This is an electronic version of an article published
in the American Journal of Gastroenterology 2007;102(2):412-429.
Does enteral nutrition affect clinical outcome?
A systematic review of the randomized trials
Ronald L. Koretz, M.D.
Olive View-UCLA Medical Center
Alison Avenell, M.D., M.R.C.P., F.R.C. Path., M.B., B.S., M.Sc.
Health Services Research Unit*
University of Aberdeen
Timothy O. Lipman, M.D.
Veterans Affairs Medical Center
Carol L. Braunschweig, Ph.D., R.D.
University of Illinois at Chicago
Anne C. Milne, M.Sc., S.R.D.
Health Services Research Unit*
University of Aberdeen
Submitted August 31, 2006
Address communication to:
Ronald L. Koretz, M.D.
Department of Medicine
Olive View-UCLA Medical Center
14445 Olive View Drive
Sylmar, California 91342
* The Health Services Research Unit is core funded by the Chief Scientist Office of the Scottish
Executive Health Department; however, the views expressed here are those of the authors.
1. Because malnourished patients have poorer outcomes than do nourished ones, artificial nutrition is
employed in order to improve the clinical outcome.
2. Randomized trials of parenteral nutrition have largely been unable to show that this therapy improves
the clinical outcome, and have even suggested that the therapy is associated with net harm (especially in
patients undergoing cancer chemo- or radiation therapy).
What is new:
1. Randomized trials of low quality have indicated that nutritional supplements, taken orally, improve
survival in malnourished geriatric patients who are in institutions (including hospitals). Two trials of high
quality found effects in the same direction, but failed to achieve statistical significance.
2. Randomized trials of low quality have suggested that enteral nutrition may reduce infection rates in
critically ill patients and improve outcome in low-birth-weight infants (when provided as “trophic
3. Randomized trials of low quality have suggested that both enteral nutrition and oral supplements may
be helpful in the postoperative situation and in patients with chronic liver disease.
4. Two large high quality randomized trials found that there is no benefit in providing enteral nutrition in
the first week to patients with strokes associated with dysphagia or in providing supplements to stroke
patients who do not have dysphagia.
5. Low quality randomized trials have failed to demonstrate any utility from EN in the non-surgical
treatment of cancer patients, or in the treatment of patients with inflammatory bowel disease or hip
6. Low quality randomized trials have failed to demonstrate any utility from oral supplements in the non-
surgical treatment of cancer patients or in patients with chronic pulmonary disease or hip fractures.
7. There are no or inadequate data from randomized trials to assess the use of enteral nutrition or oral
supplements in any other diseases.
Background: Both parenteral nutrition (PN) and enteral nutrition (EN) are widely advocated as
adjunctive care in patients with various diseases. A systematic review of 82 randomized controlled trials
(RCTs) of PN published in 2001 found little, if any, effect on mortality, morbidity, or duration of hospital
stay; in some situations, PN increased infectious complication rates. Objective: To assess the effect of
EN or volitional nutrition support (VNS) in individual disease states from available randomized
controlled trials (RCTs). Design: We conducted a systematic review. RCTs comparing EN or VNS to
untreated controls, or comparing EN to PN, were identified and separated according to the underlying
disease state. Meta-analysis was performed when at least 3 RCTs provided data. The evidence from the
RCTs was summarized into one of five grades. A or B indicated the presence of strong or weak (low
quality RCTs) evidence supporting the use of the intervention. C indicated a lack of adequate evidence to
make any decision about efficacy. D indicated that limited data could not support the intervention. E
indicated either that strong data found no effect, or that either strong or weak data suggested that the
intervention caused harm. Patients and settings: RCTs could include either hospitalized or non-
hospitalized patients. The EN or VNS had to be provided as part of a treatment plan for an underlying
disease process. Interventions: The RCT had to compare recipients of either EN or VNS to controls not
receiving any type of artificial nutrition or had to compare recipients of EN with recipients of PN.
Outcome measures: Mortality, morbidity (disease-specific), duration of hospitalization, cost, or
interventional complications. Summary of grading:
A – No indication was identified.
