total of 7546 patients underwent retrospective screening, with 2205 patients ultimately fulfilling the study’s selection criteria (figure 1). Of these, 1500 (68.0%) received EEN, and 705 (32.0%) received DEN. After matching, the baseline characteristics demonstrated satisfactory balance (table 1). Figure 2A depicts the standardised mean difference between variables both prior to and following PSM, whereas figure 2B illustrates the distribution of the propensity score.

Table 1. Baseline characteristics of the participants.

Variables	Before PSM			After PSM
	EEN	DEN	SMD	EEN	DEN	SMD
N	1500	705		603	603
Age (years)	59.46 (16.26)	57.85 (17.68)	0.095	61.00 (47.00, 71.00)	60.00 (46.00, 71.50)	0.019
Male, n (%)	949 (63.3)	411 (58.3)	0.102	356 (59.0)	349 (57.9)	0.024
BMI (kg/m²)	24.63 (23.23, 26.12)	24.50 (23.30, 26.12)	0.024	24.61 (23.22, 26.12)	24.49 (23.32, 26.12)	0.005
Admission type			0.116			<0.001
Surgical, n (%)	444 (29.6)	247 (35.0)		199 (33.0)	199 (33.0)
Medical, n (%)	1056 (70.4)	458 (65.0)		404 (67.0)	404 (67.0)
Septic shock, n (%)	382 (25.5)	279 (39.6)	0.305	209 (34.7)	217 (36.0)	0.028
Infection site
Lung, n (%)	771 (51.4)	245 (34.8)	0.333	249 (41.3)	234 (38.8)	0.051
Abdominal, n (%)	350 (23.3)	349 (49.5)	0.619	239 (39.6)	263 (43.6)	0.081
Skin and soft tissue, n (%)	135 (9.0)	42 (6.0)	0.106	49 (8.1)	42 (7.0)	0.044
Other sites, n (%)	244 (16.3)	69 (9.8)	0.176	66 (11.0)	64 (10.6)	0.011
Vital indicators
Temperature (°C)	36.70 (36.40, 37.40)	36.80 (36.40, 37.50)	0.091	36.80 (36.40, 37.50)	36.80 (36.40, 37.50)	0.016
HR (bpm)	94.00 (80.00, 109.50)	103.00 (87.00, 122.00)	0.403	101.00 (86.00, 118.00)	101.00 (85.00, 118.00)	0.007
RR (bpm)	20.00 (18.00, 22.00)	22.00 (20.00, 25.00)	0.282	20.00 (19.00, 24.00)	21.00 (20.00, 24.00)	0.026
MAP (mm Hg)	95.67 (85.00, 107.33)	92.33 (80.00, 104.00)	0.198	93.33 (82.50 104.00)	93.33 (81.50, 104.33)	0.002
Medications and interventions
IMV, n (%)	887 (59.1)	339 (48.1)	0.223	315 (52.2)	292 (48.4)	0.076
CRRT, n (%)	84 (5.6)	114 (16.2)	0.344	63 (10.5)	69 (11.4)	0.032
Vasopressors, n (%)	552 (36.8)	254 (36.0)	0.016	201 (33.3)	212 (35.2)	0.038
NEE (µg/kg/min)	0.00 (0.00, 0.20)	0.00 (0.00, 0.25)	0.175	0.00 (0.00, 0.21)	0.00 (0.00, 0.22)	0.039
Disease severity scoring system
SOFA	7.00 (5.00, 10.00)	8.00 (5.00, 11.00)	0.129	7.00 (5.00, 10.00)	8.00 (5.00, 10.00)	0.059
APACHE II	15.00 (11.00, 20.00)	14.00 (11.00, 19.00)	0.070	14.00 (10.00, 19.00)	14.00 (11.00, 19.00)	0.021
SAPS II	34.00 (28.00, 39.00)	33.00 (28.00, 38.00)	0.074	33.00 (28.00, 38.00)	33.00 (28.00, 38.00)	0.040
mNUTRIC	4.50 (3.00, 6.00)	5.00 (3.00, 6.00)	0.007	4.00 (3.00, 6.00)	4.00 (3.00, 6.00)	0.028
