Early enteral nutrition in critically ill patients: ESICM clinical practice guidelines

Question 2: Should we delay EN in patients with shock receiving vasopressors or inotropes?

No RCTs were retrieved. We identified and analysed four prospective cohort studies, four case series/retrospective cohort studies and two reviews (Supplement 5, Table 2).

There is concern that EN in shock further jeopardizes the already impaired splanchnic perfusion. Non-occlusive bowel necrosis or non-occlusive mesenteric ischaemia (NOMI) has been reported in fewer than 1% of patients [11, 12], without evidence for causal relationship between shock, vasopressors, EN and NOMI [11‐14]. In a large observational study, EEN (<48 h) in patients with ‘stable’ haemodynamics after fluid resuscitation, whilst receiving at least one vasopressor, was associated with reduced mortality compared to late EN (>48 h) [15]. These results suggest that the use of concomitant vasopressors (especially with stable or decreasing doses) should not preclude a trial of EN, despite a high prevalence of feeding intolerance [16]. In very unstable patients, EN may not have priority and potential positive effects of EN are unlikely to help improve instability. Persisting lactic acidosis may help identify uncontrolled shock.

Should we delay EN in patients with:

We found no direct evidence on these subquestions in the literature, and RCTs in this population are unlikely to become available.

The rationale to withhold EN in patients with hypoxaemia, hypercapnia and acidosis is to limit oxygen consumption and CO₂ production. However, the process of starving mobilises endogenous stores and is energy-consuming [17]. Acidosis may represent persistent shock and possibly contribute to gut dysfunction. Identifying and treating the cause of shock has priority over the initiation of EN. Similarly, in uncontrolled life-threatening hypoxaemia and hypercapnia, EN should be delayed until the symptoms are resolving.

In patients with acute lung injury, an RCT comparing trophic to full EN for up to 6 days was associated with less gastrointestinal intolerance when compared to full EN, without affecting ventilator-free days, infectious complications, physical function, or survival [7, 18]. There are no data suggesting EN in patients with chronic, subacute, compensated or permissive hypercapnia is unsafe or not feasible.

Question 4: Should we delay EN in patients receiving neuromuscular blocking agents?

One prospective study was identified (Supplement 5, Table 4), reporting similar gastric emptying as measured by gastric residual volume (GRV) in sedated patients with or without concomitant use of neuromuscular blocking agents [19]. The critical condition necessitating the use of neuromuscular blocking agents always needs to be considered, but these agents per se should not preclude EN. Analgosedation is known to slow gastric emptying [20]. Increased rate of EN intolerance is expected in deeply sedated patients with/without concomitant use of neuromuscular blocking agents.

Question 5: Should we delay EN in patients receiving therapeutic hypothermia?

One case series study addressing EN during therapeutic hypothermia was identified [21] (Supplement 5, Table 5).

During therapeutic hypothermia, energy metabolism might be markedly reduced [22, 23] when shivering is prevented. The rationale to withhold EN during therapeutic hypothermia is based on the presumed decrease in gut motility due to hypothermia [24, 25] and required analgosedation [20]. It has been suggested that EN could be successfully administered to these patients [21]. Tolerance to enteral feeding was impaired during hypothermia, but improved during rewarming [21].

Question 6: Should we delay EN in patients receiving extracorporeal membrane oxygenation (ECMO)?

No RCTs and no prospective cohort studies were identified. Four case series in adult patients with ECMO were assessed (Supplement 5, Table 6), suggesting that EN is feasible during ECMO.

Question 7: Should we delay EN during prone position?

One prospective cross-over, one cohort and three case series studies were identified (Supplement 5, Table 7).

Data on tolerance of EN in prone position are controversial. Observational studies found similar GRVs in prone and supine position [26], whereas poor feeding tolerance was improved with semi-recumbent position during supine periods and prokinetics [27, 28]. Although no RCTs on EN tolerance during prone position are available, reported studies do not support withholding EN in prone position. Gastric emptying seems not to be significantly influenced by prone position and adverse events in most studies not increased.

Question 8: Should we delay EN in patients with traumatic brain injury?

We identified a Cochrane review with two updates and one recent meta-analysis, comparing early vs. late feeding, independent on the route of nutrition (EN or PN) (Supplement 5, Table 8C). We identified three RCTs comparing EEN vs. early PN, three RCTs comparing EEN vs. delayed EN (one with restricted randomisation), and one RCT comparing early PN vs. delayed EN (Supplement 5, Table 8A).

