PURPOSE

Most critically ill patients are unable to provide their own nutrition. In these patients, artificial nutrition is often provided. The purpose of this guideline is to summarize the evidence within nutrition support to guide practitioners in their provision of artificial nutrition to critically ill patients and provide/update recommendations for several foundational questions that are central to the provision of nutrition support for most critically ill adult patients.

Existing American Society for Parenteral and Enteral Nutrition (ASPEN) clinical guidelines are reviewed for potential updating every 5 years or when significant new additions to the literature have occurred, whichever occurs first. Whereas the earlier guideline provided extensive practice guidance for a large group of clinical decisions that had few randomized controlled trials (RCTs), resulting in many “expert consensus” recommendations,1 the current guideline restricted the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) process to questions that trials had explored. This resulted in fewer questions overall and some recommendations that did not translate directly into nutrition support prescription. Following the publication of this guideline, a separate Clinical Recommendations paper will ensue to answer expert opinion questions from the previous guideline and other questions for which there is insufficient evidence. To assist the reader in making practice decisions, a “Clinical Application” row has been added beneath each GRADE question in Table 1 that provides guidance for how to incorporate the GRADE recommendations into practice. To increase the external validity and comparability to today's intensive care unit (ICU) patient, only RCTs between January 1, 2001, and July 15, 2020, were included to reflect more current nutrition support practices in the modern era, a time when routine care includes maintenance of glycemic control, avoidance of overfeeding energy, and improved catheter care. Particular importance was given to the nutrition aspects of the exposures (ie, energy, protein, and lipid injectable emulsion [ILE]) provided to patients in the trials as well as timing and route of nutrition delivery.

