Individuals of good health with normal body weight and composition can survive only about 2 months with no or minimal nutrition support. When serious, systemic illness occurs, this period may be shortened considerably as increased metabolism may occur. Although formal trials demonstrating the benefits of nutrition supplementation are lacking,1,2 the provision of artificial (nonvolitional) nutrition support is regarded as essential if significant, suboptimal nutrition intake has been present for 1-2 weeks in such patients.3 What remains controversial is the timing of its introduction and the route and regimen of its administration.
Although nonvolitional nutrition support has been employed since antiquity (the ancient Egyptians used rectal feeding4), it was not until 1598 that the use of feeding tubes was first reported. In the 18th century, John Hunter, the pioneering Rritish surgeon, used eel skin tubes for this purpose.4 The relatively large diameter of the early nasogastric feeding tubes and the absence of a soft yet durable material from which to manufacture such tubes hampered their use, particularly in patients requiring prolonged feeding. A significant leap forward was made in the early 1980s,5 when percutaneous gastrostomy feeding tubes were first introduced. Notwithstanding this, control trial data showing that percutaneous gastrostomy feeding leads to better outcomes compared with nasogastric feeding are lacking. More recently, jejunostomy feeding tubes and nasojejunal feeding have also provided a viable option for those seeking to feed their patients enterally.
Parenteral nutrition (PN), on the other hand, has a much more recent history, being first used in the late 1960s.6 It was not until Jeejeebhoy's group7 showed that patients who would otherwise have died could be kept alive for many months with parenteral feeding that its value became appreciated and its use became widespread. PN has also undergone a significant although less staggered improvement in delivery systems, biochemical composition, and early recognition and prevention of metabolic side effects.
Not only is the history of enteral nutrition (EN) and PN different, but their complications are different as well. Enteral feeding tubes can lead to nasal, pharyngeal, and upper GI trauma. Sinusitis can also occur. Persistent intolerance of enterically administered feed can both compromise nutrition status and predispose to pulmonary aspiration. EN-induced diarrhea can lead to protein, fluid, and electrolyte losses. More recently, nonobstructive intestinal ischemia has become recognized as a serious complication of enteral feeding.8 PN, on the other hand, is fraught with important although very different delivery system problems both at its inception (hemorrhage, pneumothorax, and malposition of the line) and as part of its ongoing administration (line sepsis and occlusion). Intestinal atrophy and bacterial translocation are complications, the significance of which is still open to debate.9 The only complications common to both types of nonvolitional nutrition support are metabolic (refeeding syndrome, electrolyte and fluid derangements, and hyperglycemia). Rut even here, the modulatory effect of the intestine and the first-past effect of the liver serve to substantially alter both the bioavailability and metabolism of enterally administered substrates and in general attenuate these complications.
Notwithstanding their different histories and complication profiles, there has been a concerted effort over the past 20 years to compare enteral feeding with parenteral feeding head to head. Such is the plethora of studies in this area that 5 meta-analyses2,10-13 have been conducted (Table 1). In keeping with their cities/countries of origin, the meta-analyses will be referred to as Chicago,10 Canadian,11 Aust 1,12 Aust 2,13 and USA/Scotland.2 The USA/Scotland meta-analysis is in fact a group of meta-analyses published together. To allay confusion and repetition, it will be referred to in the singular. The meta-analyses are noteworthy for the care taken in ensuring that the largest possible group of trials be considered for inclusion. In each case, a relatively small group of trials was then selected. Care was taken to examine clinical outcomes only and to identify publication bias. In Aust 2, the smallest of the meta-analyses, the methodology was the most scrupulous, with studies having < 90% follow-up being excluded. Notwithstanding this, serious shortcomings are present in all 5 meta-analyses. The trials selected in these meta-analyses are also flawed. A more detailed evaluation is presented below.
Methodological Problems of Meta-analyses Comparing EN With PN
A summary of the methodological problems related to these meta-analyses is listed in Table 2.
1. Selection of Studies: The Limitations of a Consensus Approach
In 4 of the meta-analyses,2,10-12 differences in opinion regarding which studies to include were resolved by consensus-hardly an objective or scientific process. Neither the nature of the consensus process nor the trials subjected to it was stated. Despite being published within a few years of each other, the trials included in each analysis were significantly different. Of those selected as candidates for the meta-analysis, there were trials in each of the meta-analyses other than Aust 2 that were unique to each one. These differences are indicative of the absence of an agreed-upon, objective methodology with respect to the evaluation of candidate trials.