Question 8A: EEN vs. early PN

Three RCTs (116 patients) were included. EEN compared to early PN in patients with traumatic brain injury did not affect mortality (RR 1.91; 95% CI 0.59–6.18; P = 0.279; I ² = 0%) or the risk of pneumonia (RR 1.23; 95% CI 0.79–1.90; P = 0.36; I ² = 0%). The certainty of evidence for mortality outcome was low, for pneumonia it was very low. We rated down for risk of bias and imprecision (Table 3). Supplement 7, Fig. 5.

Question 8B: EEN vs. delayed EN

For mortality, two RCTs (86 patients) were included. EEN did not affect mortality compared to delayed EN (RR 0.66; 95% CI 0.18–2.45; P = 0.53; I ² = 0%). The certainty of evidence was low. We rated down for imprecision (Table 3).

For pneumonia, three RCTs (118 patients) were included. EEN did not affect the risk of pneumonia compared to delayed EN (RR 0.86; 95% CI 0.55–1.35; P = 0.51; I ² = 0%). The certainty of evidence was very low. We rated down for risk of bias and imprecision (Table 3). Supplement 7, Fig. 6.

In addition to RCTs, five cohort studies addressing this question were identified (Supplement 5, Table 8B).

Existing evidence did not allow determining or excluding any benefit or harm of EEN, therefore our recommendation is based on expert opinion.

Question 9: Should we delay EN in patients with stroke (haemorrhagic or ischaemic)?

We identified two RCTs in patients with ischaemic stroke and one retrospective study in patients with hypertensive intracerebral haemorrhage (Supplement 5, Tables 9A, B).

One small RCT compared early vs. delayed EN and reported amelioration of cell-mediated immunity [29]; however, both groups received PN to meet caloric targets from day 1. A large RCT compared EEN (“as soon as possible”) to no nutrition within 7 days and reported a trend towards reduction of long-term mortality (6 months) with EN, with an increased risk of poor neurologic outcome in survivors [30]. An observational study reported reduction in infectious complications with EEN vs. delayed EN [31].

Question 10: Should we delay EN in patients with spinal cord injury?

One RCT addressed EEN (<72 h) vs. delayed EN in cervical spinal injury [32]. No differences in outcome variables were identified. One retrospective cohort study addressed safety of EN early after spinal cord injury and reported no major complications [33] (Supplement 5, Tables 10A, B).

Question 11: Should we delay EN in patients with severe acute pancreatitis (SAP)?

We identified five systematic reviews with meta-analyses comparing EN to PN while not considering timing (Supplement 5, Table 11B). All meta-analyses concluded that EN was beneficial in reducing infections and three reported reduced mortality [3, 34, 35].

We identified five RCTs addressing EEN (“early” as defined by the authors) vs. early PN in SAP whereas only two studies defined “early” as <48 h. Three further RCTs addressed EEN vs. early PN and one RCT EEN vs. delayed EN in “predicted SAP”. Two RCTs addressing acute pancreatitis independent of severity and one RCT studying mixed patients undergoing abdominal surgery were not included. Supplement 5, Table 11A.

We performed three separate meta-analyses all comparing EEN vs. early PN: (A) SAP and “early” as defined by the authors of the original study; (B) predicted SAP and “early” as defined by the authors of the original study; (C) predicted SAP and early defined as <48 h.

Question 11A: SAP (as stated by the authors). Early (“early” as defined by the authors) EN vs. PN

For mortality we included five RCTs (283 patients). EEN did not reduce the risk of death compared to PN (RR 0.57; 95% CI 0.23–1.38; P = 0.21; I ² = 35.1%). The certainty of evidence was low. We rated down for imprecision (Table 3).

For any infections we included five RCTs (283 patients). EEN reduced the risk of infections compared to PN (RR 0.48; 95% CI 0.23–0.98; P = 0.045; I ² = 76%). The certainty of evidence was low. We rated down for inconsistency and imprecision (Table 3).

For pancreatic infections we included four RCTs (233 patients). EEN reduced the risk of pancreatic infections compared to PN (RR 0.33; 95% CI 0.21–0.52; P < 0.0001; I ² = 0%) The certainty of evidence was low. We rated down for risk of bias and imprecision (Table 3). Supplement 7, Fig. 7.

Question 11B: Predicted SAP. Early (“early” as defined by the authors) EN vs. PN

For mortality we included eight RCTs (417 patients). EEN did not reduce the risk of death compared to PN (RR 0.50; 95% CI 0.22–1.13; P = 0.09; I ² = 38%). The certainty of evidence was low. We rated down for imprecision (Supplement 6).