TABLE 1. Guideline questions, evidence grades, recommendations summary, and clinical applications Guideline question 1. In adult critically ill patients, does provision of higher vs lower energy intake impact clinical outcomes? Evidence GRADE: Moderate GRADE recommendation: No significant difference in clinical outcomes was found between patients with higher vs lower levels of energy intake. We suggest feeding between 12 and 25 kcal/kg (ie, the range of mean energy intakes examined) in the first 7–10 days of ICU stay. Strength of GRADE recommendation: Weak Discussion on clinical application for question 1: Until data become available that enable more precise recommendations on energy intake, clinicians should rely on clinical judgment. When EN or PN is associated with problems in glycemic control, respiratory acidosis, or high serum triglyceride concentrations, consider whether feedings should be reduced. Lipid-based sedation also provides a source of energy that should be considered in the total daily intake. Gastrointestinal tolerance may limit how much EN can be provided. Feeding less than the EN formula volume needed to deliver dietary reference intake levels may risk inadequate vitamin, mineral, and trace element intake. Guideline question 2. In adult critically ill patients, does provision of higher as compared with lower protein intake impact clinical outcomes? Evidence GRADE: Low GRADE recommendation: There was no difference in clinical outcomes in the relatively limited data. Because of a paucity of trials with high-quality evidence, we cannot make a new recommendation at this time beyond the 2016 guideline suggestion for 1.2–2.0 g/kg/day. Strength of GRADE recommendation: Weak Discussion on clinical application for question 2: Few studies have investigated the impact of higher protein doses provided with equivalent energy; thus, the impact on outcomes is not known. Until more data are available, we suggest clinicians should individualize protein prescriptions based on clinician judgment of estimated needs. Guideline question 3. In adult critically ill patients who are candidates for EN, does similar energy intake by PN vs EN as the primary feeding modality in the first week of critical illness impact clinical outcomes? Evidence GRADE: High GRADE recommendation: There was no significant difference in clinical outcomes. Because similar energy intake provided as PN is not superior to EN and no differences in harm were identified, we recommend that either PN or EN is acceptable. Strength of GRADE recommendation: Strong Discussion on clinical application for question 3: Our findings indicate that when similar energy is delivered by PN or EN early in critical illness for relatively short periods of time, clinical outcomes are similar. Given these data, cost and convenience of providing EN vs PN may be larger determinants of route of feeding early in critical illness than differences in clinical outcomes. The question of PN use arises when EN is not feasible or tolerated or in patients with significant gastrointestinal disease, who were not the populations studied for question 3. The two reported trials gave ∼18–20 kcal/kg/day and 0.6–0.8 g/kg/day protein, and both used a premixed PN solution. Avoidance of energy overfeeding may be the most important decision to make regarding PN use. Optimal glycemic control and catheter care are also important factors in the provision of PN to reduce infectious complications. Clinical judgment about an individual patient's metabolic tolerance to the dextrose (monitor glycemic control), ILE (monitor serum triglyceride concentrations), and amino acid dose is key to delivery of appropriate PN feedings. Guideline question 4. In adult critically ill patients receiving EN, does provision of SPN, as compared with no SPN during the first week of critical illness, impact clinical outcomes? Evidence GRADE: High GRADE recommendation: There was no significant difference in clinical outcomes. Based on findings of no clinically important benefit in providing SPN early in the ICU admission, we recommend not initiating SPN prior to day 7 of ICU admission. Strength of GRADE recommendation: Strong Discussion on clinical application for question 4: The data in this guideline compared SPN within the first week of ICU care and excluded patients with malnutrition. These findings imply that the average critically ill patient will not be harmed by waiting a week to initiate SPN. Further, the patient's tolerance to EN may improve in that time window. However, the needs of malnourished patients or patients who have limited lean muscle mass were not included in these trials and may differ from those of nonmalnourished patients. Patient-specific clinical judgment should be used regarding the initiation of SPN in the first 7 days for these special cases. Guideline question 5A. In adult critically ill patients receiving PN, does provision of mixed-oil ILEs (ie, medium-chain triglycerides, olive oil, FO, mixtures of oils), as compared with 100% soybean-oil ILE, impact clinical outcomes? Evidence GRADE: Low GRADE recommendation: Because of limited statistically or clinically significant differences in key outcomes, we suggest that either mixed-oil ILE or 100% soybean-oil ILE be provided to critically ill patients who are appropriate candidates for initiation of PN, including within the first week of ICU admission. Strength of GRADE recommendation: Weak Guideline question 5B. In adult critically ill patients receiving PN, does provision of FO-containing ILE, as compared with non–FO-containing ILE, impact clinical outcomes? Evidence GRADE: Low GRADE recommendation: Because there was only one outcome found with a significant difference that was not supported by data covering the other key downstream outcomes, we suggest that either FO- or non–FO-containing ILE be provided to critically ill patients who are appropriate candidates for initiation of PN, including within the first week of ICU admission. Strength of GRADE recommendation: Weak Discussion on clinical application questions 5A and 5B: In addition to 100% soybean-oil ILE, mixed oil– and FO-containing ILE products are now available in the United States, but health-system formulary availability of these formulations may vary. In general, ILE is a safe and effective energy source that can be included with the PN formulation at the time of initiation, including within the first week of ICU admission. Optimizing ILE provision helps avoid excessive dextrose provision and hyperglycemia. Monitoring serum triglyceride concentrations will give information about the adequacy of lipid clearance. The energy provided by lipid-based sedation should be considered in the overall estimate of lipid and energy intake. It is also important to give adequate levels of the essential fatty acids to meet requirements if the PN will be needed for >10 days. The essential fatty acid content of the mixed-oil ILE and FO-containing ILE is lower than that of the soybean-oil ILE. Abbreviations: EN, enteral nutrition; FO, fish oil; GRADE, Grading of Recommendations, Assessment, Development, and Evaluation; ICU, intensive care unit; ILE, lipid injectable emulsion; PN, parenteral nutrition; SPN, supplemental PN. GUIDELINE LIMITATIONS

These ASPEN clinical guidelines are based on a general consensus among a group of health professionals who, in developing such guidelines, have examined benefits of nutrition practices against risks inherent with such therapy. A task force of multidisciplinary experts in clinical nutrition comprising clinical epidemiologist/methodologists, dietitians, a pharmacist, and physicians was convened by ASPEN. These individuals participated in the development of the guidelines and jointly authored this document. Any recommendations in this guideline do not constitute medical or other professional advice and should not be taken as such. To the extent that the information published herein may be used to assist in the care of patients, this is the result of the sole professional judgment of the attending healthcare professional whose judgment is the primary component of quality medical care. The information presented here is not a substitute for the exercise of such judgment by the healthcare professional. Circumstances in clinical settings and patient indications may require actions different from those recommended in this document, and in those cases, the judgment of the treating professional should prevail. This paper was approved by the ASPEN Board of Directors.

