The truth about the effectiveness of any given medical intervention is not a black or white answer. It is not a dichotomous outcome, either present or absent. Rather, it is represented by a mosaic of signals emerging across the basic and clinical studies of a particular issue so that, when systematically reviewed, the findings coalesce to give us a sense on a continuum, with one end anchored with "no(n) sense," to the other end anchored with a strong sense of truth. The whole purpose of the systematic review and critical appraisal of the literature exercise is to try to figure out where on this continuum a particular body of evidence lies. Drs Koretz and Thompson raise some interesting points about the application of evidence appraisal to nutrition therapy in the intensive care unit (ICU) setting. We are motivated to add a few points of our own.
There is a consistent signal emerging from the basic science literature and observational studies in critically ill patients that enteral nutrition (EN), compared with parenteral nutrition (PN) or no EN, has a more favorable effect on the stress response, innate immunity, and gastrointestinal structure and function.1 Unfortunately, clinical nutrition studies in critically ill patients are too small so that individually, these trials typically offer little illumination. When they are statistically aggregated, these studies converge to demonstrate that EN is associated with less infectious complications compared with PN.2 However imprecise or nonstandard the definitions used, this bias would be equally applied to both groups, and nevertheless, the signal still emerges. This signal from the clinical studies is consistent with our current understanding of the underlying pathophysiology of critical illness and the results of animal studies, and thus the dots connect!
To say that, on average, it does not matter whether you feed with EN or PN dismisses the entire evidencebased paradigm that modern medicine is built upon. However weak the evidence, it is still evidence, and we must move forward and make clinical recommendations based on the current available evidence. Yes, when applying evidence-based recommendations to individual patients, because of unique characteristics of a given patient, it may be quite appropriate to use EN and PN or neither. Alas, this is the "art" of medicine. But let's be sure we are good scientists and are appropriately applying the best evidence before we start painting!
There are established criteria that can be used to judge the scientific rigor of systematic reviews and meta-analyses.3 The review process usually begins with a focused clinical question that specifies an intervention (EN or PN), a particular patient population (critically ill), and a specific study design (only randomized controlled trials [RCTs]). If the clinical question is framed differently or if the specific criteria vary somewhat, it is not surprising that 5 different groups of investigators came up with different studies to be included in their respective reviews. This is not a flaw of meta-analyses, just that investigators had different priorities. And yet it is noteworthy that despite different methods, all 5 meta-analyses show fewer infectious complications associated with EN, a finding that increases our confidence that the finding is real.
However, if the purpose of the review exercise is to lead to clinical recommendations of specific patient populations, such as critically ill patients, then limiting the review to studies of this specific patient population is an absolute prerequisite. The treatment effect of nutrition interventions varies depending on the underlying pathophysiology, so we cannot combine studies of elective surgery, noncritically ill obese patients, and so on with studies of critically ill patients to make an inference of how best to manage critically ill patients.2,4 Mega-analyses5,6 that combines large numbers of studies with heterogeneous patient groups and nutrition interventions are not helpful in generating specific clinical recommendations.
If the purpose of the review exercise is to determine the most reliable, overall estimate of treatment effect in critically ill patients, then a comprehensive search and complete inclusion of all studies is required and is most likely to yield the best results. To exclude studies based on even some methodological criteria (such as completeness of follow-up, nonconcealed randomization, blinding, or all of the above) is to open the results up to bias. One may end up with spurious unsupported findings, like PN is associated with a survival advantage!7 The dots don't connect-there is no mechanistic rationale supported by other clinical observations or basic science to support the validity of this finding.