CCI	3.00 (2.00, 5.00)	3.00 (2.00, 5.00)	0.044	3.00 (2.00, 5.00)	3.00 (2.00, 5.00)	<0.001
Comorbidities
CHF, n (%)	254 (16.9)	109 (15.5)	0.040	100 (16.6)	95 (15.8)	0.023
Chronic pulmonary disease, n (%)	280 (18.7)	128 (18.2)	0.013	105 (17.4)	115 (19.1)	0.043
Liver disease, n (%)	299 (19.9)	201 (28.5)	0.201	164 (27.2)	153 (25.4)	0.041
Renal disease, n (%)	274 (18.3)	158 (22.4)	0.103	126 (20.9)	122 (20.2)	0.016
Diabetes, n (%)	535 (35.7)	229 (32.5)	0.067	209 (34.7)	202 (33.5)	0.024
Cancer, n (%)	185 (12.3)	81 (11.5)	0.026	59 (9.8)	69 (11.4)	0.054
Laboratory indicators
WCC (10∧9 /L)	10.70 (8.00, 14.60)	11.00 (7.80, 15.60)	0.068	11.00 (7.79, 15.10)	10.90 (7.75, 15.58)	0.007
Hb (g/L)	110.00 (94.00, 129.00)	112.00 (96.00, 131.00)	0.070	110.00 (94.00, 128.00)	112.00 (95.00, 130.00)	0.043
PLT (10∧9 /L)	177.00 (119.00, 244.00)	157.00 (90.00, 223.00)	0.220	158.00 (104.50, 232.00)	162.00 (97.50, 226.50)	0.016
LYM (10∧9 /L)	0.72 (0.48, 1.10)	0.70 (0.40, 1.10)	0.044	0.70 (0.42, 1.10)	0.70 (0.43, 1.10)	0.007
CRP (mg/L)	79.30 (29.60, 177.98)	131.97 (42.00, 196.60)	0.241	118.40 (36.20, 194.20)	120.80 (38.70, 191.70)	0.014
ALB (g/L)	30.90 (27.20, 35.10)	29.00 (25.90, 32.90)	0.345	29.30 (25.80, 33.40)	29.40 (26.20, 33.40)	0.018
PA (g/L)	0.14 (0.09, 0.19)	0.12 (0.08, 0.18)	0.214	0.12 (0.08, 0.18)	0.13 (0.08, 0.18)	0.020
Cr (μmol/L)	76.00 (54.00, 122.47)	92.00 (64.00, 157.00)	0.234	82.00 (56.50, 135.00)	85.00 (61.00, 142.25)	0.029
BUN (mmol/L)	7.20 (4.80, 12.00)	8.50 (5.30, 14.20)	0.163	8.11 (5.20, 14.60)	8.20 (5.10, 13.35)	0.024
ALT (U/L)	25.85 (15.17, 52.87)	27.40 (15.00, 71.00)	0.096	27.40 (15.30, 57.50)	26.40 (14.95, 63.70)	0.008
TBIL (μmol/L)	13.20 (8.61, 20.72)	14.90 (9.40, 25.90)	0.124	14.10 (8.70, 23.50)	14.52 (9.17, 25.55)	<0.001
Na (mmol/L)	140.60 (136.50, 144.70)	140.50 (136.20, 145.00)	0.015	140.60 (136.25, 144.80)	140.10 (136.05, 145.05)	0.019
K (mmol/L)	4.05 (3.68, 4.46)	4.08 (3.75, 4.52)	0.103	4.05 (3.72, 4.48)	4.04 (3.75, 4.50)	0.018
Ca (mmol/L)	2.05 (1.93, 2.15)	1.97 (1.85, 2.09)	0.336	2.00 (1.88, 2.11)	1.99 (1.87, 2.10)	0.030
Bicarbonate (mmol/L)	21.60 (18.90, 24.60)	20.60 (17.30, 24.00)	0.171	21.40 (18.50, 23.95)	20.90 (17.60, 24.30)	0.039
PT (s)	13.10 (12.00, 14.80)	13.60 (12.30, 15.80)	0.152	13.30 (11.90, 15.30)	13.40 (12.20, 15.50)	0.010
FIB (mg/dL)	4.07 (3.17, 4.95)	3.96 (3.02, 4.86)	0.076	4.04 (3.12, 4.88)	4.01 (3.04, 4.87)	0.054
PaO2 (mm Hg)	95.40 (80.00, 100.00)	93.22 (79.30, 99.14)	0.091	94.80 (77.70, 100.00)	93.30 (79.55, 99.25)	0.034
Lactate (mmol/L)	2.10 (1.58, 3.00)	2.47 (1.76, 3.87)	0.227	2.20 (1.68, 3.27)	2.38 (1.71, 3.48)	0.050
Glucose (mmol/L)	10.32 (7.80, 10.60)	10.32 (8.00, 10.60)	0.009	10.30 (8.20, 10.80)	10.30 (8.00, 11.10)	0.010