For any infections we included eight RCTs (417 patients). EEN reduced the risk of infections compared to PN (RR 0.53; 95% CI 0.30–0.91; P = 0.023; I ² = 63.5%). The certainty of evidence was low. We rated down for risk of bias and inconsistency (Supplement 6).

For pancreatic infections we included five RCTs (202 patients). The use of EEN reduced the risk of pancreatic infections compared to PN (RR 0.35; 95% CI 0.24–0.52; P < 0.0001; I ² = 0%). The certainty of evidence was low. We rated down for risk of bias and imprecision (Supplement 6). Supplement 7, Fig. 8.

Question 11C: Predicted SAP. Early (<48 h) EN vs. PN

For mortality we included five RCTs (232 patients). EEN (<48 h) did not reduce the risk of death compared to PN (RR 0.61; 95% CI 0.15–2.55; P = 0.50; I ² = 41%). The certainty of evidence was low. We rated down for imprecision (Supplement 6).

For any infections we included five RCT (232 patients), EEN (<48 h) reduced the risk of infections compared to PN (RR 0.49; 95% CI 0.28–0.83; P = 0.008, I ² = 9%). The certainty of evidence was low. We rated down for risk of bias, inconsistency and imprecision (Supplement 6).

For pancreatic infections we included three RCTs (167 patients). EEN (<48 h) reduced the risk of pancreatic infections compared to PN (RR 0.40; 95% CI 0.22–0.73; P = 0.003; I ² = 0%). The certainty of evidence was low. We rated down for risk of bias and imprecision (Supplement 6). Supplement 7, Fig. 9.

Taken together, the studies in different subpopulations have demonstrated a reduction of infections but no convincing effect of EEN on mortality.

Question 12: Should we delay EN in patients after GI surgery?

Out of three published meta-analyses [36‐38] addressing early postoperative feeding including early oral diet, the two more recent papers [36, 37] reached different conclusions: reduced mortality and length of stay (LOS) but increased risk of vomiting analysing 15 RCTs [37] vs. no difference in mortality and LOS, but reduced complications in early group from 13 RCTs [36].

We identified three RCTs comparing early vs. delayed EN after emergency GI surgery and six RCTs in elective GI surgery. Two RCTs compared EEN vs. early PN in patients after elective GI surgery (Supplement 5, Table 12).

Question 12A: Emergency GI surgery. EEN vs delayed EN

Three RCTs (343 patients) were included. EEN did not affect mortality compared to delayed EN (RR 0.80; 95% CI 0.46–1.40; P = 0.44; I ² = 0%). EEN reduced the risk of infections compared to delayed EN (RR 0.61; 95% CI 0.40–0.93; P = 0.02; I ² = 0%). The certainty of evidence was low. We rated down for risk of bias and imprecision (Table 3). Supplement 7, Fig. 10.

Question 12B: Elective GI surgery. EEN vs. delayed EN

For mortality three RCTs (346 patients) were included. EEN did not affect mortality compared to delayed EN in patients after elective GI surgery (RR 0.83; 95% CI 0.25–2.81; P = 0.77; I ² = 17%). The certainty of evidence was low. We rated down for imprecision (Table 3).

For any infections six RCTs (432 patients) were included. EEN reduced the risk of infections compared to delayed EN (RR 0.43; 95% CI 0.23–0.82; P = 0.01; I ² = 46%). The certainty of evidence was low. We rated down for risk of bias and imprecision (Table 3).

Five RCTs (404 patients) reported anastomotic leak. EEN reduced the risk of surgical leak compared to delayed EN (RR 0.43; 95% CI 0.20–0.93; P = 0.03; I ² = 0%). The certainty of evidence was low. We rated down for imprecision (Table 3). Supplement 7, Fig. 11.

Question 12C: Elective GI surgery. EEN vs early PN

Two RCTs (440 patients) were included. EEN did not reduce the risk of pneumonia compared to early PN (RR 0.59; 95% CI 0.31–1.14; P = 0.12, I ² = 0%), but reduced the risk of anastomotic leak compared to early PN (RR 0.42; 95% CI 0.19–0.95; P = 0.04; I ² = 63%). The certainty of evidence was low. We rated down for risk of bias, inconsistency and imprecision (Table 3). Supplement 7, Fig. 12.

Question 13: Should we delay EN in patients after abdominal aortic surgery?