The guidelines offer recommendations that are supported by review and analysis of the current literature as well as a blend of expert opinion and clinical practicality. The current literature has limitations that include variability in study design; limited description of actual intake levels of energy, protein, and ILE; heterogeneity in patient samples and treatment strategies; and limited information on nutrition status and difficulty in blinding. Because of the electrolyte and fluid instability of most critically ill patients and impracticality of blinding the type or details of feeding (enteral nutrition [EN] vs parenteral nutrition [PN], protein supplements, infusion pump rate, etc), most studies were not blinded to the ordering process or nurse administering the feedings or details of the feeding contents.

TARGET POPULATION

The target population is critically ill adult patients in surgical or medical ICUs who are unable to maintain volitional oral intake and are supported by PN or EN.

INCLUSION CRITERIA

The criteria for inclusion are RCTs that enrolled patients over 16 years of age, had an intervention that included EN or PN, reported clinically important outcomes (mortality, ICU or hospital length of stay [LOS], quality of life, or complications), and were published in English.

EXCLUSION CRITERIA

Studies that included only biochemical, nitrogen balance, metabolic, microbial, or nutrition outcomes; that included quasi-randomization; or that enrolled only patients 16 years or younger were excluded.

TARGET AUDIENCE

These guidelines are intended for use by clinicians, including but not limited to dietitians, nurses, nurse practitioners, pharmacists, physicians, and/or physician assistants who provide nutrition care for critically ill adult patients; nutrition researchers interested in critical illness; and hospital committees with a charge to evaluate nutrition support policies.

DEFINITIONS

Nutrition support refers to the provision of either EN provided by an enteral access device and/or PN provided intravenously. Critically ill patients may also receive IV fluid intake or sedative medications, some of which provide energy.

METHODS

The GRADE process was used to develop the key questions using the PICO (population, intervention, comparator, outcome) format and to plan data acquisition and assessment for these guidelines.2 The task force of experts defined keywords to be used for the literature search, developed key PICO questions that address major contemporary practice themes, and determined the time frame for the literature search, target population, and the specific outcomes to be addressed. These PICO questions defined the limits of the literature search. We plan to revisit this guideline within 5 years or as important evidence becomes available.

Literature search

All citations were culled from the PubMed/MEDLINE database and limited to those posted between January 1, 2001, and July 15, 2020. Search terms are included in Figure 1. Our search strategy was designed to collect citations if (1) they were indexed in MEDLINE and contained at least one term from both group 1a and group 1b, (2) they were indexed in MEDLINE and contained at least one term from both group 2a and group 2b within the citation title or abstract, or (3) they were indexed in the PubMed non-MEDLINE database and contained at least one term from both group 2a and group 2b. The search strategy was then further restricted to only those citations that were cross-referenced to the terms listed in group 3. Analogous strategies were used to search the Embase and Cochrane Central databases.

Search terms for literature search. lang, language; MeSH, Medical Subject Heading; ptyp, publication type

Data acquisition

Each abstract was independently screened by two authors to determine whether the study met the inclusion criteria. Articles that met all three inclusion criteria were reviewed using a standardized data abstraction form (DAF) that was developed based on specific questions for the guideline using the GRADE approach for RCTs. Data retrieval included demographic information, methods used to assess energy and protein requirements, the amount of energy and protein received (ie, exposure variables), various clinical outcome variables, and assessment of quality of the investigation. Trials with quasi-randomization were excluded. Each article was independently reviewed by two task force members, results were compared, differences were resolved by consensus, and a final DAF was created for each trial. For purposes of consistency in analysis and comparisons of findings between studies, for questions 1–4 the “intervention” group was designated as those individuals randomized to receive greater and/or earlier energy and/or higher protein intake; the “control” group was defined as patients randomized to receive less energy or protein or delayed feedings. For questions 5A and 5B, respectively, the intervention group was defined as patients provided mixed-oil ILE or fish-oil (FO) ILE vs soybean-oil (SO) ILE.