B – EN or VNS in the perioperative patient or in patients with chronic liver disease; EN in critically ill
patients or low birth weight infants (trophic feeding); VNS in malnourished geriatric patients. (The low
quality trials found a significant difference in survival favoring the VNS recipients in the malnourished
geriatric patient trials; two high quality trials found non-significant differences that favored VNS as well.)
C – EN or VNS in liver transplantation, cystic fibrosis, renal failure, pediatric conditions other than low
birth weight infants, well-nourished geriatric patients, non-stroke neurologic conditions, AIDS; EN in
acute pancreatitis, chronic obstructive pulmonary disease, non-malnourished geriatric patients; VNS in
inflammatory bowel disease, arthritis, cardiac disease, pregnancy, allergic patients, preoperative bowel
D – EN or VNS in patients receiving non-surgical cancer treatment or in patients with hip fractures; EN in
patients with inflammatory bowel disease; VNS in patients with chronic obstructive pulmonary disease
E – EN in the first week in dysphagic, or VNS at any time in non-dysphagic, stroke patients who are not
malnourished; dysphagia persisting for weeks will presumably ultimately require EN.
Conclusions: There is strong evidence for not using EN in the first week in dysphagic, and not using
VNS at all in non-dysphagic, stroke patients who are not malnourished. There is reasonable evidence for
using VNS in malnourished geriatric patients. The recommendations to consider EN/VNS in
perioperative/liver/critically ill/low birth weight patients are limited by the low quality of the RCTs. No
evidence could be identified to justify the use of EN/VNS in other disease states.
An association exists between malnutrition and a poor clinical outcome. Furthermore, deprivation
of nutrient intake for a long enough period of time will have adverse clinical consequences. These two
observations have led to the hypothesis that providing artificial nutrition to patients who are, or who are at
risk of becoming, malnourished, would be beneficial. However, artificial nutrition is a medical
intervention with associated risks and costs. As such, we need to know that it is efficacious. The best
way to establish efficacy is to demonstrate it in a randomized controlled trial (RCT). In 2001, a
systematic review of the RCTs of parenteral nutrition (PN) failed, with a few exceptions, to find outcome
That review considered the intravenous infusion of nutrients. While the rationale for artificial
nutrition is usually the same when it is delivered directly into the gastrointestinal tract (consumed by
mouth or infused through a tube), the physiologic mechanisms are different than when it is provided
intravenously. It would be inappropriate to extrapolate any conclusions from PN to another form of
artificial nutrition. The objective of this systematic review is to evaluate the clinical efficacy of medical
interventions that deliver nutrient formulations directly into the gastrointestinal tract through defined
orally consumed supplements (volitional nutritional support [VNS]) or via a tube (enteral nutrition [EN]).
In doing so, we focus on RCTs that evaluate the ability of such interventions to alter one or more
clinically important outcomes in specific disease states.
A protocol was written in which a priori decisions were made regarding methodology. The
terminologies used to describe study groups and the artificial nutrition formulations are defined in Table
RCTs were identified employing previously reported strategies (1); the details are available
electronically (“Literature Search Methodology”). This search was begun in 1975; in the intervening 3
decades, some efforts were made to contact authors to obtain more information about a particular study.
However, no systematic effort was undertaken to contact all authors for this systematic review. No
language restrictions were employed; when translation facilities were not available, the data from the
English abstract, as well as any interpretable data from tables, were employed. Abstracts from meetings
The RCTs that compared EN to no treatment, EN to PN, or VNS to no treatment were so
categorized within each disease state. We assessed the outcomes of mortality, total and/or infectious
complications, lengths of hospitalization, costs, interventional complications (e.g., nausea/vomiting,
diarrhea, hyperglycemia), or any of the disease-specific outcomes. A list of these specific diseases, as
well as the disease-specific outcomes, is available electronically (“Disease States Considered”). Meta-
analysis (Revman 4.2, Cochrane Collaboration) was performed when data were available from 3 or more
RCTs. (If only one or two trials were available, we thought that there would be no more insight available
from data combination by meta-analysis than could be gleaned by just assessing each trial individually.)
Any RCT meeting the following criteria was used in the respective meta-analysis:
1) The report explicitly stated that the groups were randomized; quasirandomized or cluster randomized
trials were excluded.