ALB, albumin; ALT, alanine aminotransferase; APACHE II, Acute Physiology and Chronic Health Evaluation II; BMI, body mass index; BUN, blood urea nitrogen; Ca, calcium; CCI, Charlson Comorbidity Index; CHF, chronic heart disease; Cr, creatinine; CRP, C reactive protein; CRRT, continuous renal replacement therapy; DEN, delayed enteral nutrition; EEN, early enteral nutrition; FIB, fibrinogen; Hb, haemoglobin; HR, heart rate; IMV, invasive mechanical ventilation; K, potassium; LYM, lymphocyte count; MAP, mean arterial pressure; mNUTRIC, Modified Nutritional Risk in Critically III; Na, sodium; NEE, norepinephrine equivalent; PA, prealbumin; PaO2, arterial oxygen partial pressure; PLT, platelet count; PSM, propensity score matching; PT, prothrombin time; RR, respiratory rate; SAPS II, Simplified Acute Physiology Score II; SMD, standardised mean difference; SOFA, Sequential Organ Failure Assessment; TBIL, total bilirubin; WCC, white cell count.

Comparison of the primary outcome before and after PSM

Prior to the application of PSM, the 28-day mortality rates were similar between the EEN group and the DEN group, recorded at 14.50% and 13.90%, respectively (p=0.710). However, postmatching analysis revealed a significant imbalance, with the EEN group exhibiting a higher mortality rate of 18.08% compared with 13.27% in the DEN group (p=0.026). This disparity was further accentuated in the 60-day hospital mortality rates, where the EEN group demonstrated a mortality rate of 22.05%, significantly exceeding the 15.42% observed in the DEN group (p=0.004) (table 2). Kaplan-Meier survival analyses supported these findings, showing that after PSM, the EEN group had a significantly elevated risk of 28-day and 60-day hospital mortality, with HRs of 1.44 (95% CI 1.08 to 1.92; p=0.014) and 1.45 (95% CI 1.11 to 1.89; p=0.006), respectively (figure 3). To strengthen the model’s robustness, we performed multivariable Cox regression and IPW analyses to corroborate the PSM results. Following multivariable adjustment, the HR for the 28-day mortality risk was determined to be 1.32 (95% CI 1.02 to 1.70; p=0.033), whereas the IPW analysis produced an HR of 1.18 (95% CI 1.01 to 1.38; p=0.038). Regarding the 60-day mortality risk, the adjusted HR was 1.31 (95% CI 1.04 to 1.66; p=0.023), with the IPW analysis indicating an HR of 1.19 (95% CI 1.03 to 1.37; p=0.023) (refer to table 3). These findings further substantiate the association between EEN and elevated mortality rates.