Evidence quality assessment

Five main factors are considered when assessing the quality of the evidence in the GRADE approach. Risk of bias refers to limitations in study design (quasi-randomization, lack of blinding, lack of comparability of groups at baseline) or execution (lack of intent-to-treat [ITT] analysis, inadequate delivery of exposure, excess loss to follow-up). The Cochrane Risk of Bias 2 tool (RoB2)3 was used to assess bias. It is important to note that bias, in the context of this new tool, is a comment not on the study itself but on the ability of the study to answer our specific question for each individual outcome (ie, a study may have high risk of bias for one outcome and low risk of bias for another, depending on the question asked and the design of the study). Inconsistency refers to substantial unexplained heterogeneity in results across the trials from which the recommendation is drawn. This is assessed by examining the consistency of study outcomes. Indirectness is a subjective term that refers to how directly applicable the available evidence is to the guideline question. This is assessed by examining limitations in our ability to apply our findings to our population and question of interest. Imprecision denotes the degree to which we are confident the estimated effect size reflects the true effect size. This is assessed by examining the width of CIs and assessing power to detect an effect. Publication bias reflects the likelihood that the nature of the study findings determined its publication status, thereby skewing the combined study findings away from the direction of the findings of studies that were omitted from the literature.

The GRADE process distinctly separates the evaluation of the quality of the body of evidence from the strength of the recommendation statements. This separation enables incorporation of the weight of the risks vs the benefits that occur from adopting the recommendation. Thus, a recommendation may be “strong” despite comparatively weak published evidence if the net benefits outweigh the harms from its adoption. Also, a forest plot that combines included trials may display a difference that is statistically significant (at P ≤ 0.05) but is not a clinically meaningful difference to support a strong recommendation. Table 2 describes the standard language and rationale for the GRADE assigned to a recommendation. Of note, the clinical applications developed for each question were not part of the GRADE process—that is, they do not represent an “expert consensus” of the GRADE data. Rather, they are provided to assist practitioners in the clinical application of our findings that did not directly translate into nutrition support prescription.

TABLE 2. Language for guideline recommendationsa Quality of evidence Weighing risk vs benefits Strength of recommendation Guideline recommendation language Very low to high Net benefits outweigh harms Strong We recommend Very low to high Net harms are considerable and may outweigh the benefits Weak We suggest Note: In very rare cases, a clinical panel may decide that they should not make a recommendation, but in almost all cases, the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) process encourages panels to make a recommendation regardless of the evidence. Wherever possible, these recommendations are evidence based. When this is not possible, they are deemed expert opinion, which is not a category of GRADE. Statistical analysis

Outcomes for which there were three or more studies with comparable data were meta-analyzed using a random-effects model and an alpha significance level of 0.05. Trials that met the inclusion criteria and had data that were pertinent to the question but presented in a dissimilar manner relative to other included trials were discussed in the text but not meta-analyzed. The risk difference (RD) between the intervention and control groups for the outcome variables was calculated. The RD provides information about the absolute effect of the exposure on the risk of the clinical outcome in those in the intervention compared with those in the control group. The risk for any group is derived by dividing the number of events by the number of patients at risk for the event. The RD is a straightforward subtraction of the risk of the occurrence of the event in the control group from the risk of the event in the intervention group. When there is no difference between the intervention and the control groups, the RD = 0. A 95% CI for an RD that contains 0 indicates the difference between the groups is not statistically significant.

This impact and clinical significance of an RD are altered by the underlying risk of having an event such that an RD for low-risk events may be more clinically significant than the same RD for events for which the underlying risk is higher. For example, if the risk in the control group is 0.02 (2%) and the risk in the intervention group is 0.04 (4%), the intervention has doubled the risk (RD = 0.02). Conversely, if the risk in the control group is 0.75 (75%) and the risk in the intervention group is 0.77 (77%), the clinical risk has barely increased, yet the RD is the same (RD = 0.02). For this reason, it is always important to consider underlying risk when interpreting a significant RD.

When trial data could not be combined to estimate the effect size, they were reported in a summary table for each question as the author presented the data. All statistical analyses were performed in Stata 16 (StataCorp; College Station, TX). Summary statistics were calculated and forest plots constructed using random-effects models. Publication bias was assessed through funnel plots and Egger statistics but only presented for main document forest plots with at least 10 studies (see Supporting Appendix). The GRADEPRO Guideline Development Tool (Cochrane Collaboration, 2020) was used to estimate the strength of the body of evidence for each outcome in each question and create summary tables. This was then used to infer the overall quality of the evidence relative to its ability to answer the respective PICO question.