2) The study compared treatment groups as defined above.
3) The study reported one or more of the outcomes being sought.
4) The therapeutic intervention was employed for at least five days, during which time the control group
received only standard therapy (or PN in the comparative trials). (Since the hypothesis underlying
artificial nutrition is that the morbidity or mortality is due to malnutrition, it was assumed that it would
require at least five days to begin to reverse that process; this same assumption was made previously .)
If a report included groups of patients randomized to an untreated control and to two or more
forms of EN or VNS, all treated groups were combined and compared to the control. If a report included
groups of patients assigned to EN and to VNS, the data from the EN group were included in the EN
analysis and the data from the VNS group were included in the VNS analysis. (The same control group
was used in both analyses in this latter case.)
If dichotomous outcomes were reported as a total number instead of the number of affected
patients, it was assumed that there was one event per patient. (If the number of events was greater than
the number of patients, each patient was assigned one event.) Numerical estimates were made from
graphs when necessary.
Meta-analysis of continuous data requires knowledge of both the mean and standard deviation for
both treatment arms. Thus, the initial analyses only included trials in which these numbers were reported.
However, one of the peer reviewers of this paper pointed out a statistical method of converting the median
and range into a mean and variance (2). Thus, post facto analyses were also performed employing
continuous data from trials in which the medians and ranges were provided. Since none of those analyses
materially changed any of the estimates or conclusions, they will not be subsequently reported.
Two reviewers independently abstracted the information from each RCT onto predesigned data
summary forms. Discrepancies were resolved by consensus between the two abstractors.
Heterogeneity (i.e., adding apples and oranges) is a limitation of meta-analysis. It can be sought
with statistical tests; the Revman software calculates both Cochran’s Q (expressed as a p value) and I2
statistics (3). Statistical heterogeneity was defined as p < 0.10 or I2 > 25%. A fixed (or random) effects
model was employed when heterogeneity was absent (or present).
Dichotomous data are presented as an absolute risk difference (ARD), namely the difference
between the incidence in the treated group and that in the control group. A 95% confidence interval (CI)
that did not include 0% was considered “significant”. The term “tendency” was applied to an analysis in
which one end of a confidence interval was 0% (i.e., the confidence interval included the possibility of no
difference). A negative ARD indicates a beneficial effect associated with the treatment.
Continuous variables are reported as weighted mean differences and 95% CIs. The same rules for
significance and tendency were applied to these calculations. A negative number represents a beneficial
effect in the treated group.
RCTs with more methodologic rigor have demonstrated lower treatment effects (4, 5); this is
presumed to represent the advertent or inadvertent introduction of bias into the less rigorous trials.
“Quality” is a term that reflects the degree of rigor that was employed. Thus, each trial was graded for
quality, with the expectation that all of the studies were randomized and the large majority of them were
not blinded. A “high quality” (lower risk of bias) study was one in which either 1) the
investigators/assessors and participants were blinded, or 2) the RCT contained both an explicit description
of an adequate allocation concealment and data that were evaluable on an intent-to-treat basis. RCTs not
meeting one or the other of these criteria were categorized as low quality.
In order to explore possible sources of heterogeneity, subgroup analyses were planned. These
planned analyses are available electronically (“Analyses and subgroup analyses planned”). Those that are
relevant will be discussed.
Our primary focus is the strength of the evidence that supports the use of EN or VNS in each
disease state. These assessments (grades) are based only on data from RCTs. Nonrandomized controlled
trials are less reliable because of potential bias (6). Expert opinion is often based on incomplete
information (7-9) and/or influenced by conflicts of interest (10, 11).
The evidence was graded by the amount and quality of the RCTs that were available as well as
what effects those trials did, or did not, demonstrate. The grades ranged from A to E, with A describing a
situation in which there was a high likelihood of benefit and E indicating a high likelihood of no benefit,
or even of harm. For some conditions, a limited amount of information was available, and that
information failed to demonstrate any significant differences in outcomes. In such situations, it may be
that a true difference was present, but the numbers were too small to establish it (i.e., a type II error), or it
may be that the failure to observe a difference was due to the intervention not having an effect.