Table 2. Clinical outcomes of patients with sepsis.

Outcome	Before PSM			After PSM
	EEN (N=1500）	DEN (N=705）	P value	EEN (N=603）	DEN (N=603）	P value
Primary outcome
28-day hospital mortality, n (%)	219 (14.50)	98 (13.90)	0.710	109 (18.08)	80 (13.27)	0.026
60-day hospital mortality, n (%)	259 (17.27)	114 (16.17)	0.562	133 (22.05)	93 (15.42)	0.004
Secondary outcomes
Hospital LOS (days)	17.80 (10.10, 30.05)	18.00 (11.00, 29.70)	0.400	16.80 (9.70, 30.65)	18.00 (10.80, 29.50)	0.363
ICU LOS (days)	15.00 (9.00, 26.00)	13.00 (8.00, 22.00)	<0.001	14.00 (8.00, 25.00)	13.00 (8.00, 22.00)	0.066
IMV duration (days)	6.00 (0.00, 14.00)	4.00 (0.00, 10.00)	<0.001	4.00 (0.00, 11.50)	4.00 (0.00, 10.00)	0.176
Diarrhoea, n (%)	275 (18.33)	130 (18.44)	0.999	116 (19.24)	103 (17.08)	0.370
Gastric retention, n (%)	149 (9.93)	39 (5.53)	0.001	56 (9.29)	33 (5.47)	0.015
GI bleeding, n (%)	52 (3.47)	32 (4.54)	0.268	28 (4.64)	24 (3.98)	0.671

DEN, delayed enteral nutrition; EEN, early enteral nutrition; GI Bleeding, gastrointestinal bleeding; ICU, intensive care unit; IMV, invasive mechanical ventilation; LOS, length of stay; PSM, propensity score matching.

Figure 3. Kaplan-Meier survival curves of the two groups at 28 (A, B) and 60 (C, D) days before and after PSM. DEN, delayed enteral nutrition; EEN, early enteral nutrition; PSM, propensity score matching.

Table 3. Association between enteral nutrition timing and mortality and complications in patients with sepsis.

Outcome	Model	HR (95% CI) or OR (95% CI)		P value
		DEN	EEN
Mortality
28-day hospital mortality	Unadjusted	Ref	1.08 (0.85 to 1.37)	0.523
	Multivariate adjusted	Ref	1.32 (1.02 to 1.70)	0.033
	PSM	Ref	1.44 (1.08 to 1.92)	0.014
	IPW	Ref	1.18 (1.01 to 1.38)	0.038
60-day hospital mortality	Unadjusted	Ref	1.09 (0.87 to 1.36)	0.449
	Multivariate adjusted	Ref	1.31 (1.04 to 1.66)	0.023
	PSM	Ref	1.45 (1.11 to 1.89)	0.006
	IPW	Ref	1.19 (1.03 to 1.37)	0.023
Complications
Diarrhoea	Unadjusted	Ref	0.99 (0.76 to 1.22)	0.952
	Multivariate adjusted	Ref	1.09 (0.86 to 1.38)	0.481
	PSM	Ref	1.16 (0.86 to 1.55)	0.332
	IPW	Ref	1.09 (0.98 to 1.21)	0.130
Gastric retention	Unadjusted	Ref	1.88 (1.27 to 2.00)	0.001
	Multivariate adjusted	Ref	1.85 (1.26 to 2.72)	0.002
	PSM	Ref	1.77 (1.14 to 2.79)	0.012
	IPW	Ref	1.48 (1.28 to 1.71)	<0.001
GI bleeding	Unadjusted	Ref	0.76 (0.27 to 1.17)	0.221
	Multivariate adjusted	Ref	1.01 (0.60 to 1.69)	0.981
	PSM	Ref	1.17 (0.67 to 2.07)	0.571
	IPW	Ref	1.06 (0.87 to 1.29)	0.587

DEN, delayed enteral nutrition; EEN, early enteral nutrition; GI, gastrointestinal; IPW, inverse probability weighting; PSM, propensity score matching.