Safety analysis

For questions 1 and 2, to facilitate recommendations for specific cut points in energy and protein provision, a post hoc safety analysis (see Supporting Appendix) was performed with the goal to assess qualitatively hospital mortality across the available range of energy per kilogram. The randomization groups for each study were separated and ordered along an x-axis by energy received per kilogram. This was then plotted against hospital mortality, and linear trend lines were qualitatively assessed. Studies were restricted to RCTs that met our inclusion criteria and to those that provided information on energy received per kilogram. Further between-study differences, such as country and medical system, and whether or not the study comprised all admissions to the ICU vs a subgroup were dealt with through stratification. We also examined stratifications by median splits of Acute Physiology and Chronic Health Evaluation (APACHE) II and body mass index (BMI). Additional forest plots were run to explore combining energy exposure trials that used different interventions (higher vs lower energy exposure, EN vs PN, supplemental PN [SPN] vs standard care) in their study designs. Stratifications of studies that intended hypocaloric interventions or intensive nutrition interventions and forest plots that were ordered by and stratified by the between-group separation of energy received per kilogram were also examined.

RESULTS

Our search strategy detected 2320 citations. Of these, 138 citations were downloaded for further assessment. After review, 80 articles met the inclusion criteria for data abstraction, of which 36 trials contained data that could be used to answer the questions posed and, thus, had DAFs completed.

Question 1. In adult critically ill patients, does provision of higher vs lower energy intake impact clinical outcomes?

Recommendation: No significant difference in clinical outcomes was found between patients with higher vs lower levels of energy intake. We suggest feeding between 12 and 25 kcal/kg (ie, the range of mean energy intakes examined) in the first 7–10 days of ICU stay.

Quality of evidence: Moderate

Strength of recommendation: Weak

Rationale: This broad-range recommendation reflects two major components of our forest plots that limit their interpretation. First, a more specific cut point for energy goals could not be generated, as this would require examination between the outcomes at different levels of energy per kilogram. Unfortunately, this was not possible because of the overlap between the trials’ energy exposure (Table 3) and limited within-study energy differences between intervention and control. Second, the forest plot analysis assumes a linear relationship between the exposures of interest (ie, energy intake) and outcomes; however, energy/outcome relationships have not been demonstrated to be linear. To address these limitations, a safety analysis (see Supporting Appendix) was performed to visually inspect our data for evidence of benefit or harm (Figures S1–S59). Our broad recommendation reflects the range of mean energy exposures in our data and the findings from the safety analysis. This decision was made based on having no evidence of benefit for and a lack of certainty regarding harms of energy provision consistent with those recommended in past guideline recommendations.

TABLE 3. Data summary for question 1: In adult critically ill patients, does provision of higher vs lower energy intake impact clinical outcomes? First author, year Population Comparison (n) of higher vs lower energy Intake provided Infections, n (%) Time on mechanical ventilation, median (IQR) or mean ± SD, days Length of stay, median (IQR) or mean ± SD, days Mortality, n (%) Arabi, 20115

Adult med/surg ICU patients

Mean BMI: 28.5 vs 28.5

Standard-care feeding (n = 120) vs 60%–70% goal energy (n = 120)

Energy (kcal/day): 1252 ± 433 vs 1067 ± 306

Protein (g/day): 48 ± 21 vs 44 ± 19

Sepsis: 56 (47) vs 53 (44) 13.2 ± 15.2 vs 10.6 ± 7.6

ICU: 14.5 ± 15.5 vs 11.7 ± 8.1

Hospital: 67.2 ± 93.6 vs 70.2 ± 106.9

28-day: 28 (23) vs 22 (18)

ICU: 26 (22) vs 21 (18)

Hospital: 51 (43) vs 36 (30)

6-month: 52 (55) vs 38 (33)

Arabi, 20154

Adult med/surg ICU patients

Mean BMI: 29.7 vs 29

25 kcal/kg (n = 446) vs 40%–60% goal (n = 448)

Energy (kcal/day): 1299 ± 467 vs 835 ± 297

Protein (g/day): 59 ± 25 vs 57 ± 24

Pneumonia: 90 (20) vs 81(18)

Any infection: 169 (38) vs 161(36)

10 (5–16) vs 9 (5–15)

ICU: 13 (8–20) vs 13 (8–21)

Hospital: 30 (14–63) vs 28 (15–54)