It was decided that, if at least three low quality RCTs had failed to show a difference, the
intervention would be judged to be ineffective. If only one or two RCTs were available, the total number
of patients were considered; if that total was less than 100, the evidence was classified as being
inadequate to determine the presence or absence of a benefit.
If at least 100 patients were studied, the direction of the difference was then considered. (If two
RCTs were available, the data were arithmetically combined.) If that direction was in favor of the control
group, the intervention was judged to be ineffective. If it was in favor of the treated group, a p value for
the difference was obtained. If that p value was < 0.20, it was concluded that a type II error might be
present, so the data were considered as inadequate. (Employing a one-sided test [since only benefit was
being sought] and an α error of 0.20, there is approximately an 80% power of seeing an ARD of 15%
between the two groups; since the trials were almost always of low quality, the real ARD would likely be
lower. Given the resources required [both economic and time], such an ARD would be the lowest one
that would be clinically meaningful.)
The evidence was thus divided into 5 grades:
A – One or more high quality RCTs demonstrated benefit.
B – The evidence of benefit was limited to low quality RCTs.
C – There were inadequate data to decide if benefit is present or absent. Three different scenarios could
result in a C grade: 1) no RCTs were available; 2) only one or two RCTs containing fewer than 100
patients were available; 3) 100 or more patients were available but a type II error could not be ruled out.
D – Limited evidence was not able to define a benefit. Either at least three low quality RCTs failed to
show a difference or a type II error was not likely to be present.
E – One or more high quality RCTs indicated that the intervention was not effective or that there was any
evidence that it caused harm.
As will be noted, none of the RCTs included severely malnourished individuals. Thus, these
grades will refer to studies in which the patients were not severely malnourished and would not be
deprived of nutrient intake for > 2 weeks. Although there are only limited data to allow us to know with
certainty the period of time for which such individuals can tolerate nutrient deprivation, it was previously
concluded that waiting at least 2 weeks in the non-severely malnourished patient was reasonable (1).
Summary of RCTs identified
The total number of assessed titles could not be determined; this information was not collected
during the handsearch of Index Medicus. However, each identified RCT was stored, so the total number
of RCTs from which those included in the meta-analyses were derived could be ascertained.
Only 33 of 376 RCTs of EN met our inclusion criteria (12-45); data for one of these trials were
extracted from two different abstracts (25, 26). Similarly, we included 48 of 115 RCTs comparing EN to
PN (19, 25, 26, 44, 46-92); data for three of these trials were extracted from two different papers (25, 26,
69, 70, 76, 77). Finally, 54 of 418 RCTs of VNS were included (30, 93-145).
Data from one of the EN versus PN RCTs (55) were published at least seven different times (55,
146-151). Different numbers of patients appear in each publication, and, in at least one instance, a portion
of the study was extended and additional patients were added (151). In order to avoid duplication, we
used the most recent paper (55) and supplemented any missing categories of data with information from
the other publications. The details of the 135 RCTs employed in the various meta-analyses (“Study
Characteristics”), as well as the excluded trials and the reason(s) for exclusion (“References Excluded
from Meta-analyses”), are available electronically.
The results of the meta-analyses of the perioperative RCTs (12, 13, 15, 19, 24, 27, 29, 32, 34, 36-
40, 44, 45, 48, 50, 53-56, 64-66, 71, 74, 75, 78, 81, 82, 84-87, 95, 102, 109, 115, 122, 134, 136, 138, 141)
are summarized in Table 2. No differences were seen with regard to mortality. EN, when compared to no
artificial nutrition, was not shown to reduce the rate of total/major/wound complications or postoperative
pneumonia. There were significantly fewer infections in the EN recipients as well as a tendency for fewer
intraabdominal or intrathoracic complications. The heterogeneity resolved, and the significant differences
persisted, in the low quality trials and in the trials of the patients without cancer.
EN produced better outcomes than PN. VNS provided across-the-board benefits. These
conclusions must also be tempered by the fact that the data were largely derived from low quality studies.