Post hoc power analysis revealed statistical power of 0.83 for 28-day mortality and 0.90 for 60-day mortality, indicating adequate power to detect the observed differences between groups.

Comparison of secondary outcomes before and after PSM

Regarding secondary endpoints, the EEN group had a significantly longer ICU stay than the DEN group before matching (15.00 vs 13.00 days, p<0.001), but this difference was not significant after matching (14.00 vs 13.00 days, p=0.066). The EEN group also required longer mechanical ventilation before matching (6.00 vs 4.00 days, p<0.001), but this difference was not significant after matching (4.00 vs 4.00 days, p=0.176). No significant differences in hospital length of stay were found between the groups in either analysis (p=0.400 and p=0.363). Gastric retention was more common in the EEN group across all analyses: unadjusted (9.93% vs 5.53%, OR 1.88, 95% CI 1.27 to 2.00, p=0.001), PSM (9.29% vs 5.47%, OR 1.77, 95% CI 1.14 to 2.79, p=0.012), multivariate-adjusted (OR 1.85, 95% CI 1.26 to 2.72, p=0.002) and IPW (OR 1.48, 95% CI 1.28 to 1.71, p<0.001). No significant differences were found between groups for diarrhoea or gastrointestinal bleeding (refer to tables2 3).

Subgroup analysis

We performed detailed subgroup analyses with propensity score-matched data to assess the link between EEN versus DEN and mortality at 28 and 60 days across relevant clinical groups (figure 4; online supplemental figure 4). The adverse impact of EEN was significantly more pronounced in patients aged 65 years or younger (HR=1.65, 95% CI 1.05 to 2.59, p=0.0292), male patients (HR=1.53, 95% CI 1.07 to 2.20, p=0.0201), those with a BMI below 24 kg/m² (HR=1.76, 95% CI 1.14 to 2.72, p=0.0105), patients with a SOFA score exceeding 8 (HR=1.81, 95% CI 1.24 to 2.64, p=0.0022), patients with APACHE II scores greater than 15 (HR=1.67, 95% CI 1.13 to 2.48, p=0.0106), individuals experiencing septic shock (HR=1.66, 95% CI 1.12 to 2.45, p=0.0117), patients with abdominal infections (HR=1.70, 95% CI 1.07 to 2.70, p=0.0243), those with NEE ≥0.1 (HR=1.66, 95% CI 1.03 to 2.65, p=0.0360) and patients with lactate levels exceeding 2 mmol/L (HR=1.56, 95% CI 1.13 to 2.17, p=0.0077). In contrast, no significant differences in mortality were observed among patients older than 65 years, those with a BMI ≥24 kg/m², SOFA scores ≤8, APACHE II scores ≤15, female patients, individuals without septic shock, those with lung, skin, or soft tissue infections, patients with NEE <0.1, lactate levels ≤2 mmol/L, and across subgroups defined by mNUTRIC score, IMV or CRRT. Consistent patterns were observed in the 60-day mortality rates, with only minor variations in the statistical significance across individual subgroups (refer to online supplemental figure 4).

Figure 4. Forest plots of subgroup analyses of 28-day mortality between EEN and DEN groups after propensity score matching. APACHE II, Acute Physiology and Chronic Health Evaluation II; BMI, body mass index; CRRT, continuous renal replacement therapy; DEN, delayed enteral nutrition; EEN, early enteral nutrition; IMV, invasive mechanical ventilation; mNUTRIC, Modified Nutrition Risk in the Critically Ill; NEE, norepinephrine equivalent; SOFA, Sequential Organ Failure Assessment.