2. Infectious Complications
No mention was made as to how infectious complications were analyzed in the Chicago or the USA/Scotland studies. The Canadian meta-analysis advised, "The nature of the infectious complications varied with the particular patient population and included pneumonia, aspiration pneumonia, urinary tract infection, bacteremia, wound infection, abdominal abscess and line sepsis." Aust 1 defined infectious complications as "the occurrence of any infection . . . related or unrelated to nutrition, recorded as the authors of the study reported." Aust 2 used a definition of infection based on positive culture results only. Even if the definition of an "infectious complication" were without controversy, simply adding up reported incidences of a wide variety of infections without regard to their site of origin, severity, and type of organism, then subjecting these results to statistical analysis is of highly questionable clinical significance. Using culture results alone is more objective but also fraught with difficulty, as the clinical impact of a cultured pathogen varies very greatly. In addition, laboratories vary in methodology and consistency in processing material for microbiological analysis.
3. Failure to Allow for Hyperglycemia
Two of the 5 meta-analyses made no mention of the potential confounding factor of hyperglycemia on infectious complications. Hyperglycemia is a more common adverse effect in PN patients and has been shown to confer an increased risk of sepsis in critically ill patients.14 The Canadian study mentioned that in a subgroup analysis, no difference in treatment effect was found between those studies in which the PN groups received more calories or had a higher incidence of hyperglycemia. Aust 1 only mentioned hyperglycemia as a possible factor in the discussion, but without elaboration. Only the USA/Scotland study systematically examined metabolic complications including hyperglycemia, concluding that PN was associated with a higher risk of hyperglycemia compared with EN in critically ill patients and patients with acute pancreatitis. The possible influence that this may have had on infectious complications and mortality was not discussed.
4. Timing of Introduction of Nutrition Support and Differences in Feeding Regimens
Considering that PN has traditionally been commenced later than EN, it is surprising that there was so little attention paid to the timing of introduction of nutrition support in any of the meta-analyses. Aust 1 included trials only if the institution of nutrition support occurred within 96 hours of hospitalization, intensive care unit admission, or surgery. Waiting for 96 hours before the introduction of EN would be considered tardy by many observers, whereas introducing PN before 96 hours would often be regarded as premature. Interestingly, in Aust 2, the survival benefit of PN over EN was present only if the EN introduction was delayed. There was limited or no discussion concerning the timing of introduction in the other 3 meta-analyses.
With respect to caloric intake and feeding regimens, the Canadian meta-analysis discussed these, albeit briefly without elaboration in the discussion, indicating that there was no treatment effect between studies in which the PN groups received more calories. The USA/Scotland study mentioned limited data, indicating that enteral formulations with putative immunonutrients led to fewer infectious complications compared with standard enteral formulations in the critically ill. There was no discussion, however, about different regimens or caloric intake with respect to trials comparing EN with PN in this context.
5. Heterogeneity of Trials (Patient Population)
The most important shortcoming of the meta-analyses is the heterogeneity of the included trials, both in clinical material and methodological quality. Trials of patients with acute pancreatitis, trauma, elective upper GI surgery, inflammatory bowel disease, and sepsis were bundled together. Only 3 meta-analyses made any attempt to allow for clinical heterogeneity. The Canadian meta-analysis included only cases of trauma, pancreatitis, and major sepsis, excluding elective surgical cases. Trauma, pancreatitis, and major sepsis, however, are likely to have very different effects on intestinal function and integrity. Aust 1 split its 30 trials into 3 subgroups, medical (10), surgical (11), and trauma (9), although it also reported statistical analysis when all 30 were included. The USA/Scotland study split the trials into several clinical subgroups: perioperative (16), acute pancreatitis (4), inflammatory bowel disease (4), critical illness (9), and pediatrics (2). Even within these subgroups, there is clinical heterogeneity. Abdominal trauma would be expected to affect gut function more significantly than head injuries. Similarly, abdominal surgery has different effects on the gut depending on the site of the operation.15 In this vein, there is now good evidence that laparoscopic surgery leads to less postoperative ileus than open abdominal surgery.16-18 In addition, baseline nutrition status differs between patients who have acute or chronic alcohol-related pancreatitis,19 for instance, compared with previously well patients admitted to the hospital due to sepsis. Statistical heterogeneity was assessed, but its absence does not validate pooling such disparate groups of patients at least in drawing clinically meaningful conclusions.