Several years ago, Lewis et al (152) published a meta-analysis of 11 RCTs (12, 15, 24, 34, 45,
153-158) that assessed early postoperative enteral nutrient delivery in surgical patients. These trials
utilized EN or VNS. Lewis et al noted that artificial nutrition reduced the infection rate, but suggested
that a large trial needed to be conducted. Six of these RCTs were excluded from our analyses. While our
analysis of 11 trials of mostly postoperative EN (12, 15, 24, 27, 29, 32, 34, 36, 38, 39, 45) demonstrated a
similar effect, when only the four high quality trials were considered, this observation could not be
confirmed (ARD –2% [95% CI –25%, + 21%]).
Specialized formulations containing putative immunonutrients (ω-3 fatty acids, arginine,
ribonucleic acid, and/or glutamine) are often employed. A limited database (24, 55, 109) precluded
meaningful conclusions. Such specialized EN resulted in fewer infections than did standard EN (159).
The durations of therapy in the VNS trials were analogous to those in the EN trials (days or a few
weeks). However, all of these patients were capable of orally consuming supplements. The beneficial
effect was limited to the trials of postoperative VNS and VNS in malnourished patients.
VNS also appeared to shorten hospitalization. The subgroup analyses indicated that this benefit
was limited to the trials of preoperative VNS and VNS provided to nourished patients. Given the data
regarding complications, this observation seems counterintuitive. The quality of the trials was low and
duration of hospitalization is a subjective outcome. Furthermore, one high quality (22) and four other low
quality (115, 136, 141, 160) trials that did not provide adequate information to be included in the meta-
analysis failed to find significant differences in the duration of hospitalization.
Non-surgical cancer treatment
PN causes net harm to cancer patients undergoing chemotherapy or radiation therapy (1). There
was no apparent positive or negative effect from the enteral provision of nutrients (either through a tube
[35, 43] or by volitional consumption [93, 96, 103, 126, 135]). The trials of VNS provided mortality data;
the estimated effect favored the control group (ADR +5%) but the CIs crossed 0% (-2%, +12%). Two of
the VNS trials reported data regarding the incidences of gastrointestinal toxicity; in one the treated group
had significantly more (96) and in the other the treated group had significantly fewer (135). No trials
assessed EN or VNS in hematologic malignancies. A systematic review of artificial nutrition in bone
marrow transplantation did not identify any data regarding the enteral delivery of nutrients (alone or in
comparison to PN) (161).
Liver disease Download full-text
Chronic liver disease: The outcomes in patients with chronic liver disease that could be assessed
with meta-analysis are summarized in Table 3 (97, 113, 125, 129, 137). VNS did not have any effect on
survival. Survival was also not affected when VNS was provided with EN as a backup (162) or when
given with an anabolic steroid (163). VNS did not have a significant effect on infectious complications or
the development of hepatic encephalopathy. Although the data could not be combined, VNS did not
appear to influence the length of hospitalization (97, 113, 129). Limited information was available from
one or two trials about other outcomes. There was no effect on the subsequent development of
hepatocellular carcinoma (125) or gastrointestinal bleeding (113, 129). The VNS recipients had
significantly lower incidences of total complications in two trials (113, 129) and also may have been less
likely to develop ascites (113, 129).
EN did not appear to have any impact on liver-disease-associated morbidity (14, 17, 21, 28, 164)
nor on the length of hospitalization (14, 17, 28). Survival with EN was not different from that with PN
(79). In light of these observations, it is curious that three low quality RCTs (14, 17, 28) of EN versus no
therapy found improved survival in the treated recipients when the fixed effects model was employed.
This model was used because the p value for heterogeneity was 0.21; however, the I2 statistic was 37%
and when the random effects model was employed, this significant difference disappeared.
Branched-chain amino acids had a modest effect in treating hepatic encephalopathy but this effect
disappeared if the analysis was limited to the high quality trials (165).
Liver transplantation: No significant differences were observed (23, 90, 120).
Pancreatic disease (acute pancreatitis)
RCTs have evaluated the use of EN in acute pancreatitis, but not in pancreatic insufficiency. The
meta-analyses of five RCTs (46, 49, 67, 72, 73) comparing EN to PN are displayed in Table 4.