Sensitivity analysis

After excluding patients with missing data, we repeated the PSM process (refer to online supplemental figure 5) and conducted a primary outcome analysis (refer to online supplemental figure 6). Additionally, we re-included patients who were admitted to the ICU for a minimum of 72 hours and subsequently died within 7 days in the PSM (refer to online supplemental figure 7) and performed a primary outcome analysis (refer to online supplemental figure 8). The analyses demonstrated that the results remained consistent.

Discussion

Our study identified a correlation between EEN and elevated 28-day and 60-day mortality rates among patients with sepsis. Subgroup analyses indicated that specific patient cohorts, such as those experiencing septic shock, younger individuals, male gender, patients with a BMI of less than 24 kg/m², abdominal infections, those with severe sepsis and patients necessitating vasopressor support at a NEE of ≥0.1 µg/kg/min, may experience more pronounced effects. Furthermore, we observed a higher incidence of gastric retention within the EEN group.

The findings of this study imply that existing clinical practice guidelines, which recommend initiating enteral nutrition within 24–48 hours of ICU admission,²⁴ may require thorough reevaluation, especially for specific septic patient populations. The data indicate that the conventional ‘one-size-fits-all’ strategy for nutritional timing could be enhanced by adopting a more nuanced, patient-specific assessment.²⁵ The variations observed among different subgroups suggest that individualised nutritional strategies, which consider patient-specific risk factors and clinical conditions, may be more suitable than the universal application of early feeding protocols in critically ill patients with sepsis.²⁶

Several pathophysiological mechanisms may elucidate the observed correlation between EEN and heightened mortality in patients with sepsis. First, sepsis-induced haemodynamic instability significantly alters splanchnic circulation, thereby compromising the physiological environment necessary for enteral nutrition. During the acute phase of sepsis, microcirculatory dysfunction emerges as a central pathophysiological characteristic, leading to the redistribution of blood flow to vital organs at the expense of gastrointestinal perfusion.²⁷ This splanchnic hypoperfusion not only restricts nutrient absorption and utilisation but also heightens the risk of mesenteric ischaemia, potentially compromising intestinal barrier integrity and facilitating bacterial translocation, which subsequently amplifies systemic inflammation.28,30 Second, sepsis-associated gastrointestinal dysfunction poses substantial challenges to feeding tolerance. Gastrointestinal dysfunction is prevalent among patients with acute illnesses and can result in delayed gastric emptying and functional intestinal obstruction.³¹ In individuals with sepsis, the elevation of inflammatory mediators such as TNFα and IL-6 is closely linked to these dysfunctions.³² This pathophysiological cascade is clinically evident as delayed gastric emptying and functional intestinal obstruction, accounting for the increased incidence of gastric retention observed in our EEN cohort. Third, premature nutritional intervention may interfere with the adaptive metabolic reprogramming crucial for sepsis tolerance. The body’s natural stress response entails coordinated metabolic adaptations, including the activation of autophagy and selective protein synthesis, which regulate iron and glucose metabolism to maintain homeostasis.^{33 34} EEN may adversely affect these protective mechanisms by modulating the AMPK pathway, potentially exacerbating metabolic-inflammatory dysregulation and undermining the body’s intrinsic protective responses during the critical early phase of sepsis.³⁵