No RCTs but two cohort studies were identified (Supplement 5, Table 13). Cohort studies both in elective [39] and emergency repair [40] did not compare EEN with any of our comparators, but showed that EEN was successful in a minority of patients. A multimodal approach has been proposed [41], including early removal of nasogastric tubes, immediate postoperative mobilisation early oral or enteral feeding, accepting GRV up to 500 ml and use of prokinetics. Although these patients are at risk of bowel ischaemia with prevalence reported between 7 and 17% [42, 43], the risk itself should not lead to withholding EN, unless bowel ischaemia is suspected (see also Recommendation 15).

Question 14: Should we delay EN in patients with abdominal trauma?

Ten RCTs and ten cohort studies addressing EEN in trauma patients (RCTs: within 6–48 h; cohort studies: within 12–96 h) were identified, but abdominal trauma specifically was addressed in six RCTs, four of them compared EEN to early PN and two EEN to delayed EN (Supplement 5, Table 14A).

Question 14A: EEN vs early PN

For mortality two RCTs (142 patients) were included. EEN did not affect mortality compared to early PN (RR 0.49; 95% CI 0.09–2.69; P = 0.41; I ² = 0%). The certainty of evidence was very low. We rated down for risk of bias and imprecision (Table 3).

For any infection four RCTs (219 patients) were included. EEN did not affect the risk of infections compared to early PN (RR 0.59; 95% CI 0.24–1.42; P = 0.24; I ² = 59%). The certainty of evidence was very low. We rated down for risk of bias, inconsistency and imprecision (Table 3). Supplement 7, Fig. 13.

Question 14B: EEN vs delayed EN

Two RCTs (101 patients) were included. EEN did not affect mortality compared to delayed EN (RR 0.74; 95% CI 0.18–3.11; P = 0.708). The certainty of evidence was very low. We rated down for risk of bias and imprecision (Table 3).

EEN did not affect the risk of infections compared to delayed EN (RR 0.83; 95% CI 0.41–1.70; P = 0.837). The certainty of evidence was very low. We rated down for risk of bias, inconsistency and imprecision (Table 3). See Supplement 7, Fig. 14.

Of note, earlier studies in this patient group almost exclusively used surgical jejunostomy for EN.

Existing evidence did not allow verifying or excluding any benefit or harm of EEN; therefore our recommendation is based on expert opinion. In addition to RCTs, nine observational studies were identified (Supplement 5, Table 14B).

An earlier meta-analysis in adult trauma patients in ICU (not specifically abdominal trauma) showed survival benefit in EEN commenced within 24 h after trauma [44].

Question 15: Should we delay EN in patients with bowel ischaemia?

We identified no clinical studies, but physiological knowledge and common sense support withholding EN in patients with overt bowel ischaemia. However, patients with endoscopic evidence of mild to moderate large bowel mucosal ischaemia, without signs of transmural ischaemia or bowel distension, might profit from low dose EN. In this case we support considering EN. In a recent retrospective study, survivors were more often fed enterally before the diagnosis of acute mesenteric ischaemia, but no independent association between EN and mortality was demonstrated [45].

Question 16: Should we delay EN in critically ill adult patients with intestinal fistula?

We identified one retrospective cohort study and two case series, all showing outcome benefit of “early” EN (Supplement 5, Table 16). However, “early” was defined as EN started within 7 days or 14 days of admission. Retrospective design further diminishes the importance of these studies.

Intolerance of EN or increasing fistula output causing skin breakdown or fluid/electrolyte imbalance are evident reasons to decrease or discontinue EN [46].

Question 17: Should we delay EN in patients with an open abdomen?

Seven observational studies (one prospective cohort study, three retrospective cohort studies and four case series) were identified; two studies compared EEN (different definitions) vs delayed EN and reported higher rate of early abdominal closure, less fistula formation and lower incidence of ventilator-associated pneumonia in the “early” EN group (Supplement 5, Table 17). The largest study comparing EN to no EN in patients with open abdomen after abdominal trauma reported independent associations between EN and ultimate fascial closure and decreased mortality rate in patients without bowel injury, but no difference in a subgroup of patients with bowel injury [47].

Question 18: Should we delay EN in patients with intra-abdominal hypertension?

Four observational studies were identified (Supplement 5, Table 18), only one addressed early vs. delayed EN [48]. All studies reported high incidence of feeding intolerance associated with intra-abdominal hypertension, but data are not conclusive regarding causality. A recently published study demonstrated that EEN did not increase intra-abdominal pressure, but values exceeding 15 mmHg were associated with higher rates of feeding intolerance in patients with severe acute pancreatitis [48].