6. Heterogeneity of Trials (Quality)
With regard to the quality of included trials, only 1 of the 13 trials (the Scarborough study20) in the Canadian meta-analysis included details regarding concealed randomization, blinded outcome adjudication, and intent-to-treat analysis. Nonetheless, the authors still proceeded with the analysis. The Chicago group graded the quality of included studies, assigning points for various methodological approaches (concealed randomization, comparability of groups at baseline, whether staff were blinded to end points, and so forth). Although there is evidence that many of these factors may alter outcomes and thereby reduce bias,21-23 the authors do not make any attempt at validating this methodology and do not enunciate a precedent for such an approach. In the methods section of Aust 1, brief mention was made that blinded quality assessment of the selected studies was made by the authors, but no further discussion of this was made in the rest of the article. In the Aust 2 analysis, none of the included trials reported allocation concealment, and only 1 reported the use of blinding. Finally, in the USA/Scotland study, trials were graded into high quality and low quality depending on whether the investigators/ assessors and participants were blinded and whether the trial contained both an explicit description of an adequate allocation concealment and data that were evaluable on an intentto-treat basis. Again, no precedent for this approach was cited, and even where most of the trials analyzed were of low quality (in the perioperative group), no adjustment or allowance for this was made other than to state that the conclusions "must be tempered by the fact that the data were largely derived from low-quality studies."
7. Patients Counted Twice
In the Chicago meta-analysis, 68 of the same trauma patients reported separately were counted twice. The authors of the trauma studies indicated explicitly in the second of their studies that these 68 patients had been previously reported.24,25
8. Inclusion of a Large Amount of Data Published in a Previous Meta-analysis for the First Turn
In the large trauma meta-analysis by Moore et al,26 a sizeable number of the patients were derived from studies that were and remain reported only in the meta-analysis. Nonetheless, the patients from this meta-analysis were included in the Canadian and Aust 1 meta-analyses.
9. Failure to Examine Separately Trials With Different Dropout Rates
Other than Aust 2, which excluded trials for which the follow-up of patients was < 90%, there was no attempt to stratify trials according to success of follow-up.
Individual Trials Comparing EN With PN
It is not just the problems of combining nutritionally based trials together that renders the results of the meta-analyses so questionable, but it is the trials themselves. Many of the problems of the meta-analyses are also evident in the individual trials. The practice of rigidly adhering to one route of nutrition support is not a reflection of sensible clinical practice in that a patient's clinical status and gut function are protean. In much the same way that it would be inappropriate to randomize patients with community-acquired pneumonia presenting to an emergency department to either intubation and ventilation for 7 days or to oxygen delivery via a facial mask (with intubation and ventilation as a backup), it is also pointless and indeed unethical to randomize patients to a single route of nutrition support unless there is genuine equipoise-a true dilemma as to which route is appropriate. Even here, as the clinical situation changes, such equipoise often evaporates, and it becomes obvious that one route of nutrition support is appropriate. This was demonstrated in the Scarborough nutrition support study.20 Of 562 patients who had received little or no nutrition in the previous 7 days, for 64 patients, it was felt on clinical grounds by the attending specialist that there was equipoise between whether EN or PN was the preferred route of nonvolitional support. These 64 patients were then randomized to receive either PN or EN. Recause of changes in the patients' clinical status, the initial choice of route of nutrition support often became inappropriate. Of the 32 patients randomized to receive EN, 10 (31%) were converted to receiving PN, and of those randomized to receive PN, 5 (15.6%) were converted to receiving EN.
The other weakness with respect to comparison of EN with PN is that traditionally, the threshold for the introduction of PN has been higher than that for EN. The dilemma in the first few days after admission or surgery is often not between EN and PN administration but between EN and no nonvolitional nutrition support. To compare them head to head therefore either implies that EN has been introduced too late or that PN has been introduced too early. Indeed, there may be 3-way equipoise, in which the dilemma resides between PN, EN, or neither (see the text box). Optimal nutrition support is complex and requires flexibility, which the studies do not acknowledge let alone allow for.