Our findings contribute to the complex body of evidence regarding the optimal timing of nutritional interventions in critically ill patients, emphasising the need for disease-specific considerations, particularly within sepsis populations. Recent studies have suggested potential benefits of EEN in certain critically ill groups. Research by Powierza et al, Haines et al and others has demonstrated improved outcomes associated with EEN.36,39 However, these positive outcomes primarily stem from studies involving mixed ICU populations or trauma cohorts, with limited representation of sepsis-specific cases. Conversely, when examining sepsis-specific populations, the evidence presents a more nuanced and concerning picture. Recent analyses of the Medical Information Mart for Intensive Care IV database conducted by Xu et al have yielded concerning findings. Their 2024 study reported no mortality benefit in a cohort of 1111 patients with sepsis.⁴⁰ Furthermore, their subsequent 2025 analysis indicated that EEN was associated with increased 28-day mortality among mechanically ventilated patients with sepsis.⁴¹ Beyond database studies, extensive prospective cohort investigations have corroborated these concerning findings. The FRANS study, a multicentre prospective cohort involving 1206 critically ill patients, demonstrated that early nutritional support was significantly associated with an increased 28-day mortality rate (adjusted OR 1.05, 95% CI 1.00 to 1.10).⁴² Additional studies from other cohorts have further supported these findings. Ortiz-Reyes et al revealed that the apparent benefits observed in circulatory shock patients were no longer significant after adjusting for illness severity.⁴³

The findings from individual studies are corroborated by systematic evaluations of the evidence. In a comprehensive systematic review, Moon et al concluded that the current evidence is inadequate to endorse routine EEN for sepsis patients.¹³ Similarly, Grillo-Ardila et al in their sepsis-focused meta-analysis identified no significant mortality benefit associated with this intervention.⁴⁴ In addition to the absence of benefits, recent evidence has underscored complications associated with EEN. Baik et al reported a significantly higher incidence of gastrointestinal complications with EEN compared with parenteral methods.⁴⁵ The FRANS study observed that feeding intolerance occurred in 34% of patients receiving enteral nutrition, with gastrointestinal complications being prevalent.⁴² These accumulating findings have contributed to a broader shift in nutritional perspectives. Stoppe et al advocate for a reassessment of current nutritional paradigms, emphasising that aggressive early nutrition may not enhance and could potentially deteriorate outcomes in critically ill patients.⁴⁶ This shift towards personalised nutrition is consistent with our findings, suggesting that nutritional timing decisions should be guided by haemodynamic status and disease-specific factors rather than by universal early feeding protocols.

One strength of our study is the relatively large sample size drawn from a real-world clinical setting, which helps capture the diversity and complexity of patients with sepsis encountered in everyday practice. Moreover, we employed several rigorous statistical methodologies, including PSM, IPW and multivariable Cox regression models, to mitigate potential indication bias and confounding, which are common challenges in observational studies.

Our study is subject to several limitations. First, our single-centre retrospective analysis may not be widely applicable due to differences in clinical practices, patient demographics and resources across healthcare systems. Second, although we used initial clinical and lab data for PSM and regression analysis, these may not fully reflect patient status when enteral nutrition began, potentially causing time-dependent confounding. Advanced methods like marginal structural models could address this but require detailed data at multiple points, which is hard to obtain retrospectively. Future prospective studies collecting data at enteral nutrition initiation would help validate our findings. Additionally, unmeasured variables may still cause residual confounding in the propensity score analysis. Third, the decade-long study period (2015–2024) may have influenced results due to changes in feeding protocols and clinical practices. Fourth, our outcome assessment was primarily focused on mortality endpoints and did not encompass other critical clinical outcomes, such as ICU-acquired infections or long-term functional outcomes. Fifthly, we did not precalculate sample size as it is a retrospective cohort study, and while post hoc power analysis showed sufficient power, it does not replace prospective planning. Future studies with predetermined sample sizes are needed. Lastly, the retrospective design limits establishing causal links between nutritional timing and outcomes, necessitating prospective randomised controlled trials for validation.

Conclusions

Our retrospective analysis of 2205 propensity score-matched patients with sepsis found that EEN is linked to higher short-term mortality, especially in younger males, those with severe illness, abdominal infections or needing moderate to high vasopressor support. This suggests that personalised nutritional timing might be better than universal early feeding. However, these findings are preliminary and need confirmation through randomised controlled trials.

Data availability statement

Data are available on reasonable request.