28-day: 97 (22) vs 93 (21)

ICU: 85 (19) vs72 (16)

Hospital: 123 (28) vs 108 (24)

90-day: 127 (29) vs 121 (27)

180-day: 140 (32) vs 131 (30)

Braunschweig, 20157

ALI patients in med ICU

BMI ≥ 30, 45% vs 47%

Intensive therapyb (n = 40) vs standard care (n = 38)

Energy (kcal/kg/day): 25.4 ± 6.6 vs 16.6 ± 5.6

Protein (g/day): 82 ± 23 vs 60.4 ± 24

NR 6 (4–10) vs 7 (3–14)

ICU: 15.5 ± 12.8 vs 16.1 ± 22.5

Hospital: 27.2 ± 18.2 vs 22.8 ± 14.3

Mortality: 16 (40) vs 6 (16) Chapman, 201812

Adult patients in 46 ICUs who were mechanically ventilated and eligible for EN

Mean BMI: 29.2 vs 29.3

1.5-kcal/ml (n = 1971) vs 1.0-kcal/ml (n = 1986) formula administered at 1 ml per kilogram of IBW

Energy based on IBW (kcal/kg/day): 30.2 ± 7.5 vs 21.9 ± 5.6

Energy based on ABW (kcal/kg/day): 23.9 (7.8) vs 17.4 (5.5)

Bacteremia: 228 (12) vs 221 (11) NR NR

Hospital: 468 (24) vs 470 (24)

28-day: 450 (23) vs 455 (23)

90-day: 523 (27) vs 505 (26)

Charles, 20146

Adult patients in surg ICU

Mean BMI: 28.1 vs 32.9

Standard 25–30 kcal/kg (n = 42) vs 12–15 kcal/kg (n = 41)

Energy (kcal/kg/day): 17.1 ± 1.1 vs 12.3 ± 0.7

Protein (g/kg/day): 1.1 ± 0.1 vs 1.1 ± 0.1

Bacteremia: 8 (19) vs 10 (24)

Pneumonia: 20 (48) vs 18 (44)

Any infection: 32 (76) vs 29 (71)

NR

ICU: 13.5 ± 1.1 vs 16.7 ± 2.7

Hospital: 31 ± 2.5 vs 35.2 ± 4.6

Mortality: 4 (10) vs 3 (7) Desachy, 20088

Mechanically ventilated patients

Mean BMI: 25 vs 27

Immediate EN (n = 50) vs gradual EN (n = 50)

Energy (kcal/day): 1715 ± 331 vs 1297 ± 331

Protein (g/day): 82 ± 23 vs 60.4 ± 24

Pneumonia: 0(0) vs 0(0) NR

ICU: 15 ± 11 vs 15 ± 11

Hospital: 56 ± 59 vs 51 ± 75

ICU: 6(12) vs 8(16)

Hospital: 14(28) vs 11(22)

Doig, 20159

Patients in 13 med/surg ICUs

Mean BMI: 28 vs 28

Standard care (n = 165) vs 2 days with 20 kcal/h then gradual increase to usual (n = 166)

Energy (kcal/kg/day): NR

Protein: NR

Pneumonia: 22 (13) vs 14 (8)

Bacteremia: 8 (5) vs 2 (1)

Any infection: 27(16) vs 13(8)

a7.45 (7.16 to 7.65) vs 7.86 (7.54 to 8.18)

ICU: a10.0 (9.2–0.9) vs 11.4(10.5–12.4)

Hospital: a21.7 (20.0–23.5) vs 27.9 (25.7–30.3)

ICU: 15(9) vs 9(5)

Hospital: 30(18) vs 15(9)

60-day: 35 (21) vs 15 (9)

90-day: 35 (21) vs 21 (13)

Peake, 201410

Adult critically ill patients in surg ICU

Mean BMI: 28.7 vs 26.2

1.5 kcal/ml (n = 57) vs 1.0 kcal/ml (n = 55)

Energy (kcal/kg/day): 27.3 ± 7.4 vs 19.0 ± 6.0

Protein (g/kg/day): 1.0 ± 0.3 vs 1.1 ± 0.3

NR NR

ICU: 9.6 (5.9–22.6) vs 11.8(6.9–22.8)

Hospital: 34.5(16.9–83.6) vs 30.6(15.2–undefined)