No prospective study addressing EN in patients with abdominal compartment syndrome [49] was identified. As abdominal compartment syndrome is an immediately life-threatening condition with jeopardized splanchnic perfusion, we suggest to withhold or stop EN and try to lower intra-abdominal pressure.

Question 19: Should we delay EN in patients with upper GI bleeding?

No studies addressing EEN were identified. One RCT in bleeding due to gastric or duodenal ulcer reported shorter hospital stay (4.2 ± 1.2 vs. 5.9 ± 1.4 days, P < 0.001) in the early oral feeding group [50].

EN as protection against stress ulceration and GI bleeding is suggested in one meta-analysis [51], one retrospective study in burns [52] and several reviews [53‐55]. An RCT comparing ranitidine and sucralfate reported EN as an independently protective factor against GI bleeding [56]. The main rationale to prohibit eating/EN is based on fear for disturbed visibility in a further endoscopy/intervention due to rebleeding. Therefore, delaying EN for 48–72 h in patients with a high risk of rebleeding has been suggested [57]. Considering the absence of evidence to support this time frame, we suggest starting EN during the first 24–48 h after bleeding has been stopped; prolonged postponement of EN is unnecessary or even harmful because of increased risk of stress ulceration. Importantly, there is no evidence that fine-bore nasogastric tubes cause variceal bleeding [57].

Question 20: Should we delay EN in patients with acute liver failure?

We could not identify any study in acute or acute-on-chronic liver failure patients. Some benefits of EN have been shown in patients with alcoholic hepatitis, malnourished patients with cirrhosis and patients with liver transplantation [58‐60], where glycogen stores may be depleted after an overnight fast and metabolic conditions resemble prolonged starvation in healthy individuals [61]. EN in fulminant acute liver failure has never been studied. These patients often present with hypoglycaemia, which should be corrected with intravenous glucose, sometimes together with insulin. Fulminant liver failure is associated with increased serum amino acid concentrations, especially glutamine [62, 63]. It seems likely that a failing liver is unable to provide effective metabolic support required for nutrition. The pathophysiological rationale to delay EN in fulminant hepatic failure would be to “spare” the severely injured liver from the duties of metabolising and storing nutrition during a period of stress and also to avoid additional increases in ammonia. Intravenous provision of nutrients except correction of hypoglycemia and appropriate provision of vitamins and trace elements may be futile or harmful early in the clinical course [64].

Question 21: Should we delay EN in patients with large gastric aspirate volumes (GAV)?

We identified no study addressing this question. Based on existing evidence from two RCTs comparing the threshold volumes to stop already started EN [65, 66], a clear threshold volume (in ranges up to 500 ml) that increased the risk of ventilator-associated pneumonia was not identified. Measurements of GAV/GRV are not a gold standard and alternative methods (like ultrasound) can be applied to diagnose overfilling of the stomach. Gross distension of the stomach is likely to be undesirable and therefore we suggest that EN should be delayed when GAV/GRV is >500 ml/6 h [65], either for a limited time period or until administration of prokinetics. For patients with persistently large GAV/GRVs the use of postpyloric feeding should be considered rather than withholding EN, unless bowel ischaemia or obstruction is suspected (see also Recommendation 15).

Question 22: Should we delay EN in patients with absent bowel sounds?

One cohort study was identified [67] (Supplement 5, Table 22). Bowel sounds are frequently absent in mechanically ventilated patients and this is associated with impaired outcome [68]. The concept that bowel sounds must be present before initiation of enteral feeding is not based on evidence and should be abandoned [69]. After laparotomy small intestinal motility is frequently preserved despite gastric and colonic paresis. The small intestine may contract silently (absence of gas), while feeding is well tolerated [69]. Gastric and colonic paresis may effectively be treated with prokinetics [70]. Initiation of EN in absence of bowel sounds might be associated with earlier return of bowel sounds, fewer episodes of vomiting, and shorter ICU and hospital stay [67].

Question 23: Should we delay EN in patients with diarrhoea?

There were no studies testing delay of EN in case of diarrhoea, but diarrhoea is often considered as a reason to delay EN [71]. Prevalence of diarrhoea in unselected ICU population is between 14 and 21% [72, 73]. Causes include impaired digestion/absorption, bacterial overgrowth or infection such as Clostridium difficile. Observational studies [74, 75] suggest that diarrhoea can be effectively managed with protocolised measures other than immediate cessation in EN. We recommend analysing the causes of diarrhoea and treat appropriately (e.g. C. difficile colitis). We also suggest considering treating bacterial overgrowth by selective decontamination, fibre-enriched or semi-elementary diet or digestive enzymes to reduce diarrhoea.