Admittedly, there are no RCTs comparing EN or PN with no treatment. We could say the same about using antibiotics in the treatment of infectious complications until recently: a Canadian group conducted a placebo-controlled trial of antibiotics in the treatment of ICU patients suspected of new infection.8 Essentially, patients with new suspicion of infection were randomized to receive either a broad-spectrum antibiotic or a placebo until culture results were available. Subsequent antibiotic use in both groups was based on the culture results. Investigators allowed for an escape clause. If the physician treating the patient in the study was worried about clinical deterioration, he or she could withdraw the patient from the study and provide open-label antibiotics. The investigators postulated that this would happen with equal rates in both groups. To their surprise, it happened in 5 patients, all in the placebo group. Moreover, there was a significant reduction in ICU-free days and a trend toward increased mortality in patients randomized to no treatment. All previous evidence regarding bacterial infection (based on our understanding of underlying pathophysiology, clinical trials comparing different drugs, etc) indicates antibiotics are needed. Designing a study that withholds such treatment to enhance scientific methodology proves little, adds nothing to our understanding, endangers patients, and is both unethical and unreasonable.
When we look across the signals emerging from the nutrition literature pertaining to critically ill patients, we would be extremely uncomfortable randomizing a critically ill septic patient to "no nutrition treatment." Imagine trying to explain to the mother of an 18-year-old anorexic female with meningitis and septic shock that she may be randomized to a group where no nutrition therapy is provided. Or try explaining to the wife of an obese man who presents with severe community-acquired pneumonia and respiratory failure, and who you know will have a prolonged ICU course and end up very weak, that protein and calories are not needed. Although we have no direct studies of nutrition vs no nutrition, we are influenced by the following bodies of literature that indirectly support the assertion that early, aggressive EN is beneficial:
1. Studies comparing early with delayed EN: These are probably the closest we will come to having an EN vs no nutrition treatment study; the nutrition intervention is suspended for a period of time. Looking across the RCTs and observational studies, we see a large signal of benefit in terms of mortality reduction and less infectious complications.9,10 If delaying EN is harmful, then "no EN" would be significantly worse!
2. Studies that employ strategies to increase the provision of EN, comparing results in improved clinical outcomes to less aggressive strategies."12
3. Experimental studies in animals showing that EN, compared with no EN, has a favorable effect on gastrointestinal structure and function, cytokine profiles, and clinical outcomes.1
And thus the dots connect, and although the clinical evidence is weak, it is an overstatement to say that "there is no evidence." Yes, there are attendant risks to EN. For this reason, our Canadian Clinical Practice guidelines13 (see www.criticalcareconnections.com for the most recent version of the CPGs) focused on strategies to minimize those risks and maximize the benefits. When used properly, the benefit suggested by this weak evidence seems to outweigh the risks, and a clinical recommendation can be made. There is a "horse" leading the cart, to borrow from Dr Koretz's analogy, albeit a small one (even a pony!). Consensus opinion supports this interpretation of the literature. International guideline committees consistently recommend EN over PN.6,7,9,14,15 An international survey of attitudes and beliefs of more than 500 physicians and dietitians indicated endorsement of recommendations to provide EN in the ICU in 88% of those surveyed. A survey of actual nutrition practices in 156 ICUs across the world showed that most patients (84%) in these ICUs are receiving EN (unpublished data). So we seem to be in good company!
Although some merit should be given to Thompson for his review of the evidence, we are uncomfortable with his recommendation for future study design. Having individual patient characteristics or conditions drive an algorithm for a study will introduce considerable complexity and variability and would lose the value of a more "rigid" prospective study design. Such an approach promotes anecdotal experience over scientific rigor and would be appropriate for bedside care but not for large clinical trials attempting to answer scientific questions. Thompson's recommendation to shift the focus to physiologic studies implies that surrogate markers of physiologic parameters are more important than clinical outcomes. This concept is incongruous with the concept of evidence-based medicine where surrogate markers of physiologic changes in response to nutrition therapy are meaningless if they do not affect clinical outcomes.
There is one thing on which we agree with Drs Koretz and Thompson. We need to improve the quality and quantity of critical care nutrition research. We disagree that we need to step backwards to address EN vs PN or EN vs no EN. Let's get over it and move on to more important questions and issues.