Discussion
In view of the inevitable differences between EN and PN in terms of indications, bioavailability, time, and method of initiation and regimens, it is remarkable that such a large number of clinical trials and 5 meta-analyses have been devoted to comparing EN with PN. Many of the outcome measures, such as infectious complications, are impossible to evaluate objectively let alone quantitate in a clinically relevant manner. Admittedly, publication bias, statistical heterogeneity, prospectivity, and clinical outcomes were assessed and where possible accommodated for. The care taken to optimize the methodology is ironic to the extent that it masks their inevitable shortcomings. The use of both EN and PN administered either sequentially or contemporaneously in selected cases has also been examined in several trials. A small meta-analysis of these trials2' failed to reveal any advantage in this approach. This meta-analysis and the trials that comprise it, however, are as open to criticism as the ones discussed above. Recently, there has also been a call to supplement EN with PN in the critically ill as there is often a proteinenergy deficit resulting from GI intolerance and temporary cessation of feeding for investigations with a failure to achieve full coverage of energy needs until day 7.28 There is also emerging evidence that individual nutrients (glutamine and antioxidants such as selenium) may have a role in and of themselves in the management of the critically ill.29,30 These nutrients may be beneficial in supraphysiological doses, and their effects to some extent may depend on the route of administration.31
Although Aust 2 showed that EN was associated with increased mortality, many of the trials and all 4 of the other meta-analyses showed that EN was associated with less morbidity than PN. Regrettably, these conclusions have contributed to a remarkably enthusiastic adoption of EN with an aversion to PN, a philosophy that has recently been described as "enteralism."32 As Lipman33 pointed out in a review in 1998 before concluding that there was little clinical evidence for favoring EN over PN: "Almost any clinical study of nutrition support starts with a seemingly obligatory introduction that enteral nutrition is the preferred method for non-volitional delivery." In another recent publication, a truly extreme position was taken in which PN was painted as a poison.34
The Future
Objective assessment of intestinal function is central to determining (preferably prospectively) whether enteral feeding should be used, particularly in the sickest patients who tolerate EN the least.35 There are few data concerning even the most basic of intestinal functions-the proportion of feed not being absorbed. A recent study indicated that 30% of critically ill patients with loose stools have malabsorption (defined as > 15% of administered feed not being absorbed).36 Other measures of intestinal function have been developed: paracetamol absorption tests,3' small-bowel motility,38 radiological markers,39 and triolein breath testing.38 Data, however, are very limited. There may be a role for abdominal ultrasound to evaluate small-bowel motility and edema, somewhat akin to its role in Crohn's disease.40 Intraperitoneal cameras incorporated into or next to drain tubes or placed transcutaneously may be another option particularly in patients who have recently undergone surgery.
It is remarkable that there are such sophisticated methods for evaluating cardiorespiratory function yet the evaluation of gut function, which is often compromised substantially in the critically ill, is so rudimentary and so dependent on subjective parameters. The paucity of methodologies with respect to objective evaluation of nutrition support approaches is reflected in the simplicity of the guidelines of major nutrition societies. The American Society for Parenteral and Enteral Nutrition practice guidelines, for instance, regarding administration of specialized nutrition support when addressing the issue of the route of nutrition, simply state that when specialized nutrition support is required, "EN should generally be used in preference to PN" and that "PN should be used when the GI tract is not functional or cannot be accessed and in patients who cannot be adequately nourished by oral diets or EN."3 If significant progress is to be made in optimizing EN and ensuring that it is administered in an optimal way, evaluating the response of the gut contemporaneously may well be required.
Similarly, very little is known about the metabolic fate of the substrates that are absorbed or that are administered parenterally. Clearly, there is no value in nutrition substrate if it is deposited in the liver as fat or passed as glucose in the urine. It may be beneficial to view nonvolitional feeding (parenteral and enteral) more in the way drugs are regarded. Indeed, if as little were known about the absorption and metabolism of a drug as is known about that of artificial nutrition support, it would probably not be licensed for use. Chemically tagging administered substrate and evaluating energy expenditure may be expensive but would give much greater insight into the physiological changes induced by nutrition support. Muscle function, as determined by electrophysiological studies41 and spectroscopic magnetic resonance,42 has been shown to be a sensitive marker to changes in nutrition status, which may also have a role in the evaluation of patients being fed nonvolitionally. Simply adding up complications and determining the prognosis in heterogeneous groups of patients will never substitute for more detailed physiological analysis. Such analyses should of course not be viewed in isolation but interpreted in conjunction with more readily accessible clinical outcomes. The funding of these more physiological studies in the context of nutrition support may not be as easy to procure as analogous funding in the testing of pharmaceutical products because nutrition companies may not be so willing to part with funds. However, if such studies can demonstrate a safer, more rational, and ultimately more efficacious way of administering nutrition support, funding is bound to follow.
Conclusion
Nonvolitional nutrition support is a vital part of the care of seriously ill patients. The attempts at comparing the 2 routes of nonvolitional feeding have used very inflexible approaches, and the reported results are of limited clinical significance. The future of this field lies in testing strategies rather than comparing fixed menus. Such strategies should encompass ongoing assessment of the ease of access and function of the vehicles of administration: the GI tract and venous system. They should also examine the absorption and metabolism of administered substrate and its physiological effects in a much more rigorous and scientific way than has hitherto been the case.
Acknowledgment
I am indebted to Professor K. N. Jeejeebhoy of the University of Toronto and Professor F. Bowden and Associate Professor L. Hillman of the Australian National University for their help in the preparation of this article.