ICU: 6 (11) vs 9 (16)

Hospital: 10(19) vs 14 (27)

28-day: 11(20) vs 18(33)

90-day: 11(20) vs 20(37)

Petros, 201611

Adult patients in med ICU

Mean BMI: 27.1 vs 28.6

Normocaloric (n = 54) vs hypocaloric (n = 46) EN or PN

Energy (kcal/kg/day): 19.7 ± 5.7 vs 11.3 ± 3.1

Protein (g/kg/day): NR

Any infection: 6(11) vs 12(26) 7.44 (2.90–16.80) vs 10.60 (4.81–28.60) NR

ICU: 12(22) vs 10(22)

Hospital: 17(32) vs 17(37)

28-day: 18(33) vs 18(39)

Rice, 201114

Adult patients with respiratory failure in med ICU

Mean BMI: 28.2 vs 29.2

Normocaloric (n = 102) vs trophic EN (n = 98)

Energy (kcal/day): 1418 ± 686 vs 300 ± 149

Protein (g/day): 54.4 ± 33.2 vs 10.9 ± 6.8

NR NR

ICU-free days: 21(9.3–24) vs 21(6.5–24)

Hospital-free days: 16.5(0–21) vs 12(0–21)

Hospital: 20 (20) normocaloric vs 22(22) trophic Rice, 201213

Adults with ALI in 44 ICUs

Mean BMI: 30.4 vs 29.9

25–30 kcal/kg full EN feeding (n = 492) vs trophic (n = 508) NR

Bacteremia: 46(9) vs 59(12)

Pneumonia: 33(7) vs 37(7)

NR ICU-free days in 28 days: a 14.7(13.8–15.6) vs 14.4(13.5–15.3) 60-day: 109(22) vs 118(23) Rugeles, 201616

Adult patients expected to require EN for >96 h in med/surg ICU

Mean BMI: 25 vs 25

Normocaloric (n = 60) vs hypocaloric (n = 60)

Energy (kcal/kg/day): a 19.2 ± 4.3 vs 12.1 ± 2.6

Protein (g/kg/day): a1.3 (0.3) vs 1.3 (0.3)

NR 9(8.3) vs 9(8.3) ICU: 10.5(8.0) vs 12(7.3) 28-day: 16(27) vs 18(30) Singer, 201115

Adult med/surg ICU

Mean BMI: 27.8 vs 27.4

Indirect calorimetry- measured requirement (n = 65) vs 25 kcal/kg/day (n = 65)

Energy (kcal/day): 2086 ± 460 vs 1480 ± 356

Protein (g/day): 76 ± 16 vs 53 ± 16

Bacteremia: 13(20) vs 8(12.3)

Pneumonia: 18(27.7) vs 9(13.8)

16.6 ± 14.7 vs 10.5 ± 8.3

ICU: 17.2 ± 14.6 vs 11.7 ± 8.4

Hospital: 33.8 ± 22.9 vs 31.8 ± 27.3

ICU: 16(25) vs 17(26) Abbreviations: ABW, actual body weight; ALI, acute lung injury; BMI, body mass index (kg/m2); EN, enteral nutrition; IBW, ideal body weight; ICU, intensive care unit; IQR, interquartile range; med, medical; NR, not reported; PN, parenteral nutrition; surg, surgical.

To be included in the analysis for this question, the trial needed to randomize energy exposure without causing a secondary competing intervention, such as a shift from EN to PN. Although this question was meant to be a broader question on general energy intake from any source, as a sensitivity analysis, protein dose was also considered as a potential competing intervention. The forest plots below therefore contain both combined results and results stratified by whether the trial was isonitrogenous between allocation group. Thirteen trials4-16 representing data from 8690 patients were included to answer this question (Table 3). In acknowledgment of the lack of difference between PN and EN, reported in question 3, the safety analysis also included a series of plots that combined these studies from questions 1, 3, and 4 in its analysis.

Methodological quality and intervention design varied between trials. One trial11 reported important baseline differences, including imbalances in baseline diabetes mellitus prevalence, that were not controlled for in their analysis. Two trials9, 17 lacked a true ITT design, removing participants from analysis after they withdrew consent. Because withdrawal counts were low and did not likely impact the outcomes, they are included in this analysis. Most trials included all patients admitted to the ICU,4-6, 9-